OBJECTIVES: The reconstruction of pseudo patient-level data (PLD) plays a critical role in the ability to accurately conduct indirect treatment comparisons (ITCs) from published survival data. The Guyot method¹ is a widely used algorithm that assumes censoring events follow a uniform distribution. However, censoring events where treatment discontinuation is due to changes in treatment regimen or adverse events may not follow a uniform distribution in practice. This study aims to explore the impact that different reconstruction methods may have on the accuracy of censoring times, and provide guidance on selecting an appropriate method based on type of ITC analysis.

METHODS: A targeted literature review was conducted to identify commonly used methodologies for generating pseudo PLD from Kaplan–Meier curves with either a constant relative hazard or a time-varying hazard ratio. This review assessed various pseudo PLD reconstruction methodologies, focusing on identifying use cases where the proportional hazard (PH) assumption was violated.

RESULTS: The choice of censoring distribution may lead to differences in the number of patients at risk throughout follow-up. When conducting time-varying ITCs for survival outcomes where the data violates the PH assumption, analysis results may be sensitive to differences in censoring times. Decision making considerations for the choice of reconstruction algorithm should be carefully considered based on ITC methodology.

CONCLUSIONS: It is important to think critically about the use case when selecting a method for pseudo-PLD reconstruction. When used in models where the PH assumption does not hold, the choice of algorithm should aim to reflect the event and censoring times of the patient population. In cases where PH does hold, analysis results may not be as strongly impacted by the number of patients at risk throughout the trial follow-up period.