Doctor: Guillaume Lonjon

Title: Use of the propensity score in observational studies evaluating a surgical procedure

Supervisors: Isabelle Boutron, Raphaël Porcher

Doctoral school: ED 393 Epidemiology and Biomedical Information Sciences, Université Paris Cité

Date of thesis defense: 17/03/2017

Jury: Bertrand Dousset, Paul Landais, Laurence Meyer, Rémy Nizard, Hugues Pascal-Mousselard, Isabelle Boutron


Surgery has changed in the last decade. Surgical techniques have advanced and are being used more frequently. Assessment of surgical procedures has become a public health priority, but evaluating a surgical procedure in randomized controlled trials (RCTs) is challenging. Observational study is an alternative, despite the presence of confusion bias. Use of propensity score (PS) analysis in an observational study is a way to limit confounding bias. The PS is defined for each participant as the probability of receiving the treatment, given baseline covariates. With the assumption of no unmeasured confounders, a treated and an untreated patient with the same PS can be considered as if they had been randomly assigned to each group. PS analysis has been increasingly used in the last 10 years, specifically in studies of surgery.

In this project, I aimed to 1) compare treatment effect estimates from non-randomized studies with PS analysis and RCTs and 2) describe and assess the reporting and potential bias of PS analysis used in a sample of published observational studies assessing surgical procedures.

In a first part, a meta-epidemiological study of 31 clinical questions in surgery revealed no statistically significant difference in treatment effect estimates between non- randomized studies with PS adjustment and RCTs, although the treatment effect estimates varied widely between the two study designs.

In a second part, from 652 reports selected, I systematically assessed the use of PS analysis in observational studies evaluating surgical procedures. The number of observational studies evaluating a surgical procedure with PS analysis increased from < 10 before 2000 to > 200 after 2013. However, the use and reporting of PS analysis in observational studies raises important concerns related to constructing the PS, accounting for missing data and cross- overs and reporting all features necessary to reproduce the analysis.

Observational studies with PS analysis of surgery are increasing in frequency and their use can be relied upon for evidence when RCTs are not possible. Specific methodological issues and weaknesses in reporting exist. Some limitations should be easily correctable, but some need more assessment.

Link to download the thesis (written in French)

Back to top