register-based research –- pros and cons a register...

10
Register-based research — pros and cons a register-based researcher’s view Paul W. Dickman 1 1 Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden 10th International Symposium on Hodgkin Lymphoma (ISHL10) 22/10/2016

Upload: vuque

Post on 07-Jun-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

Register-based research — pros and cons

a register-based researcher’s view

Paul W. Dickman1

1Department of Medical Epidemiology and Biostatistics,Karolinska Institutet, Stockholm, Sweden

10th International Symposium on Hodgkin Lymphoma (ISHL10)22/10/2016

Presenter Disclosure Information

Register-based research — pros and cons

I will not discuss off label use in my presentation.

I have no financial relationships to disclose.

I have nothing to disclose.

Paul Dickman Register-based research ISHL10 22/10/2016 2

Main points

Population-based registers are important for evaluating theeffectiveness of healthcare systems.

Register-based research can provide knowledge that can improvehuman health, for example, through identifying opportunities fordisease prevention and improved treatment.

Registers are not perfect; Register-based research is not easy!

Intricate knowledge of the registers is required.

Important to have a well-defined hypothesis.

Combining register data and clinical data (e.g., CarolineWeibull’s study) can be advantageous.

Register-enhanced trials have exciting potential.

Paul Dickman Register-based research ISHL10 22/10/2016 3

Validating the Swedish Cancer Registry [1]

We then did a detailed study of the medical records

in order to find possible reasons for the non-

reporting. Cancer cases (according to ICD-10:

C00-C96, except C77-C79) identified in the Swed-

ish Cancer Register and in the Hospital Discharge

Register are shown in Table I. There was a total of

42 770 cancer cases (in 42 160 persons) reported to

the Swedish Cancer Register in 1998 compared with

42 010 persons in the Hospital Discharge Register.

The number of persons with a malignant cancer

diagnosis reported only to the Hospital Discharge

Register was 3 429. The underreporting for breast,

female genital organs and urologic sites was low,

while it was high for soft tissue, nervous system and

for leukaemia and lymphoma.

When the non-reporting was further studied by

gender and age (Table II), we found that it was worse

in ages over 70, and for cancers of the soft tissues

and nervous system the underreporting was highest

in the 70� age group. The non-reporting was also

notable for lung cancer in the elderly and for male

breast cancer, though the number of male patients

with breast cancer was very low. There was a

tendency for underreporting of cancers in women

to be worse in the elderly, and in men to be worse in

the younger age group.

Table I. Number of malignant cancer cases, according to ICD-10: C00-C96 except C77-C79, in the Swedish Cancer Register in 1998.

Number of cancer cases in the Hospital Discharge Register in 1998 and the number of cancer cases that were not registered in the Swedish

Cancer Register in 1958�1998.

Reported to the Cancer

Registry in 1998

Reported to the Hospital

Discharge Registry in 1998

Reported to the Hospital

Discharge Registry only

Site Number of tumours Number of persons Number of persons Number of persons Ratio (%)1

Head and neck 988 982 1 140 99 10.1

Digestive organs 8 715 8 579 9 979 863 10.1

Lung 2 997 2 982 3 470 329 11.0

Soft tissue 1 316 1 312 1 527 165 12.6

Skin 4 232 4 092 1 653 398 9.7

Breast 6 225 6 091 6 118 84 1.4

Female genital organs 2 936 2 909 3 354 100 3.4

Urology 9 996 9 865 8 207 509 5.2

Nervous system 1 410 1 407 1 376 195 13.9

Leukemia and lymphoma 3 171 3 157 3 884 534 16.9

Other sites 784 784 1 302 153 19.5

Total 42 770 42 160 42 010 3 429 8.1

1Ratio in per cent of the number of persons with a tumour reported to the Swedish Cancer Registry.

Table II. Number of persons diagnosed with a malignant tumour, according to ICD-10: C00-C96 except C77-C79, reported to the

Swedish Cancer Registry in 1998 by age and sex. The ratio of persons reported to the Hospital Discharge Registry in 1998 but not to the

Swedish Cancer Registry (n�3 429), to the number of persons reported to the Swedish Cancer Registry in 1998, by site, age, and sex.

Males Females

0�69 70� 0�69 70�

Site N Ratio (%) N Ratio (%) N Ratio (%) N Ratio (%)

Head and neck 380 7.1 293 8.9 161 13.7 148 16.2

Digestive organs 1 855 7.8 2 627 10.5 1 509 5.1 2 588 14.1

Lung 850 7.4 986 13.7 639 6.1 507 18.1

Soft tissue 328 13.7 169 23.7 493 9.3 322 10.6

Skin 870 7.9 1 384 7.9 731 9.3 1 107 13.6

Breast 12 25.0 16 31.3 4 023 1.1 2 040 1.6

Female genital organs � � 1 732 1.9 1 177 5.5

Urology 3 247 3.3 5 647 5.3 382 7.1 589 12.7

Nervous system 509 5.5 148 43.9 534 7.1 216 29.6

Leukemia and lymphoma 951 13.0 797 19.7 656 11.3 753 23.8

Other sites 136 8.8 155 38.7 183 8.2 310 21.3

Total 9 138 6.8 12 222 9.6 11 043 4.4 9 757 11.8

The completeness of the Swedish Cancer Register 29

Act

a O

ncol

Dow

nloa

ded

from

info

rmah

ealth

care

.com

by

(AC

TIV

E)

Kar

olin

ska

Inst

itute

t Uni

vers

ity L

ibra

ry o

n 10

/05/

11Fo

r pe

rson

al u

se o

nly.

46% of cases in HDR only should have been reported to SCR.

Paul Dickman Register-based research ISHL10 22/10/2016 4

Validating the Swedish Cancer Registry [2]

16

CANCER INCIDENCE IN SWEDEN 2011 Table A1. Main cancer sites in not autopsied cases in the Cause of Death register 2011, but not in the Cancer Register 1958-2011

ICD-10 Site Sex

Age at death

All ages 0-49 50-74 75+

C00-C97 All sites M 1 289 25 391 873

F 1 349 21 223 1 105

C16 Stomach M 17 1 3 13

F 22 1 3 18

C18-C21 Colon and rectum M 100 3 26 71

F 142 1 16 125

C22 Liver and intraheptic bile ducts M 105 2 49 54

F 62 1 10 51

C23-C24 Gallbladder and biliary tract M 30 0 6 24

F 94 0 16 78

C25 Pancreas M 174 3 60 111

F 216 1 34 181

C26 Other and ill-defined digestive organs M 18 0 5 13

F 27 0 0 27

C34 Bronchus and lung M 233 3 81 149

F 204 2 52 150

C50 Breast M 1 0 0 1

F 41 4 6 31

C51-C58 Female genital organs M 0 0 0 0

F 55 1 13 41

C61 Prostate M 186 0 35 151

F 0 0 0 0

C64-C68 Urinary tract M 95 1 28 66

F 99 0 16 83

C69-C72 Central nervous system M 40 7 13 20

F 39 4 11 24

C76-C80 Ill-defined, secondary & unspecified sites M 157 0 46 111

F 208 3 19 186

C81-C96 Lymphoid and haematopoietic system M 76 3 20 53

F 100 2 16 82

Other Other sites M 57 2 19 36

F 40 1 11 28

Paul Dickman Register-based research ISHL10 22/10/2016 5

A typical register-based study [3];

androgen deprivation and risk of hip fracture

Compare incidence of hip fracture among,

17,731 men diagnosed with prostate cancer treated withbilateral orchiectomy.43,230 men diagnosed with prostate cancer not treated withbilateral orchiectomy.362,354 men randomly selected from the general population.

Cancer patients identified from the cancer registry.

Matched controls identified from the total population register.

Exposure (orchiectomy) and outcome (femoral neck fracture)identified from inpatient register.

Paul Dickman Register-based research ISHL10 22/10/2016 6

logical fractures in the group of men who underwent orchi-ectomy. However, men diagnosed with metastatic prostatecancer have a poor prognosis, and so most pathological frac-tures would occur within a year or 2 following diagnosis. Byexcluding the first 6 months of followup following diagnosiswe excluded men in whom tumor was diagnosed as a result ofa fracture as well as men who experienced fracture within 6months of diagnosis. We observed approximately constantrelative risks throughout followup, suggesting that our esti-mates were not heavily influenced by the occurrence ofpathological fractures. Skeletal metastases are more com-mon in the vertebrae and pelvis than in the hip, so that if ourestimates were heavily influenced by pathological fractures,we would have expected larger relative risks for these bones,which we did not observe.

The methods of diagnosing and treating prostate cancerhave changed with time. We observed similar patterns ofrelative fracture risk when analyses were restricted to indi-viduals diagnosed in 1992 or later, eg relative risk estimateswere similar for each time interval since entry. However, theabsolute fracture risk during this period was slightly higherin men who underwent orchiectomy and slightly lower inpopulation controls, leading to slightly higher estimates ofrelative risk. During the 1990s compared to earlier years asmaller proportion of men diagnosed with prostate cancer

had metastatic disease at diagnosis. As such, we would haveexpected the influence of pathological fractures on the rela-tive fracture risk comparing men with and without orchiec-tomy to be smaller during this period. That is, we would haveexpected the estimated relative risks to be larger for thisperiod, which is exactly what we observed. Estrogen therapy,which is thought to protect against osteoporosis,17 was com-monly used as a means of androgen deprivation during 1964to 1983, so that we might have expected patients diagnosedwith prostate cancer to be at decreased fracture risk com-pared to population controls during this period, which we didnot observe.

The observed difference in fracture risk between, for ex-ample, men treated with orchiectomy and population con-trols cannot be completely attributed to osteoporotic frac-tures secondary to orchiectomy. Other factors that affectfracture risk, the probability of being treated as an inpatientwhen a fracture is diagnosed, the probability that a diag-nosed fracture is recorded in the inpatient registry or selec-tion into the groups that we studied should be considered. Itis reasonable to believe that most individuals with a hipfracture have the fracture diagnosed and are admitted to thehospital for surgery. On the other hand, vertebral fracturesare not always diagnosed and even when a vertebral frac-ture is diagnosed, the patient is not always admitted to thehospital. For most fracture types including those of the hip,we believe that the sensitivity and specificity of recordingfractures was similar in the 3 study groups. Therefore, weargue that our relative risk estimates should not be substan-tially biased.

FIG. 2. Incidence proportion of femoral neck fractures in each of 3exposure groups as function of time since start of followup (6 monthsafter diagnosis).

FIG. 3. Incidence proportion of intertrochanter fractures in each of3 exposure groups as function of time since start of followup (6months after diagnosis).

TABLE 3. IRRs estimated separately for each period since entry at 6months following diagnosis

Yrs Since EntryNo Orchiectomy Orchiectomy

IRR 95% CI IRR 95% CI

Femur neck:0–1 1.45 1.25–1.68 1.81 1.50–2.191–2 1.39 1.18–1.64 2.07 1.71–2.522–3 1.42 1.19–1.71 2.41 1.96–2.963–4 1.08 0.86–1.36 2.18 1.72–2.774–5 1.32 1.04–1.67 2.55 1.97–3.285–7 1.41 1.17–1.71 2.10 1.66–2.667–10 1.32 1.07–1.63 2.17 1.65–2.87

10–14 1.20 0.90–1.59 1.27 0.68–2.3715 or Greater 1.32 0.77–2.26 5.61 1.40–22.6

Intertrochanter:0–1 1.74 1.48–2.06 2.38 1.95–2.901–2 1.42 1.18–1.71 1.84 1.46–2.322–3 1.07 0.84–1.35 2.27 1.80–2.873–4 1.22 0.95–1.57 2.73 2.13–3.504–5 1.22 0.93–1.59 1.70 1.22–2.375–7 1.21 0.97–1.51 2.01 1.55–2.617–10 1.13 0.88–1.45 1.90 1.39–2.60

10–14 1.32 0.97–1.81 2.96 1.90–4.6215 or Greater 1.64 0.98–2.77 2.86 0.40–20.4Estimates were obtained using interaction terms in Poisson regression

model and are adjusted for attained age and calendar period at diagnosis. TABLE 4. IRRs and 95% CIs estimated separately according toattained age

Attained AgeNo Orchiectomy Orchiectomy

IRR 95% CI IRR 95% CI

Femur neck:0–64 3.71 2.36–5.81 7.63 3.92–14.9

65–69 2.05 1.44–2.92 6.66 4.55–9.7570–74 1.35 1.09–1.68 2.93 2.27–3.7975–79 1.44 1.24–1.67 2.52 2.12–2.9980–84 1.27 1.12–1.45 1.91 1.64–2.2285 or Older 1.26 1.12–1.41 1.63 1.40–1.89

Intertrochanter:0–64 2.81 1.68–4.72 10.80 5.94–19.8

65–69 1.75 1.22–2.53 4.39 2.82–6.8270–74 1.57 1.23–2.00 3.65 2.78–4.8075–79 1.26 1.06–1.50 2.28 1.87–2.7880–84 1.29 1.11–1.50 1.77 1.48–2.1285 or Older 1.24 1.08–1.41 1.89 1.62–2.21Estimates were obtained using interaction terms in Poisson regression

model and are adjusted for calendar period and time since entry.

HIP FRACTURES AND PROSTATE CANCER 2211

Paul Dickman Register-based research ISHL10 22/10/2016 7

Caroline Weibull’s study

Population-based registers were not sufficient; information ontreatment and date of relapse were crucial.

We used data collected by Ingrid Glimelius, where additionalinformation (e.g., treatment and relapse) was obtained frommedical records.

Matching with the medical birth register provided good data onpregnancies.

Paul Dickman Register-based research ISHL10 22/10/2016 8

Enhancements

Supplement the register data with collections of additionalclinical data.

Registry-based randomized controlled trials [4], also called‘pragmatic trials’ [5].

Still need to consider the strengths/weaknesses of the registersand only use such studies to answer suitable research questions.

Registry-based randomized controlled trials are usually performedto address comparative effectiveness research questions inreal-world settings when high-quality registries are available.

Also useful for studying treatment side-effects in a real worldsetting.

Paul Dickman Register-based research ISHL10 22/10/2016 9

References

[1] Barlow L, Westergren K, Holmberg L, Talback M. The completeness of the Swedish CancerRegister: a sample survey for year 1998. Acta Oncol 2009;48:27–33.

[2] National Board of Health and Welfare. Cancer incidence in Sweden 2011. Official Statisticsof Sweden 2012-12-19, The National Board of Health and Welfare (Socialstyrelsen), 2012.

[3] Dickman PW, Adolfsson J, Astrom K, Steineck G. Hip fractures in men with prostatecancer treated with orchiectomy. Journal of Urology 2004;172:2208–2212.

[4] Li G, Sajobi TT, Menon BK, Korngut L, Lowerison M, James M, et al.. Registry-basedrandomized controlled trials- what are the advantages, challenges, and areas for futureresearch? J Clin Epidemiol 2016;.

[5] Thabane L, Kaczorowski J, Dolovich L, Chambers LW, Mbuagbaw L, CI. Reducing theconfusion and controversies around pragmatic trials: using the cardiovascular healthawareness program (CHAP) trial as an illustrative example. Trials 2015;16:387.

Paul Dickman Register-based research ISHL10 22/10/2016 10