ukpds

Findings Over 10 years, haemoglobin A1c (HbA1c) was 7·0%(6·2–8·2) in the intensive group compared with 7·9%(6·9–8·8) in the conventional group—an 11% reduction.There was no difference in HbA1c among agents in theintensive group. Compared with the conventional group,the risk in the intensive group was 12% lower (95% CI1–21, p=0·029) for any diabetes-related endpoint; 10%lower (–11 to 27, p=0·34) for any diabetes-related death;and 6% lower (–10 to 20, p=0·44) for all-cause mortality.Most of the risk reduction in the any diabetes-relatedaggregate endpoint was due to a 25% risk reduction(7–40, p=0·0099) in microvascular endpoints, includingthe need for retinal photocoagulation. There was nodifference for any of the three aggregate endpointsbetween the three intensive agents (chlorpropamide,glibenclamide, or insulin).

Patients in the intensive group had morehypoglycaemic episodes than those in the conventionalgroup on both types of analysis (both p<0·0001). Therates of major hypoglycaemic episodes per year were0·7% with conventional treatment, 1·0% withchlorpropamide, 1·4% with glibenclamide, and 1·8% withinsulin. Weight gain was significantly higher in theintensive group (mean 2·9 kg) than in the conventionalgroup (p<0·001), and patients assigned insulin had agreater gain in weight (4·0 kg) than those assignedchlorpropamide (2·6 kg) or glibenclamide (1·7 kg).

Interpretation Intensive blood-glucose control by eithersulphonylureas or insulin substantially decreases the riskof microvascular complications, but not macrovasculardisease, in patients with type 2 diabetes. None of theindividual drugs had an adverse effect on cardiovascularoutcomes. All intensive treatment increased the risk ofhypoglycaemia.

Lancet 1998; 352: 837–53See Commentary page xxx

IntroductionStarted in 1977, the UK Prospective Diabetes Study(UKPDS) was designed to establish whether, in patientswith type 2 diabetes, intensive blood-glucose controlreduced the risk of macrovascular or microvascularcomplications, and whether any particular therapy wasadvantageous.

Most intervention studies have assessed microvasculardisease: improved glucose control has delayed the

Summary

Background Improved blood-glucose control decreasesthe progression of diabetic microvascular disease, but the effect on macrovascular complications is unknown.There is concern that sulphonylureas may increasecardiovascular mortality in patients with type 2 diabetesand that high insulin concentrations may enhanceatheroma formation. We compared the effects of intensiveblood-glucose control with either sulphonylurea or insulinand conventional treatment on the risk of microvascularand macrovascular complications in patients with type 2diabetes in a randomised controlled trial.

Methods 3867 newly diagnosed patients with type 2diabetes, median age 54 years (IQR 48–60 years), whoafter 3 months’ diet treatment had a mean of two fastingplasma glucose (FPG) concentrations of 6·1–15·0mmol/L were randomly assigned intensive policy with a sulphonylurea (chlorpropamide, glibenclamide, orglipizide) or with insulin, or conventional policy with diet.The aim in the intensive group was FPG less than 6mmol/L. In the conventional group, the aim was the bestachievable FPG with diet alone; drugs were added only ifthere were hyperglycaemic symptoms or FPG greater than15 mmol/L. Three aggregate endpoints were used toassess differences between conventional and intensivetreatment: any diabetes-related endpoint (sudden death,death from hyperglycaemia or hypoglycaemia, fatal ornon-fatal myocardial infarction, angina, heart failure,stroke, renal failure, amputation [of at least one digit],vitreous haemorrhage, retinopathy requiringphotocoagulation, blindness in one eye, or cataractextraction); diabetes-related death (death frommyocardial infarction, stroke, peripheral vascular disease,renal disease, hyperglycaemia or hypoglycaemia, andsudden death); all-cause mortality. Single clinicalendpoints and surrogate subclinical endpoints were alsoassessed. All analyses were by intention to treat andfrequency of hypoglycaemia was also analysed by actualtherapy.

Intensive blood-glucose control with sulphonylureas or insulincompared with conventional treatment and risk of complicationsin patients with type 2 diabetes (UKPDS 33)

UK Prospective Diabetes Study (UKPDS) Group*

ARTICLES

THE LANCET • Vol 352 • September 12, 1998 837

*Study organisation given at end of paper

Correspondence to: Prof Robert Turner, UKPDS Group, Diabetes Research Laboratories, Radcliffe Infirmary, Oxford OX2 6HE, UK

Articles

ARTICLES

development and progression of retinopathy,nephropathy, and neuropathy in patients with type 1diabetes1,2 and those with type 2 diabetes.3 In the UK,9% of patients with type 2 diabetes developmicrovascular disease within 9 years of diagnosis, but20% have a macrovascular complication—andmacrovascular disease accounts for 59% of deaths inthese patients.4

Epidemiological studies of the general population haveshown an increased risk of cardiovascular disease withconcentrations of fasting glucose or haemoglobin A1c

(HbA1c) just above the normal range.5,6 The onlyprevious large-scale randomised trial in type 2 diabetes,the University Group Diabetes Program (UGDP),7

followed 1000 patients assigned different therapies forabout 5·5 years (range 3–8 years) and found no evidencethat improved glucose control, by any therapy, reducedthe risk of cardiovascular endpoints. That study did,however, report increased risk of cardiovascularmortality in patients allocated the sulphonylurea,tolbutamide, and this unexpected finding introducednew hypotheses.8 These hypotheses included increasedmyocardial damage from inhibition of ATP-K+ channelopening in the presence of myocardial ischaemia9 due tosulphonylurea binding to the cardiovascular SUR2receptor—an event that could also increase thelikelihood of ventricular arrhythmia.10 An increase inatherosclerosis with insulin treatment has also beensuggested, since plasma insulin concentrations aresupraphysiological.11,12

We report the final results of our study of intensiveblood-glucose control policy, with sulphonylurea orinsulin therapy, compared with conventional treatmentpolicy with diet, on the risk of microvascular andmacrovascular clinical complications. We alsoinvestigated whether there was any particular benefit orrisk with sulphonylurea or insulin therapy.

MethodsPatientsBetween 1977 and 1991, general practitioners in the catchmentareas of the 23 participating UKPDS hospitals were asked torefer all patients with newly diagnosed diabetes aged 25–65years. Patients generally attended a UKPDS clinic within 2weeks of referral. Patients who had a fasting plasma glucose(FPG) greater than 6 mmol/L on two mornings, 1–3 weeksapart, were eligible for the study. An FPG of 6 mmol/L wasselected because this was just above the upper limit of normalfor our reference range. The exclusion criteria were: ketonuriamore than 3 mmol/L; serum creatinine greater than175 µmol/L; myocardial infarction in the previous year; currentangina or heart failure; more than one major vascular event;retinopathy requiring laser treatment; malignant hypertension;uncorrected endocrine disorder; occupation that precludedinsulin therapy (eg, driver of heavy goods vehicle); severeconcurrent illness that would limit life or require extensivesystemic treatment; inadequate understanding; andunwillingness to enter the study.

7616 patients were referred and 5102 were recruited (58%male). The 2514 patients excluded were similar in age, sex, andglycaemic status to those recruited. The study design andprotocol amendments, which conform with the guidelines of the

838 THE LANCET • Vol 352 • September 12, 1998

Figure 1: Trial profile

Declarations of Helsinki (1975 and 1983), were approved bythe Central Oxford Research Ethics Committee and by theequivalent committees at each centre. Each patient gaveinformed witnessed consent.

Dietary run-inPatients had a 3-month dietary run-in during which theyattended a monthly UKPDS clinic and were seen by aphysician and dietician. The patients were advised to followdiets that were low saturated fat, moderately high fibre and had

about 50% of calories from carbohydrates; overweight patientswere advised to reduce energy content.13 After the run-in, amean FPG was calculated from measurements on 3 days over 2weeks.

DefinitionsMarked hyperglycaemia was defined as FPG greater than 15mmol/L, symptoms of hyperglycaemia, or both, in the absenceof intercurrent illness. Hyperglycaemic symptoms includedthirst and polyuria.

ARTICLES


Conventional (n=1138) Intensive (n=2729) All patients (n=3867)

DemographicAge (years)* 53·4 (8·6) 53·2 (8·6) 53·3 (8·6)M/F 705/433 649/444 2359/1508Ethnicity (%) Caucasian/Indian Asian/Afro-Caribbean/Other 81/11/7/1 81/10/8/1 81/10/8/1

ClinicalWeight (kg)* 78·1 (16·3) 77·3 (15·4) 77·5 (15·5)Body-mass index (kg/m2)* 27·8 (5·5) 27·5 (5·1) 27·5 (5·2)Systolic blood pressure (mm Hg)* 135 (19) 135 (20) 135 (20)Diastolic blood pressure (mm Hg)* 82 (10) 83 (10) 82 (10)Smoking (%) never/ex/current 34/35/31 35/35/30 34/35/31Alcohol (%) none/social/regular/dependent 26/56/18/2 24/56/17/1 22/56/18/1Exercise (%) sedentary/moderately active/active/fit 20/37/39/4 21/34/40/5 20/35/40/5

BiochemicalFPG (mmol/L)† 8·0 (7·1–9·6) 8·1 (7·1–9·8) 8·0 (7·1–9·7)HBA1c (%)* 7·05 (1·42) 7·09 (1·54) 7·08 (1·51)Plasma insulin (pmol/L)‡ 91 (52–159) 92 (52–159) 92 (52–160)Triglyceride (mmol/L)‡ 2·31 (0·84–6·35) 2·37 (0·85–6·63) 2·35 (0·84–6·55)Total cholesterol (mmol/L)* 5·4 (1·02) 5·4 (1·12) 5·4 (1·1)LDL-cholesterol (mmol/L)* 3·5 (0·99) 3·5 (1·0) 3·5 (1·0)HDL-cholesterol (mmol/L)* 1·08 (0·24) 1·07 (0·25) 1·07 (0·24)

MedicationsMore than one asprin daily (%) 1·5 1·7 1·6Diuretic (%) 13 13 13Others (%) digoxin/antihypertensive/lipid lowering/HRT or OC 0·9/12/0·3/0·9 1·3/12/0·3/0·7 1·1/12/0·3/0·8

Surrogate clinical endpointsRetinopathy (%) 36 36 36Proteinuria (%) 2·1 1·7 1·9Plasma creatinine (mmol/L)‡ 81 (66–99) 82 (67–100) 81 (67–100)Biothesiometer more than 25 volts (%) 11·4 11·8 11·5

Data are % of group, *mean (SD), †median (IQR), or ‡geometric mean (1 SD). HRT=hormone replacement therapy. OC=oral contraceptive therapy.Table 1: Baseline characteristics of patients in conventional and intensive-treatment groups

Conventional Chlorpropamide Glibenclamide Insulin All patients(n=896) (n=619) (n=615) (n=911) (n=3041)

DemographicAge (years)* 54 (9) 54 (9) 54 (8) 54 (8) 54 (8)M/F 555/341 359/260 381/234 656/346 1885/1156Ethnicity (%) Caucasian/Indian Asian/Afro Caribbean/Other 83/9/7/1 79/10/11/0 84/8/7/1 82/8/9/1 82/8/9/1

ClinicalWeight (kg)* 77 (16) 75 (15) 77 (14) 76 (14) 76 (15)Body-mass index (kg/m2)* 27·5 (5·3) 27·0 (4·9) 27·4 (5·0) 27·0 (4·8) 27·2 (5·0)Systolic blood pressure (mm Hg)* 136 (19) 136 (19) 136 (19) 136 (20) 136 (19)Diastolic blood pressure (mm Hg)* 83 (10) 83 (10) 83 (10) 83 (11) 83 (10)Smoking (%) never/ex/current 34/34/32 38/31/31 32/38/30 34/36/30 35/35/30Alcohol (%) none/social/regular/dependent 24/55/20/1 26/52/21/1 22/58/19/1 24/57/18/1 24/57/18/1Exercise (%) sedentary/moderately active/active/fit 18/38/40/4 19/37/40/4 18/32/44/6 21/35/40/4 19/36/41/4

BiochemicalFPG (mmol/L)† 7·9 (7·1–9·4) 8·0 (7·1–9·7) 8·0 (7·2–9·6) 8·1 (7·1–9·9) 8·0 (7·1–9·6)HBA1c (%)* 6·2 (1·2) 6·3 (1·4) 6·3 (1·3) 6·1 (1·1) 6·2 (1·2)Plasma insulin (pmol/L)‡ 89 (51–156) 90 (51–160) 91 (52–160) 90 (52–156) 90 (52–156)Triglyceride (mmol/L)‡ 2·43 (0·86–6·92) 2·58 (0·88–7·55) 2·37 (0·84–6·72) 2·48 (0·85–7·25) 2·46 (0·86–7·10)Total cholesterol (mmol/L)* 5·4 (1·03) 5·5 (1·15) 5·5 (1·11) 5·4 (1·13) 5·4 (1·10)LDL-cholesterol (mmol/L)* 3·5 (0·99) 3·5 (1·05) 3·5 (1·00) 3·5 (1·03) 3·5 (1·02)HDL-cholesterol (mmol/L)* 1·07 (0·23) 1·08 (0·25) 1·09 (0·25) 1·07 (0·25) 1·08 (0·24)

MedicationsMore than one asprin daily (%) 1·2 1·5 1·1 1·8 1·4Diuretic (%) 13 12 15 14 14Others (%) digoxin/antihypertensive/lipid lowering/ 0·5/12·2/0·1/0·3 1·0/11·2/0·3/0·3 1·3/11·3/0/0·5 1·3/10·7/0·2/0·7 1·0/11·6/0·3/0·5HRT or OC

Surrogate clinical endpointsRetinopathy (%) 38 40 30 38 38Proteinuria (%) 2·2 1·7 2·1 1·5 1·9Plasma creatinine (mmol/L)‡ 80 (66–97) 81 (67–82) 82 (67–99) 81 (67–99) 81 (67–99)Biothesiometer more than 25 volts (%) 12·1 10·1 15·2 12·1 12·3

Data are % of group, *mean (SD), †median (IQR), or ‡geometric mean (1 SD). HRT=hormone replacement therapy. OC=oral contraceptive therapy.Table 2: Baseline characteristics of patients in conventional group and individual intensive groups

ARTICLES

RandomisationThe flow of patients in the study is shown in figure 1.

Patients were stratified by ideal bodyweight (overweight was>120% ideal bodyweight).14 Non-overweight patients wererandomly assigned intensive treatment with insulin (30%),intensive treatment with sulphonylurea (40%: equalproportions in the first 15 centres to chlorpropamide orglibenclamide, and in the last eight centres to chorpropamide orglipizide), or conventional treatment with diet (30%). Thenon-balanced randomisation was chosen so that there weresufficient patients in the two sulphonylurea groups to allowcomparison between the first-generation and second-generationdrugs. Overweight patients were randomly assigned treatmentwith the additional possibility of metformin: intensive treatmentwith insulin (24%), intensive treatment with sulphonylurea withequal proportions of patients on chlorpropamide andglibenclamide (32%), intensive treatment with metformin(20%), and conventional treatment with diet (24%). The 342overweight patients who were randomly allocated metformintherapy are reported separately, as intended per protocol.15

Randomisation was by means of centrally produced,computer-generated therapy allocations in sealed, opaqueenvelopes which were opened in sequence. The numericalsequence of envelopes used, the dates they were opened, andthe therapies stipulated were monitored. The trial was openonce patients were randomised. No placebo treatments weregiven.

Conventional treatment policyThe aim in this group was to maintain FPG below 15 mmol/Lwithout symptoms of hyperglycaemia. Patients attendedUKPDS clinics every 3 months and received dietary advicefrom a dietician with the aim of maintaining near-normalbodyweight.

If marked hyperglycaemia or symptoms occurred, patientswere secondarily randomised to treatment with sulphonylureaor insulin therapy, with the additional option of metformin inoverweight patients; this was a separate stratified randomisationfrom the original randomisation, but with the same proportionsallocated sulphonylurea and insulin.13 If marked hyperglycaemiarecurred in participants secondarily allocated sulphonylurea,metformin was added, and in those secondarily allocatedmetformin, glibenclamide was added. Patients with markedhyperglycaemia or symptoms on both agents were changed toinsulin. Throughout, the aim of FPG below 15 mmol/L withoutsymptoms was maintained. Clinical centres were advised byautomatically generated letters when patients allocatedconventional treatment received inappropriate pharmacologicaltherapy.

Intensive treatment policyThe aim of intensive treatment was FPG less than 6 mmol/Land, in insulin-treated patients, pre-meal glucoseconcentrations of 4–7 mmol/L. These patients also continuedto receive dietary advice from a dietician. The daily doses of thesulphonylureas used were: chlorpropamide 100–500 mg;glibenclamide 2·5–20 mg; and glipizide 2·5–40 mg.

Whenever glucose concentrations were above target

concentrations, a letter was sent from the coordinating centerwith advice on necessary changes in therapy. Patients assignedinsulin started on once daily ultralente insulin (Ultratard HM,Novo-Nordisk, Crawley, UK or Humulin Zn, Eli-Lilly,Basingstoke, UK) or isophane insulin. If the daily dose wasmore than 14 units (U) or pre-meal or bed-time home blood-glucose measurements were more than 7 mmol/L, ashort-acting insulin, usually soluble (regular) insulin wasadded—ie, basal/bolus regimen. Patients on more than 14 Uinsulin per day, or on short-acting insulins, were particularlyencouraged to do regular home-glucose monitoring.

Protocol and amendmentsThe original protocol for the first 15 centres stipulated thatpatients continue their assigned treatment (diet,chlorpropamide, glibenclamide, metformin, or insulin) for aslong as possible to achieve maximum exposure to each therapyalone and thus find out whether there were differences inresponse to each agent. Additional therapies were added tothose allocated to diet, sulphonylurea, or metformin only whenmarked hyperglycaemia developed. For patients onsulphonylureas, metformin was added; but if markedhyperglycaemia recurred, patients were changed to insulintherapy. Metformin was used to a maximum of 2550 mg perday.

When the progressive hyperglycaemia in all groups becameapparent, the protocol was amended to allow the early additionof metformin when, on maximum doses of sulphonylurea, FPGwas greater than 6 mmol/L in symptomless patients in theintensive group. Patients were changed to insulin therapy ifmarked hyperglycaemia recurred.

When the last eight centres were recruited in 1988, patientsallocated sulphonylurea had insulin added early, rather thanmetformin, when on maximum doses of sulphonylurea FPGwas greater than 6 mmol/L.

Embedded studies1148 UKPDS patients were in the Hypertension in DiabetesStudy (HDS).16 This study, which started in 1987, randomlyallocated hypertensive patients to a tight blood-pressure-controltreatment that aimed for a blood pressure of 150/85 mm Hg orlower with either captopril or atenolol or, to a less tight blood-pressure-control treatment that aimed for a blood pressure of180/105 mm Hg or lower but avoided the use of captopril andatenolol. The UKPDS Acarbose Study17 started in 1994 andrandomly allocated 1946 patients to additional double-blind,placebo-controlled therapy with acarbose for 3 years—irrespective of their blood-glucose and blood-pressure controlallocations.

Clinic visitsPatients attended morning clinics every 3 months or morefrequently as needed to attain glycaemic control. From 1990,the routine clinic visits were every 4 months. Patients fastedfrom 2200 h the night before for plasma glucose and otherbiochemical measurements, and did not take their allocatedtreatment on the morning of the clinic visit.

At each visit plasma glucose, blood pressure, and weight


Assigned therapy in 15 centres (32 406 person-years) Assigned therapy in all 23 centres(38 263 person-years)

Conventional Chlorpropamide Glibenclamide InsulinConventional Intensive

(n=896) (n=619) (n=615) (n=911)(n=1138) (n=2729)

Total person-years 9491 6562 6573 9780 11 188 27 075

Actual therapy (person years)Diet alone 5495 (58%) 409 (6%) 432 (7%) 1896 (19%) 6490 (58%) 3206 (12%)Chlorpropamide alone or in combination 621 (7%) 5266 (80%) 126 (2%) 66 (1%) 743 (7%) 6372 (24%)Glibenclamide alone or in combination 1699 (18%) 483 (7%) 5467 (83%) 823 (8%) 1715 (15%) 6789 (25%)Glipizide alone or in combination 47 (0·5%) 28 (0·4%) 17 (0·3%) 58 (1%) 281 (3%) 1359 (5%)Metformin alone or in combination 1105 (12%) 900 (14%) 1319 (20%) 329 (3%) 1132 (10%) 2581 (10%)Insulin 1458 (15%) 615 (9%) 681 (10%) 7215 (74%) 1809 (16%) 10 413 (38%)

Table 3: Person-years of follow-up on assigned and actual therapies for first 15 and all centres

were measured, and therapy was adjusted if necessary. Froma checklist we asked about all medications, hypoglycaemicepisodes, home blood-glucose measurements, illness, timeoff work, admissions to hospital, general symptoms includingany drug side-effects, and clinical events. Hypoglycaemicepisodes were defined as minor if the patient was able totreat the symptoms unaided, or major if third-party help ormedical intervention was necessary. Details of all majorhypoglycaemic episodes were audited to ensure the coding wasappropriate.

At entry, randomisation, 6 months, 1 year, and annuallythereafter a fasting blood sample was taken for measurement ofHbA1c, plasma creatinine (annually from 1989), triglyceride,total cholesterol, LDL-cholesterol, HDL-cholesterol, insulin,and insulin antibodies. Every year, urinary albumin andcreatinine were measured in a random urine sample.

At entry and then every 3 years all patients had a full clinicalexamination. At these reviews, a 12-lead electrocardiogram wasrecorded and Minnesota coded13 and a posterior-anterior chestradiograph taken for measurement of cardiac diameter. Dopplerblood pressure was measured in both legs and in the right arm.Visual acuity was measured with a Snellen chart until 1989 andsubsequently with an Early Treatment of Diabetic RetinopathyStudy (ETDRS) chart.13 The best attainable vision was assessedwith the patient’s usual spectacles or with a pinhole. Directophthalmoscopy with pupil dilation was carried out every

3 years. Since 1982, retinal colour 30º photographs of fourfields per eye (nasal, disc, macula, and temporal-to-maculafields) were taken with additional stereo photographs of themacula; poor quality photographs were repeated. Two assessorsat a single centre reviewed the photographs for diabeticretinopathy; any fields with retinopathy were graded by twoother assessors by a modified ETDRS final scale.13

Neuropathy was assessed clinically by knee and ankle reflexesand by biothesiometer (Biomedical Instruments Co, Newbury,OH, USA) readings at the lateral malleolus and at the end ofthe big toe.13 Autonomic neuropathy was assessed by: R-Rintervals measured on electrocardiograms at expiration andinspiration on deep breathing for five cycles; change in R-Rinterval on standing; basal heart rate during deep breathing;lying and standing blood pressure; and, in men, self-reportederectile dysfunction. These assessments, including visualacuity, grading of photographs, and Minnesota coding, werecarried out by staff from whom the allocations and actualtherapies were concealed.

BiochemistryMethods have been reported previously.18 Plasma glucoseanalysers were monitored monthly in each clinical centre by the UKPDS Glucose Quality Assurance Scheme; the mean interlaboratory imprecision was 4% and values were

ARTICLES


Figure 2: Cross-sectional and 10-year cohort data for FPG, HbA1c, weight, and fasting plasma insulin in patients on intensive orconventional treatment

ARTICLES

within 0·1 mmol/L of those obtained by UK External QualityAssessment Scheme. Plasma creatinine, urea, and urate weremeasured in the clinical chemistry laboratories at the clinicalcentres. Blood, plasma and urine samples were transportedovernight at 4°C to the central biochemistry laboratory for allother measurements. HbA1c was measured by high-performanceliquid chromatography (Biorad Diamat AutomatedGlycosylated Haemoglobin Analyser, Hemel Hempstead, UK),and the normal range is 4·5–6·2%.18 By comparison with theUS National Glycohemoglobin Standardization Program,HbA1c(UKPDS)=1·104 HbA1c(DCCT)–0·7336, (r=0·99,n=40). From 1988 urine albumin was measured by animmunoturbidimetric method (reference range 1·4–36·5mg/L).18 Microalbuminuria has been defined for this study as aurinary albumin concentration greater than 50 mg/L due toinitial storage of urine samples at –20°C between 1979 and1988, and clinical-grade proteinuria as urinary albuminconcentrations greater than 300 mg/L.19 Insulin was measuredby double-antibody radioimmunoassay (Pharmacia RIA 100

Pharmacia Upjohn, Milton Keynes, UK) with 100%cross-reaction to intact proinsulin and 25% to 32/33 splitproinsulin.

Clinical endpoints21 clinical endpoints were predefined in the study protocol in198113 and are listed later. Particular disorders were defined:myocardial infarction by WHO clinical criteria withelectrocardiogram/enzyme changes or new pathological Q wave;angina by WHO clinical criteria and confirmed by a newelectrocardiogram abnormality or positive exercise test; heartfailure (not associated with myocardial infarction), by clinicalsymptoms confirmed by Kerley B lines, râles, raised jugularvenous pressure, or third heart sound; major stroke bysymptoms or signs for 1 month or longer; limb amputation asamputation of at least one digit; blindness in one eye by WHOcriteria with Snellen-chart visual acuity of 6/60 or worse, orETDRS logMAR 1·0 or worse, for 3 months; and renal failureby dialysis or plasma creatinine greater than 250 µmol/L not


Figure 3: Cross-sectional and 10-year cohort data for FPG, HbA1c, weight, and fasting plasma insulin in patients on chlorpropamide,glibenclamide, or insulin, or conventional treatment

related to any acute intercurrent illness. The clinical decisionfor photocoagulation or cataract extraction was made byophthalmologists independent of the trial.

Aggregate endpoints were defined by the Data-Monitoringand Ethics Committee in 1981 as time to the first occurrenceof: any diabetes-related endpoint (sudden death, death fromhyperglycaemia or hypoglycaemia, fatal or non-fatal myocardialinfarction, angina, heart failure, stroke, renal failure,amputation [of at least one digit], vitreous haemorrhage, retinalphotocoagulation, blindness in one eye, or cataract extraction);diabetes-related death (death from myocardial infarction,stroke, peripheral vascular disease, renal disease,hyperglycaemia or hypoglycaemia, and sudden death); all-causemortality. These aggregates were used to assess the differencebetween conventional and intensive treatment.

To investigate differences among chlorpropamide, insulin,and glibenclamide, four additional clinical-endpoint aggregateswere used: myocardial infarction (fatal and non-fatal) andsudden death; stroke (fatal and non-fatal); amputation or deathdue to peripheral vascular disease; and microvascularcomplications (retinopathy requiring photocoagulation, vitreoushaemorrhage, and or fatal or non-fatal renal failure).

Surrogate endpointsSubclinical, surrogate variables were assessed every 3 years.The criteria were: for neuropathy—loss of both ankle or bothknee reflexes or mean biothesiometer reading from both toes 25V or greater; for autonomic neuropathy—R-R interval less thanthe age-adjusted normal range (a ratio <1·03 of the longest R-Rinterval at approximately beat 30 to the shortest atapproximately beat 15); for orthostatic hypotension—systolicfall of 30 mm Hg or more, or diastolic fall of 10 mm Hg ormore; and for impotence—no ejaculation or erection.Retinopathy was defined as one microaneurysm or more in oneeye or worse retinopathy, and progression of retinopathy as atwo-step change in grade. Poor visual acuity was: logMARmore than 0·3 (unable to drive a car), more than 0·7 (USdefinition of blindness), and logMAR 1·0 or greater (WHOdefinition of blindness). Deterioration of vision was defined as a

three-line deterioration in reading an ETDRS chart. Ischaemicheart disease by Minnesota coding was either WHO grade 1(possible coronary heart disease) or grade 2 (probable coronaryheart disease). Left-ventricular hypertrophy was acardiothoracic ratio 0·5 or greater.

The study closed on Sept 30, 1997. All available informationfor each endpoint, such as admission notes, operation records,death certificates, and necropsy reports, were gathered. Thefile, with no reference to assigned or actual therapy, wasreviewed independently by two physicians who assignedappropriate International Classification of Disease–9 codes.20

Any disagreements between the two assessors were discussed andthe evidence reviewed; if agreement was not possible the filewas submitted to two different assessors for final arbitration.

Statistical analysisWhen the UKPDS started in the late 1970s, it was thought thatimproved blood-glucose control might reduce the incidence ofdiabetes-related endpoints by 40%. This seemed reasonablesince the risk of cardiovascular events in patients with diabetesis at least twice that of people with normal glucose toleranceand microvascular complications do not occur in thenormoglycaemic population. The first three aggregateendpoints were defined and, for death and major cardiovascularevents (the stopping criteria), the original power calculation tofind a 40% difference between the intensive and conventionalgroups was a sample size of 3600 with 81% power at the 1%level of significance.

However, by 1987 no risk reduction was seen in any of theseaggregates and it became obvious a 40% advantage was unlikelyto be obtained. The publication of other intervention studies ofchronic diseases in the mid 1980s suggested that a morerealistic goal would be a difference of 15%. Accordingly, thestudy was extended to include randomisation of 3867 patientswith a median time from randomisation of 11 years to the endof the study in 1997. In 1992, at the 1% level of significance,the power for any diabetes-related endpoint and for diabetes-related death was calculated as 81% and 23%, respectively.

There was the same proportion of patients in the

ARTICLES


Figure 4: Proportion of patients with aggregate and single endpoints by intensive and conventional treatment and relative risks

ARTICLES

non-overweight and overweight stratifications assignedintensive and conventional treatment, and, within the intensivegroup, sulphonylurea or insulin treatment, and thus thenon-overweight and overweight patients are analysed together.

The 3867 patients from all 23 centres were included in theanalyses of conventional and intensive treatment.

The analysis among chlorpropamide, glibenclamide, orinsulin in the intensive group used only 3041 patients from thefirst 15 centres where patients had remained for longer periodson monotherapy until marked hyperglycaemia occurred.

Intention-to-treat analysis was used to compare outcomesbetween the intensive and conventional treatment groups andbetween the patients on conventional treatment and those oneach of the intensive treatment agents.

All analyses of significance were two-sided (2p). Life-tableanalyses were done with log-rank tests. Hazard ratios, used toestimate relative risks, were obtained from Cox proportional-hazards models. In the text, the relative risks are quoted in termsof risk reduction. For the clinical endpoint aggregates, 95% CIare quoted. For single endpoints and surrogate variables 99% CIare given to make allowance for potential type I errors. Mean(SD), geometric mean (1SD interval), or median (IQR) have

been quoted for the biometric and biochemical variables, withWilcoxon, t test, or χ2 for comparison tests. Risk reductions forcategorical variables were derived from relative risks obtainedfrom frequency tables. Survival-function estimates werecalculated by the product-limit (Kaplan-Meier) method. Yearlyaveraged data for weight and FPG were calculated as the medianof three consecutive visits for each patient—ie, the annual visit,and the 3 month visit before and after this. HbA1c data were fromthe annual assessment but overall values for HbA1c during aperiod were the median for each patient for each allocation.Glucose control and HbA1c were assessed both cross-sectionallyand in the cohort with 10 years’ follow-up. Urine albumin wasassessed in mg/L with no adjustment for urine creatinineconcentration.21 Data for albuminuria at the triennial visit werethe median of that year and the years before and after.Hypoglycaemic episodes in each year were analysed both byintention to treat and by actual therapy.

SafetyThe Data-monitoring and Ethics Committee reviewed theendpoint analyses every 6 months to decide whether to stop ormodify the study according to predetermined guidelines. These


Figure 5: Proportion of patients with aggregate and single endpoints by individual intensive treatment and conventional treatmentand relative risks Key as for figure 4.

guidelines included a difference of 3 SD or more by log-ranktest in the three aggregate endpoints between intensive andconventional blood-glucose control groups.13 The stoppingcriteria were not attained.

ResultsBackground and biochemical data4763 (93%) of 5102 patients had mean FPG of 7·0mmol/L or more (American Diabetes Associationcriteria),22 and 4396 (86%) of 5102 had values greaterthan 7·8 mmol/L (WHO criteria).23

Baseline characteristics of the 3867 patients assignedconventional or intensive treatment are given in table 1.The baseline characteristics of the 3041 patients in thecomparison of conventional treatment with each of thethree intensive agents are in table 2.

The median follow-up for endpoint analyses was 10·0years (IQR 7·7–12·4). The median follow-up for thecomparison of conventional treatment with each of thethree intensive agents was 11·1 years (9·0–13·0). The

percentage of total person-years for which the assignedor other therapies were taken in the conventional orintensive groups are shown in table 3.

At the end of the trial, the vital status of 76 (2·0%)patients who had emigrated was not known; 57 and 19in intensive and conventional groups, respectively, whichreflects the 70/30 randomisation. A further 91 (2·4%)patients (65 in the intensive group) could not becontacted in the last year of the study for assessment ofclinical endpoints. The corresponding numbers forcomparison of the individual intensive agents were 69(2·7%) emigrated and 63 (2·1%) not contactable.

In the conventional group, the FPG and HbA1c

increased steadily over 10 years from randomisation inboth the cohort study of 461 patients and in thecross-sectional data at each year (figure 2). In theintensive group, there was an initial decrease in FPGand HbA1c in the first year, both in the 10 year cohort of1180 patients and in the cross-sectional data, with asubsequent increase similar to that in the conventional

ARTICLES


Figure 5: Continued

ARTICLES

group (figure 2). A difference between the assignedgroups in HbA1c was maintained throughout the study.The median HbA1c values over 10 years weresignificantly lower in the intensive than in theconventional group (7·0% [6·2–8·2] vs 7·9% [6·9–8·8],p<0·0001). Median HbA1c for 5-year periods of follow-up in the intensive and conventional groups were 6·6%(5·9–7·5) and 7·4% (6·4–8·5) for the first period, 7·5% (6·6–8·8) and 8·4% (7·2–9·4) for the second, and8·1% (7·0–9·4) and 8·7% (7·5–9·7) for the third period(all p<0·0001) .

The median HbA1c values over 10 years withchlorpropamide (6·7%), glibenclamide (7·2%), andinsulin (7·1%) were each significantly lower than thatwith conventional treatment (7·9%, p<0·0001). HbA1c

was significantly lower in the chlorpropamide groupthan in the glibenclamide group (p=0·008) but neitherdiffered from the insulin group (figure 3).

There was a significant increase in weight in theintensive group compared with the conventional group,by (mean) 3·1 kg (99% CI –0·9 to 7·0, p<0·0001) for

the cohort at 10 years (figure 2). Patients assigned eitherof the sulphonylureas gained more weight than theconventional group, whereas patients assigned insulingained more weight than those assigned a sulphonylurea(figure 3). In the cohort at 10 years, those assignedchlorpropamide gained 2·6 kg more (1·6–3·6, p<0·001);those assigned glibenclamide gained 1·7 kg more(0·7–2·7, p<0·001); and those assigned insulin gained4·0 kg more (3·1–4·9, p<0·0001) than those assignedconventional therapy (figure 3). The cross-sectional datawere similar to the cohort data.

Median fasting plasma insulin increased in theintensive group, and was 17·9 pmol/L (95% CI0·5–35·3) greater than in the conventional group overthe first 10 years (p<0·0001, figure 2). Fasting plasmainsulin in participants assigned to sulphonylureasincreased more than in those in the conventional groupover the first 3 years, and in those assigned to insulinthis increase was even greater from 6 years as higherinsulin doses were given (figure 3).

The median insulin doses at 3 years, 6 years, 9 years,


Figure 6: Kaplan-Meier plots of aggregate endpoints: any diabetes-related endpoint and diabetes-related death for conventional orintensive treatment, and by individual intensive therapyKey as for figures 3 and 4.

and 12 years in patients assigned intensive treatmentwith insulin were 22 U (IQR 14–34), 28 U (18–45),34 U (20–50), and 36 U (23–53), respectively. Mediandoses of insulin for patients with body-mass indices lessthan 25 kg/m2 and greater than 35 kg/m2 were 16 U(10–24) and 36 U (23–50) at 3 years; the correspondingdoses were 24 U (14–36) and 60 U (40–82) at 12 years.The maximum insulin dose was 400 U per day.

Systolic and diastolic blood pressure were significantlyhigher throughout the study in patients assignedchlorpropamide than in those assigned any of the othertherapies. For example, at 6 years’ follow-up the meanblood pressure in the chlorpropamide group was 143/82mm Hg compared with 138/80 mm Hg in each of theother allocations (p<0·001). The proportion of patientson therapy for hypertension was higher among thoseassigned chlorpropamide (43%) than among thoseassigned conventional treatment, glibenclamide, orinsulin (34%, 36%, and 38%, respectively; p=0·022).

Aggregate and single endpointsThe number of patients who developed aggregate orsingle clinical endpoints in the intensive andconventional groups are shown in figure 4; similarly,figure 5 shows the comparison between the threeintensive groups and conventional treatment.Kaplan-Meier plots for any diabetes-related endpoint—ie, the complication-free interval—and diabetes-relateddeaths are shown in figure 6 and those for microvascularendpoints, myocardial infarction, and stroke in figure 7.

The number needed to treat to prevent one patientdeveloping any of the single endpoints over 10 years was19·6 patients (95% CI 10–500). The complication-freeinterval, expressed as the follow-up to when 50% of thepatients had at least one diabetes-related endpoint, was14·0 years in the intensive group compared with 12·7years in the conventional group (p=0·029).

Patients assigned intensive treatment had a significant25% risk reduction in microvascular endpoints(p=0·0099) compared with conventional treatment—

ARTICLES


Figure 7: Kaplan-Meier plots of aggregate endpoints: microvascular disease, myocardial infarction, and stroke for intensive andconventional treatment and by individual intensive therapy Microvascular disease=renal failure, death from renal failure, retinal photocoagulation, or vitreous haemorrhage. Myocardial infarction=non-fatal, fatal,or sudden death. Stroke=non-fatal and fatal. Key as for figures 3 and 4.

ARTICLES

most of which was due to fewer cases of retinalphotocoagulation (figure 4): the reduction in risk was ofborderline significance for myocardial infarction(p=0·052) and cataract extraction (p=0·046).

There was no significant difference between the threeintensive treatments on microvascular endpoints or therisk reduction for retinal photocoagulation (figure 5).Few patients developed renal failure, died from renaldisease, or had vitreous haemorrhage.

Surrogate endpointsFigure 8 shows the proportion of patients with surrogateendpoints (two-step progression of retinopathy,biothesiometer threshold, microalbuminuria, protein-uria, and two-fold increase in plasma creatinine) foundat 3-year visits. After 6 years’ follow-up, a smallerproportion of patients in the intensive group than in theconventional group had a two-step deterioration inretinopathy: this finding was significant even whenretinal photocoagulation was excluded (data not shown).When the three intensive treatments were compared,patients assigned chlorpropamide did not have the samerisk reduction as those assigned glibenclamide or insulin(p=0·0056) for the progression of retinopathy at 12years’ follow-up, and adjustment for the difference inmean systolic or diastolic blood pressure by logisticregression analysis did not change this finding.

There was no difference between conventional andintensive treatments in the deterioration of visual acuitywith a mean ETDRS chart reduction of one letter per3 years in each group. At 12 years the proportion ofpatients blind in both eyes (logMAR>0·7) did not differbetween the intensive and conventional groups (6/734[0·8%]) vs 5/263 [1·9%], p=0·15). 11% of patients inboth groups did not have adequate vision for a drivinglicence (logMAR > 0·3 in both eyes).

Proportions of patients with absent ankle reflexes did

not differ between intensive and conventional groups(35 vs 37%, p=0·60); similar proportions had absentknee reflexes (11 vs 12%, p=0·42).

The heart-rate responses to deep breathing andstanding did not differ between the intensive andconventional groups, but at 12 years the basal heart ratewas significantly lower in the intensive than in theconventional group (median 69·8 [IQR 62·5–78·9] vs74·4 [65·2–83·3] bpm, p<0·001). β-blockers were takenby 16% and 19% (p=0·58) of patients in the intensiveand conventional groups.

The proportion of patients with impotence did notdiffer at 12 years in the intensive and conventionalgroups (46·8 vs 54·7%, respectively; p=0·09).

There was no difference between the intensive andconventional treatment groups, or between the threeintensive allocations, in the proportion of patients whohad a silent myocardial infarction, cardiomegaly,evidence of peripheral vascular disease by doppler bloodpressure, or absent peripheral pulses.

Hyperglycaemia and hypoglycaemiaThe proportion of patients with one or more major, orany, hypoglycaemic episode in a year was significantlyhigher in the intensive group than in the conventionalgroup (figure 9). When the three intensive treatmentswere compared by actual therapy, major hypoglycaemicepisodes or any episode were most common in patientson insulin therapy (figure 10). During the first few yearsof therapy, any hypoglycaemic episodes were alsofrequent in patients on glibenclamide orchlorpropamide, but fell as FPG increased. Byintention-to-treat analysis, there was less differencebetween the allocations as more patients in theconventional group had sulphonylurea or insulin therapyadded. One insulin-group patient died at home,unattended: this death was attributed to hypoglycaemia.


Figure 8: Proportion of patients with selected surrogate endpoints at 3-year intervals

In the conventional group, one patient died fromhyperglycaemic, coma after a febrile illness.

Over the first 10 years, the mean proportion ofpatients per year with one or more major hypoglycaemicepisodes while taking their assigned intensive orconventional treatment was 0·4% for chlorpropamide,0·6% for glibenclamide, 2·3% for insulin, and 0·1%for diet; the corresponding rates for any hypoglycaemicepisode were 11·0%, 17·7%, 36·5%, and 1·2%.

By intention-to-treat analyses, major hypoglycaemicepisodes occurred with chlorpropamide (1·0%),glibenclamide (1·4%), insulin (1·8%), and diet (0·7%)and any hypoglycaemic episodes in 16%, 21%, 28%,and 10%, respectively. Hypoglycaemic episodes inpatients on diet therapy were reactive and occurredeither after meals or after termination of glucoseinfusions given while in hospital (eg, postoperatively).

DiscussionWe found that an intensive blood-glucose-control policywith an 11% reduction in median HbA1c over the first 10years decreased the frequency of some clinicalcomplications of type 2 diabetes. The intensivetreatment group had a substantial, 25% reduction in therisk of microvascular endpoints, most of which was dueto fewer patients requiring photocoagulation. There

was evidence, albeit inconclusive, of a 16% riskreduction (p=0·052) for myocardial infarction, whichincluded non-fatal and fatal myocardial infarction andsudden death, but diabetes-related mortality and all-cause mortality did not differ between the intensive andconventional groups. The study did not have sufficientpower to exclude a beneficial effect on fatal outcomes.The progression of subclinical, surrogate variables ofmicrovascular disease was also decreased, in agreementwith other studies of improved glucose control.1–3 Themedian complication-free interval was 1·3 years longerin the intensive group.

The UGDP raised concerns that the sulphonylurea,tolbutamide, may increase the risk of cardiovasculardeath, and several mechanisms by which sulphonylureasmight have an adverse effect were suggested. However,we found no difference in the rates of myocardialinfarction or diabetes-related death between participantsassigned sulphonylurea or insulin therapies. Studies inanimals suggested that first-generation sulphonylureas,such as chlorpropamide, might increase the risk ofventricular fibrillation,10 but this suggestion was notsupported by our findings since the rate of sudden deathwas similar in the groups assigned chlorpropamide,glibenclamide, or insulin. Thus, the UKPDS data donot support the suggestion of adverse cardiovasculareffects from sulphonylureas.

ARTICLES


Figure 9: Major and any hypoglycaemic episodes per year by intention-to-treat analysis and actual therapy for intensive andconventional treatment Data from the first 15 centres. The numbers of patients studied at 5, 10, and 15 years in the intensive and conventional groups by actual therapy were1317, 395; 762, 150; and 120, 14 respectively.

ARTICLES

Exogenous insulin has also been suggested aspotentially harmful treatment because in-vitro studieswith raised insulin concentrations induced atheroma,24

and epidemiological studies showed an associationbetween high plasma insulin concentrations andmyocardial infarction.25,26 We did not find an increase inmyocardial infarctions in patients assigned insulintherapy, even though their fasting plasma insulinconcentrations were higher than those in any othergroup. The macrovascular subclinical surrogateendpoints did not differ between intensive andconventional groups, perhaps because 10 years’follow-up is too short to find changes in atheroma orbecause the endpoints were not sufficiently sensitive.Since there was no evidence, however, for a harmfulcardiovascular effect of sulphonylurea or insulin therapy,it appears that the beneficial effect of an intensiveglucose control with these agents outweighs thetheoretical risks.

The 0·9% difference in HbA1c between the intensive(7·0%) and conventional (7·9%) groups over 10 years,an 11% reduction, is smaller than the 1·9% difference(9·0% and 7·1%; 20% reduction) in HbA1c in theDCCT.2 The DCCT studied younger patients with type 1 diabetes and used slightly different methods thatfocused on surrogate variables. The risk reductions seem

proportional given the HbA1c differences: for progressionof microvascular disease, 21% for retinopathy inUKPDS and 63% in the DCCT; and, for albuminuria,34% and 54% respectively.8 Our data suggest thatclinical benefit can be obtained at lower HbA1c valuesthan those in the DCCT.

Few patients had late ophthalmic complications suchas vitreous haemorrhage or blindness and this may bebecause the follow-up was not long enough or, morelikely, because of the decrease in retinal damage andblindness after photocoagulation.27,28

The reduction in the progression of albuminuria byintensive treatment was probably accompanied bya reduced risk for development of renal failure,since there was a 67% risk reduction in the proportionof patients who had a two-fold increase in plasmacreatinine and 74% risk reduction in those whohad a doubling of their plasma urea. This result ispotentially important since, although less than 1% ofUKPDS patients developed renal failure, in manypopulations type 2 diabetes is the principal cause ofrenal failure.

No difference in the risk reduction of microvascularclinical endpoints was seen between the three intensivetreatments, and thus, improved glycaemic control,rather than any one therapy, is the principal factor.


Figure 10: Major and any hypoglycaemic episodes by intention-to-treat analysis and actual therapy by individual intensive therapyand conventional treatment Data from first 15 centres. The numbers of patients studied at 5, 10 and 15 years in the intensive groups with chlorpropamide, glibenclamide andinsulin and the conventional group by actual therapy were 380, 378, 559, 395; 171, 175, 416, 150; and 21, 16, 83, 14 respectively.

Nevertheless, patients assigned chlorpropamide did nothave the same risk reduction in progression of theretinopathy as those assigned glibenclamide or insulin,and this difference was not accounted for, statistically,by higher blood pressure in the chlorpropamide group.There was no difference in the progression ofalbuminuria between any treatment groups.

Increased blood pressure has been reported withchlorpropamide,29 which can also cause waterretention.30 Other sulphonylureas that do not raise theblood pressure may be preferred.

Intensive blood-glucose control had disadvantagessuch as greater weight gain than occurred in theconventional group. There was also an increased risk ofhypoglycaemic episodes, particularly in patients treatedwith insulin; each year about 3% had a major episodeand 40% a minor or major hypoglycaemic episode.Although the increased risk of hypoglycaemia withinsulin was less than that in the DCCT,31 this risklimited the extent to which normoglycaemia could beobtained in our patients with type 2 diabetes32—as itdoes in patients with type 1 diabetes.2

The relation between glycaemia and outcome in ourstudy is complex. Although a difference in HbA1c

between conventional and intensive groups wasmaintained throughout, HbA1c progressively increased.The risks of hypoglycaemia and of weight gain,particularly in patients treated by insulin, are perceivedby patients as difficulties that limit their ability toachieve improved glucose control (data not shown).Although early addition of other agents may havedelayed the increasing hyperglycaemia, each of theavailable oral hypoglycaemic agents (sulphonylureas,33

metformin,32,34 thiazolidinediones,35 and acarbose36 havelimited glucose-lowering efficacy and many patientseventually required insulin to avoid markedhyperglycaemia. Our patients on intensive treatmentwith insulin achieved lower HbA1c values than those seenin several studies of intensive glucose control in patientswith type 2 diabetes.37–39 Recent recommendations40 setan HbA1c below 7% as a goal but, to our knowledge, thishas been achieved only in intervention studies with highinsulin doses, generally above 100 U per day, in smallgroups of obese patients who received detailed attentionover a short period.41,42 Studies of glycaemic control intype 2 diabetes with insulin therapy in the communityreport mean HbA1c values of 8·5%43 and 9·0%.44 Currenttherapy of type 2 diabetes, including insulin regimens,may need to be reviewed. The US National Health andNutrition Education Examination Survey III, by thesame assay method as the DCCT, found that 51% ofinsulin-treated patients and 42% of those on oralhypoglycaemic agents had HbA1c values greater than 8%,(Maureen Harris, National Institute of Diabetes andDigestive and Kidney Diseases, USA; personalcommunication).

About 50% of patients with newly diagnosed type 2diabetes already have diabetic tissue damage,13 but lackof benefit in these patients from early treatment hasmeant variation in the guidelines for screeningpopulations.45,46 The UKPDS shows that improvedblood-pressure47 and glucose control reduce the risk ofthe diabetic complications that cause both morbidityand premature mortality, and increase the case forformal screening programmes for early detection ofdiabetes in the general population.

Our study, despite the median of 10 years’ follow-upis still short compared with the median life expectancy of20 years in UKPDS patients diagnosed at median age 53years. To investigate longer-term responses, we willcarry out post-study monitoring of UKPDS for a further5 years, to establish whether the improved glucosecontrol achieved will substantially decrease the risk offatal and non-fatal myocardial infarctions with longerfollow-up.

UKPDS shows that an intensive glucose-controltreatment policy that maintains an 11% lower HbA1c—ie, median 7·0% over the first 10 years after diagnosis ofdiabetes—substantially reduces the frequency ofmicrovascular endpoints but not diabetes-relatedmortality or myocardial infarction. The disadvantages ofintensive treatment are weight gain and risk ofhypoglycaemia. There was no evidence that intensivetreatment with chlorpropamide, glibenclamide, orinsulin had a specific adverse effect on macrovasculardisease.

UKPDS Study OrganisationWriting Committee—Robert C Turner, Rury R Holman, Carole A Cull,Irene M Stratton, David R Matthews, Valeria Frighi, Susan E Manley,Andrew Neil, Heather McElroy, David Wright, Eva Kohner, CharlesFox, and David Hadden.Coordinating centre—Chief investigators: R C Turner, R R Holman.Additional investigators: D R Matthews, H A W Neil. Statisticians:I M Stratton, C A Cull, H J McElroy, Z Mehta (previously A Smith,Z Nugent). Biochemist: S E Manley. Research associate: V Frighi.Consultant statistician: R Peto. Epidemiologist: A I Adler (previouslyJ I Mann). Administrator: P A Bassett, (previously S F Oakes).Endpoint assessors: D R Matthews (Oxford), A D Wright(Birmingham), T L Dornan (Salford). Retinal-photography grading:E M Kohner, S Aldington, H Lipinski, R Collum, K Harrison,C MacIntyre, S Skinner, A Mortemore, D Nelson, S Cockley, S Levien,L Bodsworth, R Willox, T Biggs, S Dove, E Beattie, M Gradwell,S Staples, R Lam, F Taylor, L Leung (Hammersmith). Dietician:E A Eeley (Oxford). Biochemistry laboratory staff: M J Payne,R D Carter, S M Brownlee, K E Fisher, K Islam, R Jelfs, P A Williams,F A Williams, P J Sutton, A Ayres, L J Logie, C Lovatt, M A Evans,L A Stowell. Consultant biochemist: I Ross (Aberdeen). Applicationsprogrammer: I A Kennedy. Database clerk: D Croft. ECG coding:A H Keen, C Rose (Guy’s Hospital). Health economists: M Raikou,A M Gray, A J McGuire, P Fenn (Oxford). Quality-of-life questionnaire:Z Mehta (Oxford), A E Fletcher, C Bulpitt, C Battersby(Hammersmith), J S Yudkin (Whittington). Mathematical modeller:R Stevens (Oxford).

Clinical centres—M R Stearn, S L Palmer, M S Hammersley,S L Franklin, R S Spivey, J C Levy, C R Tidy, N J Bell, J Steemson,

B A Barrow, R Coster, K Waring, L Nolan, E Truscott, N Walravens,L Cook, H Lampard, C Merle, P Parker, J McVittie, I Draisey (Oxford);L E Murchison, A H E Brunt, M J Williams, D W Pearson, X M P Petrie, M E J Lean D Walmsley, F Lyall, E Christie, J Church,E Thomson, A Farrow, J M Stowers, M Stowers, K McHardy,N Patterson (Aberdeen); A D Wright, N A Levi, A C I Shearer,R J W Thompson, G Taylor, S Rayton, M Bradbury, A Glover,A Smyth-Osbourne, C Parkes, J Graham, P England, S Gyde, C Eagle,B Chakrabarti, J Smith, J Sherwell (Birmingham); N W Oakley,M A Whitehead, G P Hollier, T Pilkington, J Simpson, M Anderson,S Martin, J Kean, B Rice, A Rolland, J Nisbet (London, St George’s);E M Kohner, A Dornhorst, M C Doddridge, M Dumskyj, S Walji,P Sharp, M Sleightholm, G Vanterpool, C Rose, G Frost, M Roseblade,S Elliott, S Forrester, M Foster, K Myers, R Chapman (London,Hammersmith); J R Hayes, R W Henry, M S Featherston, G P R Archbold, M Copeland, R Harper, I Richardson, S Martin,M Foster, H A Davison (City Hospital, Belfast); L Alexander, J H B Scarpello, D E Shiers, R J Tucker, J R H Worthington, S Angris,A Bates, J Walton, M Teasdale, J Browne, S Stanley, B A Davis,R C Strange (Stoke-on-Trent); D R Hadden, L Kennedy, A B Atkinson,P M Bell, D R McCance, J Rutherford, A M Culbert, C Hegan,H Tennet, N Webb, I Robinson, J Holmes, M Foster, J Rutherford,S Nesbitt (Royal Victoria Hospital, Belfast); A S Spathis, S Hyer,M E Nanson, L M James, J M Tyrell, C Davis, P Strugnell, M Booth,H Petrie, D Clark, B Rice, S Hulland, J L Barron (Carshalton);J S Yudkin, B C Gould, J Singer, A Badenoch, S Walji, M McGregor,L Isenberg, M Eckert, K Alibhai, E Marriot, C Cox, R Price,M Fernandez, A Ryle, S Clarke, G Wallace, E Mehmed, J A Lankester,E Howard, A Waite, S MacFarlane (London, Whittington);

ARTICLES


ARTICLES

R H Greenwood, J Wilson, M J Denholm, R C Temple, K Whitfield,F Johnson, C Munroe, S Gorick, E Duckworth, M Fatman, S Rainbow(Norwich); L J Borthwick, D J Wheatcroft, R J Seaman, R A Christie,W Wheatcroft, P Musk, J White, S McDougal, M Bond, P Raniga(Stevenage); J L Day, M J Doshi, J G Wilson, J R Howard-Williams, H Humphreys, A Graham, K Hicks, S Hexman, P Bayliss, D Pledger(Ipswich); R W Newton, R T Jung, C Roxburgh, B Kilgallon, L Dick, M Foster, N Waugh, S Kilby, A Ellingford, J Burns (Dundee); C V Fox,M C Holloway, H M Coghill, N Hein, A Fox, W Cowan, M Richard, K Quested, S J Evans (Northampton); R B Paisey, N P R Brown, A J Tucker, R Paisey, F Garrett, J Hogg, P Park, K Williams, P Harvey,R Wilcocks, S Mason, J Frost, C Warren, P Rocket, L Bower (Torbay); J M Roland, D J Brown, J Youens, K Stanton-King, H Mungall, V Ball,W Maddison, D Donnelly, S King, P Griffin, S Smith, S Church, G Dunn, A Wilson, K Palmer (Peterborough); P M Brown, D Humphriss, A J M Davidson, R Rose, L Armistead, S Townsend, P Poon (Scarborough); I D A Peacock, N J C Culverwell, M H Charlton, B P S Connolly, J Peacock, J Barrett, J Wain, W Beeston, G King, P G Hill (Derby); A J M Boulton, A M Robertson,V Katoulis, A Olukoga, H McDonald, S Kumar, F Abouaesha, B Abuaisha, E A Knowles, S Higgins, J Booker, J Sunter, K Breislin, R Parker, P Raval, J Curwell, H Davenport, G Shawcross, A Prest, J Grey, H Cole, C Sereviratne (Manchester); R J Young, T L Dornan,J R Clyne, M Gibson, I O’Connell, L M Wong, S J Wilson, K L Wright,C Wallace, D McDowell (Salford); A C Burden, E M Sellen, R Gregory,M Roshan, N Vaghela, M Burden, C Sherriff, S Mansingh, J Clarke, J Grenfell (Leicester); J E Tooke, K MacLeod, C Seamark, M Rammell,C Pym, J Stockman, C Yeo, J Piper, L Leighton, E Green, M Hoyle, KJones, A Hudson, A J James, A Shore, A Higham, B Martin (Exeter).

UKPDS Data Committee—C A Cull, V Frighi, R R Holman,S E Manley, D R Matthews, H A W Neil, I M Stratton, R C Turner.Data-monitoring and Ethics Committee—W J H Butterfield, W R S Doll, R Eastman, F R Ferris, R R Holman, N Kurinij, R Peto,K McPherson, R F Mahler, T W Meade, G Shafer, R C Turner,P J Watkins (Previous members: H Keen, D Siegel).Policy advisory group—C V Fox, D R Hadden, R R Holman,D R Matthews, R C Turner, A D Wright, J S Yudkin.Steering Committee for glucose study—D J Betteridge, R D Cohen,D Currie, J Darbyshire, J V Forrester, T Guppy, R R Holman,D G Johnston, A McGuire, M Murphy, A M el-Nahas, B Pentecost,D Spiegelhalter, R C Turner, (previous members: K G M M Alberti,R Denton, P D Home, S Howell, J R Jarrett, V Marks, M Marmot,J D Ward).

AcknowledgmentsWe thank the patients and many NHS and non-NHS staff at the centresfor their cooperation; Philip Bassett for editorial assistance; andCaroline Wood, Kathy Waring, and Lorraine Mallia for typing thepapers. The study was supported by grants from the UK MedicalResearch Council, British Diabetic Association, UK Department ofHealth, US National Eye Institute and US National Institute ofDiabetes, Digestive and Kidney Disease (National Institutes of Health),British Heart Foundation, Wellcome Trust, Charles Wolfson CharitableTrust, Clothworkers’ Foundation, Health Promotion Research Trust,Alan and Babette Sainsbury Trust, Oxford University Medical ResearchFund Committee, Novo-Nordisk, Bayer, Bristol-Myers Squibb,Hoechst, Lilly, Lipha, and Farmitalia Carlo Erba. We also thankBoehringer Mannheim, Becton Dickinson, Owen Mumford, Securicor,Kodak, Cortecs Diagnostics. We thank Glaxo Wellcome, Smith KlineBeecham, Pfizer, Zeneca, Pharmacia and Upjohn, and Roche forfunding epidemiological, statistical, and health-economics analyses.

References1 Reichard P, Berglund B, Britz A, Cars I, Nilsson BY, Rosenqvist U.

Intensified conventional insulin treatment retards the microvascularcomplications of insulin-dependent diabetes mellitus (IDDM): theStockholm Diabetes Intervention Study (SDIS) after 5 years. J InternMed 1991; 230: 101–08.

2 DCCT Research Group. The effect of intensive treatment of diabeteson the development and progression of long-term complicationsinsulin-dependent diabetes mellitus. N Engl J Med 1993; 329:977–86.

3 Ohkubo Y, Kishikawa H, Araki E, et al. Intensive insulin therapyprevents the progression of diabetic microvascular complications inJapanese patients with non-insulin-dependent diabetes mellitus: arandomized prospective 6-year study. Diabetes Res Clin Pract 1995;28: 103–17.

4 UKPDS Group. UK Prospective Diabetes Study 17: A nine-yearupdate of a randomized, controlled trial on the effect of improvedmetabolic control on complications in non-insulin-dependentdiabetes mellitus. Ann Intern Med 1996; 124: 136–45.

5 Fuller J, McCartney P, Jarrett RJ. Hyperglycaemia and coronaryheart disease: the Whitehall Study. J Chronic Dis 1979; 32: 721–28.

6 Balkau B, Shipley M, Jarrett RJ, et al. High blood glucoseconcentration is a risk factor for mortality in middle-agednondiabetic men. Diabetes Care 1998; 21: 360–67.

7 University Group Diabetes Program. Effects of hypoglycemic agentson vascular complications in patients with adult-onset diabetes VII:mortality and selected nonfatal events with insulin treatment. JAMA 1978; 240: 37–42.

8 University Group Diabetes Program. A study of the effects ofhypoglycemic agents on vascular complications in patients withadult-onset diabetes. Diabetes 1976; 25: 1129–53.

9 Smits P, Thien T. Cardiovascular effects of sulphonylureaderivatives. Diabetologia 1995; 38: 116–21.

10 Pogatsa G. Potassium channels in the cardiovascular system. Diabetes Res Clin Pract 1995; 28 (suppl 1): S91-S98.

11 Stout RW. Insulin and atherosclerosis. In: Stout RW, ed. Diabetesand atherosclerosis. Dordrecht: Kluwer Academic Publishers, 1992:165–201.

12 Genuth S. Exogenous insulin administration and cardiovascular riskin non-insulin-dependent and insulin-dependent diabetes mellitus.Ann Intern Med 1996; 124: 104–09.

13 UKPDS Group. UK Prospective Diabetes Study VIII: study design,progress and performance. Diabetologia 1991; 34: 877–90.

14 Metropolitan Life Insurance Company. Net weight standard for menand women. Stat Bull Metrop Insur Co 1959; 40: 1–4.

15 UKPDS Group. Effect of intensive blood-glucose control withmetformin on complications in overweight patients with type 2diabetes (UKPDS 34). Lancet 1998; 352: 854–65.

16 Hypertension in Diabetes Study IV. Therapeutic requirements tomaintain tight blood pressure control. Diabetologia 1996; 39: 1554–61.

17 Holman RR, Cull CA, Turner RC. Glycaemic improvement over oneyear in a double-blind trial of acarbose in 1,946 NIDDM patients.Diabetologia 1996; 39 (suppl 1): A44.

18 UKPDS Group. UK Prospective Diabetes Study XI: biochemicalrisk factors in type 2 diabetic patients at diagnosis compared withage-matched normal subjects. Diabet Med 1994; 11: 534–44.

19 Manley SE, Burton ME, Fisher KE, Cull CA, Turner RC. Decreasesin albumin/creatinine and N-acetylglucosaminidase/creatinine ratiosin urine samples stored at –20ºC. Clin Chem 1992; 38: 2294–99.

20 World Health Organisation. International Classification ofProcedures in Medicine. Geneva: World Health Organisation, 1978.

21 UKPDS Group. UK Prospective Diabetes Study IX: relationships ofurinary albumin and N-acetylglucosaminidase to glycaemia andhypertension at diagnosis of type 2 (non-insulin-dependent) diabetesmellitus and after 3 months diet therapy. Diabetologia 1992; 36:835–42.

22 American Diabetes Association. Report of the expert committee onthe diagnosis and classification of diabetes mellitus. Diabetes Care1998; 21 (suppl 1); 55–19.

23 World Health Organization. Diabetes mellitus. WHO technicalreport series no 727. Geneva: WHO, 1985.

24 Stout RW. Insulin and atheroma: 20-yr perspective. Diabetes Care1990; 13: 631–54.

25 Pyörälä K. Relationship of glucose tolerance and plasma insulin tothe incidence of coronary heart disease: results from two populationstudies in Finland. Diabetes Care 1979; 2: 131–41.

26 Desprès JP, Lamarche B, Mauriège P, et al. Hyperinsulinemia as anindependent risk factor for ischemic heart disease. N Engl J Med1996; 334: 952–57.

27 Davies EG, Petty RG, Kohner EM. Long term effectiveness ofphotocoagulation for diabetic maculopathy. Eye 1989; 3: 764–67.

28 British Multicentre Group. Photocoagulation for proliferativediabetic retinopathy: a randomised controlled clinical trial using thexenon-arc. Diabetologia 1984; 26: 109–15.

29 Schmitt JK, Moore JR. Hypertension secondary to chlorpropamidewith amelioration by changing to insulin. Am J Hypertens 1993; 6:317–19.

30 Melander A. Sulphonylureas in the treatment ofnon-insulin-dependent diabetes. Baillieres Clin Endocrinol Metab1988; 2: 443–53.

31 DCCT Research Group. Adverse events and their association withtreatment regimens in the Diabetes Control and Complications Trial.Diabetes Care 1995; 18: 1415–27.

32 UKPDS Group. UK Prospective Diabetes Study 16: overview of sixyears’ therapy of type 2 diabetes - a progressive disease. Diabetes1995; 44: 1249–58.

33 UKPDS Group. UK Prospective Diabetes Study 26: sulphonylureafailure in non-insulin dependent diabetic patients over 6 years. Diabet Med 1998; 15: 297–303.

34 UKPDS Group. UK Prospective Diabetes Study 28: a randomisedtrial of efficacy of early addition of metformin insulphonylurea-treated non-insulin dependent diabetes. Diabetes Care1998; 21: 87–92.


35 Kumar S, Boulton AJ, Beck-Nielsen H, et al. Troglitazone, an insulinaction enhancer, improves metabolic control in NIDDM patients.Diabetologia 1996; 39: 701–09.

36 Chiasson JL, Josse RG, Hunt JA, et al. The efficacy of acarbose inthe treatment of patients with non-insulin-dependent diabetesmellitus. A multicenter controlled clinical trial. Ann Intern Med 1996;121: 928–35.

37 Birkeland KI, Rishaug U, Hanssen KE, Vaaler S. NIDDM: a rapidprogressive disease. Diabetologia 1996; 39: 1629–33.

38 Yki-Järvinen H, Kauppila M, Kujansuu E, et al. Comparison ofinsulin regimens in patients with non-insulin-dependent diabetesmellitus. N Engl J Med 1992; 327: 1426–33.

39 Chow CC, Tsang LWW, Sorensen JP. Comparison of insulin with orwithout continuation of oral hypoglycaemic agents in the treatmentof secondary failure in NIDDM patients. Diabetes Care 1995; 18:307–14.

40 American Diabetes Association. Standards of medical care forpatients with diabetes mellitus. Diabetes Care 1998; 21 (suppl 1):S23-S31.

41 Abraira C, Colwell JA, Nuttall FQ. Veterans Affairs CooperativeStudy on glycemic control and complications in Type II diabetes(VACSDM). Diabetes Care 1995; 18: 1113–23.

42 Henry RR, Gumbiner B, Ditzler T, Wallace P, Lyon R, Glauber HS.Intensive conventional insulin therapy for type II diabetes: metaboliceffects during a 6 month outpatient trial. Diabetes Care 1993; 16:21–31.

43 Hayward RA, Manning WG, Kaplan SH, Wagner EH, Greenfield S.Starting insulin therapy in patients with Type 2 diabetes. JAMA 1997; 278: 1663–700.

44 Dunn NR, Bough P. Standards of care of diabetic patients in atypical English community. Br J Gen Pract 1996; 46: 401–05.

45 American Diabetes Association. Clinical practice recommendations1998. Diabetes Care 1998; 21 (suppl 1): S20-S22.

46 de Courten M, Zimmet P. Screening for non-insulin-dependentdiabetes mellitus: where to draw the line? Diabet Med 1997; 14: 5–98.

47 UKPDS Group. Tight blood pressure control and risk ofmacrovascular and microvascular complications in type 2 diabetes:UKPDS 38. BMJ (in press).

ARTICLES


associated with an increased risk of diabetes-relateddeath (96% increased risk [95% CI 2–275], p=0·039)compared with continued sulphonylurea alone. Acombined analysis of the main and supplementary studiesshowed fewer metformin-allocated patients havingdiabetes-related endpoints (risk reduction 19% [2–33],p=0·033). Epidemiological assessment of the possibleassociation of death from diabetes-related causes withthe concurrent therapy of diabetes in 4416 patientsdid not show an increased risk in diabetes-relateddeath in patients treated with a combination ofsulphonylurea and metformin (risk reduction 5% [233 to32], p=0·78).

Interpretation Since intensive glucose control withmetformin appears to decrease the risk of diabetes-related endpoints in overweight diabetic patients, and isassociated with less weight gain and fewerhypoglycaemic attacks than are insulin andsulphonylureas, it may be the first-line pharmacologicaltherapy of choice in these patients.

Lancet 1998; 352: 854–65See Commentary page xxx

IntroductionThe UK Prospective Diabetes Study reported thatintensive blood-glucose control with sulphonylureas orinsulin substantially reduced the risk of complicationsbut not macrovascular disease.1

Metformin is a biguanide that decreases blood glucoseconcentration by mechanisms different from those ofsulphonylurea or insulin. It lowers, rather thanincreases, fasting plasma insulin concentrations2 andacts by enhancing insulin sensitivity, inducing greaterperipheral uptake of glucose, and decreasing hepaticglucose output.3 The improved glucose control isachieved without weight gain.4 Biguanides also decreaseconcentrations of plasminogen-activator inhibitor type 1(PAI-1)5 and may thus increase fibrinolytic activity. Thiseffect may be secondary either to enhanced insulinsensitivity or to lower insulin concentrations, becausetherapy with troglitazone (a thiazolidinedione) alsodecreases production of PAI-1 and increases insulinsensitivity.6

The only long-term outcome data on biguanidesavailable were from the University Group DiabetesProgram (UGDP) study of phenformin. An unexpectedoutcome was higher mortality from cardiovascularcauses with phenformin than with placebo, and for totalmortality for phenformin than with a combination of

Summary

Background In patients with type 2 diabetes, intensiveblood-glucose control with insulin or sulphonylureatherapy decreases progression of microvascular diseaseand may also reduce the risk of heart attacks. This studyinvestigated whether intensive glucose control withmetformin has any specific advantage or disadvantage.

Methods Of 4075 patients recruited to UKPDS in 15centres, 1704 overweight (>120% ideal bodyweight)patients with newly diagnosed type 2 diabetes, mean age53 years, had raised fasting plasma glucose (FPG;6·1–15·0 mmol/L) without hyperglycaemic symptomsafter 3 months’ initial diet. 753 were included in arandomised controlled trial, median duration 10⋅7 years,of conventional policy, primarily with diet alone (n=411)versus intensive blood-glucose control policy withmetformin, aiming for FPG below 6 mmol/L (n=342). Asecondary analysis compared the 342 patients allocatedmetformin with 951 overweight patients allocatedintensive blood-glucose control with chlorpropamide(n=265), glibenclamide (n=277), or insulin (n=409). Theprimary outcome measures were aggregates of anydiabetes-related clinical endpoint, diabetes-related death,and all-cause mortality. In a supplementary randomisedcontrolled trial, 537 non-overweight and overweightpatients, mean age 59 years, who were already onmaximum sulphonylurea therapy but had raised FPG(6·1–15.0 mmol/L) were allocated continuingsulphonylurea therapy alone (n=269) or addition ofmetformin (n=268).

Findings Median glycated haemoglobin (HbA1c) was 7·4%in the metformin group compared with 8·0% in theconventional group. Patients allocated metformin,compared with the conventional group, had riskreductions of 32% (95% CI 13–47, p=0·002) for anydiabetes-related endpoint, 42% for diabetes-related death(9–63, p=0·017), and 36% for all-cause mortality (9–55,p=0·011). Among patients allocated intensive blood-glucose control, metformin showed a greater effect thanchlorpropamide, glibenclamide, or insulin for anydiabetes-related endpoint (p=0·0034), all-cause mortality(p=0·021), and stroke (p=0·032). Early addition ofmetformin in sulphonylurea-treated patients was

Effect of intensive blood-glucose control with metformin oncomplications in overweight patients with type 2 diabetes(UKPDS 34)

UK Prospective Diabetes Study (UKPDS) Group*

ARTICLES

*Study organisation given at end of paper

Correspondence to: Prof Robert Turner, UKPDS Group, Diabetes Research Laboratories, Radcliffe Infirmary, Oxford OX2 6HE, UK


insulin and placebo allocations.7 The study design didnot allow comparison of phenformin with thesulphonylurea used in the UGDP (tolbutamide). Onedeath from lactic acidosis occurred in the phenformingroup. Phenformin was withdrawn from clinical use inmany countries, partly because of the UGDP data andpartly because of the association with lactic acidosis.8

Metformin is now the only biguanide in general use,since it has a 10–20-fold lower risk of lactic acidosisthan phenformin, and is regarded as a safe drugprovided it is not used in at-risk patients, such as thosein renal failure.9

Metformin was included as a randomisation option inoverweight patients in the UK Prospective DiabetesStudy (UKPDS) from 1977 as part of the originalprotocol in the first 15 centres. The primary aim was tocompare conventional treatment (primarily with dietalone) with intensive treatment with metformin,10–12 witha secondary aim of comparing the group allocatedmetformin with overweight patients allocatedsulphonylurea or insulin therapies.

In 1990, increasing glycaemia despite maximumsulphonylurea therapy was noted. Following a UKPDSprotocol amendment, normal-weight and overweightpatients allocated sulphonylurea treatment, who hadfasting plasma glucose (FPG) concentrations of6⋅1–15⋅0 mmol/L but no symptoms on maximum doses,were then assigned either continuing treatment withsulphonylurea alone or addition of metformin tosulphonylurea.

We report here on whether addition of metforminreduces the risk of clinical complications of diabetes.

MethodsPatientsUKPDS has been described in the accompanying paper.1,10 In

brief, between 1977 and 1991, general practitioners in 23centres in the UK referred patients with newly diagnosed type2 diabetes, aged 25–65 years, for possible inclusion in UKPDS.5102 diabetic patients with FPG above 6⋅0 mmol/L on twomornings were recruited. The patients were advised to follow adiet high in carbohydrates and fibre and low in saturated fats,with energy restriction in overweight patients. After 3 monthson diet, 4209 eligible patients with FPG above 6⋅0 mmol/Lwere randomised by a stratified design: 2022 (48%) were non-overweight patients (<120% ideal bodyweight13) and 2187(52%) were overweight. Patients were allocated conventionaltreatment with diet or intensive treatment with sulphonylureaor insulin with metformin as an additional intensive therapyoption in overweight patients in the first 15 centres. We reporthere results for the overweight participants who had FPGbetween 6·1 and 15·0 mmol/L (n=1704) without symptoms ofhyperglycaemia, after diet treatment.

This paper reports on two randomised controlled trials inpatients in the first 15 centres, in which metformin was atherapeutic option.

Trial in overweight, diet-treated patients of intensiveblood-glucose control with metformin versusconventional treatmentThe 1704 overweight patients were randomly assignedconventional treatment, primarily with diet (24%), or intensivetreatment with chlorpropamide (16%), glibenclamide (16%),insulin (24%), or metformin (20%). This report primarilycompares the 411 overweight patients assigned conventionaltreatment and 342 overweight patients assigned intensivetreatment with metformin, as designated in the protocol10

(figure 1). The paper also reports the secondary analysiscomparing the outcomes between overweight patients allocatedmetformin (n=342) with the 951 patients allocated intensivetherapy with chlorpropamide (n=265), glibenclamide (n=277),or insulin (n=409).

Conventional treatment policyThe 411 overweight patients assigned the conventionalapproach continued to receive dietary advice at 3-monthly

ARTICLES


Figure 1: Trial profile for diet/metformin study in overweight diet-treated patients

ARTICLES

clinical visits with the aim of attaining normal bodyweight andFPG to the extent that is feasible in clinical practice. If markedhyperglycaemia developed (defined by the protocol as FPGabove 15 mmol/L or symptoms of hyperglycaemia1) patientswere secondarily randomised to additional non-intensivepharmacological therapy with the other four treatments(metformin, chlorpropamide, glibenclamide, and insulin) in thesame proportions as in the primary randomisations, with theaim of avoiding symptoms and maintaining FPG below 15mmol/L.1 If patients assigned sulphonylurea therapy developedmarked hyperglycaemia, metformin was added to theirregimen; if marked hyperglycaemia recurred, the allocation waschanged to insulin therapy.

Intensive treatment policy with metforminThe aim of the intensive approach for glucose control withmetformin, sulphonylurea, or insulin therapies, in addition todietary advice, was to obtain near-normal FPG (ie, <6⋅0mmol/L). If FPG increased, patients were kept on the allocatedmonotherapy alone until marked hyperglycaemia developed, sothat the clinical effects of each therapy could be assessed.

342 overweight patients were assigned intensive control withmetformin. Treatment started with one 850 mg tablet per day,then 850 mg twice daily, and then 1700 mg in the morning and850 mg with the evening meal (maximum dose=2550 mg). Ifon any dose, symptoms of diarrhoea or nausea occurred,patients were asked to reduce the dose to that which previouslydid not cause symptoms.

When marked hyperglycaemia developed in those allocatedmetformin, glibenclamide was added with the aim of

maintaining FPG below 6⋅0 mmol/L. If marked hyperglycaemiaagain developed, treatment was changed to insulin, initiallyultralente (Ultratard HM, Novo, or Humulin Zn, Lilly) orisophane (NPH) insulin, with the addition of short-acting(regular) insulin, usually soluble insulin before meals whenpremeal or bedtime blood-glucose concentrations were above 7⋅0 mmol/L. If the glucose control was not satisfactory,other regimens could be introduced (eg, soluble/isophaneregimens).

Trial in non-overweight and overweight sulphonylurea-treated patients of addition of metformin versuscontinued sulphonylurea alone1234 patients, both non-overweight and overweight, wereassigned to intensive treatment with sulphonylurea in the first15 centres. Of these, 537 who were treated with maximumdoses of sulphonylurea and had FPG of 6·1–15·0 mmol/Lwithout symptoms of hyperglycaemia, were randomly assigned in equal proportions early addition of metformin to the sulphonylurea (n=269) or continued sulphonylurea alone(n=268; figure 2). If those allocated sulphonylurea alone later developed protocol-defined marked hyperglycaemia,metformin was added. If patients with early or later addition ofmetformin developed protocol-defined marked hyperglycaemia,oral therapy was stopped and changed to insulin therapy.

Combined analysis of two randomised controlled trialsThe unexpected finding of an increased risk of mortality in


Figure 2: Trial profile for sulphonylurea-treated patients with randomisation to metformin

sulphonylurea-treated patients allocated addition of metforminled us to undertake a further statistical analysis. Following atest for heterogeneity between the two trials described above,15

a combined analysis of addition of metformin in patients ondiet therapy and in those on sulphonylurea therapy was done.The datasets were merged by taking time from randomisationto metformin or not, to an event, or to a censor date. A formalmeta-analysis16 was also done.

Epidemiological assessmentWe excluded 623 of the patients (537 in randomised controlledtrial in patients on maximum sulphonylurea treatment of earlyor late addition of metformin, and 86 patients who hadinsufficient baseline data or were not in the main three ethnicgroups). The aim of the epidemiological assessment in 4416participants was to find out whether the combination ofsulphonylurea and metformin was associated with an increasein mortality from diabetes-related causes. 457 patients weretreated by sulphonylurea and metformin: 107 patients assignedconventional therapy in the main randomisation who receivedthe combination after recurrent episodes of protocol-definedmarked hyperglycaemia; 257 patients assigned sulphonylureaor metformin in the main randomisation, or those with markedhyperglycaemia after the initial 3 months’ period, who had theother therapy added when marked hyperglycaemia developed;and 93 who refused allocated insulin. All these patients weretreated by combined therapy because of the progressivehyperglycaemia of type 2 diabetes,11 but if markedhyperglycaemia recurred, the treatment of these patients waschanged to insulin. The combination of sulphonylurea andmetformin was compared with all other therapies in terms ofdiabetes-related deaths by means of a Cox proportional-hazards model, with the actual therapy as a time-dependentcovariate, and allowance for age, sex, ethnic group, and FPGafter 3 months’ diet.

Clinic visitsPatients were seen every month for the first 3 months and thenevery 3 months or more frequently if required to attain controlcriteria. Patients attended fasting for plasma glucose and otherbiochemical measurements, blood pressure and bodyweightwere measured, and therapy was adjusted if necessary. Detailswere recorded of actual therapies, hypoglycaemic episodes, andhome blood-glucose monitoring. At each visit, patients wereasked whether they had experienced hypoglycaemic symptoms.Physicians recorded hypoglycaemic episodes as minor when thepatient was able to treat the symptoms unaided, or major ifthird-party help or medical intervention was necessary. Thenumber of patients, in an allocation and taking the allocatedtherapy, who had one or more minor or major hypoglycaemicepisodes in a year was recorded, and the mean over 10 yearscalculated. Hypoglycaemic episodes in each year were analysedboth by intention to treat and by actual therapy.

Clinical endpoint analysesThe closing date for the study was Sept 30, 1997. Endpoints were aggregated for analysis to keep to a minimumthe numbers of statistical tests.12 The three predefined primaryoutcome analyses were the time to the first occurrence of: anydiabetes-related clinical endpoint (sudden death, death fromhyperglycaemia or hypoglycaemia, fatal or non-fatal myocardialinfarction, angina, heart failure, stroke, renal failure,amputation [of at least one digit], vitreous haemorrhage,retinopathy requiring photocoagulation, blindness in one eye, or cataract extraction); diabetes-related death (death frommyocardial infarction, stroke, peripheral vascular disease, renal disease, hypoglycaemia, or hyperglycaemia, and suddendeath); and all-cause mortality. Four additional clinicalendpoint aggregates were used to assess the effect of therapies on different types of vascular disease in secondaryoutcome analyses: myocardial infarction (fatal and non-fatal

ARTICLES


Conventional Metformin Insulin Chlorpropamide Glibenclamide All patients(n=411) (n=342) (n=409) (n=265) (n=277) (n=1704)

DemographicAge (years)* 53 (9) 53 (8) 53 (8) 53 (9) 53 (9) 53 (8)M/F 193 (47%)/218 157 (46%)/185 192 (47%)/217 119 (45%)/146 127 (46%)/150 784 (46%)/920Ethnicity (%) Caucasian/Indian Asian/ 86/6/7/1 85/4/10/1 88/4/8/0 86/6/8/0 87/4/8/1 86/5/8/1Afro-Caribbean/other

ClinicalWeight (kg)* 87 (15) 87 (17) 85 (14) 85 (15) 86 (14) 86 (15)Body-mass index (kg/m2) 31·8 (4·9) 31·6 (4·8) 31·0 (4·2) 31·2 (4·5) 31·5 (4·4) 31·4 (4·6)Systolic blood pressure (mm Hg)* 140 (18) 140 (18) 139 (19) 141 (18) 139 (19) 140 (18)Diastolic blood pressure (mm Hg)* 86 (10) 85 (9) 85 (10) 86 (9) 85 (9) 86 (10)Smoking (%) never/ex/current 39/36/25 43/32/25 37/34/39 38/30/32 34/35/31 38/34/28Alcohol (%) none/social/regular/ 30/56/14/0·5 27/58/14/1·5 27/57/15/1·2 28/54/17/1·1 25/56/19/1·1 27/56/15/1·1dependentExercise (%) sedentary/moderately 24/40/34/3 29/34/35/3 24/37/36/4 21/38/38/3 21/34/40/5 24/36/36/4active/active/fit

BiochemicalFPG (mmol/L)† 8·0 (7·1–9·3) 8·1 (7·2–9·8) 8·2 (7·2–10·0) 8·0 (7·2–9·6) 8·2 (7·3–9·6) 8·1 (7·1–9·7)HBA1c (%)* 7·1 (1·5) 7·3 (1·5) 7·2 (1·5) 7·2 (1·7) 7·2 (1·5) 7·2 (1·5)Plasma insulin (pmol/L)‡ 114 (71–183) 116 (66–203) 116 (71–186) 111 (65–189) 114 (68–189) 114 (69–190)Triglyceride (mmol/L)‡ 2·96 (1·03–8·47) 2·79 (1·01–7·74) 2·89 (1·02–8·19) 2·85 (1·03–7·86) 2·65 (0·99–7·10) 2·84 (1·02–7·92)Total cholesterol (mmol/L)* 5·5 (1·0) 5·6 (1·3) 5·6 (1·1) 5·6 (1·2) 5·6 (1·2) 5·6 (1·2)LDL cholesterol (mmol/L)* 3·66 (1·04) 3·67 (1·16) 3·69 (1·04) 3·59 (1·10) 3·59 (1·07) 3·65 (1·08)HDL cholesterol (mmol/L)* 1·04 (0·22) 1·06 (0·23) 1·05 (0·23) 1·05 (0·23) 1·07 (0·26) 1·05 (0·23)

MedicationsMore than one aspirin daily (%) 1·5 1·5 2·9 1·9 1·1 1·8Diuretic (%) 20 17 20 20 19 19Others (%) digoxin/antihypertensives/ 0·5/16/0·4/0·4 0·9/15/0/0·3 1·7/12/0/0·3 1·9/15/0·7/0·4 0·4/16/0/0·7 0·9/15/0·1/0·4lipid lowering/HRT or OC

Surrogate clinical endpointsRetinopathy (%) 33 38 39 37 29 36Proteinuria (%) 3·1 2·0 1·1 2·2 2·6 2·2Plasma creatinine (mmol/L)‡ 78 (64–96) 77 (63–95) 77 (63–94) 79 (65–96) 79 (65–97) 79 (66–96)Biothesiometer more than 25 V (%) 13·6 13·7 15·4 19·9 14·3 15·2

Data are % of group, *mean (SD),†median (IQR), or ‡geometric mean (1 SD). HRT=hormone replacement therapy; OC=oral contraceptive therapy.

Table 1: Baseline characteristics of patients in conventional group and in individual intensive-treatment groups

ARTICLES

and sudden death); stroke (fatal and non-fatal); amputation (ofat least one digit) or death due to peripheral vascular disease(including death from gangrene); and microvascularcomplications (retinopathy requiring photocoagulation,vitreous haemorrhage, and fatal or non-fatal renal failure).Subclinical, surrogate variables1 were assessed every 3 years.

BiochemistryMethods have been previously reported.1,17 The normal rangefor glycated haemoglobin (HbA1c) was 4·5–6·2%.Microalbuminuria has been defined for this study as urinaryalbumin concentration above 50 mg/L and clinical gradeproteinuria as more than 300 mg/L.

AssignmentAll randomisations were done at the level of the individualpatient, by means of therapy allocations in sealed opaqueenvelopes, which were opened in sequence. The numerical

sequence of envelopes used, the dates they were opened, andthe therapies stipulated were monitored. No placebo was given.

Statistical analysisAnalyses were by intention to treat. Life-table analyses weredone with log-rank tests and hazard ratios, used to estimaterelative risks, were obtained from Cox proportional-hazardsmodels. For the primary and secondary outcome analyses ofclinical endpoint aggregates, 95% CIs are quoted. For singleendpoints 99% CIs are quoted, to make allowance for potentialtype 1 errors.1 Further details are given in the accompanyingpaper.1

ResultsIntensive blood-glucose control with metformin versusconventional treatment in overweight patientsTable 1 shows the baseline data for overweight patients


Figure 3: Median FPG, median HbA1c, mean change in bodyweight, and median change in fasting plasma insulin in cohorts ofpatients followed up to 10 years by assigned treatment (shown by continuous lines)Cross-sectional data at each year are shown by individual symbols for all patients assigned regimen.

at the time of randomisation to conventional treatmentor intensive treatment with chlorpropamide,glibenclamide, insulin, or metformin. The mean body-mass index for overweight patients with type 2 diabeteswas 31·4 kg/m2 (SD 4·6); 99⋅5% of patients had body-mass index greater than 25 kg/m2, and 54⋅0% had body-mass index greater than 30 kg/m2.

The median follow-up (to the last known date atwhich vital status was known or to the end of the trial)was 10·7 years. Vital status was not known at the end ofthe trial for 13 (1·8%) patients who had emigrated. A further 43 (2·5%) patients could not be contacted inthe last year of the study for assessment of clinicalendpoints.

Figure 3 shows the median FPG and HbA1c in thecohort of 482 patients with data available studied over10 years and cross-sectional data for all those assignedeach therapy. In the metformin group there was adecrease in FPG and HbA1c in the first year, with asubsequent gradual rise in both variables. From 10years, FPG in the metformin group approached that ofthe conventional treatment group. The median HbA1c

during the 10 years of follow-up was 7·4% in themetformin group and 8⋅0% in the conventionaltreatment group. The patients assigned intensive controlwith sulphonylurea or insulin had similar HbA1c to themetformin group. The median HbA1c values in themetformin group and conventional control group were6·7% and 7·5%, respectively, in the first 5 years offollow-up, 7·9% and 8·5% in the second 5 years, and8·3% and 8·8% in the last 5 years. The cross-sectional

data, of all patients at each year, were similar to thecohort data.

For the cohorts followed up for 10 years, the changein bodyweight was similar in the metformin andconventional control groups, and less than the increasein bodyweight observed in patients assigned intensivecontrol with sulphonylureas or insulin. There was adecrease in fasting plasma insulin in the patientsassigned metformin, which persisted throughout follow-up (figure 3).

Of the 4292 person-years of follow-up amongpatients assigned conventional control, 2395 (56%)were treated by diet. The remaining 44% of person-years required, as per protocol, additional non-intensivepharmacological therapies. Of the 3682 person-years offollow-up among the overweight patients assignedmetformin, 3035 (82%) were treated with metforminalone or in combination. The median dose of metforminwas 2550 mg/day (IQR 1700–2550). For theconventional control group, there were 3557 (83%) ofperson-years with crossover to metformin therapy.

Figure 4 shows the proportion of patients per yearwho had a major hypoglycaemic episode according toactual therapy and intention to treat. The rate of anyhypoglycaemic episodes was higher in patients takingmetformin as allocated than in those on diet alone butlower than the rates in those taking sulphonylureas asallocated. The rate of hypoglycaemic episodes increasedover time among patients treated with insulin, as higherinsulin doses were required, and decreased among thoseon sulphonylurea therapy, as glucose concentrations

ARTICLES


Figure 4: Proportion of patients who reported one or more episodes of major hypoglycaemia or any hypoglycaemia per year,assessed by actual therapy and by allocation (intention to treat) Numbers of patients studied at 5 years, 10 years, and 15 years in actual therapy analysis=168, 60, and 6 for conventional group; 220, 101, and 6 formetformin group; 235, 166, and 26 for insulin group; 148, 60, and 5 for chlorpropamide group; and 161, 71, and 6 for glibenclamide group.

ARTICLES

Aggregate and single endpoints (diet vs metforminstudy)Patients assigned intensive blood-glucose control withmetformin had a 32% lower risk (p=0·0023) ofdeveloping any diabetes-related endpoint than thoseallocated conventional blood-glucose control (figures 5and 6). These endpoints included macrovascular andmicrovascular complications and represented the effectof intensive policy with metformin on complication-freesurvival. The group assigned metformin had asignificantly greater risk reduction than those assignedintensive therapy with sulphonylurea or insulin(p=0·0034).

The metformin group had a lower risk of diabetes-related death than the conventional treatment group(figures 5 and 6), with no significant difference betweenthe metformin group and those assigned therapy withsulphonylurea or insulin. There were no deaths fromlactic acidosis.

Cardiovascular disease accounted for 62% of the totalmortality in the overweight patients in the conventionaltreatment group. The metformin group had a 36%lower risk (p=0·011) of all-cause mortality than theconventional group (figure 6). There was a greater riskreduction than in the groups assigned intensive therapywith sulphonylurea or insulin (p=0·021). Themetformin group had a 39% lower risk (p=0·010) ofmyocardial infarction than the conventional treatmentgroup, but did not differ from the other intensivetreatment group (figure 6). There were no significantdifferences between the metformin group and theconventional group in the other aggregate endpoints.For all macrovascular diseases together (myocardialinfarction, sudden death, angina, stroke, and peripheraldisease), the metformin group had a 30% (5–48,p=0·020) lower risk than the conventional treatmentgroup but did not differ significantly from the otherintensive groups.

Data for the single endpoints are shown in figures 7and 8. There was no difference in the rate of death dueto non-diabetes-related endpoints (accidents, cancer,other specified causes, or unknown causes).

Surrogate endpoints—The metformin group had alower rate of progression to retinopathy than theconventional group, of borderline significance(p=0·044), at 9 years; there was no difference at 12years. The result was similar to that in the otherintensive therapy group. The proportion of patients withurine albumin above 50 mg/L did not differ significantlybetween the intensive treatment, metformin, andconventional groups (24%, 23%, and 23% respectively).There was no difference between the treatment groupsin any of the surrogate indices of macrovascular disease.

Addition of metformin in patients receivingsulphonylureaTable 2 shows the demographic data for the patientswhose response to maximum sulphonylurea treatmentwas not adequate (FPG 6·1–15·0 mmol/L) and whowere assigned continuing intensive policy withsulphonylurea alone or with early addition ofmetformin. The mean body-mass index of normal andoverweight patients in this study was 29·6 kg/m2 (SD5·5); 17% had body-mass index below 25 kg/m2 and39% had values above 30 kg/m2.


Figure 5: Kaplan-Meier plots in diet/metformin study for anydiabetes-related clinical endpoint and diabetes-related deathIntensive, in this figure, indicates chlorpropamide, glibenclamide, andinsulin groups. Similar plots of data for sulphonylurea/metformin studyare superimposed showing relative time of commencement.

increased. Over 10 years of follow-up among patientstaking therapy as allocated, the proportions of patientsper year who had one or more major hypoglycaemicattacks in the conventional, chlorpropamide,glibenclamide, insulin, and metformin groups were0·7%, 0·6%, 2·5%, 0·3%, and 0% respectively; for anyhypoglycaemic episode the corresponding proportionswere 0·9%, 12·1%, 17·5%, 34·0%, and 4·2%.

Among all patients assigned treatments (intention-to-treat analyses), major hypoglycaemic episodesoccurred in 0·7%, 1·2%, 1·0%, 2·0%, and 0⋅6%,respectively, of the conventional, chlorpropamide,glibenclamide, insulin, and metformin groups, and anyhypoglycaemic episodes in 7·9%, 15·2%, 20·5%, 25·5%,and 8·3%, respectively. Hypoglycaemic episodes inpatients on diet therapy were reactive hypoglycaemicattacks, either after meals or, in some patients, aftertermination of glucose infusions while in hospital (eg,postoperatively).

The median duration from the initial randomisationto subsequent randomisation of addition or no additionof metformin was 7·1 years. The median follow-up afterrandomisation was 6·6 years. Vital status was not knownin ten (2%) patients who had emigrated and a furtherfive (1%) who could not be contacted.

Figure 9 shows the median FPG and HbA1c in thecohorts studied for 4 years after second randomisationto addition or no addition of metformin therapycompared with data for all the overweight patients in thecomparison of intensive control with metformin andconventional control. There was a decrease in FPG inpatients on sulphonylurea therapy who were assignedaddition of metformin, whereas FPG concentrations inthose on sulphonylurea therapy alone approached thoseof overweight patients in the conventional treatmentgroup. HbA1c values in patients with addition ofmetformin decreased initially but approached those ofthe patients remaining on sulphonylurea alone after 3

years. The median HbA1c over 4 years in the cohort withaddition of metformin was 7·7% compared with 8·2% inthose on sulphonylurea alone. There were no significantdifferences in bodyweight or plasma insulin between thegroups allocated addition of metformin or continuedsulphonylurea therapy alone.

The patients assigned addition of metformin took thisdrug for 62% of their person-years of follow-up. Forthose randomly assigned continuing sulphonylureaalone, there were 75% of person-years withoutmetformin therapy.

Aggregate and single endpoints (addition of metforminstudy)Figure 10 shows the aggregates of endpoint data andfigure 11 the single endpoint data.

The addition of metformin to sulphonylurea wasassociated with a 96% increased (p=0·039) risk ofdiabetes-related death. Addition of metformin to

ARTICLES


Figure 6: Incidence of clinical endpoints among patients assigned intensive control with metformin (n=342), intensive control withchlorpropamide, glibenclamide, or insulin (intensive; n=951), or conventional control (n=411)Relative risk (RR) is for metformin or intensive group compared with conventional group.

Figure 7: Kaplan-Meier plots in diet/metformin study for microvascular disease (renal failure or death from renal failure, retinopathyrequiring photocoagulation, or vitreous haemorrhage), myocardial infarction (non-fatal and fatal, including sudden death), stroke(non-fatal and fatal) and cataract extractionSimilar plots of data for sulphonylurea/metformin study are superimposed showing relative time of commencement.

ARTICLES

sulphonylurea therapy also increased the risk of deathfrom any cause (60% increase, p=0·041). There were nosignificant differences between the groups for the otheraggregate endpoints. In a subgroup analysis, there wasno significant difference between patients allocatedmetformin in addition to chlorpropamide orglibenclamide (data not shown).

The data for the single endpoints are shown in figure 11.

Combined analysis of both trialsHeterogeneity tests confirmed the different outcomesbetween the two trials for any diabetes-related endpoint(p=0·034), diabetes-related death (p=0·00256), and all-cause mortality (p=0·0173), with a non-significant trendfor myocardial infarction (p=0·068). Figure 10 shows

the results for the two trials combined, with a 12%reduced risk for any diabetes-related endpoint(p=0·033). A formal meta-analysis gave similar resultsfor diabetes-related endpoints (observed minus expected22·7, variance 104·9, p=0·026) and for myocardialinfarction (observed minus expected 12·2, variance43·9, p=0·065).

Epidemiological analysisThe 4417 patients had 45 527 person-years of follow-up; 5181 (11%) of these person-years were treated withsulphonylurea plus metformin therapy. 39 (8%) of the490 diabetes-related deaths occurred while patients were receiving sulphonylurea plus metformin therapy. ACox proportional-hazards model, with adjustment forage, sex, ethnic group, and FPG after 3 months’ diet,


Figure 8: Incidence of single endpoints in diet vs metformin studyRelative risk (RR) is for comparison with conventional control.

Sulphonylurea alone (n=269) Sulphonylurea plus metformin (n=268) All patients (n=537)

DemographicAge (years)* 58 (9) 59 (8) 59 (9)M/F 164 (61%)/108 158 (59%)/118 322 (60%)/226Ethnicity (%) Caucasian/Indian Asian/Afro-Caribbean/other 77/13/10/0 77/11/12/0 77/11/11/1

ClinicalWeight (kg)* 82 (16) 83 (16) 83 (16)Body-mass index (kg/m2) 29·4 (5·7) 29·7 (5·3) 29·6 (5·5)Systolic blood pressure (mm Hg)* 138 (21) 140 (20) 139 (21)Diastolic blood pressure (mm Hg)* 81 (11) 83 (11) 82 (11)Smoking (%) never/ex/current 31/40/29 35/40/24 33/40/27Alcohol (%) none/social/regular/dependent 37/44/18/0·4 32/51/16/1·1 34/52/13/0·8Exercise (%) sedentary/moderately active/active/fit 14/38/45/3 22/35/39/4 18/37/42/3

BiochemicalFPG (mmol/L)† 9·2 (7·8–10·9) 9·0 (7·6–11·3) 9·1 (7·7–11·1)HBA1c (%)* 7·6 (1·8) 7·5 (1·7) 7·5 (1·7)Plasma insulin (pmol/L)‡ 102 (58–180) 102 (58–181) 102 (58–181)Triglyceride (mmol/L)‡ 1·61 (0·91–2·86) 1·64 (0·89–3·04) 1·63 (0·90–2·95)Total cholesterol (mmol/L)* 5·9 (1·0) 5·6 (1·1) 5·6 (1·1)LDL cholesterol (mmol/L)* 3·67 (0·96) 3·53 (0·93) 3·60 (0·95)HDL cholesterol (mmol/L)* 1·08 (0·28) 1·10 (0·30) 1·09 (0·29)

MedicationsMore than one aspirin daily (%) 6·4 4·5 5·5Diuretic (%) 13 16 15Others (%) digoxin/antihypertensives/lipid lowering/HRT or OC 1·9/24/0·4/0·8 1·5/25/0/0·4 1·7/25/0·4/0·8

Data are % of group, *mean (SD), †median (IQR), or ‡geometric mean ( 1 SD). HRT=hormone replacement therapy; OC=oral contraceptive therapy.

Table 2: Baseline characteristics of patients assigned sulphonylurea treatment and subsequently randomised to continuingsulphonylurea treatment alone or with early addition of metformin

with current therapies as a time-dependent variable,showed a non-significant risk reduction in diabetes-related death for sulphonylurea plus metformincompared with all other treatments of 5% (95%CI -33 to 32, p=0·78).

DiscussionThe main trial reported in this paper evaluated theeffect of metformin in diet-treated overweight patientswith type 2 diabetes. The study design parallels that inthe accompanying paper,1 comparing conventionalblood-glucose control primarily with diet alone andintensive treatment with sulphonylurea or insulin. Thedata shown here suggest that metformin therapy in diet-treated overweight patients reduced the risk for anydiabetes-related endpoint, diabetes-related death, and

all-cause mortality. These possible benefits were notseen in the second trial reported here, which suggests anincreased risk for diabetes-related deaths and all-causemortality when metformin is given in addition tosulphonylurea therapy in non-overweight andoverweight patients. Because the difference in the effectof metformin between diet-treated and sulphonylurea-treated patients could be extremes of the play of chance,a combined analysis of all the data was undertaken. Thisshowed that addition of metformin had a comparableeffect to that seen with intensive therapy withsulphonylurea or insulin reported in the accompanyingpaper1 with a net reduction of 19% in any diabetes-related endpoint (p=0·033).

The trend to a reduced risk for microvascularendpoints with metformin therapy was comparable to

ARTICLES


Figure 9: Median FPG and median HbA1c in cohorts of patients followed to 10 years from primary randomisation in diet vs metforminstudy, and cohorts of patients followed to 4 years from second randomisation to sulphonylurea alone or sulphonylurea plusmetformin in sulphonylurea vs metformin study

Figure 10: Incidence of clinical endpoints in sulphonylurea vs metformin study and diet vs metformin studyRelative risk (RR) is for comparison with conventional or sulphonylurea alone. Results of a combined analysis of these two studies shown also.

ARTICLES

that reported in the accompanying paper for intensiveglucose control1 but did not achieve statisticalsignificance.

Clinical use of metformin in overweight patientsIn diet-treated overweight patients metformin similarlyimproved HbA1c levels as with sulphonylurea and insulintherapy but did not induce weight gain and wasassociated with fewer episodes of hypoglycaemia. Giventhe equivalent HbA1c levels obtained, the possibleadditional benefit of metformin observed in overweightdiet-treated patients, of a reduced risk for any diabetes-related endpoint, all mortality, and stroke is notexplicable on the basis of glycaemic control. Theimprovements in the predominantly cardiovascularoutcomes seen with metformin may be due to thedecrease in PAI-1 that accompanies the metformin-induced increase in insulin sensitivity.3 PAI-1 caninhibit fibrinolysis; thus decrease in PAI-1 could lessenthe likelihood of extension of a thrombolysis. In addition, metformin lowers systemic methylglyoxalconcentrations in patients with type 2 diabetes,18 whichsuggests that it may have an aminoguanidine-like action.However, these postulated mechanisms may not berelevant since, in the combined analysis, the effect ofmetformin on cardiovascular outcomes was notsubstantiated.

Clinical use of metformin in patients already treatedwith sulphonylureaWhen metformin was prescribed in the trial in bothnon-overweight and overweight patients already treatedwith sulphonylurea there was a significant increase inrisk of diabetes-related death and all-cause mortalityrather than a beneficial effect on the primary outcome.The different outcomes seen in these two trials may beexplained by differences in the patients studied. Thesulphonylurea-treated patients were on average 5 yearsolder; more hyperglycaemic (baseline median FPG 9·1vs 8·1 mmol/L); less overweight; and followed up on

average for 5 years less. Secondly, it is important to notethat the differences in outcome relate to a relativelysmall number of endpoints. The epidemiologicalanalysis did not corroborate an association of diabetes-related deaths with combined sulphonylurea andmetformin therapy although the CIs were wide.

The UKPDS studied metformin primarily in obesepatients, since when the study started (1970s),metformin was generally prescribed only in suchpatients. Obesity is common among patients with type 2diabetes.19 At entry to UKPDS, body-mass index wasabove 25 kg/m2 in 75% of patients and above 30 kg/m2

in 35%.Since metformin seems to give risk reduction of

diabetes-related endpoints in overweight patients withtype 2 diabetes, does not induce weight gain, and isassociated with fewer hypoglycaemic attacks thansulphonylurea or insulin therapy,10 it could be chosen asthe first-line pharmacological therapy in such patients.Although these findings may not apply to non-overweight patients, metformin seems to lowerglycaemia in patients with type 2 diabetes, irrespectiveof the degree of obesity.1

ConclusionThe addition of metformin in patients already treatedwith sulphonylureas requires further study. On balance,metformin treatment appears to be advantageous as afirst-line pharmacological therapy in diet-treatedoverweight patients with type 2 diabetes.

UKPDS Study OrganisationParticipating centres—Radcliffe Infirmary, Oxford; Royal Infirmary,Aberdeen; Birmingham General Hospital; St George’s Hospital,London; Hammersmith Hospital, London; Belfast City Hospital; NorthStaffordshire Royal Infirmary, Stoke-on-Trent; Royal Victoria Hospital,Belfast; St Helier Hospital, Carshalton; Whittington Hospital, London;Norfolk and Norwich Hospital, Norwich; Lister Hospital, Stevenage;Ipswich Hospital; Ninewells Hospital, Dundee; Northampton Hospital;Torbay Hospital; Peterborough General Hospital; ScarboroughHospital; Derbyshire Royal Infirmary; Manchester Royal Infirmary;


Figure 11: Incidence of single endpoints in sulphonylurea vs metformin studyRelative risk (RR) is for sulphonylurea plus metformin vs sulphonylurea alone.

Hope Hospital, Salford; Leicester General Hospital; Royal Devon andExeter Hospital.

Writing committee—Robert C Turner, Rury R Holman, Irene MStratton, Carole A Cull, David R Matthews, Susan E Manley, ValeriaFrighi, David Wright, Andrew Neil, Eva Kohner, Heather McElroy,Charles Fox, David Hadden

AcknowledgmentsWe thank the patients and many NHS and non-NHS staff at the centresfor their cooperation. Major grants for this study were obtained from the UK MedicalResearch Council, British Diabetic Association, the UK Department ofHealth, the National Eye Institute and the National Institute ofDigestive, Diabetes and Kidney Disease in the National Institutes ofHealth, USA, the British Heart Foundation, Novo-Nordisk, Bayer,Bristol Myers Squibb, Hoechst, Lilly, Lipha, and Farmitalia Carlo Erba.Other funding companies and agencies, the supervising committees, andall participating staff are listed in reference 11.

References1 UKPDS Group. Intensive blood-glucose control with sulphonylureas

or insulin compared with conventional treatment and risk ofcomplications in patients with type 2 diabetes (UKPDS 33). Lancet1998; 352: 837–53.

2 UKPDS Group. UK Prospective Diabetes Study 24: relative efficacyof sulfonylurea, insulin and metformin therapy in newly diagnosednon-insulin dependent diabetes with primary diet failure followed forsix years. Ann Intern Med 1998; 128: 165–75.

3 Cusi K, Consoli A, DeFronzo RA. Metabolic effects of metforminon glucose and lactate metabolism in noninsulin-dependent diabetesmellitus. J Clin Endocrinol Metab 1996; 81: 4059–67.

4 Bailey CJ. Biguanides and NIDDM. Diabetes Care 1992; 15:755–72.

5 Nagi DK, Yudkin JS. Effects of metformin on insulin resistance, riskfactors for cardiovascular disease, and plasminogen activatorinhibitor in NIDDM subjects: a study of two ethnic groups. DiabetesCare 1993; 16: 621–29.

6 Nolan JJ, Ludvik B, Beersden P, Joyce M, Olefsky J. Improvement inglucose tolerance and insulin resistance in obese subjects treated

with troglitazone. N Engl J Med 1994; 331: 1188–93.7 University Group Diabetes Program. A study of the effects of

hypoglycemic agents on vascular complications on patients withadult-onset diabetes: V– evaluation of phenformin therapy. Diabetes1975; 24 (suppl 1): 65–184.

8 Nattrass M, Alberti KG. Biguanides. Diabetologia 1978; 14: 71–74.

9 Bailey CJ, Turner RC. Metformin. N Engl J Med 1996; 334:574–79.

10 UKPDS Group. UK Prospective Diabetes Study VIII: study design,progress and performance. Diabetologia 1991; 34: 877–90.

11 UKPDS Group. UK Prospective Diabetes Study 16: overview of sixyears’ therapy of type 2 diabetes—a progressive disease. Diabetes1995; 44: 1249–58.

12 UKPDS Group. UK Prospective Diabetes Study 17: a nine-yearupdate of a randomized, controlled trial on the effect of improvedmetabolic control on complications in non-insulin-dependentdiabetes mellitus. Ann Intern Med 1996; 124: 136–45.

13 Metropolitan Life Insurance Company. Net weight standard for menand women. Statist Bull 1959; 40: 1–4.

14 UKPDS Group. UK Prospective Diabetes Study 28: a randomisedtrial of efficacy of early addition of metformin in sulphonylurea-treated non-insulin dependent diabetes. Diabetes Care 1998; 21:87–92.

15 Rothman KJ. Modern epidemiology. Boston: Little, Brown, 1986.16 Early Breast Cancer Trialists Collaborative Group. Treatment of

early breast cancer. Oxford: Oxford University Press, 1990.17 UKPDS Group. UK Prospective Diabetes Study XI: biochemical

risk factors in type 2 diabetic patients at diagnosis compared withage-matched normal subjects. Diabet Med 1994; 11: 534–44.

18 Beisswenger P, Howell S, Touchette A, Lal S, Szwergold B, Rohlf J.Metformin reduces systemic methylgloxal levels in NIDDM. Diabetes1997; 46 (suppl 1): 74A (abstr).

19 Modan M, Karasik A, Halkin H, et al. Effect of past and concurrentbody mass index on prevalence of glucose intolerance and type 2(non-insulin-dependent) diabetes and on insulin response: the Israelstudy of glucose intolerance, obesity and hypertension. Diabetologia1986; 29: 82–89.

20 Yki-Jarvinen H, Nikkil K, Ryysy L, Tulokas T, Vanamo R, HekkilM. Comparison of bedtime insulin regimens in NIDDM: metforminprevents insulin-induced weight gain. Diabetologia 1996; 39 (suppl1): A33 (abstr).

ARTICLES


Combination Therapies With Insulin inType 2 DiabetesHANNELE YKI-JARVINEN, MD, FRCP

1

The U.K. Prospective Diabetes Study(UKPDS) demonstrated that inten-sive glucose control with insulin or

sulfonylureas markedly reduces the riskof microvascular complications (1). Formyocardial infarction, the reduction inrisk (16% for a 0.9% decrease in HbA1c)was of borderline significance but corre-sponded closely to epidemiological pre-dictions (14% decrease for a 1% drop inHbA1c) (2). These data demonstrated thatneither insulin nor sulfonylureas, despitecausing hyperinsulinemia and weight gain,have adverse effects on cardiovascular out-come. Glycemic control deteriorated con-tinuously, however, even in intensivelytreated patients in the UKPDS (1).

In the UKPDS, the worsening of gly-cemic control has been attributed to thenatural course of type 2 diabetes and lackof efficacy of current antihyperglycemictherapies (1). Insulin therapy consisted ofa single injection of ultralente or isophaneinsulin. If the daily dose exceeded 14 U,regular insulin was added and home-glucose monitoring was encouraged (1).Combination therapy regimens with in-sulin and oral agents were not used. Wenow know that 14 U of long-acting insu-lin is insufficient to control fasting glyce-mia in most type 2 diabetic patients (3).Since 1977, when the UKPDS was started,several studies have tried to define the op-timal insulin treatment regimen for type 2diabetic patients. These studies are the fo-cus of this review and include studiescomparing insulin alone to combinationtherapy with insulin and sulfonylureas(subject to meta-analyses in 1991 and1992) (4,5) and more recent trials usingmetformin, glitazones, or acarbose in in-sulin combination therapy regimens.They do not contain data on cardiovascu-

lar end points but only on surrogatemarkers of risk of micro- and macrovas-cular complications, mostly data on gly-cemia, body weight, insulin doses, lipids,and in a few studies, also accurate data onthe frequency of hypoglycemias.

According to a Medline search (1966–2000), insulin alone has been comparedwith insulin combination therapy in a to-tal of 34 prospective studies that lasted atleast 2 months and reported data on HbA1or HbA1c in type 2 diabetic patients. Stud-ies comparing glycemic control, weightgain, hypoglycemias, and insulin require-ments between the two modes of treat-ment in insulin-naıve patients are listed inTable 1 and in previously insulin-treatedpatients are listed in Table 2. The studieshave been ranked according to glycemiccontrol at the end of the trial.

GLYCEMIC CONTROL ANDINSULIN REQUIREMENTS

Insulin-naıve and previously insulin-treated patients.In insulin-naıve patients in a total of 15comparisons (10 studies), glycemic con-trol was similar in most (11 of 15) com-parisons and better with the insulincombination than the insulin-alone regi-men in four comparisons (Table 1). In allstudies, the daily insulin dose was lowerwith insulin combination therapy thanwith insulin alone. The weighted meanfor the insulin-sparing effect of two drugs(sulfonylureas and metformin) in addi-tion to insulin was 62%, i.e., 1.5–2.0 timesthat with regimens combining either met-formin alone (232%) or sulfonylureasalone (242%) (Table 1) with insulin. Thesedata imply that oral agents still have sig-nificant glucose-lowering effects even in

patients who are poorly controlled on oraldrugs. One may also predict from thesedata that if the insulin dose is lowered lessthan ;30% when patients are transferredfrom insulin alone to insulin combinedwith sulfonylurea or metformin, glycemiccontrol will be better during insulin com-bination therapy. This is documented byanalysis of data from studies in previouslyinsulin-treated patients (Table 2). In thesecomparisons, glycemic control was betterin most (19 of 25) comparisons, but theinsulin dose was decreased by only 19%in the combination regimens using met-formin and by 21% in comparisons usinginsulin and sulfonylureas (Table 2). Thus,although in most comparisons (30 of 45)glycemic control has been better with in-sulin combination therapy regimens thanwith insulin alone, the difference may beat least partly explained by how the insu-lin dose has been decreased during insu-lin combination therapy. All glitazoneshave improved glycemic control whenadded to previous insulin treatment (Ta-ble 2). In a study directly comparing theinsulin-sparing effects of troglitazone(600 mg/day) and metformin (1,700 mg/day), troglitazone had a greater (253%)insulin-sparing effect than metformin(231%). This was explained by insulin-sensitizing effects of troglitazone but notmetformin (6).

WEIGHT GAIN

What determines weight gain duringinsulin therapy?Although most patients with type 2 dia-betes are overweight, weight loss pre-cedes the diagnosis of type 2 diabetes (7).This weight loss is due to hyperglycemia-induced wasting of energy, as glucosuriaand as energy used to overproduce glu-cose (8). When glucose control is im-proved with insulin and/or sulfonylureas,energy loss in the urine decreases orceases, weight increases, and basal meta-bolic rate (kJ/min) (8–10) and dietary in-take (11) remain unchanged. The increasein body weight increases basal metabolicrate, but this is counterbalanced by im-proved glycemic control, which decreasesbasal metabolic rate because less energy is

● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●

From the 1Department of Medicine, Division of Diabetes, University of Helsinki, Helsinki, Finland.Address correspondence and reprint requests to Hannele Yki-Jarvinen, MD, FRCP, Department of Med-

icine, Division of Diabetes, P.O. Box 340, 00029 HUS, Helsinki, Finland. E-mail: [email protected] for publication 19 October 2000 and accepted in revised form 2 January 2001.Abbreviations: GADA, GAD antibodies; UKPDS, U.K. Prospective Diabetes Study.A table elsewhere in this issue shows conventional and Systeme International (SI) units and conversion

factors for many substances.

R e v i e w s / C o m m e n t a r i e s / P o s i t i o n S t a t e m e n t sR E V I E W A R T I C L E

758 DIABETES CARE, VOLUME 24, NUMBER 4, APRIL 2001

needed for glucose overproduction. Be-cause dietary intake remains unchanged(11), weight gain is proportional to re-duction of glucosuria and can indeed bepredicted based on fasting glucose con-centrations (11). Because glucosuriaappears when the fasting glucose concen-tration exceeds 10 –12 mmol/l, weightgain is inevitable if insulin therapy is post-poned until significant glucosuria occurs.In our experience, a 5-mmol/l (90-mg/dl)decrease in fasting glucose or a decreasein HbA1c by 2.5% from a baseline of 15mmol/l (270 mg/dl) is associated with a5-kg weight gain during 1 year (or 2kg/1% decrease in HbA1c) (11). Thus, themain predictors of weight gain are initialglycemia and its response to treatment(11). The patient with poor glycemic con-trol before initiation of insulin therapybut with a good treatment response is atgreatest risk for weight gain.

Choice of oral agent and weight gainin insulin-naıve patients.Only one trial has compared thecombination of insulin with that of in-sulin and metformin alone in previouslyinsulin-naıve patients (12). In thisstudy, which lasted 12 months, the bed-time insulin-metformin regimen wassuperior to three other bedtime insulinregimens with respect to glycemic con-trol, weight gain, and hypoglycemias(Table 1) (12). The ability of metforminto counteract weight gain and improveglycemia, when combined with insulin,has been confirmed in abstract reports(13,14). In these studies, weight gainwas less despite comparable glycemia(13) or weight gain was similar despitebetter glycemic control (14) in patientsusing metformin and insulin comparedwith those using insulin and sulfonyl-ureas or insulin alone. Data are heteroge-nous regarding the ability of metformin toinfluence weight gain when combinedwith both insulin and sulfonylureas, com-pared with regimens containing insulinalone (Table 1). In two comparisons,weight gain was less with the combinationregimen than with insulin alone (15,16),whereas two other comparisons revealedno difference (12,15) (Table 1, Fig. 1).The ability of metformin to counteractweight gain during insulin combinationtherapy has been attributed to a decreasein dietary intake (11).T

able

1—St

udie

sco

mpa

ring

com

bina

tion

trea

tmen

tre

gim

ens

wit

hin

suli

nto

insu

lin

alon

ein

insu

lin-

naiv

ety

pe2

diab

etic

pati

ents

Ref

eren

ceno

.C

ombi

nati

onre

gim

enD

urat

ion

(mon

ths)

End

HbA

1c*

Gly

cem

iaPl

aceb

oco

ntro

lC

ross

over

/pa

ralle

lW

eigh

tga

inH

ypog

ly-

cem

iaD

iffer

ence

inin

sulin

dose

(%)†

Met

form

inal

one

12M

ET

1be

dtim

eN

PH12

7.2%

Com

bBe

tter

wit

hM

ET

‡Ye

sPa

ralle

lLe

ssw

ith

ME

T‡

No

diff

eren

ce2

32M

etfo

rmin

and

sulfo

nylu

reas

12G

LYB

1M

ET

1be

dtim

eN

PH12

7.6%

Com

bN

odi

ffer

ence

Yes

Para

llel

No

diff

eren

ceN

odi

ffer

ence

262

15G

LYB

1M

ET

1m

orni

ngN

PH3

7.7%

Com

bN

odi

ffer

ence

No

Para

llel

No

diff

eren

ceN

odi

ffer

ence

258

15G

LYB

1M

ET

1be

dtim

eN

PH3

8.0%

Com

bN

odi

ffer

ence

No

Para

llel

Less

wit

hM

ET

No

diff

eren

ce2

4416

GLY

B1

ME

T1

bedt

ime

NPH

68.

4%In

sN

odi

ffer

ence

No

Para

llel

Less

wit

hM

ET

No

diff

eren

ce2

74W

eigh

ted

mea

n2

62Su

lfony

lure

are

gim

ens

25G

LIM

EP

1be

dtim

e30

/70

67.

6%C

omb

No

diff

eren

ceYe

sPa

ralle

lN

odi

ffer

ence

Mor

ew

ith

GLY

B2

3712

GLY

B1

bedt

ime

NPH

127.

8%C

omb

No

diff

eren

ceYe

sPa

ralle

lN

odi

ffer

ence

No

diff

eren

ce2

5529

GLY

B1

bedt

ime

NPH

68.

1%C

omb

No

diff

eren

ceN

oPa

ralle

lN

odi

ffer

ence

No

diff

eren

ce2

3829

GLY

B1

mor

ning

NPH

68.

2%In

sN

odi

ffer

ence

No

Para

llel

No

diff

eren

ceN

odi

ffer

ence

233

55G

LYB

1In

s6

8.4%

Com

bN

odi

ffer

ence

Yes

Para

llel

No

diff

eren

ceN

D2

4332

GLY

B1

lispr

ot.

i.d.

28.

4%C

omb

No

diff

eren

ceN

oPa

ralle

lN

odi

ffer

ence

No

diff

eren

ce2

3632

GLY

B1

bedt

ime

NPH

28.

5%In

sN

odi

ffer

ence

No

Para

llel

Less

wit

hG

LYB

No

diff

eren

ce2

5628

GLY

B1

Ins

48.

8%C

omb

Bett

erw

ith

GLY

BYe

sC

ross

over

No

diff

eren

ce—

250

27G

LYB

1In

s4

9.8%

Com

bBe

tter

wit

hG

LYB

Yes

Cro

ssov

erLe

ssw

ith

Ins

—2

2156

GLI

CL

1In

s12

11.8

%In

s§Be

tter

wit

hG

LIC

LN

oPa

ralle

l—

—2

35W

eigh

ted

mea

n2

42

The

tria

lsar

egr

oupe

dac

cord

ing

toth

eor

alag

entu

sed

and

then

rank

edw

ithi

nth

ese

grou

psba

sed

ongl

ycem

icco

ntro

latt

heen

dof

trea

tmen

twit

hth

ebe

tter

regi

men

.Onl

ytr

ials

last

ing

2m

onth

sor

long

erar

ein

clud

ed;

*HbA

1c

atth

een

dof

trea

tmen

tin

the

grou

pw

ith

bett

erco

ntro

l(e

ven

ifno

tsi

gnifi

cant

lybe

tter

inon

egr

oup

vers

usth

eot

her)

;†%

diff

eren

cein

insu

lindo

ses

atth

een

dof

trea

tmen

tw

ith

aco

mbi

nati

onre

gim

enve

rsus

insu

linal

one;

‡sig

nific

antd

iffer

ence

;§H

bA1

Com

b,co

mbi

nati

onre

gim

en;G

LIC

L,gl

icla

zide

;GLI

ME

P,gl

imep

irid

e;G

LYB,

glyb

urid

e;In

s,re

gim

enco

ntai

ning

insu

linal

one;

ME

T,

met

form

in;3

0/70

5an

insu

linm

ixtu

reco

ntai

ning

30%

regu

lar

insu

linan

d70

%N

PH.

Yki-Jarvinen

DIABETES CARE, VOLUME 24, NUMBER 4, APRIL 2001 759

Tab

le2—

Stud

ies

com

pari

ngco

mbi

nati

ontr

eatm

ent

regi

men

sw

ith

insu

lin

toin

suli

nal

one

inpr

evio

usly

insu

lin-

trea

ted

type

2di

abet

icpa

tien

ts

Ref

eren

ceno

.C

ombi

nati

onre

gim

enD

urat

ion

(mon

ths)

End

HbA

1c*

orH

bAI

Plac

ebo

Para

llel/

cros

sove

rG

lyce

mia

Wei

ght

gain

Hyp

ogly

cem

ias

Diff

eren

cein

insu

lindo

se(%

)†

Met

form

inre

gim

ens

24M

ET

1in

sulin

66.

5%C

omb

Yes

Para

llel

Bett

er†

wit

hM

ET

Less

wit

hM

ET

‡Le

ssw

ith

ME

T‡

223

42M

ET

1in

sulin

47.

8%C

omb

No

Para

llel

Bett

erw

ith

ME

TLe

ssw

ith

ME

T—

226

43M

ET

1in

sulin

37.

8%C

omb

Yes

Cro

ssov

erBe

tter

wit

hM

ET

No

diff

eren

ce—

23

57M

ET

1in

sulin

69.

8%C

omb

Yes

Para

llel

Bett

erw

ith

ME

TN

D—

220

Wei

ghte

dm

ean

219

Sulfo

nylu

rea

regi

men

s45

GLY

B1

insu

lin3

6.0%

Com

bN

oC

ross

over

No

diff

eren

ceN

odi

ffer

ence

—2

2558

GLY

B1

insu

lin3

7.0%

Com

bYe

sPa

ralle

lBe

tter

wit

hG

LYB

No

diff

eren

ce—

220

46SU

1in

sulin

127.

5%C

omb

No

Para

llel

No

diff

eren

ceN

odi

ffer

ence

—2

3526

GLY

B1

insu

lin4

8.3%

Com

b§Ye

sC

ross

over

Bett

erw

ith

GLY

BN

odi

ffer

ence

Mor

ew

ith

GLY

220

47T

OLA

Z1

insu

lin3

8.8%

Com

b§Ye

sC

ross

over

Bett

erw

ith

GLY

BFi

xed

—2

2359

GLY

B2

insu

lin12

8.8%

Com

b§Ye

sPa

ralle

lBe

tter

wit

hG

LYB

——

247

60G

LYB

1in

sulin

39.

1%C

omb

Yes

Cro

ssov

erBe

tter

wit

hG

LYB

——

27

61G

LYB

1in

sulin

49.

6%C

omb

Yes

Para

llel

Bett

erw

ith

GLY

B—

—Fi

xed

62G

LIP

1in

sulin

39.

8%C

omb§

Yes

Cro

ssov

erN

odi

ffer

ence

——

235

49G

LYB

1in

sulin

410

.2C

omb§

Yes

Cro

ssov

erBe

tter

wit

hG

LYB

No

diff

eren

ce—

23

48G

LYB

1in

sulin

1110

.3%

Ins§

Yes

Cro

ssov

erBe

tter

wit

hG

LYB

No

diff

eren

ce—

27

50G

LYB

insu

lin2

11.0

%C

omb§

Yes

Cro

ssov

erN

odi

ffer

ence

No

diff

eren

ce—

22

44G

LYB

1in

sulin

212

.4%

Com

b§Ye

sC

ross

over

Bett

erw

ith

GLY

BN

odi

ffer

ence

—Fi

xed

63T

OLA

Z1

insu

lin2

12.6

%C

omb§

Yes

Cro

ssov

erN

odi

ffer

ence

——

Fixe

d64

GLY

B1

insu

lin12

12.9

%In

sYe

sC

ross

over

No

diff

eren

ceN

odi

ffer

ence

—2

2251

GLY

B1

insu

lin2

13.0

%C

omb§

Yes

Cro

ssov

erBe

tter

wit

hG

LYB

No

diff

eren

ce—

60

Wei

ghte

dm

ean

221

\G

litaz

one

54R

OSI

1in

sulin

67.

8%Ye

sPa

ralle

lBe

tter

wit

hR

OSI

Mor

ew

ith

RO

SIM

ore

wit

hR

OSI

60

52T

RO

1in

sulin

67.

9%Ye

sPa

ralle

lBe

tter

wit

hT

RO

Mor

ew

ith

TR

OM

ore

wit

hT

RO

246

53PI

O1

insu

lin4

8.6%

Yes

Para

llel

Bett

erw

ith

PIO

Mor

ew

ith

PIO

Mor

ew

ith

PIO

—a

-Glu

cosi

dase

inhi

bito

r65

AC

AR

B1

insu

lin6

8.3%

Yes

Para

llel

Bett

erw

ith

AC

AR

B—

No

diff

eren

ceFi

xed

66A

CA

RB

1in

sulin

127.

3%Ye

sPa

ralle

lBe

tter

wit

hA

CA

RB

No

diff

eren

ceN

odi

ffer

ence

—

The

tria

lsar

egr

oupe

dac

cord

ing

toth

eor

alag

entu

sed

and

then

rank

edw

ithi

nth

ese

grou

psba

sed

ongl

ycem

icco

ntro

latt

heen

dof

trea

tmen

twit

hth

ebe

tter

regi

men

.Onl

ytr

ials

last

ing

2m

onth

sor

long

erar

ein

clud

ed.

*HbA

1c

atth

een

dof

trea

tmen

tin

the

grou

pw

ith

bett

erco

ntro

l(e

ven

ifno

tsi

gnifi

cant

lybe

tter

inon

egr

oup

vers

usth

eot

her)

;†%

diff

eren

cein

insu

lindo

ses

atth

een

dof

trea

tmen

tw

ith

aco

mbi

nati

onre

gim

enve

rsus

insu

linal

one;

‡sta

tist

ical

lysi

gnifi

cant

diff

eren

cebe

twee

nin

sulin

com

bina

tion

ther

apy

vers

usin

sulin

alon

e;§H

bA1

refe

renc

era

nge

high

erth

anth

atfo

rH

bA1

c;\t

rial

sin

whi

chth

ein

sulin

dose

was

fixed

are

not

incl

uded

inth

eca

lcul

atio

nof

the

wei

ghte

dm

ean.

AC

AR

B,ac

arbo

se;

Com

b,co

mbi

nati

onre

gim

en;

GLI

P,gl

ipiz

ide;

Irs,

regi

men

cont

aini

ngin

sulin

;PI

O,

piog

litaz

one;

RO

SI,

rosi

glit

azon

e;SU

,var

ious

sulfo

nylu

reas

;TO

LAZ

,tol

azam

ide;

TR

O,t

rogl

itaz

one.

Combination therapies with insulin in type 2 diabetes


Choice of oral agent and weight gainin previously insulin-treated patients.Switching patients from treatment withinsulin alone to insulin combinationtherapy with metformin has been asso-ciated with less weight gain in two ofthree studies, whereas no difference wasfound in any of the 16 comparisons inwhich insulin plus sulfonylurea wascompared with insulin alone, despitebetter control in 10 of 16 comparisons(Table 2). It is unclear whether this isbecause weight was not accurately re-corded or because the larger dose of ex-ogenous insulin or the greater numberof insulin injections used in insulinalone as compared with the insulincombination regimen had independentweight-promoting effects. In studiescomparing insulin-glitazone treatmentwith insulin alone, glycemic control wasbetter in each study with the insulin-glitazone combination than with insulinalone. Better glycemic control was alsoassociated with greater weight gain ineach of the three studies. Although theamount of weight gain relative to theimprovement in glycemic controlseemed slightly greater than with sulfo-nylureas (Fig. 1), data on insulin-glitazone combination therapy are stillsparse. The significance of weight gainwith glitazones is also difficult to judge,because glitazones may be beneficial inredistributing fat from visceral to sub-

cutaneous sites (17–19). A small frac-tion of weight gain with glitazonescould be due to peripheral edema(20,21).

HYPOGLYCEMIAS

Frequency of hypoglycemias in type1 versus type 2 diabetes.In both patients with type 1 (22) and type 2(12) diabetes, the frequency of hypoglyce-mias is inversely proportional to glycemiccontrol. In the Diabetes Control and Com-plications Trial, in patients with HbA1c be-tween 7 and 8%, the frequency of severehypoglycemias requiring assistance in theprovision of treatment was 0.62 per patientper year (22). In contrast, in the FINFATstudy (12), the Kumamoto study (23), orother studies of type 2 diabetes in whichcomparable glycemic control was achieved(24,25), there were no severe hypoglyce-mias. The frequency of biochemical hypo-glycemias (blood glucose ,3.5 mmol/l)was 1.9 per patient per year in patientstreated with insulin plus metformin and ap-proximately twice as high in the othergroups in the FINFAT study (12). In thelatter study, HbA1c averaged between 7and 8% in all groups for 1 year. Thesedata, although derived from separatestudies, suggest that hypoglycemias aremuch less of a problem in type 2 diabeticpatients than in type 1 diabetic patients.

Does the oral agent influence theoccurrence of hypoglycemiasindependent of glycemic control?The occurrence of hypoglycemia has beensparsely reported [eight comparisons ininsulin-naıve patients (Table 1) and fivecomparisons in previously insulin-treatedpatients (Table 2)]. In insulin-naıve pa-tients, use of insulin combination therapywith metformin has been associated withless hypoglycemias than with insulinalone, despite better glycemic controlwith the insulin-combination regimen(12). No difference was observed betweeninsulin-alone and insulin-sulfonylurearegimens in five of seven studies; in twostudies (25,26), there were more cases ofhypoglycemia with insulin and sulfonyl-urea than with insulin alone (Tables 1 and2). No difference in the incidence of hy-poglycemia was observed between insulinalone compared with insulin plus sulfo-nylurea and metformin regimens (Tables1 and 2). In the latter studies, there wasalso no difference in glycemic control. Inall three studies comparing insulin-glitazone combination therapy to insulinalone, the frequency of hypoglycemia washigher and glycemic control was betterwith the insulin-combination regimen.These data suggest that with the possibleexception of metformin, use of insulincombination therapy is accompanied by asimilar frequency of hypoglycemia than isuse of insulin alone.

CHANGES IN SERUMTRIGLYCERIDES ANDOTHER LIPIDS ANDLIPOPROTEINS

Insulin-naıve patients.Data on changes in serum triglyceridesand glycemia in insulin-naıve patients aresummarized in Table 3. As judged fromthe weighted means of insulin-alone reg-imens, a decrease in HbA1c from ;10 to8% (i.e., by 2%) is associated with a 0.7-to 0.8-mmol/l decrease in serum triglyc-erides from an initial concentration of2.4–2.7 mmol/l. With insulin combina-tion therapy regimens, a decrease ofHbA1c by 2% decreases serum triglycer-ides by 0.4–0.6 mmol/l (Table 3). In allexcept one study, insulin alone loweredserum triglycerides slightly more thaninsulin combination therapy, althoughthere was no significant difference in thelowering of serum triglycerides with thetwo modes of therapy in any of the studies

Figure 1—Weight gain in previously insulin-treated patients during treatment with insulin com-bination regimens containing metformin (MET) (24,42,43), various sulfonylureas (SU) (26,44–51), and glitazones (GLIT) (52–54) and in insulin-naıve patients treated with insulin and MET(12), SU1MET (12,15,16), and SU (12,25,27–29).

Yki-Jarvinen


(Table 3). LDL and HDL cholesterol con-centrations remained unchanged in allstudies, with no differences between reg-imens (12,15,16,25,27–29).

Previously insulin-treated patients.As summarized in Table 4, the greater im-provement in glycemic control with insu-lin combination therapy than with insulinalone in 11 of 14 studies has not beenconsistently (4 of 11 studies) associatedwith a greater decrease in serum triglyc-erides. These data demonstrate that factorsother than average glucose concentrationsdetermine the degree of lowering of se-rum triglycerides. Overall, the availablecomparisons of changes in serum lipidand lipoprotein concentrations in bothinsulin-naıve and previously insulin-treated patients do not allow definitiveconclusions and do not support choice ofone treatment regimen over another.

BLOOD PRESSURERegarding blood pressure, in a follow-upstudy of the patients participating in theFINMIS study (15,30), blood pressure in-creased significantly in the entire group of100 patients during 1 year. Weight gaincorrelated both with the increase in bloodpressure and with an increase in the LDL

cholesterol concentrations (30). Threeshorter studies reported data on bloodpressure (15,25,27) but found nochanges in blood pressure or differencesbetween regimens.

SPECIAL QUESTIONS

Choice of insulin regimen duringinsulin combination therapy:NPH insulin or insulin glargine?Regarding basal insulinization, the com-monly used intermediate-acting insulin(NPH) is not ideal for once-daily use. Inthe FINMIS study, in patients with type 2diabetes with a mean BMI of 29 kg/m2,injection of NPH insulin at 9:00 P.M. re-sulted in maximal glucose lowering be-tween 4:00 and 8:00 A.M., but the effectwas gone by 3:00 P.M., i.e., 18 h after theinjection, and dinnertime glucose con-centrations were unnecessarily high. Therecently approved long-acting insulin an-alog insulin glargine seems to overcomethese problems. In a study comparingNPH plus oral agents to insulin glargineplus oral agents in 423 insulin-naıve type2 diabetic patients for 1 year, all hypogly-cemias were 35% lower and nocturnal hy-poglycemias were 56% lower with insulinglargine than with NPH (Fig. 2). Dinner-

time glucose levels were also significantlylower with insulin glargine than withNPH (Fig. 2).

Regular insulin or short-actinginsulin analogs compared with NPHduring combination therapy.Regular insulin three times per day plus asulfonylurea has been compared with asingle injection of NPH taken at bedtimeand a sulfonylurea. No difference in gly-cemic control was found, but weight gainwas significantly greater with three injec-tions of regular insulin than with a singleinjection of bedtime NPH insulin (31).Greater weight gain with no difference inglycemic control has also been reportedwith three injections of lispro plus sulfo-nylurea compared with NPH plus sulfo-nylurea (32) (Table 5).

Timing of the intermediate-actinginsulin injection.The pros and cons of timing of the inter-mediate-acting insulin injection has beenexamined in three studies (15,33,34). Intwo studies, a bedtime injection was rec-ommended because it resulted in lessweight gain (15) or less hypoglycemias(34) than a morning injection. In the third

Table 3—Changes in glycosylated hemoglobin and serum triglycerides during treatment with insulin alone as compared with combinationtherapies with insulin and oral agents in insulin-naive patients

Ref.no.

Combinationregimen

Duration(months)

BaselineHbA1c

Ins

Changein HbA1c

Ins

BaselineHbA1c

Comb

Changein HbA1c

Comb

BaselineS-TgIns

Changein S-Tg

Ins

BaselineS-Tg

Comb

Changein S-TgComb

Better†regimen

S-Tg

Better†regimenglycemia

Metformin alone12 MET 1 bedtime NPH 12 9.9 22.0‡ 9.8 22.5‡ 2.6 20.9‡ 2.4 20.7‡ NS Comb

Metformin and sulfonylureas12 GLYB 1 MET 1 bedtime NPH 12 9.9 22.0‡ 9.9 22.1‡ 2.6 20.9‡ 2.3 20.4‡ NS NS15 GLYB 1 MET 1 morning NPH 3 9.6 21.6‡ 9.5 21.7‡ 2.4 20.6‡ 2.6 20.3 NS NS15 GLYB 1 MET 1 bedtime NPH 3 9.6 21.6‡ 9.9 21.9‡ 2.4 20.6‡ 2.5 20.6‡ NS NS16 GLYB 6 MET 1 bedtime NPH 6 10.7 22.3‡ 10.2 21.5‡ 1.8 20.4‡ 2.1 20.2 NS NS

Weighted mean 9.9 21.9 9.9 21.9 2.4 20.7 2.4 20.4

Sulfonylurea regimens25 GLIMEP 1 bedtime 30/70 6 9.9 22.0‡ 9.7 22.1‡ 3.1 21.0‡ 3.2 20.3‡ NS NS12 GLYB 1 bedtime NPH 12 9.9 22.0‡ 9.8 22.0‡ 2.6 20.8‡ 2.7 20.8‡ NS NS29 GLYB 1 bedtime NPH 6 11.2 23.0‡ 10.5 22.3‡ 2.4 20.7‡ 2.0 20.3‡ NS NS29 GLYB 1 morning NPH 6 11.2 23.0‡ 11.1 22.6‡ 2.4 20.7‡ 2.2 20.3 NS NS55 GLYB 1 insulin 6 11.5* 22.2‡ 10.4* 22.2‡ 2.0 20.4 2.2 20.3 NS NS28 GLYB 1 insulin 4 9.7* 20.8 10.1* 21.3 3.1 20.6 2.8 20.8‡ Comb Comb27 GLYB 1 insulin 4 10.4* 0.2 10.6* 20.8‡ 3.6 21.1‡ 3.6 21.2‡ NS Comb

Weighted mean 10.5 22.2 10.2 22.1 2.7 20.8 2.7 20.6

The trials are grouped according to the oral agent used and then ranked within these groups based on glycemic control at the end of treatment with the better regimen.Only trials lasting at least 2 months are included. *HbA1 value; †denotes a statistically significant difference between insulin combination therapy versus insulinalone; ‡significant difference at the end versus start of the treatment period. Comb, combination regimen; Ins, regimen containing insulin alone; S-Tg, serumtriglycerides (mmol/l). For other abbreviations, see Table 1.



study, no differences in glycemia orweight gain were found (33).

PREDICTION OF INSULINREQUIREMENTSVariation in hepatic insulin sensitivityseems to be much more important thaninsulin absorption in determining insulinrequirements during combination ther-apy with NPH insulin (3). Hepatic insulinsensitivity cannot be routinely measuredbut correlates with various indexes ofobesity (3). In type 2 diabetic patientswith a mean BMI of 29 kg/m2, to achievean average HbA1c of ;7.5% from a base-line value of 10%, the mean bedtime NPHinsulin dose for body weights of 70, 80,90, and 100 kg has been 0.2, 0.3, 0.4, and0.5 IU/kg (3,11). However, interindi-vidual variation is large and has varied20-fold between 8 and 168 IU per day(12), which implies that these averagepredictions are not accurate enough to beused on an individual level.

Autoantibodies to glutamic acid de-

carboxylase (GADA) predict an increasedlikelihood of insulin requirement in bothyoung and old adults with type 2 diabetes(35). In 3,672 newly diagnosed patientsin the UKPDS, 34% of those aged 25–34years and 7% of those aged 55–65 yearshad GADA (35). Among patients olderthan 55 years at diagnosis, 34% of thosewith GADA and 5% with autoantibodiesto neither GADA nor islet cell cytoplasmrequired insulin therapy. In these olderpatients, only the presence of GADA butnot phenotypic features such as BMI pre-dicted insulin requirement. There are nostudies comparing insulin combinationregimens with insulin alone in these pa-tients, who are often classified as havingtype 2 diabetes but actually have type 1diabetes (36). In patients in whom signsof absolute insulin deficiency (rapidweight loss, ketonuria) ultimately de-velop, the presence of GADA may guidethe choice of a basal-bolus–type full insu-lin-replacement regimen (37).

PREDICTORS OF AGLYCEMIC RESPONSE TOINSULIN COMBINATIONTHERAPYIn studies in which the insulin dose is ti-trated aggressively to reach glycemic tar-gets, the decrease in HbA1c will be directlyproportional to its initial level. Of otherfactors, obesity predicts a poor responseto any type of insulin therapy, especially ifinsufficient doses of insulin are used(30,38). In addition, and as discussedabove, GADA may predict poor responseto combination therapy.

PRACTICAL ALGORITHM TOINITIATE INSULIN THERAPYInitiation of insulin therapy on an ambu-latory basis in type 2 diabetic patients hasbeen shown to be as safe and effective asan inpatient program (39). Regardless ofthe insulin treatment regimen chosen, theinsulin dose should be adjusted to reachglycemic targets. Considering the largeinterindividual variation in insulin re-

Table 4—Changes in glycosylated hemoglobin and serum triglycerides during treatment with insulin alone as compared with insulin combi-nation therapy in previously insulin-treated patients

Ref.no.

Combinationregimen

Duration(months)

BaselineHbA1c

Ins

Changein HbA1c

Ins

BaselineHbA1c

Comb

Changein HbA1c

Comb

BaselineS-TgIns

Changein S-Tg

Ins

BaselineS-Tg

Comb

Changein S-TgComb

Better†regimen

S-Tg

Better†regimenglycemia

Metformin24 MET 1 insulin 6 9 21.6§ 9.1 22.5§ 2.5 20.4 2.3 20.1 NS Comb42 MET 1 insulin 4 9.6 0.0 9.6 21.9§ 2.4 20.1 2.0 20.4§ Comb Comb43 MET 1 insulin 6 11.5 20.2 11.7 21.9 2.8 20.0 2.9 20.3§ Comb Comb57 MET 1 insulin 3 8.9 20.5§ 8.9 21.1§ 2.2 21.0§ 2.2 20.9§ NS Comb

Weighted mean 9.8 20.6 10.0 21.9 2.5 20.4 2.4 20.4

Sulfonylurea45 GLYB 1 insulin 3 6.7 20.4 6.3 20.3 1.5 20.08 1.5 0.2 NS NS46 SU 1 insulin 12 10.2 22.4§ 9.8 22.3§ 2.3 20.6 2.5 20.8 NS NS26 GLYB 1 insulin 4 9.2* 20.1 9.2* 20.9§ 1.2 0.1 1.2 0.1§ Ins Comb47 TOLAZ 1 insulin 3 10.7* 21.5§ 10.7* 21.9 2.1 0.0 2.1 20.5§ Comb Comb48 GLYB 1 insulin 11 10.3 21.3§ 11.1 22.0§ 1.7 0.1 2.4 20.7 NS NS61 GLYB 1 insulin 4 10.4 0.0 10.9 21.3§ 1.4 20.1 1.8 20.1 NS Comb51 GLYB 1 insulin 2 14.0* 20.6 14.0* 21.0 2.1 20.4 2.1 20.2 NS Comb

Weighted mean 10.1 21.0 10.1 21.4 1.8 20.1 2.0 20.2

Glitazones52 TRO 1 insulin 6 8.9 0.1 9.0 21.2 2.6 20.5 2.5 20.1 NS Comb54 ROSI 1 insulin 6 9.4 20.1 9.3 20.4 3.0 20.3 2.7 20.4 NS Comb

Weighted mean 9.2 20.0 9.2 21.3 2.8 0.1 2.6 20.2

a-Glucosidase inhibitor65 ACARB 1 insulin 6 8.7 0.1 8.8 20.6§ 2.1† — 2.2‡ — Comb Comb

The trials are grouped as in Table 3. Only trials lasting at least 2 months are included. *HbA1; †statistically significant difference between insulin combination therapyversus insulin alone; ‡serum triglycerides 120 min after a standardized meal challenge; §significant difference at the end versus start of the treatment period. Therewere no significant differences between changes in serum triglycerides with insulin alone versus insulin combination therapy. Abbreviations as in Tables 2 and 3.

Yki-Jarvinen


quirements, it is difficult to define the cor-rect insulin dose by performing doseadjustments only at outpatient visits, un-less these are very frequent. In our expe-rience of treating insulin-naıve patients(12,15), the best method of defining theinsulin dose is teaching the patient to self-adjust the dose based on results of homeglucose monitoring. This is easiest to per-form if the dose adjustment is based on

measurement of fasting plasma glucoseonly. Fasting plasma glucose is not in-fluenced by size, composition, or rate ofabsorption of meals as much as by post-prandial glucose levels, and its measure-ment does not interfere with daily activi-ties. The maximal action of NPH given atbedtime is exerted on fasting glucose,which, therefore, is a particularly suitabletarget for titration of the dose when insu-

lin combination therapy with NPH or in-sulin glargine is used. However, becauseespecially NPH insulin is unable to ade-quately control postdinner glycemia, fast-ing glucose must be in the normal range(4–6 mmol/l) for average glycemic con-trol, measured using HbA1c to be ,7.5%.A simple method of initiating insulin ther-apy, developed based on experience fromthe FINFAT study, is shown in Table 5 (12).The patient is assumed to be insulin-naıveand on maximal doses of sulfonylureasand metformin. The recommendation todiscontinue the sulfonylurea (glyburide)but not metformin after insulin combina-tion therapy is started is based on the in-ability of some patients to adequatelytitrate the dose of bedtime NPH becauseof hypoglycemia (12). An increase in mildhypoglycemias was also reported by Rid-dle and Schneider (25) with glimepiridecombined with a single injection of 30/70insulin at 6:00 P.M. compared with two in-jections of 30/70 insulin. Hypoglycemiasmay not be a problem with peakless insu-lins such as insulin glargine (40). Discon-tinuation of sulfonylurea when insulintherapy is started may retard achievementof good glycemic control unless the in-sulin dose is rapidly increased (12,25).Glitazones could be an additional or al-ternative component in the oral hypogly-cemic agent regimen, but there are nostudies in insulin-naıve patients.

CONCLUDING REMARKSAgainst the emerging epidemic of type 2 di-abetes, studies comparing different insulintreatment regimens are sparse and includeonly a small number of patients treated for amaximum of 1 year (Tables 1 and 2). Dataon effects of insulin-combination therapyversus insulin alone on diabetic microvas-cular and macrovascular complications arenonexistent. The main reason for the pau-city of data may be the reluctance of privatefunding agencies to support studies usingpharmacological agents and the reluctanceof industry to support studies with estab-lished preparations. The developmentof new agents such as glitazones andinsulin analogs have increased the num-ber of patients included in various trials,but many company-initiated trials are de-signed to fulfill licensing requirementsand must be performed in multiple cen-ters to save time. Although some compa-ny-initiated trials are of superb quality,others suffer from inadequate glycemiccontrol and may lack the comparisons the

Figure 2—Upper panel: Diurnal glucose profiles after 52 weeks of treatment of 423 patientsusing oral hypoglycemic agents with either insulin glargine (F) or NPH (E). Lower panel: Thepercentage of patients experiencing any symptomatic hypoglycemia (ALL) or nocturnal hypogly-cemia (NOCTURNAL) in the same study (40).



clinicians would be interested in. Despitethese deficiencies, some conclusions re-garding the role of insulin combinationtherapy in the treatment of type 2 diabeticpatients seem justified.

No study reported worse glycemiccontrol with insulin combination therapythan with insulin alone. Glycemic controlwas better with insulin combination ther-apy than with insulin alone in most stud-ies of previously insulin-treated patients,but this could be explained by a smallerdifference (;20% for metformin or sulfo-nylureas) (Table 2) in the insulin dose be-tween the two modes of treatment than in

studies performed in insulin-naıve pa-tients (30–40%, Table 1). Combinationregimens allow use of less insulin injec-tions, which may ease titration of the in-sulin dose and compliance (12,15,41).These benefits must be balanced againstthe side effects of oral drugs and, in somecountries, their cost. Abnormal renal orliver function also limits the use of manyoral agents. Weight gain seems propor-tional to the number of insulin injectionsused (12,15,31,32) and can be counter-acted by inclusion of metformin in thetreatment regimen. Metformin also seemsto reduce the incidence of hypoglycemias

(12), as does the use of the peakless long-acting insulin analog insulin glarginecompared with NPH (40). These consid-erations and the need to treat not onlyhyperglycemia but also other risk factorsin type 2 diabetes support the use of sim-ple insulin combination regimens such asinsulin glargine and metformin and or asulfonylurea (40). The prevailing viewthat patients who are poorly responsive tosuch a regimen benefit from adding addi-tional insulin injections is not supportedby existing data. Instead, special empha-sis should be placed on increasing thedose of the single long-acting insulin to a

Table 5—Studies comparing insulin combination regimens with different insulin injection regimens in type 2 diabetic patients (oral agentssimilar in all regimens)

Referenceno. Regimen 1 n Regimen 2 n Glycemia Weight gain Hypoglycemia

40 Bedtime NPH 1 OHA* 208 Bedtime glargine 1 OHA* 214 No difference No difference Less with glargine†15 Bedtime NPH 1 MET

1 SU28 Morning NPH 1 MET

1 SU32 No difference Less† with bedtime NPH No difference

34 Bedtime NPH 1 SU 15 Morning NPH 1 SU 14 No difference No difference Less with bedtimeNPH

33 Bedtime NPH 1 SU 24 Morning NPH 1 SU 24 No difference No difference No difference31 Bedtime NPH 1 SU 39 3 3 regular 1 SU 41 No difference Less with bedtime NPH No difference32 Bedtime NPH 1 SU 135 3 3 lispro 1 SU 139 No difference Less with bedtime NPH No difference

*OHA, oral hypoglycemic agents, 59% SU 1 MET, no differences in OHA between groups using NPH versus insulin glargine; †statistically significant differencebetween regimen 1 and regimen 2.

Table 6—Simple algorithm to start insulin combination therapy in an insulin-naive patient treated with oral combination therapy

Objectives Details

Visit–1, before start of insulin therapyz Teach home-glucose monitoring Home glucose monitoringz Correct gross errors in diet z Measure fasting glucose daily during first weeks or months; after

reaching target frequency can be even once a week

Visit 0, initiation of insulin therapyz Stop sulfonylurea, continue metformin 2 g/day† Initial dose of insulin (insulin glargine, NPH, or ultralente)z Teach insulin injection techniquez Define initial dose of insulin (glargine, NPH or 30/70 at

6:00 P.M. or later)z Give written instructions regarding self-adjustment of the

insulin dosez Teach symptoms of hypoglycemiasz Schedule a phone call after 1 week and visit after 2–4 weeks

Subsequent visitsz Individualize frequency—consider electronic transfer of home

glucose–monitoring

z Irrelevant if adjusted by patientz Safe starting dose 5 fasting glucose (mmol/l). i.e., 10 IU if fasting

glucose is 10 mmol/lSelf-adjustment of insulin dosesz If fasting glucose exceeds 5.5 mmol/l (100 mg/dl) on three consec-

utive measurements, increase bedtime insulin dose by 2 IUz During combination therapy with NPH and oral agents

(ref. FINFAT), or fasting glucose of #6 mmol/l corresponds to#7.5% HbA1c

z Results and phone calls instead of outpatient visits

*There are no data on use of glitazones in combination therapy with insulin in insulin-naive patients; †based on the FINFAT study, in which glyburide and NPHinsulin were used and use of this combination prevented adequate titration of the insulin dose (12); a higher incidence of symptoms of mild hypoglycemia was foundusing glimepiride combined with 30/70 insulin given at 6:00 P.M. Similar problems were not reported in another study in which glimepiride was combined with30–70 insulin at 6:00 P.M. (25) and may not be a problem with insulin glargine (40). Note that stopping a sulfonylurea necessitates a rapid increase in the insulindose, which can be performed by teaching the patient self-adjustment of the insulin dose.

Yki-Jarvinen


dose that normalizes the fasting glucoseconcentration.

References1. U.K. Prospective Diabetes Study (UKPDS)

Group: Intensive blood-glucose controlwith sulphonylureas or insulin comparedwith conventional treatment and risk ofcomplications in patients with type 2 di-abetes (UKPDS 33). Lancet 352:837–852,1998

2. Stratton IM, Adler AI, Neil HA, MatthewsDR, Manley SE, Cull CA, Hadden D,Turner RC, Holman RR: Association ofglycaemia with macrovascular and micro-vascular complications of type 2 diabetes(UKPDS 35): prospective observationalstudy. BMJ 321:405–412, 2000

3. Ryysy L, Hakkinen AM, Goto T, Veh-kavaara S, Westerbacka J, Halavaara J,Yki-Jarvinen H: Hepatic fat content andinsulin action on free fatty acids and glu-cose metabolism rather than insulin ab-sorption are associated with insulin re-quirements during insulin therapy in type2 diabetic patients. Diabetes 49:749–758,2000

4. Peters AL, Davidson MB: Insulin plus asulfonylurea agent for treating type 2 dia-betes. Ann Intern Med 115:45–53, 1991

5. Pugh JA, Wagner ML, Sawyer J, RamirezG, Tuley M, Friedberg SJ: Is combinationsulfonylurea and insulin therapy useful inNIDDM patients? Diabetes Care 15:953–959, 1992

6. Yu JG, Kruszynska YT, Mulford MI, Olef-sky JM: A comparison of troglitazone andmetformin on insulin requirements in eu-glycemic intensively insulin-treated type2 diabetic patients. Diabetes 48:2414–2421, 1999

7. Knowler WC, Pettitt DJ, Savage RJ, Ben-nett PH: Diabetes incidence in Pima Indi-ans: contributions of obesity and parentaldiabetes. Am J Epidemiol 113:144–156,1981

8. Bogardus C, Taskinen M-R, Zawadzki J,Lillioja S, Mott D, Howard BV: Increasedresting metabolic rates in obese subjectswith non-insulin-dependent diabetes melli-tus and the effect of sulfonylurea therapy.Diabetes 35:1–5, 1986

9. Welle S, Nair KS, Lockwood D: Effect of asulfonylurea and insulin on energy ex-penditure in type II diabetes mellitus.J Clin Endocrinol Metab 66:593–597, 1988

10. Franssila-Kallunki A, Groop L: Factorsassociated with basal metabolic rate in pa-tients with type 2 (non-insulin-depen-dent) diabetes mellitus. Diabetologia 35:962–966, 1992

11. Makimattila S, Nikkila K, Yki-Jarvinen H:Causes of weight gain during insulin ther-apy with and without metformin in pa-

tients with non-insulin-dependent diabe-tes mellitus. Diabetologia 42:406–412, 1999

12. Yki-Jarvinen H, Ryysy L, Nikkila K, Tulo-kas T, Vanamo R, Heikkila M: Compar-ison of bedtime insulin regimens inpatients with type 2 diabetes mellitus: arandomized, controlled trial. Ann InternMed 130:389–396, 1999

13. O’Brien SV, Mangnall M: Metformin su-perior to gliclazide when used in combi-nation with NPH insulin in type 2 diabeticpatients inadequately controlled on oralhypoglycemic therapy: median follow-up13 months (Abstract). Diabetes 48:A115,1999

14. Schernthaner G, Biesenbach G, Kacer-ovsky G, Mihaljevic R, Pieber TR: Evalu-ation of combination therapy in NIDDMpatients with sulfonylurea failure: theAustrian intervention study (Abstract).Diabetes 46 (Suppl. 1):166A, 1997

15. Yki-Jarvinen H, Kauppila M, Kujansuu E,Lahti J, Marjanen T, Niskanen L, RajalaS, Ryysy L, Salo S, Seppala P, TulokasT, Viikari J, Karjalainen J, Taskinen M-R:Comparison of insulin regimens in pa-tients with non-insulin-dependent diabe-tes mellitus. N Engl J Med 327:1426–1433, 1992

16. Chow C-C, Tsang LWW, Sorensen JP,Cockram CS: Comparison of insulin withor without continuation of oral hypogly-cemic agents in the treatment of second-ary failure in NIDDM patients. DiabetesCare 18:307–314, 1995

17. Akazawa S, Sun F, Ito M, Kawasaki E,Eguchi K: Efficacy of troglitazone on bodyfat distribution in type 2 diabetes. Diabe-tes Care 23:1067–1071, 2000

18. Kawai T, Takei I, Oguma Y, Ohashi N,Tokui M, Oguchi S, Katsukawa F, HiroseH, Shimada A, Watanabe K, Saruta T: Ef-fects of troglitazone on fat distribution inthe treatment of male type 2 diabetes. Me-tabolism 48:1102–1107, 1999

19. Mori Y, Murakawa Y, Okada K, HorikoshiH, Yokoyama J, Tajima N, Ikeda Y: Effectof troglitazone on body fat distribution intype 2 diabetic patients. Diabetes Care 22:908–912, 1999

20. Balfour JA, Plosker GL: Rosiglitazone.Drugs 57:921–930, 1999

21. Gillies PS, Dunn CJ: Pioglitazone. Drugs60:333–343, 2000

22. The Diabetes Control and ComplicationsTrial Research Group: The effect of inten-sive treatment of diabetes on the develop-ment and progression of long-termcomplications in insulin-dependent dia-betes mellitus. N Engl J Med 329:977–986,1993

23. Ohkubo Y, Kishikawa H, Araki E, MiytaT, Isami S, Motoyoshi S, Kojima Y, Fu-ruyoshi N, Shichiri M: Intensive insulintherapy prevents the progression of di-abetic microvascular complications in

Japanese patients with non-insulin-de-pendent mellitus: a randomized prospec-tive 6-year study. Diabetes Res Clin Pract28:103–117, 1995

24. Aviles-Santa L, Sinding J, Raskin P: Effectsof metformin in patients with poorly con-trolled, insulin-treated type 2 diabetesmellitus: a randomized, double-blind,placebo-controlled trial. Ann Intern Med131:182–188, 1999

25. Riddle MC, Schneider J: Beginning in-sulin treatment of obese patients withevening 70/30 insulin plus glimepirideversus insulin alone: Glimepiride Combi-nation Group. Diabetes Care 21:1052–1057, 1998

26. Stenman S, Groop P-H, Saloranta C, Tot-terman KJ, Fyhrqvist F, Groop L: Effectsof the combination of insulin and gliben-clamide in type 2 (non-insulin-depen-dent) diabetic patients with secondaryfailure to oral hypoglycaemic agents. Dia-betologia 31:206–213, 1988

27. Riddle MC, Hart JS, Bouma DJ, PhillipsonBE, Youker G: Efficacy of bedtime NPHinsulin with daytime sulfonylurea for sub-population of type II diabetic subjects. Di-abetes Care 12:623–629, 1989

28. Riddle M, Hart J, Bingham P, Garrison C,McDaniel P: Combined therapy for obesetype 2 diabetes: suppertime mixed insulinwith daytime sulfonylurea. Am J Med Sci303:151–156, 1992

29. Wolffenbuttel BH, Sels JP, Rondas-Col-bers GJ, Menheere PP, NieuwenhuijzenKA: Comparison of different insulin regi-mens in elderly patients with NIDDM. Di-abetes Care 19:1326–1332, 1996

30. Yki-Jarvinen H, Ryysy L, Kauppila M, Ku-jansuu M, Lahti J, Marjanen T, NiskanenL, Rajala S, Salo S, Seppala P, TulokasT, Viikari J, Taskinen M-R: Effect ofobesity on the response to insulin therapyin noninsulin-dependent diabetes melli-tus. J Clin Endocrinol Metab 82:4037–4043, 1997

31. Landstedt-Hallin L, Adamson U, Arner P,Bolinder J, Lins P-E: Comparison of bed-time NPH or preprandial regular insulincombined with glibenclamide in second-ary sulfonylurea failure. Diabetes Care 18:1183–1186, 1995

32. Bastyr EJ, Johnson ME, Trautmann ME,Anderson JHJ, Vignati L: Insulin lispro inthe treatment of patients with type 2 dia-betes mellitus after oral agent failure. ClinTher 21:1703–1714, 1999

33. Groop LC, Widen E, Ekstrand A, Salo-ranta C, Franssila-Kallunki A, ErikssonJGS-J: Morning or bedtime NPH insulincombined with sulfonylurea in treatmentof NIDDM. Diabetes Care 15:831–834, 1992

34. Soneru IL, Agrawal L, Murphy JC, Law-rence AM, Abraira C: Comparison of morn-ing or bedtime insulin with and withoutglyburide in secondary sulfonylurea fail-



ure. Diabetes Care 16:896–901, 199335. Turner R, Stratton I, Horton V, Manley S,

Zimmet P, Mackay IR, Shattock M, Bot-tazzo GF, Holman R: UKPDS 25: auto-antibodies to islet-cell cytoplasm and glu-tamic acid decarboxylase for prediction ofinsulin requirement in type 2 diabetes:UK Prospective Diabetes Study Group.Lancet 350:1288–1293, 1997

36. World Health Organization: Report of aWHO Consultation: Definition, Diagnosisand Classification of Diabetes Mellitus andits Complications. Part 1: Diagnosis andClassification of Diabetes Mellitus. Geneva,World Health Org., 1999

37. Landstedt-Hallin L, Arner P, Lins PE,Bolinder J, Olsen H, Groop L: The role ofsulphonylurea in combination therapyassessed in a trial of sulphonylurea with-drawal: Scandinavian Insulin-Sulphon-ylurea Study Group Research Team. Dia-bet Med 16:827–834, 1999

38. Wolffenbuttel BH, Sels JP, Rondas-Col-bers GJ, Menheere PP: Prognostic factorsfor successful insulin therapy in subjectswith type 2 diabetes. Neth J Med 54:63–69, 1999

39. Muller UA, Muller R, Starrach A, Hunger-Dathe W, Schiel R, Jorgens V, Grusser M:Should insulin therapy in type 2 diabeticpatients be started on an out- or inpatientbasis? Results of a prospective controlledtrial using the same treatment and teach-ing programme in ambulatory care and auniversityhospital.DiabeteMetab24:251–255, 1998

40. Yki-Jarvinen H, Dressler A, Ziemen M:Less nocturnal hypoglycemia and betterpost-dinner glucose control with bedtimeinsulin glargine compared with bedtimeNPH insulin during insulin combina-tion therapy in type 2 diabetes: HOE 901/3002 Study Group. Diabetes Care 23:1130–1136, 2000

41. Bergenstal RJM, Whipple D, Noller D,Boyce K, Roth L, Upham P, Fish L, DeboldR: Advantages of adding metformin tomultiple dose insulin therapy in type 2diabetes (Abstract). Diabetes 47:A89, 1998

42. Relimpio F, Pumar A, Losada F, MangasMA, Acosta D, Astorga R: Adding met-formin versus insulin dose increase in in-sulin-treated but poorly controlled type 2diabetes mellitus: an open-label random-ized trial. Diabet Med 15:997–1002, 1998

43. Robinson AC, Burke J, Robinson S, John-ston DG, Elkeles RS: The effects of met-formin on glycemic control and serumlipids in insulin-treated NIDDM patientswith suboptimal metabolic control. Dia-betes Care 21:701–705, 1998

44. Kyllastinen M, Groop L: Combination ofinsulin and glibenclamide in the treat-ment of elderly non-insulin dependent

(type 2) diabetic patients. Ann Clin Res 17:100–104, 1985

45. Lindstrom T, Nystrom FH, Olsson AG,Ottosson AM, Arnqvist HJ: The lipopro-tein profile differs during insulin treat-ment alone and combination therapy withinsulin and sulphonylureas in patientswith type 2 diabetes mellitus. Diabet Med16:820–826, 1999

46. Clauson PG, Linde B: Absorption of rap-id-acting insulin in obese and nonobeseNIDDM patients. Diabetes Care 18:986–991, 1995

47. Kitabchi AE, Soria AG, Radparvar A, Law-son-Grant V: Combined therapy of insu-lin and tolazamide decreases insulin re-quirement and serum triglycerides in obesepatients with noninsulin-dependent dia-betes mellitus. Am J Med Sci 294:10–14,1987

48. Gutniak M, Karlander S-G, Efendic S:Glyburide decreases insulin requirement,increases b-cell response to mixed meal,and does not affect insulin sensitivity: ef-fects of short- and long-term combinedtreatment in secondary failure to sulfonyl-urea. Diabetes Care 10:545–554, 1987

49. Schade DS, Mitchell WJ, Griego G: Addi-tion of sulfonylurea to insulin treatmentin poorly controlled type II diabetes: adouble-blind, randomized clinical trial.JAMA 257:2441–2445, 1987

50. Groop L, Harno K, Tolppanen EM: Thecombination of insulin and sulphonyl-urea in the treatment of secondary drugfailure in patients with type II diabetes.Acta Endocrinol (Copenh) 106:97–101, 1984

51. Groop L, Harno K, Nikkila EA, PelkonenR, Tolppanen EM: Transient effect of thecombination of insulin and sulfonylurea(glibenclamide) on glycemic control innon-insulin dependent diabetics poorlycontrolled with insulin alone. Acta MedScand 217:33–39, 1985

52. Schwartz S, Raskin P, Fonseca V, Grave-line JF: Effect of troglitazone in insulin-treated patients with type II diabetes melli-tus: Troglitazone and Exogenous InsulinStudy Group. N Engl J Med 338:861–866,1998

53. Rubin C, Egan J, Schneider R, Pioglita-zone 014 Study Group: Combinationtherapy with pioglitazone and insulin inpatients with type 2 diabetes (Abstract).Diabetes 48:A110, 1999

54. Raskin P, Dole JF, Rappaport EB: Rosigli-tazone (RSG) improves glycemic controlin poorly controlled, insulin-treated type2 diabetes (Abstract). Diabetes 48:A94, 1999

55. Bachmann W, Lotz N, Mehnert H, RosakC, Schoffling K: Effectiveness of com-bined treatment with glibenclamide andinsulin in secondary sulfonylurea failure:a controlled multicenter double-blind

clinical trial. Dtsch Med Wochenschr 113:631–636, 1988

56. Quatraro A, Consoli G, Ceriello A, Giugli-ano D: Combined insulin and sulfonyl-urea therapy in non-insulin-dependent dia-betics with secondary failure to oral drugs: aone year follow-up. Diabete Metab 12:315–318, 1986

57. Giugliano D, Quatraro A, Consoli G, Mi-nei A, Ceriello A, DeRosa N, D’Onofrio F:Metformin for obese, insulin-treated dia-betic patients: improvement in glycaemiccontrol and reduction of metabolic riskfactors. Eur J Clin Pharmacol 44:107–112,1993

58. Lins PE, Lundblad S, Persson-Trotzig E,Adamson U: Glibenclamide improves theresponse to insulin treatment in non-in-sulin-dependent diabetics with secondfailure to sulfonylurea therapy. Acta MedScand 223:171–179, 1988

59. Reich A, Abraira C, Lawrence AM: Com-bined glyburide and insulin therapy in typeII diabetes. Diabetes Res 6:99–104, 1987

60. Lewitt MS, Yu VK, Rennie GC, Carter JN,Marel GM, Yue DK, Hooper MJ: Effects ofcombined insulin-sulfonylurea therapy intype II patients. Diabetes Care 12:379–383, 1989

61. Osei K, O’Dorisio TM, Falko JM: Con-comitant insulin and sulfonylurea therapyin patients with type II diabetes: effectson glucoregulation and lipid metabolism.Am J Med 77:1002–1009, 1984

62. Feinglos MN, Thacker CR, Lobaugh B,DeAtkine DD, McNeill DB, English JS,Bursey DL: Combination insulin and sul-fonylurea therapy in insulin-requiringtype 2 diabetes mellitus. Diabetes Res ClinPract 39:193–199, 1998

63. Longnecker MP, Elsenhans VD, LeimanSM, Owen OE, Boden G: Insulin and asulfonylurea agent in non-insulin-depen-dent diabetes mellitus. Arch Intern Med146:673–676, 1986

64. Casner PR: Insulin-glyburide combina-tion therapy for non-insulin-dependentdiabetes mellitus: a long-term double-blind, placebo-controlled trial. Clin Phar-macol Ther 44:594–603, 1988

65. Kelley DE, Bidot P, Freedman Z, Haag B,Podlecki D, Rendell M, Schimel D, WeissS, Taylor T, Krol A, Magner J: Efficacy andsafety of acarbose in insulin-treated pa-tients with type 2 diabetes. Diabetes Care21:2056–2061, 1998

66. Chiasson J-L, Josse RG, Hunt JA, Palma-son C, Rodger NW, Ross SA, Ryan EA,Tan MH, Wolever TMS: The efficacy ofacarbose in the treatment of patients withnon-insulin-dependent diabetes mellitus:a multicenter controlled clinical trial. AnnIntern Med 121:928–935, 1994

Yki-Jarvinen


ukpds

Documents