Treatment and intention-to-treat propensity score analysis to evaluate the impact of video-assisted thoracic surgery on 90-day mortality after anatomical resection for lung cancer.

Abstract

The aim of this study was to know the treatment effect of video-assisted thoracic surgery (VATS) on 90-day mortality after anatomical lung resection based on a nationwide cohort. This is a multicentre prospective cohort of 2721 anatomical resections for lung cancer from December 2016 to March 2018. Treatment and intention-to-treat (ITT) analyses were performed after inverse probability score weighting and different propensity score matching algorithms. Covariate balance was assessed by standardized mean differences. The estimators reported were the average treatment effect, the average treatment effect on the treated and odds ratios after conditional logistic models with 95% confidence intervals. The unconfoundedness assumption was evaluated by sensitivity analysis for average treatment effect (c-dependence) and average treatment effect on the treated (Γ). VATS was the initial approach in 1911 patients (70.2%), though 273 cases (14.3%) had to be converted to thoracotomy. Ninety-day mortality rates were: treatment analysis (VATS 1.16% vs open 3.9%, P < 0.001), ITT analysis (VATS 1.78% vs open 3.36%, P = 0.012). After inverse probability score weighting and propensity score matching, in the treatment analysis, VATS meant absolute risk reductions between 2.25% and 2.96% and relative risk reductions between 65% and 70% [OR = 0.34 (95% confidence interval 0.15-0.79), all P-values <0.004). However, all the estimators turned out to be non-significant in the ITT analyses. A high sensitivity to unobservable confounders was proved (c-dependence 0.135, Γ = 1.5). VATS can reduce the risk of 90-day mortality after anatomical lung resection. However, the implications of conversion to thoracotomy, comparing ITT versus treatment analysis, and the potential impact of hidden bias should deserve further attention in the future.