Thiazolidinediones show lower hepatocellular carcinoma risk than DPP-4 inhibitors, GLP-1RAs, insulin, and sulfonylureas in this network meta-analysis

This network meta-analysis examined the association between glucose-lowering drug classes and major adverse liver outcomes in a population of 7124845 adults with type 2 diabetes. The study design included observational studies, which precludes causal inference. The analysis compared thiazolidinediones, dipeptidyl peptidase 4 inhibitors, glucagon-like peptide 1 receptor agonists, insulin, sulfonylureas, and sodium-glucose cotransporter 2 inhibitors against all other drug classes. The setting and specific publication details were not reported in the source data. The primary outcome was major adverse liver outcomes, with secondary outcomes including hepatocellular carcinoma, decompensation, cirrhosis, variceal bleeding, hepatic encephalopathy, and liver-related mortality.

Regarding hepatocellular carcinoma, thiazolidinediones were associated with the lowest risk. The hazard ratio was 0.50 versus dipeptidyl peptidase 4 inhibitors, 0.72 versus glucagon-like peptide 1 receptor agonists, 0.20 versus insulin, and 0.69 versus sulfonylureas. All these comparisons were statistically significant. In contrast, dipeptidyl peptidase 4 inhibitors, glucagon-like peptide 1 receptor agonists, insulin, and sulfonylureas showed higher associations with this outcome compared to thiazolidinediones. Absolute numbers for these events were not reported.

For liver decompensation, glucagon-like peptide 1 receptor agonists were associated with the lowest hazard compared with all other drug classes. The hazard ratios ranged from 0.16 to 0.91, and all comparisons were statistically significant. Sodium-glucose cotransporter 2 inhibitors were least associated with cirrhosis, showing a hazard ratio of 0.66 versus dipeptidyl peptidase 4 inhibitors and 0.66 versus glucagon-like peptide 1 receptor agonists. The confidence intervals for this specific comparison were not reported.

The analysis also assessed variceal bleeding and hepatic encephalopathy. Glucagon-like peptide 1 receptor agonists were least associated with variceal bleeding, and they were also least associated with hepatic encephalopathy. Specific effect sizes, absolute numbers, and confidence intervals for these two outcomes were not reported in the source data. Similarly, sodium-glucose cotransporter 2 inhibitors were least associated with liver-related mortality, though the specific effect size, absolute numbers, and confidence intervals for this outcome were not reported.

Safety and tolerability data were not reported in the source material. Discontinuation rates and specific adverse event profiles were not provided. The study limitations are significant because all included studies were observational. This design precludes causal inference regarding the drug effects on liver outcomes. Randomized trials are needed to determine whether these associations reflect true drug effects rather than confounding factors. The certainty of the evidence was not reported.

Clinical implications suggest that liver-specific risk reduction is not uniform across antihyperglycemic drug classes. While thiazolidinediones appear favorable for hepatocellular carcinoma risk in this observational synthesis, the lack of randomized trial data means clinicians cannot definitively attribute these benefits to the drug mechanism. The findings highlight the need for caution when interpreting observational associations as causal. Questions remain unanswered regarding the long-term safety of these agents in patients with pre-existing liver disease. Further research is required to validate these associations in controlled settings before they can guide prescribing decisions.