Cancer Risk in Patients With Biopsy-Confirmed Nonalcoholic Fatty Liver Disease: A Population-Based Cohort Study

Abstract

Recent studies link nonalcoholic fatty liver disease (NAFLD) to an increased incidence of hepatocellular carcinoma (HCC) and extrahepatic cancers. However, prior studies were small or lacked liver histology, which remains the gold standard for staging NAFLD severity. We conducted a population-based cohort study of all adults with histologically defined NAFLD in Sweden from 1966 to 2016 (N=8,892). NAFLD was defined from prospectively recorded liver histopathology submitted to all 28 Swedish pathology departments and categorized as simple steatosis, nonfibrotic nonalcoholic steatohepatitis (NASH), noncirrhotic fibrosis, and cirrhosis. NAFLD patients were individually matched to ≤5 general population controls without NAFLD by age, sex, calendar year, and county (N=39,907). Using Cox proportional hazards modeling, we calculated multivariable adjusted hazard ratios (aHRs) and 95% confidence intervals (CIs). Over a median of 13.8 years, we documented 1,691 incident cancers among NAFLD patients and 6,733 among controls. Compared with controls, NAFLD patients had significantly increased overall cancer incidence (10.9 vs. 13.8/1,000 person-years [PYs]; difference=2.9/1,000 PYs; aHR, 1.27 [95%CI, 1.18–1.36]), driven primarily by HCC (difference=1.1/1,000 PYs; aHR, 17.08 [95%CI, 11.56–25.25]). HCC incidence rates increased monotonically across categories of simple steatosis, nonfibrotic NASH, noncirrhotic fibrosis, and cirrhosis (0.8/1,000 PYs, 1.2/1,000 PYs, 2.3/1,000 PYs, and 6.2/1,000 PYs, respectively) (P_trend<0.01) and were further amplified by diabetes (1.2/1,000 PYs, 2.9/1,000 PYs, 7.2/1,000 PYs, and 15.7/1,000 PYs, respectively). In contrast, NAFLD was associated with modestly increased rates of pancreatic cancer, kidney/bladder cancer, and melanoma (differences=0.2/1,000 PYs, 0.1/1,000 PYs, and 0.2/1,000 PYs, respectively) but no other cancers. Compared with controls, patients with biopsy-proven NAFLD had significantly increased cancer incidence, due primarily to HCC, whereas the contribution of extrahepatic cancers was modest. Although HCC risk was highest with cirrhosis, substantial excess risk was also found with noncirrhotic fibrosis and comorbid diabetes.

Nonalcoholic fatty liver disease (NAFLD) represents a leading cause of chronic liver disease in the United States and Europe, affecting more than 100 million adults.^(1,2) Between 20% and 33% of patients with NAFLD develop progressive nonalcoholic steatohepatitis (NASH) with fibrosis, which in turn can lead to cirrhosis, decompensated liver disease, and liver-related mortality.^(3–7) More recent observational studies have also linked NAFLD to an increased risk of developing cancer.^(8,9) Among patients with NAFLD, mortality rates from both hepatocellular carcinoma (HCC) and extrahepatic cancers have increased by an alarming 2% to 4% per year over the past decade.⁽¹⁰⁾ In contrast, within the general population, overall cancer mortality has declined by 2% to 6% per year over the same period of time,^(11–13) highlighting the need to improve our understanding of cancer risk and identify opportunities for cancer prevention among patients with NAFLD.

Evidence regarding the precise risk of cancer among patients with NAFLD is limited. Although several studies have linked NAFLD with a 4- to 8-fold higher risk of incident HCC^(8,14,15) and a 2-fold higher risk of some extrahepatic cancers,^{(8,9,15–18)} others have found null associations.^(19–22) Moreover, studies have been limited either by small populations with poor generalizability or have lacked liver histology, which remains the gold standard for identifying NASH and staging NAFLD fibrosis severity. Thus, large-scale epidemiological studies leveraging liver histology are necessary to determine if cancer risks vary according to the presence and histological severity of NAFLD. Moreover, quantifying the magnitude of these risks is important for developing effective cancer prevention and screening strategies for this rapidly growing population.

We hypothesized that the excess risk of overall cancer in patients with NAFLD is driven primarily by incident HCC and that this risk increases progressively with worsening NAFLD histological severity. Thus, we constructed a nationwide cohort of all adults in Sweden with biopsy-confirmed NAFLD to examine the risk of incident hepatic and extrahepatic cancers according to the presence and histological severity of NAFLD.

Methods

Study Population and NAFLD Ascertainment

We conducted a population-based, matched cohort study using the Epidemiology Strengthened by histoPathology Reports in Sweden (ESPRESSO) cohort. ESPRESSO includes prospectively recorded liver histopathology data from all 28 Swedish pathology departments (1966–2016) and thus is complete for the entire country.^(23,24) Each report includes a unique personal identity number (PIN) and biopsy date and describes topography within the liver and morphology using the Systematized Nomenclature of Medicine (SNOMED) system. We linked ESPRESSO to validated, nationwide registers with data regarding demographics, comorbidities, prescribed medications, incident cancers, and death. ESPRESSO was approved by the Stockholm Regional Ethics Committee on August 27, 2014 (#2014/1287–31/4). Informed consent was waived because the study was register-based.⁽²⁵⁾

Using a validated algorithm,⁽²³⁾ we identified all liver histopathology specimens between 1966 and 2016 that confirmed a diagnosis of NAFLD from adults aged ≥18 years without another competing etiology of liver disease (Supporting Methods; Supporting Tables S1 and S2). We excluded anyone with another form of liver disease, prior record of alcohol abuse/misuse, liver transplantation, or emigration from Sweden before the liver biopsy date (i.e., the index date). Consistent with prior publications,^(26,27) we also excluded anyone with a cancer diagnosis at baseline, except nonmelanoma skin cancer, which generally has little clinical impact and therefore is not expected to be associated with recall bias or influence lifestyle changes. Additionally, we excluded anyone with <90 days of follow-up. In our previous validation study in this cohort (Supporting Methods), this methodology yielded a positive predictive value (PPV) of 92% for NAFLD.⁽²³⁾

Patients meeting our criteria for NAFLD were subsequently categorized into four histological groups (simple steatosis, NASH without fibrosis, noncirrhotic fibrosis, cirrhosis) using Systematized Nomenclature of Medicine (SNOMED) definitions based on nationwide reporting recommendations for all pathologists in Sweden⁽²⁸⁾ and which we have previously validated in this cohort.⁽²³⁾ (Supporting Methods) Briefly, simple steatosis was defined by at least one morphology code for steatosis and no additional codes for inflammation/ballooning (M5400x or M4-) or fibrosis (M4900x) or cirrhosis (M4950x). NASH without fibrosis was defined by at least one code for steatosis plus at least one code indicating inflammation or ballooning (M5400x or M4-) without any codes for fibrosis or cirrhosis. Noncirrhotic fibrosis (F1-F3 fibrosis, with or without NASH) was defined by the presence of at least one code for steatosis plus at least one code for fibrosis (M4900x) but no codes for cirrhosis. Cirrhosis was defined by at least one code for cirrhosis (M4950x).

Each NAFLD patient was individually matched to ≤5 general population controls without recorded NAFLD according to age, sex, calendar year, and county of residence, consistent with prior work.⁽²³⁾ Controls were derived from the Total Population Register,⁽²⁹⁾ and identical exclusion criteria were applied (Supporting Table S1).

Incident overall cancer was ascertained from the Swedish Cancer Register, which is complete for >96% of all cancers⁽³⁰⁾ and in which all cases are confirmed and classified by specialists using established histopathological or radiographic criteria. Secondary outcomes included specific cancer types (HCC, non-HCC liver cancer, and extrahepatic cancers), which were further subclassified as hematologic cancers and extrahepatic solid organ (EHSO) cancers (cancer of the prostate, breast, esophagus/stomach, colon/rectum, lung, pancreas, kidney/bladder, cervical/ovarian/uterine, melanoma, and a composite group of all others). All-cause mortality was ascertained from the Total Population Register, which records 93% of all deaths within 10 days and the remaining 7% within 30 days.⁽³¹⁾

We collected detailed information regarding demographics, comorbidities, and medication use (Supporting Methods; Supporting Table S3). Briefly, age at the index date (i.e., date of biopsy among NAFLD patients or the corresponding matching date for controls), sex, date of birth, and emigration status were ascertained from the Total Population Register,⁽²⁹⁾ and education level was obtained from the Longitudinal Integrated Database for Health Insurance and Labour Market Studies (LISA).⁽³²⁾ Comorbidities were ascertained using established ICD algorithms extracted from the validated Patient Register, which prospectively records data from hospitalizations (including surgeries), discharge diagnoses (since 1964), and specialty outpatient care (since 2001), with PPVs between 85% and 95% (²⁵) (for ICD definitions, see Supporting Methods; Supporting Table S3). Medication data, including statins, low-dose aspirin (<163 mg), antidiabetic medications, and use of angiotensin-converting enzyme (ACE) inhibitors and angiotensin II receptor blockers (ARBs), were ascertained from the Prescribed Drug Register, a well-validated and virtually complete register with prospectively recorded data regarding all dispensed prescriptions from Swedish pharmacies that has been available since July 2005.⁽³³⁾

Statistical Analysis

Our primary analysis examined the incidence of overall cancer according to the presence and histological severity of NAFLD compared with matched controls. Follow-up began ≥90 days after the index biopsy date and continued to the first diagnosis date of incident cancer, death, emigration, or end of follow-up (December 31, 2016).

Unadjusted incidence rates and absolute rate differences (RDs), with 95% confidence intervals (CIs), were calculated by dividing the number of events by the number of person-years (PYs). We also calculated unadjusted 20-year absolute risks and corresponding risk differences, with 95% CIs approximated by the normal distribution. Using Cox proportional hazard models, we estimated multivariable adjusted hazard ratios (aHRs) for incident cancers, accounting for matching factors (age at the index date, sex, calendar year, and county) and also for covariates defined up to and including the index date (cardiovascular disease, diabetes, hypertension, dyslipidemia, obesity, end-stage renal disease, family history of cancer at age <50 years, education, the number of recorded hospital encounters in the year prior to the index date), and time-varying alcohol abuse/misuse, consistent with prior work.^(23,24) (Supporting Methods; Supporting Table S3) The proportional hazards assumption was tested by visually examining and formally testing the association between Schoenfeld residuals and time.

In stratified models, we evaluated the associations between NAFLD and the development of cancer according to risk strata defined by age, sex, index date, follow-up time, Nordic vs. non-Nordic birth country, education level, cardiovascular disease, diabetes, hypertension, dyslipidemia, and the metabolic syndrome, and we tested the significance of effect modification using the log likelihood ratio test. To better characterize the gradient of risk associated with progressive NAFLD and to minimize potential misclassification of controls or bias related to the indication for liver biopsy, we excluded controls and restricted the analysis to patients with histologically defined NAFLD, with simple steatosis as the comparator.

We conducted several sensitivity analyses to test the robustness of our results. First, to address potential overestimation of cancer incidence in the setting of competing events, we accounted for competing risks of noncancer mortality.⁽³⁴⁾ Second, to more carefully account for differences in clinical comorbidities between NAFLD patients and population comparators at the index date, we propensity score (PS)-matched NAFLD patients to population controls without NAFLD in a 1:1 manner on the index date using a nearest-neighbor PS algorithm and included age at the index date (±2 years), sex, calendar year (±2 years), county of residence, cardiovascular disease, diabetes, hypertension, dyslipidemia, obesity, and the metabolic syndrome. After applying this algorithm, 100% of the 8,982 NAFLD patients were successfully matched to 8,982 population controls.

Third, because a widely used histological scoring system for NAFLD was published in 2005,⁽³⁵⁾ the year prescription medication data were first available in the Prescribed Drug Register, we restricted the cohort to patients with an index date on or after January 1, 2006, and constructed multivariable models further accounting for time-varying use of relevant medications, including aspirin, statins, antidiabetic drugs, and ACE inhibitors or ARBs (Supporting Methods). Fourth, to address potential confounding from shared genetic or early environmental factors, we restricted the cohort to NAFLD patients with ≥1 full sibling without recorded NAFLD⁽²³⁾ and compared each NAFLD patient with his or her sibling(s) after conditioning on matching set within each family (i.e. 1 stratum per family) and further adjusting for all covariates. Fifth, we compared our findings after applying an alternative, broader definition of HCC⁽³⁶⁾ and, again, after excluding anyone diagnosed with cancer within ≤2 years. We also repeated our primary analysis using a control exposure with an expected positive association (the metabolic syndrome). Finally, using an array-based approach,⁽³⁷⁾ we tested the sensitivity of our models to potential unmeasured confounding.

Analyses were conducted using R software (version 3.6.1 and survival package version 2.44, R Foundation for Statistical Computing, Vienna, Austria). P < 0.05 was considered statistically significant.

Results

Table 1 outlines the baseline characteristics of 8,892 adults with histologically confirmed NAFLD and 39,907 matched population controls. Among NAFLD patients, the average age at index biopsy was 51 years, and 42% were female. Simple steatosis was found in 5,939 (66.8%), 1,050 (11.8%) had NASH without fibrosis, 1,400 (15.7%) had noncirrhotic fibrosis, and 503 (5.7%) had cirrhosis. NAFLD patients were more likely than controls to have cardiovascular disease, diabetes, hypertension, and dyslipidemia.

Table 1.

Baseline Characteristics of Adults With Histologically Defined NAFLD and Matched Population Comparators

Characteristic	Population Comparators N = 39,907	All NAFLD N = 8,892	Simple Steatosis N = 5,939	NASH Without Fibrosis N = 1,050	Noncirrhotic Fibrosis N = 1,400	Cirrhosis N = 503	P Value3
Male, %	56.7	58.3	59.4	54.0	57.4	57.3	0.02
Age at the index date, years (SD)	49.9 (14.4)	50.6 (14.5)	49.3 (14.4)	51.1 (14.8)	53.3 (14.0)	57.9 (12.1)	<0.001
Years of follow-up (IQR)	15.5 (8.6)	13.8 (8.5)	15.3 (8.5)	12.1 (7.9)	10.3 (7.4)	8.9 (7.3)	<0.001
Start of follow-up, %							0.10
1966–1989	8,274 (20.7)	1,784 (20.1)	1,388 (23.4)	145 (13.8)	139 (9.9)	112 (22.3)	NA
1990–2000	18,404 (46.1)	4,050 (45.6)	2,933 (49.4)	439 (41.8)	463 (33.1)	215 (42.7)	NA
2001–2010	9,449 (23.7)	2,159 (24.3)	1,218 (20.5)	314 (29.9)	516 (36.9)	111 (22.1)	NA
2011–2016	3,780 (9.5)	899 (10.1)	400 (6.7)	152 (14.5)	282 (20.1)	65 (12.9)	NA
Nordic country of birth (%)	36,343 (91.1)	7,919 (89.1)	5,328 (89.7)	920 (87.6)	1219 (87.1)	452 (89.9)	<0.001
Highest education level1 (%)	(among n = 31,633)	(among n = 7,108)	(among n = 4,551)	(among n = 905)	(among n = 1,261)	(among n = 391)	<0.001
≤9 years	8,625 (27.3)	2,093 (29.5)	1,307 (28.7)	259 (28.6)	375 (29.7)	152 (38.9)	NA
10–12 years	13,394 (42.3)	3,251 (45.7)	2,079 (45.7)	426 (47.1)	568 (45.0)	178 (45.5)	NA
≥13 years	9,228 (29.2)	1,699 (23.9)	1,132 (24.9)	210 (23.2)	305 (24.2)	52 (13.3)	NA
Unknown	386 (1.2)	65 (0.9)	33 (0.7)	10 (1.1)	13 (1.0)	9 (2.3)	NA
Cardiovascular disease (%)	1,939 (4.9)	961 (10.8)	542 (9.1)	129 (12.3)	197 (14.1)	93 (18.5)	<0.001
Dyslipidemia (%)	1,066 (2.7)	531 (6.0)	215 (3.6)	86 (8.2)	179 (12.8)	51 (10.1)	<0.001
Diabetes (%)	703 (1.7)	687 (7.7)	303 (5.1)	96 (9.1)	194 (13.9)	94 (18.7)	<0.001
Hypertension (%)	1,770 (4.4)	1,040 (11.7)	466 (7.9)	156 (14.9)	304 (21.7)	114 (22.7)	<0.001
Obesity (%)	59 (0.1)	136 (1.5)	67 (1.5)	19 (1.8)	31 (2.2)	19 (3.8)	<0.001
Metabolic syndrome2 (%)	215 (0.5)	207 (2.3)	66 (1.1)	37 (3.5)	67 (4.8)	37 (7.4)	<0.001
Number of hospital encounters in the year prior to the index date (SD)	0.4 (1.6)	2.6 (4.0)	2.3 (3.3)	3.0 (3.8)	3.7 (6.4)	2.8 (3.4)	<0.001
Family history of cancer at age <50 years (%)	2,284 (5.7)	511 (5.8)	328 (5.5)	78 (7.4)	84 (6.0)	21 (4.2)	<0.001

All variables reported as mean (SD) or % unless described otherwise. For definitions of the NAFLD histological groups and all covariates, see Supporting Methods and Supporting Table S3.

¹Education categories based on compulsory school, high school, and college (see Supporting Methods). Education level was recorded beginning in 1990; therefore, data presented are for persons with index dates on or after January 1, 1990.

²Metabolic syndrome was defined as ≥3 metabolic risk factors (dyslipidemia, diabetes, hypertension, and/or obesity), as outlined in the Methods section and Supporting Table S3.

³P values signify the comparison between all NAFLD patients versus population comparators.

Abbreviations: IQR, interquartile range; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis; SD, standard deviation; NA, not applicable.

Overall Cancer Incidence

Over a median of 13.8 years, we documented 1,691 incident cancers among NAFLD patients (13.8/1,000 PYs) and 6,733 incident cancers among controls (10.9/1,000 PYs), yielding an absolute RD of 2.9/1,000 PYs and a 20-year absolute risk difference of 5.4% (95% CI, 3.7–7.0) (Table 2). After multivariable adjustment, the aHR for incident overall cancer was 27% higher among patients with NAFLD compared with controls (95% CI, 1.18–1.36) (Fig. 1; Table 2). This significant, positive association was similar among women and men and across all clinical risk strata (all P_{heterogeneity} > 0.05) (Supporting Table S4).

TABLE 2.

Risk of Incident Cancer Among Adults With Histologically Confirmed NAFLD and Matched Population Comparators

	Population Comparators N = 39,907	NAFLD*
Incident Cancer		All NAFLD N = 8,892	Simple Steatosis N = 5,939	NASH Without Fibrosis N = 1,050	Noncirrhotic Fibrosis N = 1,400	Cirrhosis N = 503
Overall cancers, N.	6,733	1,691	1,160	187	233	111
Incidence rate1 per 1,000 PYs (95% CI)	10.9 (10.6–11.2)	13.8 (13.2–14.5)	12.8 (12.1–13.6)	14.7 (12.7–17.0)	16.2 (14.1–18.4)	24.8 (20.4–29.9)
Incidence RD1 (95% CI)	0 (ref.)	2.9 (2.2–3.7)	1.9 (1.1–2.7)	3.8 (1.7–6.0)	5.3 (3.2–7.4)	14.0 (9.3–18.6)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	5.4 (3.7–7.0)	3.4 (1.6–5.2)	7.8 (2.5–13.2)	10.0 (4.4–15.7)	20.7 (9.3–32.1)
Multivariable aHR3 (95% CI)	1 (ref.)	1.27 (1.18–1.36)	1.22 (1.12–1.32)	1.27 (1.03–1.56)	1.35 (1.13–1.61)	1.95 (1.48–2.55)
P value3	NA	<0.001	<0.001	0.024	0.001	<0.001
HCC, N.	56	153	72	16	35	30
Incidence rate1 per 1,000 PYs (95% CI)	0.1 (0.1–0.1)	1.2 (1.0–1.4)	0.8 (0.6–1.0)	1.2 (0.7–1.9)	2.3 (1.6–3.2)	6.2 (4.2–8.8)
Incidence RD1 (95% CI)	0 (ref.)	1.1 (0.9–1.3)	0.7 (0.5–0.8)	1.1 (0.5–1.7)	2.2 (1.4–3.0)	6.1 (3.9–8.3)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	2.1 (1.7–2.5)	1.2 (0.8–1.6)	2.1 (0.7–3.6)	4.6 (2.7–6.5)	11.4 (6.5–16.3)
Multivariable aHR3 (95% CI)	1 (ref.)	17.08 (11.56–25.25)	9.25 (5.63–15.20)	55.66 (14.43–214.72)	72.72 (19.20–275.51)	49.48 (15.71–155.81)
P value3	NA	<0.001	<0.001	<0.001	<0.001	<0.001
Non-HCC liver cancers, N.	82	25	16	2	5	2
Incidence rate1 per 1,000 PYs (95% CI)	0.1 (0.1–0.2)	0.2 (0.1–0.3)	0.2 (0.1–0.3)	0.2 (0.0–0.5)	0.3 (0.1–0.8)	0.4 (0.1–1.5)
Incidence RD1 (95% CI)	0 (ref.)	0.1 (0.0–0.2)	0.0 (0.0–0.1)	0.02 (−0.2–0.2)	0.2 (−0.1–0.5)	0.3 (−0.3–0.9)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	0.1 (0.0–0.3)	0.1 (−0.1–0.3)	0.1 (−0.4–0.6)	0.2 (−0.3–0.7)	0.8 (−0.8–2.4)
Multivariable aHR3 (95% CI)	1 (ref.)	1.75 (0.98–3.14)	1.26 (0.62–2.58)	3.49 (0.36–34.00)	4.20 (1.05–16.86)	3.08 (0.35–27.48)
P value3	NA	0.062	0.535	0.261	0.048	0.175
EHSO cancers, N.	6,160	1,390	987	152	174	77
Incidence rate1 per 1,000 PYs (95% CI)	9.9 (9.7–10.2)	11.3 (10.7–11.9)	10.8 (10.2–11.5)	11.8 (10.0–13.9)	12.0 (10.3–13.9)	17.1 (13.5–21.3)
Incidence RD1 (95% CI)	0 (ref.)	1.4 (0.7–2.0)	0.9 (0.2–1.6)	1.9 (0.0–3.8)	2.0 (0.3–3.8)	7.1 (3.3–11.0)
20-year absolute risk ifference,2 % (95% CI)	0 (ref.)	3.0 (1.5–4.5)	1.9 (0.2–3.6)	5.2 (0.4–10.1)	4.7 (−0.2–9.5)	11.4 (2.0–20.7)
Multivariable aHR3 (95% CI)	1 (ref.)	1.12 (1.04–1.20)	1.11 (1.02–1.22)	1.11 (0.89–1.39)	1.05 (0.86–1.28)	1.41 (1.04–1.91)
P value3	NA	0.002	0.013	0.26	0.529	0.018
Hematologic cancers, N.	435	123	85	17	19	2
Incidence rate1 per 1,000 PYs (95% CI)	0.7 (0.6–0.7)	1.0 (0.8–1.1)	0.9 (0.7–1.1)	1.3 (0.7–2.0)	1.2 (0.8–1.9)	0.4 (0.0–1.5)
Incidence RD1 (95% CI)	0 (ref.)	0.3 (0.1–0.5)	0.2 (0.0–0.4)	0.6 (0.0–1.2)	0.6 (0.0–1.1)	−0.3 (−0.8–0.3)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	0.6 (0.2–1.0)	0.5 (0.0–0.9)	0.8 (−0.3–1.9)	1.8 (0.1–3.5)	−0.3 (−1.9–1.4)
Multivariable aHR3 (95% CI)	1 (ref.)	1.46 (1.12–1.90)	1.47 (1.06–2.03)	1.05 (0.51–2.15)	1.86 (0.96–3.62)	1.07 (0.19–5.93)
P value3	NA	0.004	0.014	0.874	0.083	0.652

^*NAFLD was defined by liver histology. For definitions and algorithm, see the Methods section and Supporting Information.

¹Incidence rates were calculated by dividing the number of cases by PYs; CIs for incidence rates and absolute RDs were approximated by the normal distribution (Methods).

²20-year absolute risks and absolute risk differences (percentage points) were calculated based on Kaplan-Meier estimates (Methods).

³The multivariable-adjusted model accounted for age at the index date, sex, calendar year, county of residence, cardiovascular disease, diabetes, hypertension, dyslipidemia, obesity, end-stage renal disease, family history of cancer at age <50 years, education (3 groups + missing category), the number of recorded hospital encounters in the year prior to the index biopsy date (or corresponding matching date), and alcohol abuse/misuse defined as a time-varying covariate. All clinical covariates were ascertained from the Patient Register using established ICD codes, and defined by the presence of ≥1 inpatient diagnosis or by the second of ≥2 outpatient diagnoses (as outlined in the Supporting Methods and Supporting Table S3). P values signify the comparison between each NAFLD category and population comparators.

Abbreviations: aHR, adjusted hazard ratio; CI, confidence interval; EHSO, extrahepatic solid organ; HCC, hepatocellular carcinoma; N., number; PYs, person-years; RD, rate difference; ref., referent.

FIG. 1.

Cumulative incidence of overall cancer according to the prevalence and histological severity of NAFLD. NAFLD histological severity was defined in four categories: simple steatosis, NASH without fibrosis, noncirrhotic fibrosis, and cirrhosis. (For details, see Supporting Methods.) P values for the absolute incidence RDs for overall cancer between population comparators versus simple steatosis, NASH without fibrosis, noncirrhotic fibrosis, and cirrhosis were all <0.001. P values were approximated using the normal distribution. Abbreviations: NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis; RDs, rate differences; ref., referent.

Rates of incident overall cancer increased progressively with worsening NAFLD histological severity (P_trend < 0.01) (Fig. 1; Table 2). Compared with controls, the absolute RDs and corresponding aHRs with simple steatosis, NASH without fibrosis, noncirrhotic fibrosis, and cirrhosis were 1.9/1,000 PYs (aHR, 1.22 [95% CI, 1.12–1.32]), 3.8/1,000 PYs (aHR, 1.27 [95% CI, 1.03–1.56]), 5.3/1,000 PYs (aHR, 1.35 [95% CI, 1.13–1.61]), and 14.0/1,000 PYs (aHR, 1.95 [95% CI, 1.48–2.55]), respectively. These findings were similar in men and women and across all clinical strata (all P_{heterogeneity} > 0.05).

HCC Incidence

Compared with controls, NAFLD patients had significantly higher rates of developing incident HCC (0.1 vs. 1.2/1,000 PYs; aHR, 17.08 [95% CI, 11.56–25.25]), corresponding to a 20-year absolute excess risk of 2.1% (Table 2). Rates of incident HCC also increased monotonically with worsening NAFLD histological severity (P_trend < 0.01) (Table 2; Fig. 2). Compared with controls, the absolute RDs and corresponding aHRs for incident HCC with simple steatosis, NASH without fibrosis, noncirrhotic fibrosis, and cirrhosis were 0.8/1,000 PYs (aHR, 9.25 [95% CI, 5.63–15.20]), 1.2/1,000 PYs (aHR, 55.66 [95% CI, 14.43–214.72]), 2.3/1,000 PYs (aHR, 72.72 [95% CI, 19.20–275.51]), and 6.2/1,000 PYs (aHR, 49.48 [95%CI, 15.71–155.81]), respectively. This significant, positive, dose-dependent gradient was similar among women and men (both P_trend < 0.01) and after further accounting for competing events (P_trend < 0.01) (Supporting Table S5).

FIG. 2.

Cumulative incidence of HCC according to the presence and histological severity of NAFLD. NAFLD histological severity was defined in four categories: simple steatosis, NASH without fibrosis, noncirrhotic fibrosis, and cirrhosis. (For details, see Supporting Methods.) P values for the absolute incidence RDs for incident HCC between population comparators versus simple fibrosis, noncirrhotic fibrosis, and cirrhosis were all <0.001. P values were approximated using the normal distribution. Abbreviation: HCC, hepatocellular carcinoma.

To further explore variation in annual HCC incidence rates according to clinical factors, we conducted subgroup analyses among patients with NASH without fibrosis, noncirrhotic fibrosis, and cirrhosis (Supporting Fig. S1). Among patients with diabetes and NASH without fibrosis, noncirrhotic fibrosis, or cirrhosis, the observed HCC incidence rates were 2.9/1,000 PYs, 7.2/1,000 PYs, and 15.7/1,000 PYs, respectively. In contrast, among nondiabetic patients with NASH without fibrosis, noncirrhotic fibrosis, or cirrhosis, the corresponding annual HCC incidence rates were 1.1/1,000 PYs, 1.9/1,000 PYs, and 5.0/1,000 PYs (Supporting Fig. S1). Similarly, annual incidence rates of EHSO cancers among diabetic patients with NASH without fibrosis, noncirrhotic fibrosis, or cirrhosis were 15.4/1,000 PYs, 17.8/1,000 PYs, and 18.2/1,000 PYs, respectively. In contrast, the corresponding annual incidence rates among nondiabetic patients were 11.8/1,000 PYs, 11.6/1,000 PYs, and 17.1/1,000 PYs, respectively.

Non-HCC Cancer Incidence

Compared with controls, NAFLD patients had modest yet significantly higher rates of developing EHSO cancers (9.9 vs. 11.3/1,000 PYs; aHR, 1.12 [95% CI, 1.04–1.20]) and hematologic cancers (0.7 vs. 1.0/1,000 PYs; aHR, 1.46 [95% CI, 1.12–1.90]), but significant differences were not observed with worsening NAFLD severity (both P_trend > 0.05) (Table 2).

In analyses of individual EHSO cancers (Supporting Table S6), NAFLD patients had significantly higher rates of developing pancreatic cancer (aHR, 2.15 [95% CI, 1.40–3.30]), kidney/bladder cancer (aHR, 1.41 [95% CI, 1.07–1.86]), and melanoma (aHR, 1.30 [95% CI, 1.08–1.57]) compared with controls. In contrast, no significant differences were observed for prostate, lung, breast, colorectal, esophageal/stomach, or cervical/ovarian/uterine cancers. Although there was a trend toward higher rates of non-HCC liver cancer among NAFLD patients, this did not achieve statistical significance (aHR, 1.75 [95% CI, 0.98–3.14]).

NAFLD-Only Subgroup

To address potential misclassification of controls and bias related to the indication for liver biopsy, we restricted the analyses just to patients with biopsy-confirmed NAFLD, with simple steatosis as the comparator. Consistent with our primary analyses, we observed progressively higher rates of incident overall cancer with worsening NAFLD histological severity (P_trend = 0.03), particularly incident HCC (P_trend < 0.01) (Table 3). Specifically, compared with simple steatosis, the absolute RDs and aHRs for incident HCC in patients with NASH without fibrosis, noncirrhotic fibrosis, and cirrhosis were 0.7/1,000 PYs (aHR, 1.61 [95% CI, 0.91–2.84]), 1.5/1,000 PYs (aHR, 2.45 [95% CI, 1.57–3.84]), and 5.4/1,000 PYs (aHR, 4.95 [95%CI, 3.08–7.97]), respectively. In contrast, dose-dependent associations were not observed for EHSO cancers. These findings persisted after rematching subjects with simple steatosis to all others with NAFLD (not shown).

TABLE 3.

Risk of Incident Cancer With Increasing NAFLD Histological Severity, Compared With Simple Steatosis^*

	NAFLD*
Types of Incident Cancer, N.	Simple Steatosis N = 5,939	NASH Without Fibrosis N = 1,050	Noncirrhotic Fibrosis N = 1,400	Cirrhosis N = 503
All incident cancers	1,160	187	233	111
Incidence RD1 (95% CI)	0 (ref.)	2.0 (−0.3–4.2)	3.4 (1.2–5.6)	12.1 (7.4–16.8)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	4.4 (−1.1–10.0)	6.7 (0.8–12.5)	17.3 (5.8–28.9)
Multivariable aHR3 (95% CI)	1 (ref.)	1.10 (0.94–1.29)	1.08 (0.93–1.26)	1.46 (1.19–1.79)
HCC	72	16	35	30
Incidence RD1 (95% CI)	0 (ref.)	0.7 (0.0–1.3)	1.5 (0.8–2.3)	5.4 (3.3–7.6)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	1.0 (−0.5–2.4)	3.4 (1.5–5.3)	10.2 (5.3–15.2)
Multivariable aHR3 (95% CI)	1 (ref.)	1.61 (0.91–2.84)	2.45 (1.57–3.84)	4.95 (3.08–7.97)
Non-HCC liver cancers	16	2	5	2
Incidence RD1 (95% CI)	0 (ref.)	0.0 (−0.2–0.2)	0.2 (−0.1–0.5)	0.2 (−0.3–0.8)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	0.0 (−0.6–0.5)	0.1 (−0.5–0.6)	0.7 (−1.0–2.3)
Multivariable aHR3 (95% CI)	1 (ref.)	0.76 (0.16–3.52)	1.97 (0.66–5.88)	1.56 (0.35–7.08)
EHSO cancers	987	152	175	77
Incidence RD1 (95% CI)	0 (ref.)	1.0 (−1.0–3.0)	1.2 (−0.7–3.1)	6.3 (2.4–10.1)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	3.3 (−1.7–8.4)	2.9 (−2.2–7.9)	9.5 (0.0–19.0)
Multivariable aHR3 (95% CI)	1 (ref.)	1.04 (0.88–1.25)	0.94 (0.80–1.12)	1.23 (0.97–1.56)
Hematologic cancers	85	17	19	2
Incidence RD1 (95% CI)	0 (ref.)	0.4 (−0.3–1.0)	0.4 (−0.2–1.0)	−0.5 (−1.1–0.1)
20-year absolute risk difference,2 % (95% CI)	0 (ref.)	0.3 (−0.9–1.5)	1.3 (−0.5–3.0)	−0.7 (−2.4–1.0)
Multivariable aHR3 (95% CI)	1 (ref.)	1.38 (0.80–2.36)	1.39 (0.81–2.37)	0.35 (0.08–1.46)

^*NAFLD was defined by liver histology. For definitions and algorithm, see the Methods section and Supporting Information.

¹Incidence rates were calculated by dividing the number of cases by PYs; CIs for incidence rates and absolute RDs were approximated by the normal distribution (Methods).

²20-year absolute risks and absolute risk differences (percentage points) were calculated based on Kaplan-Meier estimates (Methods).

³The multivariable aHR model accounted for the covariates outlined in Table 2.

Sensitivity Analyses

Our findings were robust across all sensitivity analyses. After PS matching NAFLD patients (N = 8,982) to population controls (N = 8,982), all clinical comorbidities were well balanced between groups (standardized mean differences <0.1) (Supporting Table S7), and our findings remained consistent with our primary analyses (Supporting Table S8). Similarly, after restricting the cohort to NAFLD patients with ≥1 full sibling without recorded NAFLD and then comparing each NAFLD patient with his or her full sibling(s), our findings were consistent (Supporting Table S9). Among the subgroup of NAFLD patients with an index date on or after January 1, 2006 (N = 1,826), and matched comparators (N = 7,785) (Supporting Table S10), NAFLD remained significantly and positively associated with both overall cancer (aHR, 1.53 [95% CI, 1.19–1.96]) and incident HCC (aHR, 38.20 [95% CI, 8,29–176.09]) in the fully adjusted multivariable model after further accounting for relevant medication use (Supporting Table S11). Our findings also persisted after further excluding anyone diagnosed with cancer within <2 years (Supporting Table S12) and after applying the alternative definition of HCC (P_trend < 0.01) (Supporting Table S13). When we repeated the NAFLD-only subgroup analysis and the full-sibling analysis after accounting for the potential competing risk of death without cancer, we observed similar, significant associations for HCC but not for other cancers (Supporting Table S14A,B). In analyses focused on a control exposure with an expected positive relation to overall cancer risk (the metabolic syndrome), we observed a modest but statistically significant increased risk (aHR, 1.07 [95% CI, 1.00–1.15]), consistent with published data.^(38,39) Finally, we found that an unmeasured confounder would have to be both very strongly associated with incident cancer risk and highly imbalanced (aHR ≥3.0, with ≥30% difference in prevalence) to fully attenuate our results (Supporting Table S15).

Discussion

In this population-based cohort comprising all Swedish adults with biopsy-confirmed NAFLD and matched population controls, NAFLD was associated with a significantly increased risk of developing overall cancer, driven primarily by an increased incidence of HCC. Compared with controls, patients with NAFLD had a 17-fold higher rate of developing HCC and a 20-year absolute excess risk of 2.1%. Significant excess HCC incidence was apparent at all ages, in both men and women, and after accounting for both established HCC risk factors and competing risks. Moreover, HCC incidence increased monotonically with worsening NAFLD histological severity, and it was further amplified in the setting of diabetes; thus, the highest HCC incidence rates were found in patients with comorbid diabetes and noncirrhotic fibrosis (7.2/1,000 PYs) or cirrhosis (15.7/1,000 PYs). Although NAFLD was also associated with the development of some EHSO cancers, those associations were relatively modest. In contrast, the 20-year absolute excess risk of patients with noncirrhotic NAFLD fibrosis (4.6%) or cirrhosis (11.4%) developing HCC was comparable to that of all EHSO cancers combined (4.7%−11.4%).

Our findings extend prior evidence linking NAFLD to an increased incidence of overall cancer,^(8,9) including HCC,^(14,40) by quantifying these risks for the first time in a nationwide, population-based histopathology cohort. Beyond the association with HCC, NAFLD was also associated with a small but significantly increased risk of developing pancreatic cancer, kidney/bladder cancer, and melanoma, a finding supported by some observational studies.^(8,16) However, for each of those cancers, significant associations were no longer present after we restricted the population just to patients with NAFLD, suggesting that other factors, such as obesity, may have confounded those relationships. Additionally, we detected no significant associations between NAFLD and cancers of the lung, prostate, breast, colon, rectum, esophagus, stomach, cervix, ovaries, or the uterus and only weak evidence for hematologic cancers. For some of those cancer sites, small associations have been described,^(8,9,16) yet those estimates are included within our CIs. In contrast to some reports, the current study benefitted from prospectively recorded histopathology collected from all adults with biopsy-confirmed NAFLD in Sweden, which therefore enabled a more comprehensive evaluation of cancer risk across the NAFLD histological spectrum.

Previous large-scale studies of NAFLD and HCC have relied on billing codes or serum markers to estimate NAFLD severity,^(14,40,41) and the few published studies with liver histology have had small populations of fewer than 500 participants, resulting in imprecise risk estimates and poor generalizability. Thus, although it has been suggested that a proportion of NAFLD-HCC tumors might arise in the absence of cirrhosis,^(42,43) the precise relationship between NAFLD severity and HCC risk remains unclear.^(44,45) In this nationwide population, rates of incident HCC increased in a dose-dependent fashion with worsening NAFLD severity. Patients with simple steatosis had minimally elevated annual HCC incidence rates (0.8/1000 PYs), most likely related to the small proportion who progress to advanced fibrosis over time. In contrast, patients with cirrhosis had HCC incidence rates of 6 to 9/1000 PYs. Importantly, these estimates varied according to key clinical factors, and the greatest magnitude of risk was found in patients with both diabetes and cirrhosis (annual incidence, 15.7/1000 PYs) or diabetes and noncirrhotic fibrosis (7.2/1000 PYs). Both of these subgroups had upper CIs that met or exceeded the threshold range, beyond which HCC surveillance is considered cost-effective (i.e., at or beyond the 95% CI of 0.8–2.3/100 PYs).^(44,46)_ENREF_31 Among patients with noncirrhotic fibrosis, diabetes was the only subgroup with CIs that approached this threshold range. In contrast, all subgroups with cirrhosis had CIs that included this range; furthermore, because many patients with established or decompensated cirrhosis will not undergo biopsy, their estimates are likely to be conservative. Collectively, our findings support future research to personalize HCC surveillance protocols and to test whether such protocols should include high-risk patients with diabetes and noncirrhotic NAFLD fibrosis.

Our study is strengthened by the population-based cohort with complete and prospectively recorded histopathology and by our strict, validated definitions of NAFLD and confounding variables in registers with near-complete follow-up for the entire country. This large sample size and prolonged follow-up time permitted calculation of more precise risk estimates across the complete NAFLD spectrum. Additionally, accounting for ranges of time between liver biopsy and incident cancer minimized potential reverse causation, and conducting analyses restricted to patients with histologically defined NAFLD reduced potential misclassification and bias from the underlying indication for biopsy. Moreover, we applied robust analytical techniques to address potential bias from residual confounding, shared familial factors, and competing events.

We considered whether the relationship between NAFLD and cancer risk merely reflected an association with the metabolic syndrome or its components. Consistent with other administrative data sets,⁽⁴⁷⁾ the recorded prevalence of the metabolic syndrome was low, which could lead to unmeasured confounding. Nevertheless, our findings remained similar in patients with and without this diagnosis, and our results also persisted in PS-matched analyses. Furthermore, substantial prior evidence demonstrates that the full metabolic syndrome contributes only modestly to the excess risk of developing overall cancer (pooled aHRs for the full metabolic syndrome, 1.10–1.61, depending on cancer site),⁽³⁹⁾ and our sensitivity analysis demonstrated that our results were robust to unmeasured confounding. Specifically, a confounder would need to have an aHR ≥3.0 for incident cancer risk and also simultaneously have a >30% difference in prevalence between groups to attenuate our results. Thus, the excess cancer risk associated with NAFLD likely exceeds that which is explained by the metabolic syndrome alone.

We acknowledge several limitations to our study. First, NAFLD was defined histologically, which could introduce selection bias, and therefore our findings may not extend to patients with NAFLD who do not undergo biopsy. However, our case distribution across histological categories accords with data from smaller histology cohorts,^(15,48–50) and our estimates are broadly consistent with prior studies,^(8,40) including a population-based study that defined NAFLD by ultrasound,⁽⁹⁾ underscoring the generalizability of our results. Second, our controls may have included patients with undiagnosed NAFLD, which could have attenuated our estimates; however, the gradients of risk for overall cancer and for HCC across NAFLD stages remained similar in the NAFLD-only subgroup analysis. Third, although histopathology is subject to interobserver variability, our validation study (Supporting Methods) demonstrates the accuracy of our exposure definitions.⁽²³⁾ Fourth, despite careful matching, some residual confounding cannot be excluded in this administrative data set, and we lacked laboratory values or data regarding smoking, alcohol consumption, body mass index (BMI), or confirmation of the absence of viral hepatitis. This highlights the need for additional large-scale histopathology cohorts with long-term follow-up that also include more detailed clinical phenotyping. Nevertheless, our findings were robust after multivariable adjustment and PS matching and in models stratified by clinical comorbidities and the NAFLD histological group—approaches that were not possible in previous, smaller studies. Fifth, although HCC cases could have been diagnosed more frequently among patients with NAFLD compared with controls, our estimates were similar when we applied an alternative, validated algorithm for HCC ascertainment.⁽³⁶⁾ Sixth, the Swedish population is primarily Caucasian, and we lacked specific data regarding individual cancer risks across different ethnic groups. Moreover, despite confirming a high PPV for NASH (87%) in our previous validation study,⁽²³⁾ our data did not distinguish between inflammation and hepatocyte ballooning or between individual fibrosis stages. Further, some of our subgroups were small in size with a limited number of events. Thus, there is a need for future studies that further characterize individual cancer risks according to specific NAFLD histological features as well as those focused on the type and severity of HCC, rare non-HCC liver cancers, and cancer-related mortality. Finally, we acknowledge changing trends in noninvasive NAFLD diagnostic tools; however, all models accounted for calendar year, and our results were similar in recent time periods and in the histology-only subgroup.

In conclusion, within this population-based cohort of 8,982 adults in Sweden with biopsy-confirmed NAFLD and matched population controls, NAFLD was associated with significantly increased risk of developing overall cancer, which was due primarily to a significantly increased incidence of HCC. For the first time on a nationwide scale, this study provides quantitative estimates regarding the risks of developing overall and site-specific cancers according to the presence and histological severity of NAFLD. We demonstrate that significant excess HCC risk is evident across all stages of NAFLD, and the magnitude of that risk is amplified by comorbid diabetes. Our findings underscore the need for improved, personalized HCC surveillance strategies for high-risk patients with NAFLD.

Supplementary Material

2

Supporting FIG. S1. Annual HCC incidence rates according to NAFLD histology and clinical factors. (For definitions of NAFLD histological groups, see Supporting Methods.)*HCC surveillance in patients with NAFLD is considered cost-effective if the annual incidence rate meets or exceeds 1.5% per year (95% CI, 0.8–2.3),⁽⁴⁶⁾ as outlined in the Discussion. Abbreviations: CI, confidence interval; HCC, hepatocellular carcinoma; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis.

Abbreviations:

aHR

adjusted hazard ratio

CI

confidence interval

EHSO

extrahepatic solid organ

ESPRESSO

Epidemiology Strengthened by histoPathology Reports in Sweden

HCC

hepatocellular carcinoma

ICD

International Classification of Diseases

LISA

Longitudinal Integrated Database for Health Insurance and Labour Market Studies

NAFLD

nonalcoholic fatty liver disease

NASH

nonalcoholic steatohepatitis

PIN

personal identity number

PPV

positive predictive value

PS

propensity score

PY

person-year

RD

rate difference

SNOMED

Systematized Nomenclature of Medicine