Mortality in Biopsy-Confirmed Nonalcoholic Fatty Liver Disease

Objective: Population-based data are lacking regarding the risk of overall and cause-specific mortality across the complete histological spectrum of nonalcoholic fatty liver disease (NAFLD). Design: This nationwide, matched cohort study included all individuals in Sweden with biopsy-confirmed NAFLD (1966–2017; n=10,568). NAFLD was confirmed histologically from all liver biopsies submitted to Sweden’s 28 pathology departments, after excluding other etiologies of liver disease, and further categorized as, simple steatosis, non-fibrotic steatohepatitis (NASH), non-cirrhotic fibrosis and cirrhosis. NAFLD cases were matched to ≤5 general population comparators by age, sex, calendar year and county (n=49,925). Using Cox regression, we estimated multivariable-adjusted hazard ratios (aHRs) and 95%CIs. Results: Over a median of 14.2 years, 4,338 NAFLD patients died. Compared to controls, NAFLD patients had significantly increased overall mortality (16.9 vs. 28.6/1000 person-years [PY]; difference=11.7/1000PY; aHR=1.93, 95%CI=1.86–2.00). Compared to controls, significant excess mortality risk was observed with simple steatosis (8.3/1000PY, aHR=1.71, 95%CI=1.64–1.79), non-fibrotic NASH (13.4/1000PY, aHR=2.14, 95%CI=1.93–2.38), non-cirrhotic fibrosis (18.4/1000PY, aHR=2.44, 95%CI=2.22–2.69) and cirrhosis (53.6/1000PY, aHR=3.79, 95%CI=3.34–4.30)(Ptrend<0.01). This dose-dependent gradient was similar when simple steatosis was the reference (Ptrend<0.01). The excess mortality associated with NAFLD was primarily from extra-hepatic cancer (4.5/1000PY; aHR=2.16, 95%CI=2.03–2.30), followed by cirrhosis (2.7/1000PY; aHR=18.15, 95%CI=14.78–22.30), cardiovascular disease (1.4/1000PY; aHR=1.35, 95%CI=1.26–1.44) and hepatocellular carcinoma (HCC)(1.2/1000PY; aHR=11.12, 95%CI=8.65–14.30). Conclusions: All NAFLD histological stages were associated with significantly increased overall mortality, and this risk increased progressively with worsening NAFLD histology. Most of this excess mortality was from extra-hepatic cancer and cirrhosis, while in contrast, the contributions of cardiovascular disease and HCC were modest.

. Construction of the Nationwide Matched Cohort

Definition of NAFLD and NAFLD Histological Categories
Consistent with nationwide liver histopathology reporting recommendations provided to all pathologists in Sweden 1 , we defined NAFLD from histopathology reports using an established algorithm of SNOMED topography and morphology codes, after excluding other etiologies of liver disease. Specifically, patients were identified by a liver biopsy histopathology report that included a topography code for liver (T56), and at least one morphology code for steatosis (M5008x or M5520x), consistent with our validation study in ESPRESSO (see below). In Sweden, clinicallyindicated liver biopsies are conducted with a single pass of the liver, unless a satisfactory specimen is not obtained with the initial pass. Furthermore, consistent with Swedish liver histopathology reporting recommendations, it is documented if any biopsy is too short (i.e. <15mm in length), contains fewer than 5 portal tracts, or is fragmented 1 .
Among individuals with multiple liver biopsies that demonstrated steatosis, we included the first such biopsy. We excluded anyone with at least one primary diagnosis or the second of two secondary diagnoses of another etiology of liver disease (all outlined in Table S1; i.e. alcoholrelated liver disease, drug-induced liver disease, viral hepatitis B or C infection, Budd-Chiari, liver abscess, HIV/AIDS, autoimmune hepatitis, primary biliary cholangitis, other cholangitis, alpha-1 antitrypsin deficiency, glycogen storage disease), or a primary diagnosis or the second or two secondary diagnoses of alcohol abuse/misuse or drug abuse. Additionally, we further excluded anyone with use of a steatogenic medication or a medication specifically used to treat an alternative etiology of liver disease (as per Table S1). Medication use was defined by the Prescribed Drug Register as a filled prescription for at least 30 cumulative defined daily doses (cDDD) of a given medication, at any time prior to the index date (or corresponding matching date), with the exception of systemic steroids (for which a person was excluded if they used systemic steroids within 0 to 3 months prior to the index date).
All patients who met our criteria for NAFLD were further classified at the time of their index biopsy into one of 4 mutually exclusive categories of histological severity (i.e. simple steatosis, NASH without fibrosis, non-cirrhotic fibrosis and cirrhosis), using validated definitions (see Methods for our validation study). Specifically, simple steatosis was defined by at least 1 code for steatosis and no additional codes for inflammation (i.e. M5400x or M4-) or fibrosis (i.e. M4900x) or cirrhosis (i.e. M4950x). NASH without fibrosis was defined broadly by the presence of at least 1 code for steatosis plus at least 1 code for inflammation (i.e. M5400x or M4-), without any codes for fibrosis or cirrhosis. Non-cirrhotic fibrosis (i.e. F1-F3 fibrosis, with or without NASH) was defined by the presence of at least 1 code for steatosis plus at least 1 code for fibrosis (i.e. M4900x), but no codes for cirrhosis. Cirrhosis was defined by at least 1 code for cirrhosis (i.e. M4950x).

Definitions of Secondary Outcomes
The primary outcome was all-cause mortality (see Methods for details). Secondary outcomes included specific causes of mortality, ascertained from the Cause of Death Register. We categorized cause-specific mortality in 5 groups. Cardiovascular-specific mortality was defined from the Cause of Death Register by a primary ICD code for cardiovascular disease (ICD-8 / 9: 390-459; ICD-10: I01-I99). Hepatocellular carcinoma (HCC)-specific mortality was defined from the Cause of Death Register by a primary ICD code for HCC (ICD-7: 155.0, 155,0, 1550; ICD9: 155A, 155.0; ICD-10: C22.0, C22.9, C22.99). Cirrhosis-specific mortality was defined as death from any non-HCC cause of liver disease or liver failure, defined by a primary ICD code for cirrhosis, liver failure, decompensated liver disease (i.e. variceal bleeding, spontaneous bacterial peritonitis, encephalopathy or hepatorenal syndrome), or death from liver transplantation, as outlined in Table S2. Cancer-specific mortality was defined as death from any non-HCC cancers, defined from the Cause of Death Register by a primary ICD code for any cancer (ICD7 / 8: 140-239; ICD10: C00-C99; with the exclusion of any HCC-specific ICD code). All other deaths were classified as, "other causes of mortality".

Definitions of Covariates and Construction of Multivariable Models
Definitions of clinical, demographic and prescription medication covariates are outlined in the Methods and in Table S2. Since 1990, the longitudinal integrated database for health insurance and labour market studies (LISA) database 2 has prospectively recorded and annually-updated detailed data from the Swedish labor market, and also from the educational and social sectors, for all Swedish residents aged 16 years and older. Education level, a proxy for socioeconomic status, was obtained from the validated LISA database 2 .
For the primary analysis, our main multivariable model adjusted for matching factors (i.e. age at the index date/corresponding matching date, sex, calendar year and county in Sweden) as well as the following a priori defined prognostic covariates: education level, cardiovascular disease, and the components of the metabolic syndrome (i.e. diabetes, obesity, hypertension and dyslipidemia).
We also conducted a series of sensitivity analyses, to further address potential residual confounding. In one analysis, we repeated the primary analysis after restricting the cohort to persons with an index date on or after January 1, 2006, and then constructed an additional multivariable-adjusted model that further accounted for the following a priori time-varying medication covariates: low-dose aspirin (<163mg dosage), statin medications, metformin, other antidiabetic medications and anti-hypertensive medications (as outlined in Table S2). For each medication, use was defined from the Prescription Drug Register, as at least 30 cumulative defined daily doses (cDDD) of filled prescriptions for that medication from a Swedish pharmacy, and this information was updated over each monthly interval of study follow-up.
In an additional sensitivity analysis, we constructed a separate multivariable model that further adjusted for a modified Charlson Comorbidity Index (CCI) score. This modified CCI score was constructed specifically to incorporate known or putative confounders not previously included in our main multivariable model, and thereby further address potential residual confounding. This modified CCI score included: Chronic Obstructive Pulmonary Disease (COPD), rheumatic or connective tissue disease (i.e. lupus, polymyalgia rheumatica, polymyositis, mixed connective tissue disease, and/or moderate-to-severe rheumatoid arthritis), Peptic ulcer disease, Hemiplegia, moderate-to-severe chronic kidney disease (CKD), solid tumor, Leukemia and Lymphoma. We excluded conditions that already were accounted for in the main multivariable model (i.e. age, myocardial infarction, congestive heart failure, peripheral vascular disease, stroke or transient ischemic attack). We also did not include acquired immune deficiency syndrome (AIDS), because we previously had excluded anyone with HIV/AIDS from this analysis. Finally, because liver disease severity could represent intermediates that lie on the causal pathway between NAFLD and the study outcome (mortality), and because those factors were already accounted for by our assessments of NAFLD histological severity, we did not include them in the modified CCI Score.

Missing Data
Data were missing for education level in <2.5% of adults with an index date on or after January 1,1990, which was the first year in which the LISA database was available 2 . This included 2.1% of population comparators and 2.3% of NAFLD patients, as outlined in Table 1. For our multivariable models, we included a separate missing category for missingness on education level.
The nationwide Swedish registers provide complete information on liver biopsies and deaths including cause of death, as mandated by the government, and the validity and completeness of clinical comorbidities are well-established (see Methods). Despite this, missing data do arise, primarily from under-reporting. For example, if an individual does not have a coded diagnosis for obesity, this would be classified as "no obesity", however the absence of a diagnostic code for obesity could be due to either (1) a true negative (i.e. the individual does not have obesity), or (2) obesity was not recorded. Thus, consistent with prior work 3 , no other data were considered missing, with the sole exception of education level. Abbreviations: ICD, International Classification of Disease; ATC, anatomic therapeutic chemical classification *For study flowchart outlining exclusion criteria, please see Figure S1 1 We excluded any person with a first recorded primary diagnosis or the second of two secondary diagnoses for another etiology of liver disease, or alcohol abuse/misuse or alcohol-related liver disease, defined on or prior to the index date. 2 We excluded any person with use of a steatogenic medication or a medication used to treat an alternative etiology of liver disease. Medication use was defined by a filled prescription for at least 30 cumulative defined daily doses (cDDD), in the Prescribed Drug Register, at any time prior to the index date (or corresponding matching date), with the exception of systemic steroids (for which a person was excluded if they used systemic steroids within 0 to 3 months prior to the index date). As the Prescribed Drug Register began on July 1, 2005, these exclusion criteria were only applied to the subgroup analysis in which the cohort was restricted to persons with an index biopsy (or corresponding matching date) on or after January 1, 2006.

Diabetes
Hospital discharge letters containing ICD-8 / ICD-9: 250 either a primary or a secondary ICD-10: E10.0-E14.9 diagnosis for diabetes or outpatient specialty Medication  Abbreviations: HCC, hepatocellular carcinoma; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis; aHR, adjusted hazard ratio; CI, confidence interval; ref., referent *For each specific cause of death, competing events included death from any other cause, as outlined in the Methods. Because HCC-specific mortality was assessed separately, cancer mortality did not include deaths from HCC; similarly, cirrhosis-related mortality also did not include deaths from HCC (for details, see eMethods).