Identifying Effect of a Treatment from Observational Data in Absence of Contemporaneous Control: Example of the French Study Tosca, Comparing Cemiplimab Vs. Historical Systemic Treatment in Cutaneous Squamous Cell Carcinoma

OBJECTIVES: The anti-PD1 antibody cemiplimab was approved for patients with advanced cutaneous squamous cell carcinoma (CSCC). In the absence of evidence from Randomized Controlled Trial, the TOSCA study aims to evaluate the extent of benefit of cemiplimab versus Historical Systemic Treatment (HST) from observational retrospective data.

METHODS: TOSCA is a retrospective, multicenter study (NCT05302297) in patients with advanced-CSCC treated with cemiplimab via the Early Access Program (2018-2019) or with HST (2013-2018). Primary endpoint was overall survival (OS). As most eligible patients were treated with cemiplimab once available in 2018, use of an historical cohort was necessary. The main challenge was to control for imbalance in confounders between treatment groups. Propensity Score of exposure to treatment of interest was based on a set of confounders identified through systematic literature review and validated by a Scientific Committee. A bootstrapping method was used to ensure comparability on the number of treatment lines. Inverse probability weighting (IPW) approach was used for effectiveness analysis. Further sensitivity analyses including various estimands in terms of population set – “trial-like” in main analysis vs. larger “real-life” set – were prespecified to ensure robustness of the results.

RESULTS: Data from 199 patients in 28 sites (cemiplimab, n=129 and HST, n=70) were included in main effectiveness analysis. With a median follow-up of 20 and 10 months for cemiplimab and HST arms, and after controlling for confounding using IPW, the median OS was 21 and 10 months (hazard ratio [95%CI]: 0.57 [0.45-0.73]; P-value <0.0001), respectively. Sensitivity analyses were consistent with the main results.

CONCLUSIONS: TOSCA study showed longer survival in patients treated with cemiplimab versus HST. Since the absence of contemporaneous control could induce large bias in observational studies, the use of historical data requires appropriate methodology for limiting bias and ensuring robustness of the results.