Abstract
OBJECTIVE: Previous systematic reviews of retrospective cohorts (RSC) indicate that statin use decreases the risk of liver cancer. However, the summary effect size (sES) of the randomized controlled trials was not statistically significant. This study aimed to conduct a subgroup meta-analysis based on the types of constructed cohorts. METHODS: RSCs were selected from previous systematic reviews. Based on the characteristics of the source database (national vs. hospital) and the selection criteria of the subjects (population vs. patients), RSCs were categorized into three types of study cohorts: a national-based population cohort (NPo), national-based patient cohort (NPa), and hospital-based patient cohort (HPa). The sES and 95% confidence intervals were calculated using a random-effects model. RESULT: The 28 cohorts from 23 RSC were classified into 15 NPa, 7 NPo, and 6 HPa. The sES of 15 NPa decreased the liver cancer risk with statin intake history with statistical significance, but 7 NPo lost statistical significance. CONCLUSION: The lack of statistical significance in NPo supports the argument that the conclusions of existing systematic reviews on RSC have low validity. It is necessary to conduct a subgroup meta-analysis of the NPo, NPa, and HPa proposed in this study when conducting a systematic review of RSCs, which will evaluate various outcomes of a specific drug intake with time-varying exposure.