recorded diagnosis of nafld and the independent risk of · 9department of medical informatics,...

Confidential: For Review OnlyRecorded diagnosis of NAFLD and the independent risk of incident acute myocardial infarction and stroke: findings

from analyses of 18 million European adults

Journal: BMJ

Manuscript ID BMJ-2019-049068.R1

Article Type: Research

BMJ Journal: BMJ

Date Submitted by the Author: 14-May-2019

Complete List of Authors: Alexander, Myriam; GlaxoSmithKline, UK, Real-World DataLoomis, Katrina; Pfizer, van der lei, Johan; Erasmus MC, Department of Medical InformaticsDuarte-Salles, Talita; Institut Universitari d'Investigació en Atenció Primària Jordi Gol (IDIAP Jordi Gol), Gran Via de les Corts Catalanes, 587,Prieto-Alhambra, Daniel; university of oxford, ndormsAnsell, David; IQVIAPasqua, Alessandro; Italian College of General Practicioners, Lapi, Francesco; Health Search, Rijnbeek, Peter; Erasmus University Medical Centers, Medical InformaticsMosseveld, Mees; Erasmus University Medical Center, Department of Medical InformaticsAvillach, Paul; Harvard Medical School, Egger, Peter; GlaxoSmithKline, UK, Real-World DataDhalwani, Nafeesa; Evidera Market Access LtdKendrick, Stuart; GlaxoSmithKline, Medicines Research CentreCelis-Morales, Carlos; University of Glasgow, Institute of Cardiovascular and Medical SciencesWaterworth, Dawn; GSK, Alazawi, William; Queen Mary, University of London, Sattar, Naveed; University of Glasgow, Institute of Cardiovascular and Medical Sciences, British Heart Foundation Glasgow Cardiovascular Research Centre

Keywords: non-alcoholic fatty liver disease, non-alcoholic steatohepatitis, cardiovascular disease risk, risk factors, electronic health records

https://mc.manuscriptcentral.com/bmj

BMJ

Confidential: For Review OnlyRecorded diagnosis of NAFLD and the independent risk of incident acute myocardial infarction and stroke: findings from analyses of 18 million European adults

First author’s surname: Alexander. Short title: NAFLD and incident AMI OR STROKE risks

Myriam Alexander, PhD,1 A Katrina Loomis, MA,2 Johan van der Lei, PhD,3 Talita Duarte-Salles, PhD,4 Daniel Prieto-Alhambra, PhD,5 David Ansell, MD,6 Alessandro Pasqua, MSc,7 Francesco Lapi, PhD,8 Peter Rijnbeek, PhD,9 Mees Mosseveld, MSc,10 Paul Avillach, PhD,11 Peter Egger, PhD,12 Nafeesa N Dhalwani,, PhD,13 Stuart Kendrick, PhD,14 Carlos Celis-Morales, PhD,15 Dawn M. Waterworth, PhD,16

William Alazawi, PhD,17¶ Naveed Sattar, FMedSci*¶18

¶ These authors contributed equally to the study

Affiliations:1Real World Evidence & Epidemiology, GlaxoSmithKline, Uxbridge, Middlesex UB11 1BS, UK. Job title: Clinical data analyst – Scientist. Email: [email protected] 2Worldwide Research and Development, Pfizer, Genome Sciences and Technologies, Groton, United States. Job title: Director, Head of Cardiovascular and Metabolic Disease Human Genetics. Email: [email protected] 3Department of Medical Informatics, Erasmus University Medical Center Rotterdam,‘s-Gravendijkwal, 230, Rotterdam. 3015 CE, Netherlands. Job title: Professor of Medical Informatics. Email: [email protected] 4Institut Universitari d'Investigació en Atenció Primària Jordi Gol (IDIAP Jordi Gol). Gran Via de les Corts Catalanes, 587, Barcelona. 08007, Spain. Job title: Epidemiologist. Email: [email protected] 5Pharmaco- and Device Epidemiology, Centre for Statistics in Medicine, NDORMS, University of Oxford, UK. Job title: Associate Professor and NIHR Clinician Scientist. Email: [email protected] 6IQVIA, 210 Pentonville Road, London N1 9JY, UK. Job title: Medical Director. And: Birmingham University, Institute of Applied Health Research, UK. Job title: NIHR Academic Clinical Lecturer. Email: [email protected] 7Health Search, Italian College of General Practitioners and Primary Care. Via Sestese, 61, Firenze. 50141, Italy. Job title: Statistician. Email: [email protected] 8Health Search, Italian College of General Practitioners and Primary Care. Via Sestese, 61, Firenze. 50141, Italy. Job title: Epidemiologist. Email: [email protected] 9Department of Medical Informatics, Erasmus University Medical Center Rotterdam, ‘s-Gravendijkwal, 230, Rotterdam. 3015 CE, Netherlands. Job title: Professor of Clinical Informatics. Email: [email protected] 10Department of Medical Informatics, Erasmus University Medical Center Rotterdam, ‘s-Gravendijkwal, 230, Rotterdam. 3015 CE, Netherlands. Job title: System developer. Email: [email protected] 11Harvard Medical School, Harvard, Boston, Massachusetts, United States. Job title: Assistant Professor of Biomedical Informatics. And: Erasmus Universitair Medisch Centrum Rotterdam, ‘s-Gravendijkwal, 230, Rotterdam. 3015 CE, Netherlands. Job title: Scientist. Email: [email protected] 12Real World Evidence & Epidemiology, GlaxoSmithKline, Stockley Park West, 1-3 Ironbridge Road, Uxbridge, Middlesex UB11 1BS, UK. Job title: Senior Director. Email: [email protected] 13Diabetes Research Centre, University of Leicester, UK. Job title: Lecturer in Epidemiology. Email: [email protected] 14GlaxoSmithKline, Medicines Research Centre, Gunnels Wood Road, Stevenage, Hertfordshire, SG1 2NY, UK. Job title: Director Clinical Development, Experimental Clinical Development Physician. Email: [email protected]

Page 1 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

mailto:[email protected]














Confidential: For Review Only

2

15University of Glasgow, Institute of Cardiovascular & Medical Sciences, BHF Glasgow Cardiovascular Research Centre. 126 University Place, Glasgow. G12 8TA, UK. Job title: Research Fellow. Email: [email protected] 16GlaxoSmithKline, 709 Swedeland Road King of Prussia, PA 19406, USA. Job title: Director of Genetics and Target Sciences and Therapy Area Head. Email: [email protected] 17The Blizard Institute, Queen Mary, University of London, 4 Newark Street, London, E1 2AT, UK. Job title: Reader in Hepatology. Email: [email protected] 18University of Glasgow, Institute of Cardiovascular & Medical Sciences, BHF Glasgow Cardiovascular Research Centre. 126 University Place, Glasgow. G12 8TA, UK. Job title: Professor of Metabolic Medicine. Email: [email protected]

Corresponding authors:Professor Naveed SattarInstitute of Cardiovascular and Medical Sciences, University of GlasgowBHF Glasgow Cardiovascular Research Centre126 University Place, Glasgow, G12 [email protected]: +44 141 330 3419

Dr William AlazawiBarts Liver CentreBlizard InstituteQueen Mary, University of London4 Newark StreetLondonE1 [email protected] +44 20 7882 2308

Word count: 4,090

Page 2 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960








3

Abstract

Objective: Whether non-alcoholic fatty liver disease (NAFLD) is associated with excess acute

myocardial infarction (AMI) or stroke risk beyond associated risk factors is uncertain. Prior evidence

derives from cohort studies variably adjusted for established risk factors. We estimated risk of

acquiring diagnoses of AMI or stroke in patients with NAFLD or non-alcoholic steatohepatitis (NASH)

diagnoses compared to individually-matched unexposed patients.

Design: Matched cohort study.

Setting: Patients enrolled in population-based, electronic primary healthcare databases before

31/12/2015 from four European countries: Italy (n=1,542,672), Netherlands (n=2,225,925), Spain

(n=5,488,397), and UK (n=12,695,046).

Participants: Patients with recorded diagnosis of NAFLD or NASH (NAFLD/NASH), and not other liver

diseases, each matched at the time of NAFLD diagnosis (index date) by age, sex, practice site, and

visit recorded at ±6 months of date of diagnosis, to up to 100 patients without NAFLD/NASH in the

same database.

Intervention: NAFLD and unexposed patients were followed from index date to incident acute

myocardial infarction (AMI) or stroke.

Main outcome measures: Incident fatal or non-fatal AMI, ischaemic or unspecified stroke. Hazard

ratios (HRs) were estimated using Cox models and pooled across databases by random-effect meta-

analyses.

Results: 120,795 patients with recorded NAFLD/NASH diagnoses were identified with mean follow-

up 2.1-5.5 years. Pooled HR for AMI adjusting for age and smoking was 1.17 (95% CI: 1.05-1.30;

1,035 events in NAFLD/NASH, 67,823 in matched-unexposed). This was attenuated with

progressive adjustments for type 2 diabetes, systolic blood pressure, total cholesterol, statin use and

hypertension to 1.01 (95% CI: 0.91-1.12; 747 events in NAFLD/NASH, 37,462 in matched-

unexposed). Pooled HR for stroke adjusting for age and smoking status was 1.18 (95% CI: 1.11-1.24;

2,187events in NAFLD/NASH, 134,001 in matched-unexposed), attenuating to 1.04 (95% CI: 0.99-

1.09; 1,666 events in NAFLD, 83,882 in matched-unexposed) after further adjustment for risk

factors.

Conclusions: The diagnosis of NAFLD in current routine care of 17.7 million patients, adjusted for

conventional risk factors does not equate to a high independent, excess AMI or stroke risk.

Cardiovascular risk assessment in patients diagnosed with NAFLD is important but should be

conducted in the same way as in the in general population and without a risk multiplier.

Page 3 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


4

Summary Box:

WHAT IS ALREADY KNOWN ON THIS TOPIC

Non-alcoholic fatty liver disease (NAFLD) is associated with metabolic syndrome and other risk

factors for acute myocardial infarction (AMI) or stroke. NAFLD is associated with increased AMI

and stroke risk, and cardiovascular surrogate markers. However, the independent association of

NAFLD for AMI and stroke after comprehensively adjusting for established risk factors has yet to

be established.

WHAT THIS STUDY ADDS

This the largest prospective epidemiological study to evaluate the incidence risk of AMI and stroke

in patients with a diagnosis of NAFLD, compared to the general population, taking into account all

key potential confounders. In four large European databases of adults, the adjusted hazard ratios

for incident AMI or stroke diagnoses were modest and not significantly greater than those in age,

sex, and practice site matched unexposed individuals.

Page 4 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


5

Introduction

Over the last several years, researchers have proposed that non-alcoholic fatty liver disease (NAFLD),

in addition to being a marker of pathological ectopic fat accumulation, and diabetes risk (which is

unambiguous), may have important associations with cardiovascular outcomes.[1] The incidence of

NAFLD has risen alongside that of obesity and diabetes worldwide, however its impact on the

complications of those conditions, including on the risk of cardiovascular disease, has not yet been

firmly established. In some ways, this notion is not surprising since NAFLD patients often have

abnormal glucose and lipid levels, and are often overweight or obese. Other mechanisms that may

explain a possible association include increased oxidative stress, deranged adipokine profile, and

hypercoagulability are more present in NAFLD patients,[2] giving rise to AMI or stroke risk beyond

traditional risk factors. With respect to surrogate markers, NAFLD patients have been shown to have

increased prevalence of: subclinical atherosclerosis,[3–5] subclinical AMI or stroke in the

Framingham study,[3] and carotid atherosclerotic plaques.[6,7] The severity of coronary artery

disease was also increased in NAFLD patients referred for coronary angiography.[8]

In terms of clinical outcomes, headline results from recent meta-analyses seem to concur with a

meaningful independent AMI or stroke risk. For example, Musso et al[9] noted that NAFLD had an

odds ratio of 2.05 [1.81 to 2.31] for incident cardiovascular (CVD) events in patients with ultrasound-

defined NAFLD compared to controls without. Similarly, Targher et al[10] reported an odds ratio of

1.64 [1.26 to 2.13] for combined fatal and non-fatal AMI or stroke events in their recent meta-

analysis of over 34,000 subjects inclusive of 2,600 CVD outcomes. These two meta-analyses had

moderate to high heterogeneity, and the authors admit potential for bias given variable and often

incomplete adjustment for all usual risk factors. Even so, their top line findings seem to strengthen

the notion that patients with NAFLD have risk levels for AMI or stroke approaching those for patients

with type 2 diabetes. Such findings seem to back up proposals that all patients with NAFLD should be

aggressively treated for CVD prevention.[1]

However, the level of independent contribution of NAFLD to the increased incidence of AMI or

stroke remains strongly debated[11] since the vast majority of studies included in the

aforementioned meta-analyses[9,10] only partially adjusted for all traditional risk factors, such as

diabetes and lipid levels, which often co-exist with NAFLD, and few considered geographical and

other socioeconomic sources of heterogeneity. Furthermore, robust assessments of AMI or stroke

risk in NAFLD compared to the general population are important to establish in routine ‘real world’

health care, since such information will dictate the extent to which physicians in routine clinical care

Page 5 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


6

should aggressively manage CVD risk in patients diagnosed with NAFLD. Such data would also

determine whether there is a need for a AMI or stroke risk multiplier for patients with NAFLD, as is

the case now in patients with diabetes or rheumatoid arthritis.[12] In this context, we undertook a

longitudinal analysis of NAFLD patients recorded in four European primary care databases, as part of

the European Medical Information Framework (EMIF). We estimated the incident risk of developing

myocardial infarction (AMI) and stroke in those cohorts of patients identified in routine practice,

progressively adjusting for traditional cardiovascular risk factors and in sensitivity analyses we

investigated the associations in NAFLD patients without a subsequent NASH diagnosis.

Methods

Databases

We included patient data from four primary care databases available via the EMIF network: The

Health Improvement Network (THIN - UK), Health Search Database (HSD - Italy), Information System

for Research in Primary Care (SIDIAP - Spain), and Integrated Primary Care Information (IPCI -

Netherlands).[13–17] All databases provided ethical approval of the study protocol. All data were

de-identified at the database source and the databases are compliant with local data protection

laws. Data extraction was performed locally by each data custodian liaising with the European

Medicine Innovative Framework (EMIF)-Platform [www.emif.eu]. The data were then uploaded via a

secure server onto a private remote secure research environment and analysed centrally.

Study design

We adopted a matched cohort design. Code lists for all clinical diagnoses (exclusion criteria,

exposure, covariates and events of interest) were generated using a semantic harmonization process

that involved mapping concepts in each terminology – International Classification Disease (ICD)

version 9 for HSD, ICPC Dutch for IPCI, ICD version 10 for SIDIAP and Read Codes for THIN – to

Unified Medical Language System concepts to ensure comparability between databases. [18] All

patients with a diagnosis of NAFLD (including non-alcoholic steatohepatitis NASH) prior to

01/01/2016 were identified in the four databases. Due to difference in coding terminology,

recording of NASH diagnoses distinctively from NAFLD diagnoses was only possible in Spain (SIDIAP

database) and the UK (THIN database). In the main analyses in these databases, NAFLD and NASH

patients were grouped together, as is the case in IPCI and HSD by virtue of the coding (ICPC Dutch

codes and ICD9 do not have distinctive codes for NAFLD and NASH). Sensitivity analyses excluding

NASH patients were conducted in SIDIAP and THIN.

Page 6 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


7

Each NAFLD patient was matched with up to 100 so-called “non-exposed” patients who did not have

a NAFLD or NASH diagnosis. Index date was the date of NAFLD diagnosis for NAFLD and matched

non-exposed patient sets. Matching was done on practice site (as a proxy for socioeconomic

deprivation),[19] age at index date +/- 5 years, sex, and presence of a record of a general

practitioner visit date at index date +/- 6 months.

Patients with NAFLD and matched-unexposed were included in the analysis if they were aged ≥18 at

diagnosis, remained active in the database for at least 12 months from registration and 6 months

prior to index date; and had at least 6 months of follow-up post index date. Patients a record of

alcohol abuse at any time prior to diagnosis, a past AMI or stroke event were excluded. Flow-chart of

inclusion and exclusion criteria is in Table 1.

Patients were followed up from index date until the earliest of: occurrence of an event, end of study

period (31/12/2015), loss to follow-up due to exit out of the database or death. Events of interest

were fatal or non-fatal AMI, and ischaemic or unspecified stroke.

Variables

In Europe, primary care physicians act as gatekeepers of access to healthcare and store information

on clinical diagnoses, therapeutic prescriptions, lifestyle (smoking behaviours), and vital signs

measurements, procedures, and sometimes socio-economic information. Information filters from

secondary care via referral letters and results from laboratories were either automatically included

in the patient’s clinical history or sent back to the general practitioner for inclusion into medical

records. As such, demographic information, lifestyle, and medical history on relevant morbidities

could be extracted from patients’ records. Levels of total cholesterol, and systolic blood pressure

were extracted from 2 years prior to index date to 6 months after index. Smoking status was derived

as “current” if the patient had a record of being a smoker up to 5 years prior to index date or

anytime post index date, and “non-current” otherwise. Statin use was coded as “yes” if a patient had

a record of statin prescription in the 2 years prior and up to 6 months after index date. History of

type 2 diabetes and hypertension were defined as a record occurring any time prior to or at index

date.

Data analysis

Analyses were performed using a 2-step approach for data synthesis as described previously. First,

each of the four databases was analysed separately and then the estimates for each of these studies

Page 7 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


8

were polled together using a Random-effect meta-analysis. The NAFLD versus matched unexposed

group were described using percentages for categorical variables, mean and standard deviation for

normally distributed variables, and median and interquartile range for skewed variables. Incidence

rates of AMI and stroke were estimated within each group by dividing the number of incident events

by the total number of person-years at risk. 95% confidence intervals for incidence rates were

estimated assuming a Poisson distribution. Hazard ratios for incident AMI or stroke associated with

having a diagnosis of NAFLD were estimated using Cox proportional hazard models for each study

independently. The models were stratified by matching ID (see study design for variables for

matching which include sex) and progressively adjusted in multivariate models for (i) age and

smoking status; and (ii) age, smoking status, type 2 diabetes, statin use, diagnosis of hypertension,

systolic blood pressure, and total cholesterol (in subsets of patients with data available). In

sensitivity analyses, we also further adjusted for Body Mass Index (BMI), and high-density

lipoprotein cholesterol (HDL-C). Hazard ratios were then pooled across studies by random-effects

meta-analysis. Heterogeneity across databases was tested using the Q statistic,[20] which has a chi-

square distribution with k – 2 degrees (k = 4 databases) of freedom on the null hypothesis of no

heterogeneity, and the corresponding p-value was obtained. We also reported the I2 statistic which

gives the percentage of variation among studies that is due to heterogeneity across databases,

rather than to variation among individual patients within a database.[21] Hazard ratios were

estimated by pre-specified subgroups according to sex, BMI (obese - BMI≥30kg/m2 - versus normal

weight), smoking status, age group (<55 years old versus ≥55 years old), hypertension status, type 2

diabetes status. For this, an interaction term was added to the models between NAFLD diagnosis and

subgroup; and hazard ratios for each subgroup were then pooled across databases by random-

effects meta-analyses. We excluded values that were physiologically implausible: BMI below

15kg/m2; laboratory values greater than the mean in the database plus 3 times the standard

deviation; AST and ALT less than 5 iU/L; and platelet counts below 5x109/L. Missing data were not

imputed and analyses were run in the samples of patients with non-missing data for all variables in

the models.

The data were locally extracted within each centre after quality control checks using a standardized

script and centrally analysed using Stata v14 on the secure remote research environment of the

EMIF Platform.

Patient and public involvement

Page 8 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


9

No patients were involved in setting the research question or the outcome measures. However,

patients were involved in the setup of the overall European Medical Information Framework

consortium which underpinned this work. No patients were asked to advise on interpretation or

writing up of results. The results of the research will be disseminated to patients through the emif.eu

website.

Results

We accessed primary care records of a total number of 21,952,040 patients resident in four

European countries: Italy (HSD, n=1,542,672), Netherlands (IPCI, n=2,225,925), Catalonia (Spain;

SIDIAP, n=5,488,397), and UK (THIN, n=12,695,046) (Table 1).[13–16] After excluding patients with a

history of alcohol abuse, a past AMI or stroke event, who had less than 1 year of enrolment, and less

than 6 months of follow-up prior and post index date; we identified 120,795 patients with an

incident NAFLD diagnosis (21,627 in HSD; 12,595 in IPCI; 67,109 in SIDIAP; and 19,464 in THIN).

Baseline characteristics

The duration of follow-up before and after the index date, age distribution, and percentage of males

were comparable in the NAFLD and non-NAFLD groups of patients in each of the four databases

(Table 2).[13–16] Average follow-up post index date was lowest in IPCI (median: 2.1 years;

interquartile range: 1.2-3.4 years in NAFLD patients) and highest in HSD (5.5 years; 3.0-8.1).

Traditional cardiovascular risk factors were more common in NAFLD compared to non-NAFLD

patients: proportions of current smokers (except for the THIN database), patients with a history of

type 2 diabetes or hypertension, levels of BMI, and levels of systolic blood pressure were higher in

NAFLD versus non-NAFLD patients within each of the four databases.

Outcome Incidence rates

The total number of person-years’ follow up for NAFLD/NASH patients ranged from 85,361 in the

THIN database to 259,008 in the SIDIAP database (Supplementary Table 1). Unadjusted incidence

rates of AMI and stroke were higher in NAFLD versus non-NAFLD patients and differed across

databases: rates of AMI were highest in IPCI (4.36 [95% CI: 3.66 to 5.15] and 3.17 [3.11 to 3.24] per

1,000 person-years in NAFLD and non-NAFLD patients respectively); whilst rates of stroke were

highest in HSD (7.88 [95% CI: 7.39 to 8.39] and 6.27 [6.22 to 6.32] per 1,000 person-years

respectively) (Supplementary Table 1). In the patients with a diagnosis of NAFLD, number of incident

AMI events ranged from 137 (in IPCI) to 414 (in SIDIAP); and stroke events from 156 (in IPCI) to 962

in HSD.

Page 9 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


10

To investigate whether these associations were modified by common CVD risk factors, we defined

subsets of patients in whom we had data on total cholesterol and systolic blood pressure (“subset

1”); and patients in whom we additionally held data on BMI and HDL-C (“subset 2”). The total

number of patients with NAFLD in “subset 1” was 86,098 and in “subset 2” it was 61,718 patients;

who experienced respectively 747 and 542 MI events, and 1,666 and 1,223 stroke events during

follow-up. (Supplementary Table 2). Patients in subsets were more likely to have type 2 diabetes,

hypertension, and were more likely to be prescribed statins and be current smokers compared to

the entire NAFLD cohort in each database (Supplementary Table 3).

Hazard ratios for Incident AMI

When adjusting for age, sex, and smoking, the hazard ratio for incident AMI in patients with NAFLD

ranged from 1.03 [95% CI: 0.90 to 1.18] in HSD to 1.31 [95% CI: 1.16 to 1.49] in THIN; the pooled

hazard ratio estimate was 1.17 ([95% CI: 1.05 to 1.30], I-squared: 66%, p-value heterogeneity [p-

het]=0.032) (Figure 1). When analyses were adjusted for systolic blood pressure, type 2 diabetes,

total cholesterol, statin use, and hypertension the association between NAFLD and incident AMI was

abolished (fully adjusted hazard ratio 1.01 [0.91 to 1.12], I-squared=48.4%, p-het=0.121). Further

adjustments for BMI and HDL-cholesterol (in subset 2) also revealed no association between NAFLD

and AMI (HR: 0.96 ([0.84 to 1.10], I-squared=52.7%, p=0.096) (Supplementary Figure 1). Excluding

NASH patients did not alter the lack of association between NAFLD and AMI (Supplementary Figure

2). In subgroup analyses, pooled HRs did not significantly differ according to history of type 2

diabetes, hypertension, smoking status, age group, obesity, and sex (although estimates were

slightly higher in females compared to males) (Supplementary Figure 3).

Hazard ratios for Incident stroke

For the minimally adjusted model (adjusted for age and smoking) the pooled hazard ratio for

incident stroke was 1.18 [95% CI: 1.11 to 1.24] with low levels of heterogeneity across databases (I-

squared: 29.3% and p=0.236) (Figure 2). In subset 1, the pooled hazard ratio for stroke were

attenuated at 1.04 ([0.99 to 1.09], I-squared=0.0%, p-het=0.922) after adjustment for type 2

diabetes, systolic blood pressure, total cholesterol, statin prescription, and hypertension. Similarly,

in subset 2, the association between NAFLD and stroke became 1.05 ([0.99 to 1.11], I-squared: 0.0%,

p-het=0.959) (Supplementary Figure 4). Excluding NASH patients; associations between NAFLD and

incident stroke were unchanged (Supplementary Figure 5). Subgroup analyses did not identify any

Page 10 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


11

significant differences; although again the hazard ratio was marginally higher for females versus

males (1.15 versus 1.04) (Supplementary Figure 6).

Sensitivity analyses including recurrent AMI and stroke events, and including patients with less than

6 months of follow-up, did not produce materially different HRs (Supplementary Figures 7-8).

Discussion

Our real-world primary care record study of 205,046 cardiovascular events from 120, 795 adult

patients and 9,647,644 matched-unexposed individuals show that a recorded diagnosis of

NAFLD/NASH is more weakly associated with any excess AMI and stroke risk beyond associated risk

factors than previously thought. In the current study, age and sex adjusted hazard ratio was around

1.2 rather than 1.6 to 2.0-fold reported in recent meta-analyses of previous cohorts.[9,10] When we

adjusted for other covariates, the hazard ratio moved increasingly towards the null for both AMI and

stroke. These important data suggest that a diagnosis of NAFLD in routine clinical practice across

Europe does not necessarily indicate the need for AMI or stroke preventive therapies. Rather, our

results suggest CVD risk should be assessed in these individuals in the normal way using risk scores

with no need to consider an additional risk multiplier. This means for NAFLD patients to be identified

at high risk, the coexistence of other robust risk factors (e.g. diabetes or hypertension, dyslipidaemia

etc.) is required, which a reasonable proportion will have and such risk factors should be addressed

as per usual guideline recommendations. This is analogous to the situation for pre-diabetes where

CVD risk should be based on usual CVD risk scores without a risk multiplier.[22]

Our study represents the largest multi-cohort study on NAFLD patients to date. Due to the large

scale of the databases, we could match each patient with a recorded diagnosis of NAFLD to several

matched-unexposed individuals who did not have this diagnostic record, from the same GP practice,

sex, and same age ±5 years as the NAFLD patients. We conducted our study concurrently in four

European databases holding primary care data that have been extensively used for research, each

one of them with multiple publications, and all part of the EU-ADR Alliance which conducts post-

marketing safety studies mandated by regulatory authorities.[13–16,18] Furthermore, other

important diagnoses have been validated in these databases, for example AMI,[23] strengthening

our belief that these EHRs capture recording of clinical diagnoses in primary care.

Why should our routine care data show differential – weaker – associations of NAFLD with an excess

incident AMI or stroke outcomes over and above associated risk factors – as compared to meta-

Page 11 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


12

analyses of AMI or stroke event data in prior observation cohorts?[9,10] One potential is that, unlike

prior observation cohorts, we were comprehensive in our adjustment for routine risk factors (when

these were available). We also adjusted using continuous rather than categorical measures, as well

as considered current lipid-lowering therapies. Unexposed patients were matched according to the

GP practice which limited confounding by social class, something cohort studies have also rarely

considered. Social class is an often overlooked but important confounder since there are clear

gradients in obesity and diabetes risks by social class,[24] factors which predict differences in NAFLD

occurrence, suggesting NAFLD is also strongly socially patterned. Social class is also a strong

predictor of AMI or stroke events and, accordingly, is now included in several validated risk

scores.[25,26] Cardiovascular and NAFLD risk vary by ethnicity, but we were unable to assess this in

the current study because these data are not held in HSD, SIDIAP or IPCI.[27] Although progressive

multivariate adjustment took into account confounding for several potential factors (smoking,

medical history, obesity), and despite our extensive matching, there may still be residual

confounding due to other factors which we have not included here, but these may work to increase

rather nullify hazards. Moreover, the size of the cohorts goes some way to counterbalance this

potential bias.

We excluded patients with established diagnoses if other chronic liver conditions including alcoholic

liver disease. We also excluded patients with a coded diagnosis of ‘alcohol abuse’ in order to

exclude patients whose liver disease was very likely to be driven, at least in part, by alcohol. Alcohol

consumption is difficult to determine accurately in clinical practice and is therefore unreliably

recorded in routine care records. A recent major alcohol study combining 83 prospective cohorts in

which alcohol consumption was carefully evaluated and recorded, did not show an overall lower risk

for total CVD with alcohol.[28] Non-fatal AMI risk was slightly lower (HR 0.94 for 100g/ week higher

alcohol consumption), and the risk of all other adverse vascular outcomes including stroke (HR 1.14

for 100g/ week higher alcohol consumption) was higher. Hence, if the NAFLD/NASH patients

included in the current study were consuming moderate amounts of alcohol more often than the

matched-unexposed individuals, our results for AMI may have been biased towards the null, but

stroke risk should have been biased the other way. That the HRs associated with alcohol are modest

and that results for our two main outcomes of AMI and stroke show broadly consistent results,

however, suggest any confounding is likely minimal.

We have recently used these databases to show that the recorded prevalence of NAFLD is much

lower than expected, with less than 1% of the total number of patients registered in the databases

Page 12 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


13

having a recorded diagnosis of NAFLD.[17] Therefore, it is possible that we did not identify a

representative sample of patients with NAFLD. Even if cohort studies have over-estimated the

prevalence of NAFLD in the general population, there are likely to be many patients who actually do

have NAFLD but for whom the diagnosis has not been made or recorded, who we included in our

matched-unexposed population. However, by matching patients with NAFLD to unexposed patients

at a rate of 100:1, any such effects should be significantly diluted. That noted, the characteristics of

the NAFLD patients identified in our study are extremely consistent with published cohort studies. In

a recent meta-analysis of 86 cohort studies in 22 countries, metabolic comorbidities associated with

NAFLD included obesity (51.3%; 95% CI: 41.4 to 61.2), type 2 diabetes (22.5%; 95% CI: 17.9 to 27.9),

and hypertension (39.3%; 95% CI: 33.2 to 45.9).[29] In our summary data in Table 2, average BMI

was over 30kg/m2 in three of the EHRs for NAFLD, average diabetes percentages were around 19%

and average proportion with hypertension was around 40%, results near identical to this

aforementioned meta-analysis, lending strong external validity to our cohort make-up and hence

CVD results. We therefore believe that those with coded NAFLD/NASH have been correctly identified

as they have all the hallmarks of the condition and at levels near identical to those proven to have

NAFLD using imaging techniques. We accept a proportion in the unexposed population will have

undiagnosed NAFLD but these are diluted out by others without NAFLD: evidenced by the average

control characteristics. We also accept that we cannot determine how doctors made the diagnosis of

NAFLD in each case, but these data suggest that those identified as having NAFLD, did indeed have

NAFLD.[17] From a practical point of view, it is not possible to apply a cardiovascular risk multiplier

(if appropriate) to a particular condition in people who have not been diagnosed with that condition.

Therefore, despite the low prevalence, our data represent the reality of AMI and stroke risk in

people with recorded diagnoses of NAFLD/NASH.

To limit heterogeneity across studies, we harmonized code lists for all clinical events and ensured

that codes in multiple terminologies all mapped to the same UMLS concepts. After local data

extraction, data were formatted and analysed in the same way by a single analyst for the four

databases[13–16] on the EMIF remote server. However, we still observed significant heterogeneity

across studies, which was only partially accounted for by progressive adjustment. This is probably

due to significant differences between healthcare systems of the 4 countries (e.g. NAFLD being

diagnosed at a more advanced stage in THIN) as well as database specificities (terminology used to

record NAFLD and outcomes). Variation in the methods used to diagnose NAFLD/NASH[30,31] and

the extent to which coding was completed also contributes a degree of heterogeneity. We recognise

heterogeneity for data linked to incident AMI where data for HSD seemed at odds with the other

Page 13 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


14

three cohorts. We were careful to recheck these findings and they appeared robust. Sensitivity

analyses on incident AMI including only the other three more congruent cohorts, however, shows

similarly low hazards for AMI in patients with NAFLD. Hence, our conclusions remain the same

whether we include data from HSD or not.

In conclusion, associations of an existing and recorded diagnosis of NAFLD in routine EHRs with both

incident AMI and stroke were modest in age and sex adjusted models and were rendered non-

significant with further adjustment for usual CVD risk factors. Of course, a diagnosis of NAFLD does

warrant risk assessment for the stage of liver disease, and behaviour and lifestyle advice not only to

lessen liver fat, but also because of the myriad benefits of weight loss on AMI and stroke risk factors

including systolic blood pressure and the chance of developing diabetes (if not known to have

diabetes). Among the large numbers of patients who have NAFLD, some, perhaps many, may be at

heightened risk of AMI and stroke outcomes, however further study is needed to identify such

patients and quantify that risk, if it exists. For the time being, we suggest that patients who are

currently diagnosed with NAFLD should not automatically be considered to be at increased risk of

AMI or stroke. That noted, cardiovascular risk assessment in patients diagnosed with NAFLD is

important to conduct if not already done (as is testing for diabetes). However, for the time being

such cardiovascular risk assessment should be conducted in the same way as in the general

population and without a risk multiplier.

Page 14 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


15

Contributors:

Study Design: MA, AKL, JvdL, PA, PE, SK, DW, WA, NS,

Extracted data: TDS, DP-A, DA, AP, FL, PR (data transformation and federated data analysis), MM

Analysed Data: MA, ND, CC-M

Interpreted results: All authors

Wrote manuscript: NS, MA, WA,

Edited manuscript: All authors

Approved for submission: All authors

Guarantors: NS, WA

Acknowledgments: The European Medical Information Framework (EMIF) is a collaboration

between industry and academic partners that aims to develop common technical and governance

solutions to facilitate access to diverse electronic medical and research data sources. The authors

would like to acknowledge Nicholas Galwey for his advice on the statistical methods, Alba Jene for

her administrative support and support during submission to ethical review boards, and Derek Nunez

for support early on a protocol design stage.

Funding: European Federation of Pharmaceutical Industries Associations (EFPIA), Innovative

Medicines Initiative Joint Undertaking, European Medical Information Framework (EMIF) grant

number 115372. WA holds a New Investigator Research Award from the Medical Research Council.

The funders had no role in study design, data collection, data analysis, data interpretation, writing of

the report, or in the decision to submit the article for publication. The researchers were independent

from the funding source and all authors had full access to all of the data (including statistical reports

and tables) in the study and can take responsibility for the integrity of the data and the accuracy of

the data analysis.

Competing interests: All authors have completed the ICMJE uniform disclosure form at

www.icmje.org/coi_disclosure.pdf and declare: JvdL, TD-S, AP, PR, PA, CCM report no support from

any organisation for the submitted work. MA reports contracted by SRG to work for GlaxoSmithKline

and was receiving a salary from GSK, including bonus. JF-B, DW are paid employees at

GlaxoSmithKline and are receiving salaries, including bonuses. AKL is a paid employee at Pfizer and is

receiving a salary, including bonus. DP-A reports unrestricted research grants from UCB, Amgen,

Page 15 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


16

Servier, and consultancy fees from UCB Pharma paid to his department. DA reports as a paid

employee of IQVIA has provided consultancy and advice to many pharmaceutical companies on

undertaking outcomes studies using real world evidence. PE, SK report they are paid employees and

stock holders of GlaxoSmithKline. NS reports personal fees from AstraZeneca, Amgen, Boehringer

Ingelheim, Eli Lilly, Novo Nordisk, Janssen, Sanofi and grant from Boehringer Ingelheim. WA reports

consultancy and sponsored lectures from Gilead, GlaxoSmithKline, Intercept IQVIA and UCB Pharma.

Ethical approval: Not required

Data sharing: No additional data available

Transparency: The lead authors (WA and NS) affirm that the manuscript is an honest, accurate, and

transparent account of the study being reported; that no important aspects of the study have been

omitted; and that any discrepancies from the study as originally planned (and, if relevant, registered)

have been explained.

Page 16 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


17

References1 Targher G, Day CP, Bonora E. Risk of cardiovascular disease in patients with nonalcoholic fatty

liver disease. N Engl J Med 2010;363:1341–50. doi:10.1056/NEJMra09120632 Bhatia LS, Curzen NP, Calder PC, et al. Non-alcoholic fatty liver disease: a new and important

cardiovascular risk factor? Eur Heart J 2012;33:1190–200. doi:10.1093/eurheartj/ehr4533 Mellinger JL, Pencina KM, Massaro JM, et al. Hepatic steatosis and cardiovascular disease

outcomes: An analysis of the Framingham Heart Study. J Hepatol 2015;63:470–6. doi:10.1016/j.jhep.2015.02.045

4 Salvi P, Ruffini R, Agnoletti D, et al. Increased arterial stiffness in nonalcoholic fatty liver disease: the Cardio-GOOSE study. J Hypertens 2010;28:1699–707. doi:10.1097/HJH.0b013e32833a7de6

5 Lee Y-J, Shim J-Y, Moon B-S, et al. The Relationship Between Arterial Stiffness and Nonalcoholic Fatty Liver Disease. Dig Dis Sci 2012;57:196–203. doi:10.1007/s10620-011-1819-3

6 Cai J, Zhang S, Huang W. Association between nonalcoholic fatty liver disease and carotid atherosclerosis: a meta-analysis. Int J Clin Exp Med 2015;8:7673–8.

7 Sookoian S, Pirola CJ. Non-alcoholic fatty liver disease is strongly associated with carotid atherosclerosis: A systematic review. J Hepatol 2008;49:600–7. doi:10.1016/j.jhep.2008.06.012

8 Ong JP, Pitts A, Younossi ZM. Increased overall mortality and liver-related mortality in non-alcoholic fatty liver disease. J Hepatol 2008;49:608–12. doi:10.1016/j.jhep.2008.06.018

9 Musso G, Gambino R, Cassader M, et al. Meta-analysis: natural history of non-alcoholic fatty liver disease (NAFLD) and diagnostic accuracy of non-invasive tests for liver disease severity. Ann Med 2011;43:617–49. doi:10.3109/07853890.2010.518623

10 Targher G, Byrne CD, Lonardo A, et al. Non-alcoholic fatty liver disease and risk of incident cardiovascular disease: A meta-analysis. J Hepatol 2016;65:589–600. doi:10.1016/j.jhep.2016.05.013

11 Ghouri N, Preiss D, Sattar N. Liver enzymes, nonalcoholic fatty liver disease, and incident cardiovascular disease: a narrative review and clinical perspective of prospective data. Hepatology 2010;52:1156–61. doi:10.1002/hep.23789

12 Hippisley-Cox J, Coupland C, Vinogradova Y, et al. Predicting cardiovascular risk in England and Wales: Prospective derivation and validation of QRISK2. BMJ 2008;336:1475–82. doi:10.1136/bmj.39609.449676.25

13 Gini R, Francesconi P, Mazzaglia G, et al. Chronic disease prevalence from Italian administrative databases in the VALORE project: a validation through comparison of population estimates with general practice databases and national survey. BMC Public Health 2013;13:15. doi:10.1186/1471-2458-13-15

14 Vlug AE, van der Lei J, Mosseveld BM, et al. Postmarketing surveillance based on electronic patient records: the IPCI project. Methods Inf Med 1999;38:339–44.

15 García-Gil MDM, Hermosilla E, Prieto-Alhambra D, et al. Construction and validation of a scoring system for the selection of high-quality data in a Spanish population primary care database (SIDIAP). Inform Prim Care 2011;19:135–45.

16 Blak BT, Thompson M, Dattani H, et al. Generalisability of The Health Improvement Network (THIN) database: demographics, chronic disease prevalence and mortality rates. Inform Prim Care 2011;19:251–5.

17 Alexander M, Loomis AK, Fairburn-Beech J, et al. Real-world data reveal a diagnostic gap in non-alcoholic fatty liver disease. BMC Med 2018;16:130. doi:10.1186/s12916-018-1103-x

18 Avillach P, Coloma PM, Gini R, et al. Harmonization process for the identification of medical events in eight European healthcare databases: the experience from the EU-ADR project. J Am Med Inform Assoc 2013;20:184–92. doi:10.1136/amiajnl-2012-000933

19 Strong M, Maheswaran R, Pearson T. A comparison of methods for calculating general

Page 17 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


18

practice level socioeconomic deprivation. Int J Health Geogr 2006;5:29. doi:10.1186/1476-072X-5-29

20 Whitehead A. Meta-Analysis Of Controlled Clinical Trials. Chichester, UK: : John Wiley & Sons, Ltd 2002. doi:10.1002/0470854200

21 Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med 2002;21:1539–58. doi:10.1002/sim.1186

22 Vistisen D, Witte DR, Brunner EJ, et al. Risk of Cardiovascular Disease and Death in Individuals With Prediabetes Defined by Different Criteria: The Whitehall II Study. Diabetes Care 2018;:dc172530. doi:10.2337/dc17-2530

23 Coloma PM, Valkhoff VE, Mazzaglia G, et al. Identification of acute myocardial infarction from electronic healthcare records using different disease coding systems: a validation study in three European countries. BMJ Open 2013;3:e002862. doi:10.1136/bmjopen-2013-002862

24 Tang KL, Rashid R, Godley J, et al. Association between subjective social status and cardiovascular disease and cardiovascular risk factors: a systematic review and meta-analysis. BMJ Open 2016;6:e010137. doi:10.1136/bmjopen-2015-010137

25 Woodward M, Brindle P, Tunstall-Pedoe H, et al. Adding social deprivation and family history to cardiovascular risk assessment: the ASSIGN score from the Scottish Heart Health Extended Cohort (SHHEC). Heart 2007;93:172–6. doi:10.1136/hrt.2006.108167

26 JBS3 Board. Joint British Societies’ consensus recommendations for the prevention of cardiovascular disease (JBS3). Heart 2014;100:ii1-ii67. doi:10.1136/heartjnl-2014-305693

27 Alazawi W, Mathur R, Abeysekera K, et al. Ethnicity and the diagnosis gap in liver disease: a population-based study. Br J Gen Pract J R Coll Gen Pract 2014;64:e694-702. doi:10.3399/bjgp14X682273

28 Wood AM, Kaptoge S, Butterworth AS, et al. Risk thresholds for alcohol consumption: combined analysis of individual-participant data for 599 912 current drinkers in 83 prospective studies. Lancet (London, England) 2018;391:1513–23. doi:10.1016/S0140-6736(18)30134-X

29 Younossi ZM, Koenig AB, Abdelatif D, et al. Global epidemiology of nonalcoholic fatty liver disease-Meta-analytic assessment of prevalence, incidence, and outcomes. Hepatology 2016;64:73–84. doi:10.1002/hep.28431

30 van Asten M, Verhaegh P, Jonkers D, et al. A survey on non-alcoholic fatty liver disease amongst general practitioners: time to bridge the gap between hepatologists and primary care. J Hepatol 2017;66:S411. doi:10.1016/S0168-8278(17)31181-9

31 Sheridan DA, Aithal G, Alazawi W, et al. Care standards for non-alcoholic fatty liver disease in the United Kingdom 2016: a cross-sectional survey. Frontline Gastroenterol 2017;8:252–9. doi:10.1136/flgastro-2017-100806

Page 18 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


19

Figure legends

Figure 1: Hazard ratios (95% CI) for myocardial infarction in patients with NAFLD.

Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age,

smoking status, type 2 diabetes, systolic blood pressure, total cholesterol, statin use and

hypertension. Data for age and smoking status was available for 120,795 patients with a diagnosis of

NAFLD and 9,647,644 matched “unexposed” patients. The subset* analyses was restricted to those

participants with data available for age, smoking, diabetes diagnosed, systolic blood pressure, total

cholesterol, statin medication and hypertension (respectively 86,098 NAFLD and 4,664,988 non-

exposed patients).

Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of

the estimated hazard ratios (therefore proportional to the number of events contributing the hazard

ratios).

Note: statin imputed as missing in THIN

Figure 2: Hazard ratios (95% CI) for stroke in patients with NAFLD.

Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age,

smoking status, type 2 diabetes, systolic blood pressure, total cholesterol, statin use and

hypertension. Data for age and smoking status was available for 120,795 patients with a diagnosis of

NAFLD and 9,647,644 matched “unexposed” patients. The subset* analyses was restricted to those

participants with data available for age, smoking, diabetes diagnosed, systolic blood pressure, total

cholesterol, statin medication and hypertension (respectively 86,098 NAFLD and 4,664,988 non-

exposed patients). Note: Weights are from random-effect meta-analysis and inversely proportional

to the variance of the estimated hazard ratios (therefore proportional to the number of events

contributing the hazard ratios). Note: statin imputed as missing in THIN.

Page 19 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


20

Table 1: Attrition table

Attrition criteria HSD (Italy) IPCI (The Netherlands)

SIDIAP (Spain)

THIN (UK) Total

a) Total ever enrolled by 31/12/2015

1,542,672 2,225,925 5,488,397 12,695,046 21,952,040

b) Total adults with ≥1 year enrolment from registration

1,544,573 1,780,500 5,259,575 9,085,325 17,669,973

c) NAFLD patients after exclusion of individuals with a history of alcohol abuse, number (%)

NAFLD: 24,027(1.56%)

NAFLD: 18,865(1.06%)

NAFLD: 77,107(1.47%)

NAFLD:23,385(0.26%)

NAFLD: 143,384(0.81%)

d) NAFLD patients after exclusion because of less than 6 months of follow-up post NAFLD diagnosis, number (%)

NAFLD: 23,131(1.50%)

NAFLD: 15,669

(incident patients post registration

into IPCI database)

(0.88%)

NAFLD: 71,672(1.36%)

NAFLD:21,039(0.23%)

NAFLD: 131,511(0.74%)

e) NAFLD patients after exclusion if less than 6 months of medical history prior to NAFLD diagnosis, number (%)

NAFLD: 22,708(1.47%)

NAFLD: 13,386(0.75%)

NAFLD: 69,451(1.32%)

NAFLD:20,346(0.22%)

NAFLD: 125,891(0.71%)

f) NAFLD patients after exclusion if history of MI or stroke, number (%)

NAFLD: 21,627(1.40%)

NAFLD: 12,595(0.71%)

NAFLD: 67,109(1.28%)

NAFLD:19,464(0.21%)

NAFLD: 120,795(0.68%)

g) Number of matched unexposed patients (ratio unexposed / exposed) after applying all exclusion criteria

Non-NAFLD:

1,707,510 (ratio: 79)

Non-NAFLD: 1,207,378 (ratio: 96)

Non-NAFLD: 4,830,700 (ratio: 72)

1,902,056 (ratio: 98)

Non-NAFLD: 9,647,644

Denominators for all percentages are values in row b): Total adults with ≥1 year enrolment from

registration.

Page 20 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


21

Table 2: Descriptive characteristics of individuals with NAFLD and matched patients in four European primary care databases

HSD - Italy IPCI – The Netherlands SIDIAP-Spain THIN - UKCharacteristicsNAFLD Matched non-

NAFLDNAFLD Matched

non- NAFLDNAFLD Matched

non- NAFLDNAFLD Matched non-

NAFLDFollow-up prior to index date: Median (IQR)

7.5 (4.7-10.4) 7.6 (4.8 – 10.4) 2.5 (1.4-3.9) 2.5 (1.4 – 3.9)

5.1 (3.1 – 6.8)

5.1 (3.1 – 6.8)

13.4 (5.8 – 22.9) 14.3 (6.6 – 23.2)

Follow-up post index date: Median (IQR)

5.5 (3.0 – 8.1)

5.4 (3.0 – 8.1) 2.1 (1.2 – 3.4)

2.2 (1.2 – 3.4)

3.7 (2.0 – 5.6)

3.7 (2.0 – 5.7)

3.5 (1.8 – 6.1) 3.5 (1.8 – 6.1)

Age in years, mean (SD) 55.6 (14.2) 54.6 (13.5) 56.1 (13.6) 55.6 (13.3) 55.6 (13.3) 54.2 (12.9) 53.3 (13.1) 52.9 (13.2)

Gender, % of Males 57.2% 54.9% 48.6% 48.1% 52.5% 48.8% 51.1% 50.4%Current smokers, % 11.3% 9.1% 17.2% 11.1% 17.8% 15.4% 17.3% 18.7%Body Mass Index in kg/m2, mean (SD)

29.7 (5.0) 27.5 (5.0) 31.0 (5.4) 28.3 (5.2) 31.4 (5.1) 28.7 (5.1) 32.4 (5.9) 28.5 (5.9)

History of Type 2 diabetes, %

17.0% 10.7% 19.8% 8.6% 19.4% 9.9% 20.1% 6.5%

History of hypertension, % 46.2% 35.7% 34.6% 25.0% 42.0% 28.3% 40.0% 24.8%Aspartate transaminase (IU/L), median (IQR)

24 (19 – 33) 20.7 (17 – 25) 29 (22 - 40) 23 (20 – 28) 29 (22 – 40) 21 (18 – 27) 32 (24 – 47) 22 (19 – 27)

Alanine transaminase (IU/L), median (IQR)

30 (20 – 49) 21 (16 – 30) 37 (25 – 56) 25 (18 – 33) 35 (23 – 54) 20 (15 – 28) 46 (29 – 69) 23 (17 – 31)

Total cholesterol (mmol/l), N; mean (SD)

5.41 (1.06) 5.43 (1.03) 5.31 (1.16) 5.35 (1.10) 5.40 (1.01) 5.37 (0.97) 5.23 (1.24) 5.16 (1.16)

HDL-cholesterol (mmol/l), N; mean (SD)

1.31 (0.34) 1.43 (0.38) 1.21 (0.31) 1.36 (0.36) 1.27 (0.32) 1.42 (0.37) 1.25 (0.36) 1.43 (0.77)

Systolic blood pressure (mmHg), N; mean (SD)

132.8 (15.2) 131.7 (15.7 138.2 (17.5) 136.7 (17.7) 131.7 (13.6) 129.2 (14.2) 134.3 (14.8) 131.9 (15.8)

*After imputation of missing as non-smokers. N: Number of individuals. For laboratory values, we exclude outlier values greater than mean + 3 X SD (mean and SD computed separately in NAFLD and non-NAFLD separately).

Page 21 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Figure 1: Hazard ratios (95% CI) for myocardial infarction in patients with NAFLD.

Page 22 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


23

Figure 2: Hazard ratios (95% CI) for stroke in patients with NAFLD.

Page 23 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


1

SUPPLEMENTARY MATERIAL

Supplementary Figure 1. Hazard ratios for myocardial in patients with NAFLD.

Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for

age, smoking status, type 2 diabetes, SBP, total cholesterol, statin use, hypertension, BMI and

HDL-cholesterol. Data was available on a subset of 61,718 patients with a diagnosis of

NAFLD and 2,970,649 matched “unexposed” patients. Note: Weights are from random-effect

meta-analysis and inversely proportional to the variance of the estimated hazard ratios

(therefore proportional to the number of events contributing the hazard ratios).

Page 24 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


2

Supplementary Figure 2. Hazard ratios for myocardial infarction in NAFLD patients without a NASH records only. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking status, type 2 diabetes, SBP, total cholesterol, statin use and hypertension.

Note: it was only possible to run this analysis in the SIDIAP and THIN databases as separate codes are available for NAFLD and NASH.

Page 25 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


3

Supplementary Figure 3. Hazard ratio for myocardial infarction in subgroup and pooled by multivariate meta-analysis. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

Data is presented as hazard ratio and their 95% CI. Analyses were adjusted for age, smoking status, SBP, total cholesterol. Estimates were pooled by random effects meta-analysis within each subgroup.

Page 26 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


4

Supplementary Figure 4. Hazard ratio for stroke in patients with NAFLD. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking status, type 2 diabetes, SBP, total cholesterol, statin use, hypertension, BMI and HDL-cholesterol. Data was available on a subset of 61,718 patients with a diagnosis of NAFLD and 2,970,649 matched “unexposed” patients.

Page 27 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


5

Supplementary Figure 5. Hazard ratio for stroke in NAFLD patients without a NASH records. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).


Note: it was only possible to run this analysis in the SIDIAP and THIN databases as separate codes are available for NAFLD and NASH.

Page 28 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


6

Supplementary Figure 6. Hazard ratio for stroke in NAFLD patients without a NASH records by subgroup and pooled across databases by multivariate meta-analysis. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

Data is presented as hazard ratio and their 95% CI. Analyses adjusted for age, smoking status, SBP and total cholesterol. Estimates were pooled by random effects meta-analysis within each subgroup.

Page 29 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


7

Supplementary Figure 7. Sensitivity analyses - Hazard ratio for myocardial infarction with in NAFLD patients including patients with less than 6 months of medical history prior and follow-up post index date, or who had a history of stroke or MI. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).


Page 30 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


8

Supplementary Figure 8. Sensitivity analyses for hazard ratio for stroke in patients with NAFLD including patients with less than 6 months of medical history prior and follow-up post index date, or who had a history of stroke or MI. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).


Page 31 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


9

Supplementary Figure 9. Association of NAFLD/NASH with myocardial infarction excluding patients with less than 6 months follow-up post index date (excluding events happening in the first 6 months after index date). Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking status, type 2 diabetes, SBP, total cholesterol, statin use and hypertension. Data for age and smoking (total population data set) was available for 59,881 (patients without NAFLD n=58,970; patients with NAFLD n=911). A subset* of participants have full data available for age, smoking, type 2 diabetes, SBP, total cholesterol, statin use and hypertension, therefore the analyses were restricted to 32,481 (Non NAFLD patients n=31,829; patients with NAFLD n=652).

Page 32 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


10

Supplementary Table 1: Incidence rate of MI and stroke in four primary care databases

Categories HSD IPCI SIDIAP THIN Overall

Acute myocardial infarction, NAFLD patients Total number of person-years 124,525 31,426 259,008 85,361 500,320 Number of events 221 137 414 263 1,035 Incidence rate (95%CI) per 1,000 person-years

1.77(1.55: 2.02)

4.36(3.66: 5.15)

1.6(1.45: 1.76)

3.08(2.72: 3.48)

2.07 (1.94: 2.20)

Acute myocardial infarction, non-NAFLD patients Total number of person-years 9,728,567 3,032,175 18,700,000 8,379,073 39,839,815 Number of events 15,014 9,625 23,238 19,946 67,823 Incidence rate (95%CI) per 1,000 person-years

1.54(1.52: 1.57)

3.17(3.11: 3.24)

1.24 (1.23: 1.26)

2.38(2.35: 2.41)

1.70 (1.69: 1.71)

Stroke, NAFLD patients Total number of person-years 122,105 31,422 258,006 85,467 497,000 Number of events 962 156 854 215 2,187 Incidence rate (95%CI) per 1,000 person-years

7.88(7.39: 8.39)

4.96 (4.22: 5.81)

3.31 (3.09: 3.54)

2.52(2.19: 2.88)

4.40 (4.22: 4.59)

Stroke, non-NAFLD patients Total number of person-years 9,586,232 3,030,972 18,700,000 8,393,764 39,710,968 Number of events 60,082 11,902 45,658 16,359 134,001 Incidence rate (95%CI) per 1,000 person-years

6.27(6.22: 6.32)

3.93(3.86: 4)

2.45 (2.42: 2.47)

1.95(1.92: 1.98)

3.37 (3.35: 3.39)

Data presented as incidence rate and their 95% confidence intervals (CI). Overall incidence rates are estimated by dividing the total number of events by the total number of person-years. 95% CI are estimated using an exact Poisson model.

Page 33 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


11

Supplementary Table 2: Description of subsets used in statistical models

Number of patients Number of MI events

Number of stroke events

Sample subsetNAFLD

Matched non-

NAFLD NAFLD

Matched non-

NAFLD NAFLD

Matched non-

NAFLD

a) HSD databaseWhole Study Population 21,627 1,707,510 221 15,014 962 60,082

Subset 1 12,647 662,099 126 7,329 719 37,606

Subset 2 7,609 353,306 83 3,969 479 21,763

With data on BMI 10,837 581,981 114 6,186 603 29,169

b) IPCI databaseWhole Study Population 12,595 1,207,378 137 9,625 156 11,902

Subset 1 6,977 438,582 90 4,704 101 6,059

Subset 2 4,710 270,454 68 2,986 74 4,097

With data on BMI 5,685 335,158 79 3,697 92 4,975

c) SIDIAP databaseWhole Study Population 67,109 4,830,700 414 23,134 854 45,605

Subset 1 52,188 2,728,743 334 14,877 702 31,539

Subset 2 38,227 1,775,342 245 10,743 569 23,645

With data on BMI 46,599 2,469,013 290 13,749 657 29,694

c) THIN databaseWhole Study Population 19,464 1,902,056 263 19,946 215 16,359

Subset 1 14,286 835,564 197 10,496 144 8,656

Subset 2 11,172 571,547 146 6,887 101 5,557

With data on BMI 15,096 993,974 202 11,406 149 9,476

d) All databases combinedWhole Study Population 120,795 9,647,644 1,035 67,719 2,187 133,948

Subset 1 86,098 4,664,988 747 37,406 1,666 83,860

Subset 2 61,718 2,970,649 542 24,585 1,223 55,062

With data on BMI 78,217 4,380,126 685 35,038 1,501 73,314Subset 1 includes individuals with information on total cholesterol, SBP and history of hypertension. Subset 2 includes individuals with information on total cholesterol, SBP, history of hypertension, BMI and HDL-cholesterol.

Page 34 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


12

Supplementary Table 3: Descriptive characteristics of patients in the four databases in the whole study population and in subsets

Subset of patients with data on …

… Total cholesterol, SBP

and history of

hypertension

… Total cholesterol, SBP,

history of hypertension, BMI

and HDL-cholesterol.

… FIB4 … BMICharacteristics

Matched controls NAFLD Matched

controls NAFLD Matched controls NAFLD Matched

controls NAFLD

a) HSD database

Mean age in years (SD)

60.1 (11.6)

58.8 (12.9)

60.0 (11.3)

58.4 (12.8)

58.3 (12.9)

57.8 (13.8)

57.2 (12.7)

56.5 (13.6)

Gender, Males % 49% 53% 50% 55% 50% 55% 52% 56%

Current smoker, % 13% 13.40% 17% 17% 11% 12% 20% 18%History of Type 2 diabetes, % 17% 22% 21% 25% 15% 19% 17% 22%

History of hypertension, % 55% 58% 57% 59% 45% 49% 47% 52%

Statin use, % 24% 25% 27% 27% 21% 23% 20% 23%

b) IPCI database


61.6 (10.7)

59.2 (12.2)

62.5 (10.2)

60.0 (11.8)

58.8 (12.8)

55.9 (13.8)

61.4 (11.1)

59.0 (12.5)

Gender, Males % 45% 47% 45% 46% 43% 48% 45% 46%

Current smoker, % 17% 20% 20% 23% 16% 19% 21% 23%History of Type 2 diabetes, % 20% 32% 32% 44% 13% 21% 27% 39%


Statin use, % 39% 45% 49% 53% 28% 31% 42% 47%

c) SIDIAP database


57.9 (12.1)

56.8 (13.0)

60.0 (11.5)

58.1 (12.8)

56.7 (12.4)

55.0 (13.0)

57.6 (12.5)

57.2 (13.1)

Gender, Males % 43% 50% 42% 49% 45% 53% 44% 50%

Current smoker, % 18% 19% 19% 20% 17% 18% 20% 20%History of Type 2 diabetes, % 15% 23% 21% 28% 14% 20% 17% 26%

History of 40% 47% 49% 53% 35% 42% 42% 50%

Page 35 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


13

hypertension, %

Statin use, % 32% 37% 38% 40% 31% 34% 32% 38%

d) THIN database


58.4 (11.1)

55.2 (12.2)

58.6 (11.0)

55.5 (12.1)

56.5 (12.0)

53.7 (12.7)

55.1 (12.8)

53.9 (13.0)

Gender, Males % 50% 51% 49% 50% 44% 49% 45% 50%

Current smoker 18% 18% 19% 18% 21% 18% 21% 18%History of Type 2 diabetes, % 14% 27% 19% 31% 12% 23% 12% 25%


Note that the percentage of current smokers is estimated after imputation of patients with missing smoking status as non-current smokers.

Page 36 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Figure 1: Hazard ratios (95% CI) for myocardial infarction in patients with NAFLD. Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking status, type 2 diabetes, systolic blood pressure, total cholesterol, statin use and hypertension. Data for age and smoking status was available for 120,795 patients with a diagnosis of NAFLD and 9,647,644 matched

“unexposed” patients. The subset* analyses was restricted to those participants with data available for age, smoking, diabetes diagnosed, systolic blood pressure, total cholesterol, statin medication and hypertension

(respectively 86,098 NAFLD and 4,664,988 non-exposed patients). Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the

estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios). Note: statin imputed as missing in THIN

60x81mm (300 x 300 DPI)

Page 37 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Page 38 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Figure 2: Hazard ratios (95% CI) for stroke in patients with NAFLD. Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking status, type 2 diabetes, systolic blood pressure, total cholesterol, statin use and hypertension. Data for age and smoking status was available for 120,795 patients with a diagnosis of NAFLD and 9,647,644 matched

“unexposed” patients. The subset* analyses was restricted to those participants with data available for age, smoking, diabetes diagnosed, systolic blood pressure, total cholesterol, statin medication and hypertension (respectively 86,098 NAFLD and 4,664,988 non-exposed patients). Note: Weights are from random-effect

meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios). Note: statin imputed as missing in

THIN.

60x81mm (300 x 300 DPI)

Page 39 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 1. Hazard ratios for myocardial in patients with NAFLD. Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking status, type 2 diabetes, SBP, total cholesterol, statin use, hypertension, BMI and HDL-cholesterol. Data was

available on a subset of 61,718 patients with a diagnosis of NAFLD and 2,970,649 matched “unexposed” patients. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

60x52mm (300 x 300 DPI)

Page 40 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 2. Hazard ratios for myocardial infarction in NAFLD patients without a NASH records only. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the

estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios). Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking

status, type 2 diabetes, SBP, total cholesterol, statin use and hypertension. Note: it was only possible to run this analysis in the SIDIAP and THIN databases as separate codes are

available for NAFLD and NASH.

58x48mm (300 x 300 DPI)

Page 41 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 3. Hazard ratio for myocardial infarction in subgroup and pooled by multivariate meta-analysis. Note: Weights are from random-effect meta-analysis and inversely proportional to the

variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

Data is presented as hazard ratio and their 95% CI. Analyses were adjusted for age, smoking status, SBP, total cholesterol. Estimates were pooled by random effects meta-analysis within each subgroup.

101x72mm (300 x 300 DPI)

Page 42 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 4. Hazard ratio for stroke in patients with NAFLD. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore

proportional to the number of events contributing the hazard ratios). Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking status, type 2 diabetes, SBP, total cholesterol, statin use, hypertension, BMI and HDL-cholesterol. Data was

available on a subset of 61,718 patients with a diagnosis of NAFLD and 2,970,649 matched “unexposed” patients.

60x50mm (300 x 300 DPI)

Page 43 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 5. Hazard ratio for stroke in NAFLD patients without a NASH records. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard

ratios (therefore proportional to the number of events contributing the hazard ratios). Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking

status, type 2 diabetes, SBP, total cholesterol, statin use and hypertension. Note: it was only possible to run this analysis in the SIDIAP and THIN databases as separate codes are

available for NAFLD and NASH.

58x48mm (300 x 300 DPI)

Page 44 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 6. Hazard ratio for stroke in NAFLD patients without a NASH records by subgroup and pooled across databases by multivariate meta-analysis. Note: Weights are from random-effect meta-

analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

Data is presented as hazard ratio and their 95% CI. Analyses adjusted for age, smoking status, SBP and total cholesterol. Estimates were pooled by random effects meta-analysis within each subgroup.

100x70mm (300 x 300 DPI)

Page 45 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 7. Sensitivity analyses - Hazard ratio for myocardial infarction with in NAFLD patients including patients with less than 6 months of medical history prior and follow-up post index date, or who had a history of stroke or MI. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the

hazard ratios). Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking

status, type 2 diabetes, SBP, total cholesterol, statin use and hypertension.

58x62mm (300 x 300 DPI)

Page 46 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 8. Sensitivity analyses for hazard ratio for stroke in patients with NAFLD including patients with less than 6 months of medical history prior and follow-up post index date, or who had a history

of stroke or MI. Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the

hazard ratios). Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking

status, type 2 diabetes, SBP, total cholesterol, statin use and hypertension.

54x57mm (300 x 300 DPI)

Page 47 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960


Supplementary Figure 9. Association of NAFLD/NASH with myocardial infarction excluding patients with less than 6 months follow-up post index date (excluding events happening in the first 6 months after index

date). Note: Weights are from random-effect meta-analysis and inversely proportional to the variance of the estimated hazard ratios (therefore proportional to the number of events contributing the hazard ratios).

Data is presented as hazard ratio and their 95% CI. Analyses were progressively adjusted for age, smoking status, type 2 diabetes, SBP, total cholesterol, statin use and hypertension. Data for age and smoking (total

population data set) was available for 59,881 (patients without NAFLD n=58,970; patients with NAFLD n=911). A subset* of participants have full data available for age, smoking, type 2 diabetes, SBP, total cholesterol, statin use and hypertension, therefore the analyses were restricted to 32,481 (Non NAFLD

patients n=31,829; patients with NAFLD n=652).

59x61mm (300 x 300 DPI)

Page 48 of 47


BMJ

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960

recorded diagnosis of nafld and the independent risk of · 9department of medical informatics,...

Documents