Metformin and Risk of Hepatocellular Carcinoma in Patients With Type 2 Diabetes

Chin-Hsiao Tseng


Liver International. 2018;38(11):2018-2027. 

In This Article


This study found that metformin use in patients with type 2 diabetes mellitus was associated with a significantly lower risk of HCC. In addition, the findings supported an interaction between metformin and aspirin or statin.

Although analyses by Chi square test showed significant P–values for nephropathy, insulin, sulfonylurea and acarbose in the matched cohort, these probably would not lead to residual confounding because all their values of standardized difference were <10% (Table 1). In addition, a significantly lower risk of HCC associated with metformin use could be seen in all subgroups of patients with or without these characteristics in secondary analyses (data not shown).

It is worth noting that only approximately 11.2% of the patients were never-users of metformin (Figure 1). This low percentage is understandable, for the following reasons. Before 1995, only insulin, sulfonylurea and metformin were available in Taiwan. In 1995, acarbose was introduced in Taiwan, and several newer drugs including meglitinide, rosiglitazone and pioglitazone were introduced during the NHI enrolment period from 1999 to 2005. Although more choices were available during the enrolment period, sulfonylurea and metformin remained the main oral antidiabetic drugs prescribed by physicians at that time. In addition, metformin has always been recommended as the first-line therapy for type 2 diabetes in several guidelines, including the guideline of the American Diabetes Association,[21] which is always followed in clinical practice in Taiwan. In consideration of the fact that the guidelines for the use of antidiabetic drugs have evolved over the years, secondary analyses were conducted to calculate the proportions of never-users of metformin and to estimate the PS-weighted hazard ratios for patients enrolled during each specific year from 1999 to 2005 in the unmatched original cohort. It was noted that the proportions of never-users of metformin increased gradually from 7.4% in 1999 to 18.3% in 2005. However, the hazard ratios remained statistically significant for each specific year, ranging from 0.43 to 0.67 (data not shown).

It is not known whether some unmeasured confounders could be associated with the reasons for the 21 900 never-users in the study never being treated with metformin. The main contraindication for the use of metformin is renal dysfunction that may potentially lead to fatal lactic acidosis. However, this did not seem to be a potential confounder because, as discussed previously, the lower risk of HCC associated with metformin use could also be observed in patients with or without a diagnosis of nephropathy in the secondary analyses. For patients with a diagnosis of nephropathy, the estimated hazard ratio (95% confidence interval) was 0.56 (0.47–0.67), and for those without such a diagnosis, it was 0.48 (0.43–0.53).

Obesity is an important risk factor for HCC.[3] In the tertile analysis of the matched cohort, a significantly increased risk was observed in the first tertile (Table 2). Since metformin is always considered as first-line treatment for obese patients, a residual confounding by obesity could not be excluded in this subgroup of patients who had used metformin for a short period, and the impact of obesity on HCC could be carried over from a non-pharmacological control status to metformin initiation. However, the prevalence of obesity (defined by ICD–9–CM code 278) in the secondary analyses did not differ among the 3 subgroups of tertiles, in both the unmatched cohort (3.3% in each subgroup) and the matched cohort (2.0% in each subgroup), so the potential impact of some unknown factors could not be completely excluded.

Statin has been shown to reduce the risk of HCC,[22] and combined metformin and statin treatment may also reduce the risk of HCC.[23] It is true that patients with type 2 diabetes mellitus have a high risk of dyslipidaemia and the use of statin is very common. Therefore, the potential confounding effect of statin and the interaction between metformin and statin should not be neglected. Although the effect of aspirin on HCC is not well established,[24] this study suggested a potential beneficial effect of aspirin and its interaction with metformin (Table 4).

This study has carefully considered the adjustment for potential confounders (Table 1), and has addressed several potential biases that have not been fully considered in previous studies. These may include selection bias, prevalent user bias, immortal time bias and confounding by indication.

Selection bias can be avoided using Taiwan's NHI, a universal healthcare plan that covers more than 99% of Taiwan's population. Prevalent user bias can be avoided by enrolling new users of the drug. Immortal time bias may be introduced by inappropriately assigning treatment status and follow-up time to the patients.[25] By including patients who had been followed up and prescribed antidiabetic drugs 2 or more times in outpatient clinics (Figure 1), the diagnosis of diabetes and the assignment of treatment status should not be erroneous. Patients with immortal time between diabetes diagnosis and the use of antidiabetic drugs were not included in the analyses. Furthermore, in the calculation of person-years, the chance of inappropriate assignment of follow-up time within the initial period of antidiabetic treatment was further reduced by including only those patients who had been followed up for at least 6 months (Figure 1).

Confounding by indication was less likely in the matched cohort analyses because never-users and ever-users were balanced in potential confounders as indicated by the fact that none of the covariates had a standardized difference >10% (Table 1). The use of Cox regression incorporated with IPTW was also aimed at reducing such a potential confounding by indication.

This study has the merit of using the large NHI database, so the findings can be generalized to the whole population. Potential bias related to self-reporting can also be reduced using medical records. Biases from disparity in healthcare accessibility and differences in disease detection are less likely in Taiwan than in many other countries, because the NHI is a universal healthcare system and drug cost-sharing is low. Furthermore, the Bureau of the NHI considers cancer a catastrophic illness and most copayments by patients with cancer can be waived.

Limitations of the study may include the lack of measurement data on some potential risk factors such as biochemistry, abdominal ultrasound data, anthropometric factors, lifestyle, smoking, alcohol drinking, nutritional status, dietary patterns, family history, biomarkers of viral hepatitis and genetic markers. This study did not have data on the pathology, grading and staging of HCC. Whether the findings derived from the diabetes patients can be applied to non-diabetes patients awaits additional confirmation.

In summary, this study supports a beneficial effect of metformin use and a synergistic effect of metformin and aspirin/statin use on the prevention of HCC in patients with type 2 diabetes mellitus. Since metformin, aspirin and statin are inexpensive and commonly used in clinical practice, additional studies, especially clinical trials, are urgently needed to confirm their beneficial effects on the prevention and treatment of HCC in both diabetes and non-diabetes patients.