Background: When comparing the efficacy of surgical and non-surgical therapies for hepatocellular carcinoma (HCC), a major limitation is the causal inference problem. This concerns the impossibility of seeing both outcomes of two different treatments for the same individual at the same time because one is inevitably missing. This aspect can be addressed methodologically by estimating the so-called average treatment effect (ATE).
Methods: To estimate the ATE of hepatic resection over locoregional therapies for HCC, data from patients treated in two tertiary care settings between August 2000 and December 2014 were used to obtain counterfactual outcomes using an inverse probability weight survival adjustment.
Results: A total of 1585 patients were enrolled: 815 underwent hepatic resection, 337 radiofrequency ablation (RFA) and 433 transarterial chemoembolization (TACE). The option of operating on all patients who had tumour ablation returned an ATE of +9·8 months for resection (effect size 0·111; adjusted P = 0·064). The option of operating on all patients who had TACE returned an ATE of +27·9 months (effect size 0·383; adjusted P < 0·001). The ATE of surgery was negligible in patients undergoing ablation for very early HCCs (effect size 0·027; adjusted P = 0·627), independently of albumin-bilirubin (ALBI) grade; or in patients with ALBI liver function grade 2 (effect size 0·083; adjusted P = 0·213), independently of tumour stage. In all other instances, the ATE of surgery was notably greater. Operating on patients who had TACE with multinodular HCC beyond the Milan criteria resulted in a mild ATE (effect size 0·140; adjusted P = 0·037).
Conclusion: ATE estimation suggests that hepatic resection is a better treatment option than ablation and TACE in patients with HCC.
© 2017 BJS Society Ltd Published by John Wiley & Sons Ltd.