1709P - Influence of censoring on conclusions of FDA-approved cancer drugs using the modified time to treatment failure

Background

Time-to-treatment failure (TTF), in which both disease progression and treatment interruption are regarded as events, is a common measure in clinical trials that can be used as a sensitivity analysis for informative censoring. However, TTF is inappropriate for excessive censoring in the control group as it exaggerates the effect size instead of penalizing the bias. We propose a novel measure, the modified TTF (mTTF), that considers censoring as events only for excessive censoring in the intervention group. Here we examine whether excessive censoring in the intervention could affect the results of randomized controlled trials (RCTs) leading to FDA-approved drugs using mTFF.

Methods

We conducted a cross-sectional study collecting data from phase II and phase III RCTs of FDA drugs approved based on surrogate endpoints between 2010 and 2020. We reconstructed individual patient data from the published Kaplan-Meier curves of the primary surrogate endpoint. The reverse Kaplan-Meier method (i.e., events and censoring are flipped) was used to quantify excess censoring between study groups. The mTTF sensitivity analysis was conducted and statistical significance was calculated using the log-rank test.

Results

One hundred and one RCTs comprising 60,697 patients were included. The most common surrogate endpoint was progression-free survival with 86 (85.1%) of the studies. The median HR for surrogate endpoints was 0.54 (IQR [interquartile range], 0.41 to 0.63) and the median p-value was 3.0*10^-5 (IQR, 2.8*10^-8 to 1.5*10^-3). Following adjustment for excess censoring using the mTTF the median HR was 0.68 (IQR, 0.52 to 0.84) and the median p-value was 4.2*10^-3 (IQR, 2.5*10^-6 to 0.12). Across all trials, 35 (34.6%) lost statistical significance using the mTTF.

Conclusions

We introduce mTFF as a practical clinical endpoint that is less conservative than the TTF and is sensitive to excessive censoring in the intervention group. Our findings suggest that some studies are associated with an excessive drop-out rate favouring the intervention group that may lead to a false conclusion regarding the efficacy of FDA-approved treatments. Meaningful shared decision-making requires careful consideration of informative censoring.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

G. Markel: Financial Interests, Personal, Financially compensated role: 4c Biomed, Purple Biotech, Biond Biologics, Ella therapeutics, Nucleai, Beyond Air, Stareget; Non-Financial Interests, Personal, Advisory Board: Merck Sharp & Dohme, Sanofi SA, Novartis International AG, Bristol Myers Squib. T. Meirson: Financial Interests, Personal, Full or part-time Employment, I report receiving personal fees from Purple Biotech: Purple Biotech. All other authors have declared no conflicts of interest.