| |
Cancer Risk in Patients With Biopsy‐Confirmed Nonalcoholic Fatty Liver Disease: A Population‐Based Cohort Study
|
| |
| |
Download the PDF here
03 April 2021 Hepatology
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) (Ptrend<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.
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, colon, or the uterus and only weak evidence for For some of those cancer sites, small associations have been described, yet those estimates are our CIs. In contrast to some reports, the current study benefitted from from all adults with biopsy-confirmed NAFLD in Sweden, which there 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) 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.
| |
| |
| |
|
|
|
