Abstract
Background and Aims:
In patients with NAFLD, those with type 2 diabetes mellitus (DM) have a high risk of progression to HCC. However, the determinants of HCC risk in these patients remain unclear.
Approach and Results:
We assembled a retrospective cohort of patients with NAFLD and DM diagnosed at 130 facilities in the Veterans Administration between 1/1/2004 and 12/31/2008. We followed patients from the date of NAFLD diagnosis to HCC, death, or 12/31/2018. We used landmark Cox proportional hazards models to determine the effects of anti-DM medications (metformin, insulin, sulfonylureas) and glycemic control (percent of follow-up time with hemoglobin A1c < 7%) on the risk of HCC while adjusting for demographics and other metabolic traits (hypertension, obesity, dyslipidemia). We identified 85,963 patients with NAFLD and DM. In total, 524 patients developed HCC during a mean of 10.3 years of follow-up. Most common treatments were metformin monotherapy (19.7%), metformin-sulfonylureas (19.6%), insulin (9.3%), and sulfonylureas monotherapy (13.6%). Compared with no medication, metformin was associated with 20% lower risk of HCC (HR, 0.80; 95% CI, 0.93–0.98). Insulin had no effect on HCC risk (HR, 1.02; 95% CI, 0.85–1.22; p = 0.85). Insulin in combination with other oral medications was associated with a 1.6 to 1.7-fold higher risk of HCC. Adequate glycemic control was associated with a 31% lower risk of HCC (HR, 0.69; 95% CI, 0.62–0.78).
Conclusions:
In this large cohort of patients with NAFLD and DM, use of metformin was associated with a reduced risk of HCC, whereas use of combination therapy was associated with increased risk. Glycemic control can serve as a biomarker for HCC risk stratification in patients with NAFLD and diabetes.
INTRODUCTION
NAFLD affects 20%–30% of the US population.[1-3] Diabetes mellitus is an important risk factor for NAFLD. Patients with both NAFLD and type 2 diabetes are at an increased risk of progression from steatosis to NASH to advanced fibrosis.[4-6] Diabetes also conferred the highest risk of progression to HCC in patients with NAFLD in our recent study.[7] Importantly, as many as 40%–50% of patients with NAFLD may have diabetes.[8] Understanding the determinants of HCC in patients with NAFLD and diabetes is necessary for additional risk stratification and to aid in secondary prevention.
Many studies have examined the effect of anti-diabetes medications on HCC risk.[9] Substantial preclinical evidence suggests that metformin has anticancer properties, including mammalian target of rapamycin inhibition, cytotoxic effects, and immuno-modulation.[10] Studies also show that insulin increases the risk of disparate cancers by stimulating cell growth through insulin receptors.[11] A recent meta-analysis (5 case control, 3 cohort, 2 randomized controlled trials) showed a 50% reduction in HCC incidence with metformin use but a 62% and 161% increase with sulfonylurea and insulin use, respectively, among patients with type 2 diabetes.[12] However, recent cohort studies of patients with diabetes found no difference in HCC risk by metformin use.[13-15] Only one study of 299 patients with NAFLD cirrhosis showed a trend in favor of metformin, but the study was limited by a small cohort size with few HCC cases.[16] There are no studies that have examined the cancer-preventive or cancer-promoting effects of anti-diabetes medications in general patients with NAFLD, especially the larger group without cirrhosis.
In patients with diabetes, the benefits of maintaining glycemic control are well established for lowering the risk of diabetes-related complications, including microvascular complications, cardiovascular events, and mortality.[17-19] Whether and to what extent glycemic control is independently associated with future risk for HCC in patients with NAFLD is not fully known. If glycemic control is found to lower the risk of HCC, then these results could have significant implications for patients with NAFLD and their clinicians.
We conducted a large retrospective cohort study of over 85,000 patients with NAFLD limited to those with concomitant diabetes. Patients were followed for an average of 10 years to evaluate the independent effects of anti-diabetes medications and glycemic control on the risk of incident HCC while accounting for diabetes duration, severity, and other metabolic traits (hypertension, obesity, dyslipidemia).
MATERIALS AND METHODS
Data source
We used data from the national US Department of Veterans Affairs (VA) Corporate Data Warehouse (CDW) and VA Central Cancer Registry (CCR). CDW includes all laboratory test results, pharmacy information, and inpatient and outpatient utilization, including International Classification of Diseases (ICD) 9 and 10 diagnoses codes from VA and Fee Basis files. Fee Basis file includes information on care delivered outside but paid by the VA. CDW also contains information from annual Alcohol Use Disorders Identification Test consumption questions (AUDIT-C) screen and Vital Status files.[20,21] AUDIT-C has been used to screen over 90% of VA outpatients nationwide since 2004.[21] Vital Status combines data from Medicare, VA, Social Security, and VA Compensation and Pension Benefits to determine date of death (sensitivity 98.3% and specificity 99.8% relative to National Death Index). Finally, CCR is a centralized repository for VA patients with cancer and includes information on date of diagnosis, primary site, and histology. The research conducted in this manuscript received IRB approval and a waiver of informed consent from the IRB at Baylor College of Medicine.
Study cohort
We evaluated all patients 18 to 80 years old who had at least one visit to any VA hospital in the nation between January 1, 2003, and December 31, 2011, for the presence of NAFLD. As we reported previously,[22] patients were classified as having NAFLD if they had two or more elevated ALT values (≥40 IU/ml for men and >31 IU/ml for women) in the ambulatory setting more than 6 months apart, with no positive serologic laboratory testing for HBV (i.e., HBV surface antigen) or HCV (i.e., HCV RNA). We excluded patients with any alcohol related ICD codes or positive AUDIT-C scores (≥4 in men and ≥3 in women) any time before or during study follow-up. We also excluded patients with other rare chronic liver disorders (hereditary hemochromatosis, primary biliary cirrhosis, primary sclerosing cholangitis, alpha-1 antitrypsin disease, or autoimmune hepatitis) defined based on ICD codes. This definition was highly predictive of NAFLD diagnosis based on explicit chart review (positive predictive value, 80.8%; negative predictive value, 78%).[22] We used the date of first elevated ALT as the NAFLD index date. We included patients with NAFLD with an index date from January 1, 2004, to December 31, 2008, in this analysis because AUDIT-C was implemented in the VA in 2004. We used 2008 as the cutoff for enrollment to allow up to 10 years of follow-up.
We limited this analysis to patients with NAFLD with a concomitant diagnosis for type 2 diabetes (referred to as diabetes from here on) before or on the NAFLD index date. We defined diabetes as presence of ≥2 outpatient or ≥1 inpatient ICD-9/10 code for diabetes or >1 filled prescription of diabetes medications (oral hypoglycemic medications or insulin). The earliest date of diabetes codes or medications was used for diabetes diagnosis date. We chose the NAFLD index date as the start of study follow-up (in lieu of diabetes diagnosis) because that corresponds with the time patients and their clinicians first become aware of patients’ liver disease. We followed patients to HCC, death, or 12/31/2018, which-ever occurred first. Because our goal was to examine the risk of incident HCC, we excluded patients with prevalent diagnoses of HCC, defined as having a diagnosis of HCC any time before to 12 months after the NAFLD index.
Variable specification
Outcome
We used a hierarchical approach to define HCC, as described previously.[4] Briefly, we classified patients as subjects with HCC if they had 2 instances of diagnosis codes for HCC (ICD-9: 155.0 in the absence of 155.1; ICD-10: C22.0, C22.2, C22.7, C22.8, C22.9) in the inpatient, outpatient, or Fee Basis files of the CDW. We then examined the VA CCR for patients with HCC diagnosis. For patients who had an ICD-9/10 code but were not identified as having HCC in the CCR data, we conducted a manual review of the VA electronic medical record to determine their true HCC status. This hierarchical approach ensured high validity of all the captured HCC cases. We obtained all-cause mortality data including date of death from VA Vital Status file.
Diabetes-specific factors
The exposure variables of interest included diabetes medications and glycemic control as measured by hemoglobin A1c (HbA1c). We used patients’ full history of filled prescriptions (from index to end of follow-up) to derive time updating (every 90 days) treatment status as no treatment, metformin users, sulfonylurea users, insulin users, thiazolidinediones users, or users of other anti-diabetes medications. Patients had to receive at least one prescription of a given medication in a 90-day period to be classified as a user. For each patient, we extracted HbA1c values longitudinally. We derived a time-updating variable for glycemic control, defined as HbA1c < 7.
Other diabetes-specific factors included time-updating variables for diabetes complications and diabetes duration. We defined diabetes complications based on ICD codes for diabetes with complications (ICD9: 250.4/2, 250.5/2, 250.6/2, 250.7/2, 250.9/2) and used the first instance of these codes as the date of diabetes with complications. At each time point of interest, we defined duration of diabetes as the time elapsed since the first diabetes diagnosis.
Other metabolic traits
We used height and weight values any time before to 1 year after and nearest to the index date to define baseline body mass index (BMI). We defined hypertension by ≥2 outpatient or ≥1 inpatient ICD-9 code or >1 filled prescription of antihypertensives any time before index to the end of follow-up; we used the first evidence hypertension as the date of diagnosis and updated patients’ status each year during the follow-up time. We defined dyslipidemia as abnormal high serum triglycerides (≥200 mg/dL) and/or low HDL (<40 mg/dL) starting from the value any time before or 1 year after index (choosing the closest to the index) as the baseline value and then updated each year during the follow-up time. Once present, we assumed patients had these conditions for the duration of follow-up.
Other covariates
Other covariates included time updating age (<55, 55–65, ≥65 years), sex, and race/ethnicity (White, Black, Hispanic, and other). We defined likelihood of advanced fibrosis using a noninvasive marker, fibrosis-4 score (FIB-4). We calculated FIB-4 using age at index and values of aspartate aminotransferases, alanine aminotransferase, and platelets 1 year before or 1 year after index date. In the event of multiple values, we chose the values closest to index date. We also derived serial annual FIB-4 values during follow-up. Patients with NAFLD and diabetes have a high burden of cardiovascular diseases (CVD) and chronic kidney diseases (CVD) that could compete with HCC. We defined CVD and CKD based on ≥2 outpatient or ≥1 inpatient ICD-9 codes any time before index. Health care utilization was measured as presence of an outpatient clinic visit in the 1 year prior and 1 year after the NAFLD index date.
Statistical analyses
We used landmark Cox proportional hazard models, which consisted of a series of models defined at four landmark times (1, 2, 3, and 4 years after the index date).[23-26] At each landmark time, we examined the associations between diabetes treatment and glycemic control and subsequent risk of HCC through a cause-specific Cox proportional hazard model. Our main outcome was the time from the landmark to HCC; we treated death as competing risk in the cause-specific Cox modeling. At each landmark, the data included only at-risk patients (e.g., those who were alive without HCC) at that time. This approach accounts for immortal time bias by defining exposure based on longitudinal history before the landmark with outcome events that occurred after the landmark.[27] Patients who did not develop events (HCC or death) by the end of follow-up were censored. The use of four landmark times allows us to evaluate whether the relationship between prior exposure and subsequent outcome is consistent over the follow-up period. When consistent, these landmark models can be combined into a single “super Cox model” for increased statistical efficiency.[23] To do so, we stacked the working datasets for the four landmark models and fit a single Cox model to the stacked dataset, with stratified baseline hazards by the landmark times. The statistical estimation is consistent under working independence among multiple records per patient.
At each landmark time, we classified patients as users of each anti-diabetes medication if they received it for ≥75% time during the year before the landmark time. Because patients could receive none, one, or more than one medication, we compared different subgroups defined based on use of the more common combination of medications, such as metformin-sulfonylureas, sulfonylureas-thiazolidinedione, insulin-metformin, insulin-sulfonylureas, and insulin-metformin-sulfonylureas, by using linear contrasts of the regression coefficients of medication indicators and their interactions. We chose the top 10 medication subgroups based on percentage of the cohort taking the medications. The remaining combinations were taken by <1% of the cohort and were not examined. We defined glycemic control if patients’ serial HbA1c values were under 7% for >80% of the time during the year before each landmark. We chose 7% as it represents the optimal cutoff associated with improved diabetes endpoints in previous studies; it also corresponds with the target for glycemic control set by the American Diabetes Association.[28] Other covariates included sex, race/ethnicity, age, FIB-4, diabetes complications, diabetes duration, and other metabolic traits updated at each landmark time.
We repeated the analyses limited to patients who received at least one anti-diabetes treatment >75% of the time to examine the effect of medication-induced glycemic control. We also limited the analysis to patients with adequate diabetes control (defined as HbA1c less than 7 >80% of the time) to examine the robustness of the effect of diabetes medications in this subgroup.
RESULTS
Patient characteristics
We identified 85,963 patients with NAFLD and diabetes who were seen at 130 VA facilities between 1/1/2004 and 12/31/2008 (Table 1). The baseline mean age of our cohort was 60.5 years (SD 9.7). Most were male (96.2%) and White (69.6%); 11.8% were African American, and 5.5% were Hispanic. The mean BMI was 33.2 kg/m2 (SD 5.8). Most patients also had other metabolic traits including hypertension (93.3%) or dyslipidemia (71.2%). The cohort had relatively mild liver disease; the mean FIB-4 score was 1.3 (SD 0.99), and over half (62.1%) had FIB-4 <1.45, indicating low probability of having advanced fibrosis at baseline. On average, patients had diabetes for 2.1 years (SD 2.3) and 26.1% had diabetic complications, 12.1% had CVD, and 3.8% had CKD before NAFLD index. At baseline, 46.3% of patients had HbA1c < 7%, 19.7% were on metformin monotherapy, 19.6% were on metformin-sulfonylureas, 13.6% were on sulfonylurea monotherapy, 9.3% were on insulin alone, and 16.3% were on no medication. These proportions changed over time. By year 4, 14.4% of patients were taking metformin and 12.1% were on insulin.
TABLE 1.
Baseline characteristics of patients with NAFLD and type 2 diabetes
| Characteristic, n (%) | NAFLD diabetes (n = 85,963) |
|---|---|
| Age, mean (SD) | 60.5 (9.7) |
| Sex | |
| Men | 82,701 (96.2) |
| Women | 3,262 (3.8) |
| Race/ethnicity | |
| White | 59,836 (69.6) |
| African American | 10,186 (11.8) |
| Hispanic | 4,725 (5.5) |
| Other races | 2,343 (2.7) |
| Unreported | 8,873 (10.3) |
| Comorbidities | |
| BMI, mean (SD) | 33.2 (5.8) |
| Hypertension | 80,182 (93.3) |
| Dyslipidemia | 58,871 (71.2) |
| Chronic kidney disease | 3,298 (3.8) |
| Cardiovascular disease | 10,420 (12.1) |
| FIB-4, mean (SD) | 1.3 (0.99) |
| Diabetes-related variables | |
| HbA1c < 7% | 46.3 |
| HbA1c ≥ 7% | 53.6 |
| Diabetes complications | 22,412 (26.1) |
| Diabetes duration, mean (SD) | 2.1 (2.3) |
| Medications at first landmark time | |
| No medications at all | 13,818 (16.3) |
| Insulin alone | 7,995 (9.3) |
| Metformin alone | 16,935 (19.7) |
| Sulfonylureas alone | 11,691 (13.6) |
| Metformin-sulfonylureas | 16,849 (19.6) |
| Sulfonylureas-thiazolidinedione | 430 (0.5) |
| Insulin-metformin | 3,872 (4.4) |
| Insulin-sulfonylureas | 2,235 (2.6) |
| Insulin-metformin-sulfonylureas | 2,665 (3.1) |
| Metformin-sulfonylureas-thiazolidinedione | 774 (0.9) |
Note: In total, 3,259 (3.8%) patients had missing data on lipids and 14,156 (16.5%) had missing data regarding FIB-4 at baseline. No patient was on thiazolidinedione in the absence of other medications. Of all patients, 8.8% were on at least one of the diabetes medications for >75% of the time during the year before the first landmark time.
Effect of diabetes-specific factors on HCC risk
During the mean follow-up time of 10.3 (SD 3.36) years, 524 patients developed HCC and 30,622 patients died. Table 2 presents the results from the combined multivariable landmark Cox model. Compared with no medication, metformin use was associated with 22% lower risk of HCC (HR, 0.77; 95% CI, 0.65–0.90; p = 0.001). Use of insulin alone was not significantly associated with risk of HCC (HR, 1.05; 95% CI, 0.88–1.27; p = 0.57). Patients treated with combination treatments, such as insulin and metformin or those treated with insulin, metformin, and sulfonylureas, were associated with a 1.5 to 1.7-fold higher risk of HCC (HR for insulin and metformin, 1.53; 95% CI, 1.26–1.86; p < 0.0001; HR for insulin, metformin, and sulfonylureas, 1.71; 95% CI, 1.41–2.08; p < 0.0001). Combinations that included a thiazolidinedione did not significantly affect the risk of HCC. Likewise, sulfonylureas alone and in combination with other oral anti-diabetes medications did not significantly reduce risk of HCC.
TABLE 2.
Multivariable model examining the effect of diabetes medication groups and glycemic control on risk of HCC in patients with NAFLD diabetes in the combined landmark model; range of number of patients for each landmark time is 84,616 to 77,102
| Risk factor | HR (95% CI) | p value |
|---|---|---|
| Age (ref: <55) | ||
| 55–65 | 1.72 (1.40, 2.12) | <0.0001 |
| >65 | 2.06 (1.66, 2.56) | <0.0001 |
| Sex (ref: male) | ||
| Female | 0.32 (0.18, 0.55) | <0.0001 |
| Race (ref: White) | ||
| African American | 0.38 (0.29, 0.49) | <0.0001 |
| Hispanic | 1.61 (1.36, 1.92) | <0.0001 |
| Other | 1.13 (0.85, 1.52) | 0.41 |
| Comorbidities | ||
| Obesity | 1.20 (1.08, 1.35) | 0.001 |
| Hypertension | 1.22 (0.92, 1.61) | 0.16 |
| Dyslipidemia | 0.89 (0.78, 1.01) | 0.06 |
| Chronic kidney disease | 0.71 (0.57, 0.89) | 0.003 |
| Cardiovascular disease | 0.99 (0.87, 1.11) | 0.83 |
| FIB4 ≥ 1.45 | 5.33 (4.73, 6.00) | <0.0001 |
| Diabetes-related variables | ||
| Duration of diabetes (per year) | 0.94 (0.92, 0.97) | <0.0001 |
| Diabetes complications | 1.24 (1.12, 1.38) | <0.0001 |
| HbA1c <7 ≥80% of time | 0.68 (0.60, 0.77) | <0.0001 |
| Diabetes medications (ref: none) | ||
| Metformin alone | 0.77 (0.65, 0.90) | 0.001 |
| Sulfonylureas alone | 0.98 (0.84, 1.16) | 0.84 |
| Insulin alone | 1.05 (0.88, 1.27) | 0.57 |
| Metformin-sulfonylureas | 1.02 (0.88, 1.19) | 0.81 |
| Sulfonylureas-thiazolidinedione | 0.97 (0.56, 1.69) | 0.92 |
| Metformin-sulfonylureas-thiazolidinedione | 0.84 (0.54, 1.30) | 0.42 |
| Insulin-metformin | 1.53 (1.26, 1.86) | <0.0001 |
| Insulin-sulfonylureas | 0.87 (0.66, 1.16) | 0.34 |
| Insulin-metformin-sulfonylureas | 1.71 (1.41, 2.08) | <0.0001 |
Note: Models adjusted for health care utilization. The models also included patients who were on at least one of the diabetes medications for >75% of the time during the year before each landmark time.
Patients with good glycemic control (defined as HbA1c < 7% for >80% time) were associated with a 32% lower risk of HCC than patients who had suboptimal glycemic control (HR, 0.68; 95% CI, 0.60–0.77; p < 0.0001) (Table 2). Patients with diabetes complications was associated with a 24% higher risk of HCC than patients without diabetes complications (HR, 1.24; 95% CI, 1.12–1.38; p < 0.0001). Duration of diabetes before NAFLD diagnosis was associated with a slightly lower risk of HCC in the adjusted models (HR, 0.94; 95% CI, 0.92–0.97; p < 0.0001).
HCC risk was associated with increasing age. Women had a lower risk of HCC than men (HR, 0.32; 95% CI, 0.18–0.55; p < 0.0001). Risk of HCC was lower in African Americans (HR, 0.38; 95% CI, 0.29–0.49; p < 0.0001) and higher in Hispanic patients (HR, 1.61; 95% CI, 1.36–1.92; p < 0.0001) than in White patients. FIB-4 was strongly associated with HCC risk; the risk was 5-fold higher in patients with FIB-4 ≥1.45 than those with FIB-4 <1.45 (HR, 5.33; 95% CI, 4.73–6.00; p < 0.0001).
Table S1 presents the results for each landmark time separately. The magnitude and direction of effects were similar for each landmark year, with few exceptions.
Sensitivity and subgroup analyses
Subgroup analysis limited to patients on at least one medication at each landmark time showed an association with an increased risk of HCC in users of all other medications and combinations compared with metformin monotherapy, although some of the associations were not statistically significant given power limitations in the subgroups analysis (Table 3). Restricting the analysis to patients who received at least one anti-diabetes treatment did not result in any substantial change in the direction or magnitude of the glycemic control effect (HR, 0.78; 95% CI, 0.68–0.91; p = 0.001). Evaluation of subgroup of patients with adequate glycemic control (HbA1c < 7 ≥80% time) yielded similar results to our main analysis.
TABLE 3.
Subgroup analyses for effect of diabetes medication groups and glycemic control on risk of HCC in patients with NAFLD diabetes; n reflects the range of patients in each landmark time
| Patients who received at least one diabetes medication (n = 63,370–58,162) |
Patients with adequate glycemic control (n = 31,293–22,204) |
|||
|---|---|---|---|---|
| Risk factor | HR (95% CI) | p value | HR (95% CI) | p value |
| Age (ref: <55) | ||||
| 55–65 | 1.57 (1.25, 1.98) | 0.0001 | 1.25 (0.82, 1.90) | 0.30 |
| >65 | 1.98 (1.56, 2.52) | <0.0001 | 1.67 (1.09, 2.56) | 0.02 |
| Sex (ref: male) | ||||
| Female | 0.43 (0.25, 0.75) | 0.003 | 0.09 (0.01, 0.62) | 0.01 |
| Race (ref: White) | ||||
| African American | 0.38 (0.28, 0.52) | <0.0001 | 0.48 (0.30, 0.76) | 0.002 |
| Hispanic | 1.60 (1.32, 1.95) | <0.0001 | 0.92 (0.59, 1.45) | 0.73 |
| Other | 1.41 (1.04, 1.93) | 0.03 | 0.56 (0.25, 1.26) | 0.16 |
| Unreported | 1.09 (0.90, 1.32) | 0.39 | 1.18 (0.88, 1.57) | 0.28 |
| Comorbidities | ||||
| Obesity | 1.26 (1.10, 1.44) | 0.0006 | 1.25 (1.01, 1.55) | 0.04 |
| Hypertension | 1.39 (0.93, 2.07) | 0.11 | 1.08 (0.67, 1.72) | 0.76 |
| Dyslipidemia | 0.93 (0.80, 1.08) | 0.33 | 0.81 (0.64, 1.01) | 0.06 |
| Chronic kidney disease | 0.72 (0.56, 0.93) | 0.01 | 0.87 (0.57, 1.32) | 0.51 |
| Cardiovascular disease | 1.09 (0.95, 1.25) | 0.20 | 0.98 (0.77, 1.25) | 0.87 |
| FIB4 ≥ 1.45 | 4.88 (4.28, 5.57) | <0.0001 | 5.98 (4.66, 7.67) | <0.0001 |
| Diabetes-related variables | ||||
| Duration of diabetes | 0.93 (0.90, 0.95) | <0.0001 | 0.95 (0.90, 1.00) | 0.03 |
| Baseline diabetes complications | 1.18 (1.04, 1.32) | 0.008 | 1.33 (1.07, 1.65) | 0.01 |
| HbA1c <7 for ≥80% time | 0.78 (0.68, 0.91) | 0.001 | N/A | N/A |
| Diabetes medications (ref: none) | ||||
| Metformin | Reference | 0.81 (0.62, 1.04) | 0.10 | |
| Sulfonylureas | 1.31 (1.06, 1.61) | 0.01 | 1.00 (0.75, 1.33) | 0.99 |
| Insulin | 1.44 (1.14, 1.84) | 0.003 | 1.34 (0.87, 2.06) | 0.19 |
| Metformin-sulfonylureas | 1.70 (1.42, 2.04) | <0.0001 | 1.18 (0.88, 1.59) | 0.26 |
| Sulfonylureas-thiazolidinedionea | 1.81 (1.05, 3.12) | 0.03 | N/A | N/A |
| Metformin-sulfonylureas-Thiazolidinedionea | 1.33 (0.84, 2.10) | 0.22 | N/A | N/A |
| Insulin-metformin | 2.86 (2.30, 3.55) | <0.0001 | 1.93 (1.15, 3.23) | 0.01 |
| Insulin-sulfonylureas | 2.26 (1.68, 3.04) | <0.0001 | 1.80 (0.93, 3.48) | 0.08 |
| Insulin-metformin-sulfonylureas | 2.29 (1.78, 2.95) | <0.0001 | 3.80 (2.16, 6.70) | <0.0001 |
Note: Bolded estimates have a p value <0.05.
Models adjusted for health care utilization. The models also included patients who were on at least one of the diabetes medications for >75% of the time during the year before each landmark time.
Sulfonylureas-thiazolidinedione and metformin-sulfonylureas-thiazolidinedione groups not included in some subgroup analyses due to small numbers in the stratified samples.
DISCUSSION
In this large cohort of patients with diabetes and NAFLD, we found that use of metformin was associated with a significant reduction in the risk of HCC. Indicators of disease severity and control, such as baseline diabetes complications, were associated with an increased HCC risk, but maintaining adequate glycemic control (defined as HbA1c less than 7 >80% of the time) was associated with a reduced risk of HCC.
Metformin is a widely used first-line therapy for patients with type 2 diabetes. Clinical evidence for metformin effectiveness in NAFLD is scarce; only one study of 299 patients with NAFLD cirrhosis and few HCC cases showed a promising trend in favor of metformin.[16] Our study extends this evidence base to the larger population of patients with NAFLD and diabetes, most of whom did not have cirrhosis. We found that compared with no treatment, metformin use was associated with 21% reduction in the risk of HCC. We chose no treatment as the comparison group for metformin to minimize confounding by indication. This comparative advantage was more pronounced when we compared metformin to other diabetes medications (i.e., insulin or sulphonylureas, Table 3), although some of this difference could be due to inherent time lag when comparing the first-line treatment (metformin) with second- or third-line treatments (sulfonylureas or insulin). We also found that metformin’s effect was independent of the degree of glycemic control. Metformin has pleiotropic anticancer properties. Studies show metformin inhibits the mammalian target of the rapamycin pathway through activation of adenosine monophosphate–activated protein kinase and its upstream regulator, liver kinase B1. It also inhibits angiogenesis, blocks cell cycle, induces p53-independent apoptosis,[29-31] and has anti-inflammatory effects in both in vitro and in vivo studies.[32] Our results provide support to a direct chemo-preventive effect of metformin.
Adequate glycemic control can prevent adverse events in patients with diabetes, including macrovascular and microvascular complications, hospitalizations, and death. Our data suggest that this benefit may extend to prevention of HCC. Patients with good glycemic control were associated with 32% lower risk of HCC than those with suboptimal glycemic control. It is possible that glycemic control acts as a surrogate for presence or absence of severe diabetes and that our results reflect the well-known associations between insulin resistance, progressive hyperglycemia, and cancer. The observed association may also guide HCC risk stratification and screening efforts. Based on our results, prioritizing patients with diabetes and NAFLD with suboptimal glycemic control may be the way to start making inroads into the mass-screening problem.
We also found racial/ethnic disparities in the risk of progression to HCC. Risk of HCC was significantly higher in Hispanics compared with White and African American patients (Table 2). We previously showed HCC incidence is increasing faster in Hispanics than other ethnicities, a trend that could continue in the absence of targeted interventions.[33] Our data also call for studies that examine why these disparities exist—a gap that has hindered development of effective interventions.
Our study has several limitations. We used an observational design which precludes casual inferences. Ideally, large randomized controlled trials with a control group should compare the effectiveness and harms of HCC chemoprevention. However, such a trial would be premature in the absence of robust observational data demonstrating an association between the selected medications and reduced risk of HCC. We believe the strength of association, consistency with previous studies, temporality, and biological plausibility lend support for causality. We accounted for ascertainment bias and immortal time bias in our study. We also controlled for demographic factors, diabetes severity, and liver fibrosis and performed a range of sensitivity and subgroup analyses to minimize confounding by indication. Few patients in our cohort were on thiazolidinedione, precluding evaluation of these agents outside of combination therapies. We were not able to evaluate for true adherence to therapy. We were unable to evaluate for coutilization of Medicare or prescriptions filled at non-VA facilities. However, there was a high level of health care utilization of VA services in our cohort. We were also unable to examine the effect of lifestyle changes, including physical exercise, in our study. Our study population was predominantly male, limiting generalizability to women.[34] Lastly, our study was limited to Veterans. Although likely the biological processes are similar in Veteran and non-Veteran populations, future studies need to validate these results in non-VA cohorts.
Our results provide support of a chemo-preventive effect of metformin in patients with NAFLD and diabetes, especially among those who are at a risk of advanced fibrosis. These data are foundational for future studies that seek to extend the indications for use of metformin for chemoprevention in NAFLD. Glycemic control was associated with HCC risk. Our data point to the importance of monitoring glycemic control in patients with NAFLD and diabetes. Patients with suboptimal glycemic control may serve as an important subgroup for close monitoring for future risk of HCC. Given their high risk, these patients should ideally be managed by a multidisciplinary team including an internist, hepatologist, and endocrinologist.[35]
Supplementary Material
Funding information
This material is based on work supported by Cancer Prevention and Research Institute of Texas grant (RP150587). The work is also supported in part by the National Cancer Institute U01 CA230997-01, R01 CA256977, US Department of Veterans Affairs HSR&D Center for Innovations in Quality, Effectiveness and Safety (CIN 13-413), Michael E. DeBakey VA Medical Center and the Center for Gastrointestinal Development, Infection and Injury (NIDDK P30 DK 56338). The work of Dr. Li was partially funded by R01DK118079. The work of Drs. Li and Dai is partially funded by P30CA016672
Abbreviations:
- AUDIT-C
Alcohol Use Disorders Identification Test consumption questions
- BMI
body mass index
- CCR
Central Cancer Registry
- CDW
Corporate Data Warehouse
- CKD
chronic kidney disease
- CVD
cardiovascular disease
- FIB-4
fibrosis-4 score
- HbA1c
hemoglobin A1c
- ICD
International Classification of Diseases
- VA
US Department of Veterans Affairs
Footnotes
CONFLICT OF INTEREST
Nothing to report.
DISCLAIMER
The opinions and assertions contained herein are the sole views of the authors and are not to be construed as official or as reflecting the views of the US Department of Veterans Affairs or the United States.
SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher’s website.
REFERENCES
- 1.Williams CD, Stengel J, Asike MI, Torres DM, Shaw J, Contreras M, et al. Prevalence of nonalcoholic fatty liver disease and nonalcoholic steatohepatitis among a largely middle-aged population utilizing ultrasound and liver biopsy: a prospective study. Gastroenterology. 2011;140(1):124–31. [DOI] [PubMed] [Google Scholar]
- 2.Caballería L, Pera G, Arteaga I, Rodríguez L, Alumà A, Morillas RM, et al. High prevalence of liver fibrosis among European adults with unknown liver disease: a population-based study. Clin Gastroenterol Hepatol. 2018;16(7):1138–45.e5. [DOI] [PubMed] [Google Scholar]
- 3.Le P, Chaitoff A, Rothberg MB, McCullough A, Gupta NM, Alkhouri N. Population-based trends in prevalence of nonalcoholic fatty liver disease in US adults with type 2 diabetes. Clin Gastroenterol Hepatol. 2019;17:2377–8. [DOI] [PubMed] [Google Scholar]
- 4.Lomonaco R, Godinez Leiva E, Bril F, Shrestha S, Mansour L, Budd J, et al. Advanced liver fibrosis is common in patients with type 2 diabetes followed in the outpatient setting: The need for systematic screening. Diabetes Care. 2021;44(2):399–406. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Doycheva I, Cui J, Nguyen P, Costa EA, Hooker J, Hofflich H, et al. Non-invasive screening of diabetics in primary care for NAFLD and advanced fibrosis by MRI and MRE. Aliment Pharmacol Ther. 2016;43(1):83–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.El-Serag HB, Tran T, Everhart JE. Diabetes increases the risk of chronic liver disease and hepatocellular carcinoma. Gastroenterology. 2004;126:460–8. [DOI] [PubMed] [Google Scholar]
- 7.Kanwal F, Kramer JR, Li L, Dai J, Natarajan Y, Yu X, et al. Effect of metabolic traits on the risk of cirrhosis and hepatocellular cancer in nonalcoholic fatty liver disease. Hepatology. 2020;71(3):808–19. [DOI] [PubMed] [Google Scholar]
- 8.Yang JD, Ahmed F, Mara KC, Addissie BD, Allen AM, Gores GJ, et al. Diabetes is associated with increased risk of hepatocellular carcinoma in patients with cirrhosis from nonalcoholic fatty liver disease. Hepatology. 2020;71(3):907–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Li X, Wang X, Gao P. Diabetes mellitus and risk of hepatocellular carcinoma. Biomed Res Int. 2017;2017:5202684. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Chae YK, Arya A, Malecek MK, Shin DS, Carneiro B, Chandra S, et al. Repurposing metformin for cancer treatment: current clinical studies. Oncotarget. 2016;7(26):40767–80. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Vigneri R, Goldfine ID, Frittitta L. Insulin, insulin receptors, and cancer. J Endocrinol Invest. 2016;39:1365–76. [DOI] [PubMed] [Google Scholar]
- 12.Singh S, Singh PP, Singh AG, Murad MH, Sanchez W. Anti-diabetic medications and the risk of hepatocellular cancer: a systematic review and meta-analysis. Am J Gastroenterol. 2013;108:881–91; quiz 892. [DOI] [PubMed] [Google Scholar]
- 13.Hagberg KW, Mcglynn KA, Sahasrabuddhe VV, Jick S. Anti-diabetic medications and risk of primary liver cancer in persons with type II diabetes. Br J Cancer. 2014;111(9):1710–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Oh TK, Song IA. Metformin use and the risk of cancer in patients with diabetes: a nationwide sample cohort study. Cancer Prev Res. 2020;13(2):195–202. [DOI] [PubMed] [Google Scholar]
- 15.Miele L, Bosetti C, Turati F, Rapaccini G, Gasbarrini A, La Vecchia C, et al. Diabetes and insulin therapy, but not metformin, are related to hepatocellular cancer risk. Gastroenterol Res Pract. 2015;2015:570356. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Vilar-Gomez E, Calzadilla-Bertot L, Wong VWS, Castellanos M, Aller-de la Fuente R, Eslam M, et al. Type 2 diabetes and metformin use associate with outcomes of patients with nonalcoholic steatohepatitis–related, Child–Pugh A cirrhosis. Clin Gastroenterol Hepatol. 2021;19(1):136–45.e6. [DOI] [PubMed] [Google Scholar]
- 17.UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). Lancet. 1998;352:837–53. [PubMed] [Google Scholar]
- 18.The Diabetes Control and Complications Trial Research Group. The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. N Engl J Med. 1993;329:977–86. [DOI] [PubMed] [Google Scholar]
- 19.Epidemiology of Diabetes Interventions and Complications (EDIC) Research Group. Design, implementation, and preliminary results of a long-term follow-up of the Diabetes Control and Complications Trial cohort. Diabetes Care. 1999;22:99–111. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Sohn MW, Arnold N, Maynard C, Hynes DM. Accuracy and completeness of mortality data in the Department of Veterans Affairs. Popul Health Metr. 2006;4(1). 10.1186/1478-7954-4-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Lapham GT, Achtmeyer CE, Williams EC, Hawkins EJ, Kivlahan DR, Bradley KA. Increased documented brief alcohol interventions with a performance measure and electronic decision support. Med Care. 2012;50(2):179–87. [DOI] [PubMed] [Google Scholar]
- 22.Husain N, Blais P, Kramer J, Kowalkowski M, Richardson P, El-Serag HB, et al. Nonalcoholic fatty liver disease (NAFLD) in the Veterans Administration population: development and validation of an algorithm for NAFLD using automated data. Aliment Pharmacol Ther. 2014;40:949–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.van Houwelingen JC, Putter H. Dynamic Prediction in Clinical Survival Analysis. CRC Press; 2012. [Google Scholar]
- 24.Li L, Luo S, Hu B, Greene T. Dynamic prediction of renal failure using longitudinal biomarkers in a cohort study of chronic kidney disease. Stat Biosci. 2017;9:357–78. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Dafni U. Landmark analysis at the 25-year landmark point. Circ Cardiovasc Qual Outcomes. 2011;4:363–71. [DOI] [PubMed] [Google Scholar]
- 26.Wu C, Li L, Li R. Dynamic prediction of competing risk events using landmark sub-distribution hazard model with multiple longitudinal biomarkers. Stat Methods Med Res. 2020;29(11):3179–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Gleiss A, Oberbauer R, Heinze G. An unjustified benefit: Immortal time bias in the analysis of time-dependent events. Transpl Int. 2018;31:125–30. [DOI] [PubMed] [Google Scholar]
- 28.American Diabetes Association. 6. Glycemic targets: Standards of medical care in diabetes—2019. Diabetes Care. 2019;42:S61–S70. [DOI] [PubMed] [Google Scholar]
- 29.Gao C, Fang L, Zhang H, Zhang WS, Li XO, Du SY. Metformin induces autophagy via the AMPK-mTOR signaling pathway in human hepatocellular carcinoma cells. Cancer Manag Res. 2020;12:5803–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Ben Sahra I, Laurent K, Loubat A, Giorgetti-Peraldi S, Colosetti P, Auberger P, et al. The antidiabetic drug metformin exerts an antitumoral effect in vitro and in vivo through a decrease of cyclin D1 level. Oncogene. 2008;27(25):3576–86. [DOI] [PubMed] [Google Scholar]
- 31.Sun Y, Tao C, Huang X, He H, Shi H, Zhang Q, et al. Metformin induces apoptosis of human hepatocellular carcinoma HepG2 cells by activating an AMPK/p53/miR-23a/FOXA1 pathway. Onco Targets Ther. 2016;9:2845–53. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Cameron AR, Morrison VL, Levin D, Mohan M, Forteath C, Beall C, et al. Anti-inflammatory effects of metformin irrespective of diabetes status. Circ Res. 2016;119:652–65. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.White DL, Thrift AP, Kanwal F, Davila J, El-Serag HB. Incidence of hepatocellular carcinoma in all 50 United States, from 2000 through 2012. Gastroenterology. 2017;152(4):812–20.e5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Balakrishnan M, Patel P, Dunn-Valadez S, Dao C, Khan V, Ali H, et al. Women have a lower risk of nonalcoholic fatty liver disease but a higher risk of progression vs men: A systematic review and meta-analysis. Clin Gastroenterol Hepatol. 2021;19(1):61–71.e15. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Kanwal F, Shubrook JH, Adams LA, Pfotenhauer K, Wai-Sun Wong V, Wright E, et al. Clinical care pathway for the risk stratification and management of patients with nonalcoholic fatty liver disease. Gastroenterology. 2021;161(5):1657–69. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.