A comparative analysis of efficacy results in clinical trial publications and US prescribing information demonstrates a high degree of discordance in reported response endpoints for Acute Myeloid Leukemia

Abstract The efficacy results in clinical trial publications inform medical practice, build a reliable knowledge base, and provide statistics for hypotheses in future clinical trials. The US Prescribing Information (USPI) contains the efficacy results confirmed independently by FDA based on patient-level data submitted in support of a marketing application. Because the clinical trial publications and USPIs show efficacy results based on prespecified analyses, the efficacy results reported in publications and the USPI should be the same, but this has not been evaluated previously. We identified clinical trials that supported FDA drug approvals for treatment of acute leukemias from 2014 to 2025 by reviewing USPIs. The indication, study identifier, patient age group, endpoint used as the basis of efficacy (BOE), sample size for the efficacy population, and results of the analysis were extracted from the USPI. The study identifier was then used to retrieve the clinical trial publication. The prespecified efficacy endpoints, sample size for the efficacy population, and results of the efficacy analyses were extracted from the publication and/or information at clinicaltrials.gov for comparison to the data in the USPI. There were 34 trials identified from the USPIs. No publication was found for 1 trial, and for 1 approval, data were pooled from 2 trials, so 32 USPI-publication pairs were available for comparison. Nineteen (59%) trials were for AML, 12 (38%) were for ALL, and 1 (3%) was for any acute leukemia. Twenty-eight (88%) were for adult patients, 2 (6%) for pediatric patients, and 2 (6%) for both age groups. The BOE endpoint was response (CR, CR/CRh, MRD-negative CR) for 19 (59%) trials and time to event (TTE) (OS, EFS) for 13 (41%) trials. The BOE endpoint in the USPI was not the primary endpoint for 17 (53%) trials and was not a prespecified endpoint for 7 (22%) trials. The sample size of the efficacy population in the USPI did not match that in the trial publication for 16 (50%) trials; the efficacy population size was different more frequently with response endpoints (63%) than with TTE endpoints (31%). The point estimate of the BOE endpoint in the USPI did not match that reported in the trial publication for 20 (63%) trials, more frequently with response endpoints (74%) than with TTE endpoints (46%). Overall, there were only 9 (28%) trials where the BOE endpoint in the USPI was a prespecified endpoint in the trial in addition to the efficacy population sizes and the point estimates of the BOE being the same in the USPI and trial publication; of these, 6 used OS as the endpoint and 3 used a response endpoint. Among the subgroup of 19 AML trials, the BOE endpoint was response for 9 (47%) trials and TTE for 10 (53%). The sample size of the efficacy population in the USPI did not match that in the trial publication for 9 (47%) of the trials and more frequently with response endpoints (67%). The BOE endpoint in the USPI was not the primary endpoint for 9 (47%) trials and was not a prespecified endpoint in 2 (11%) trials. The point estimate of the BOE endpoint in the USPI did not match that reported in the trial publication for 10 (52%) of the trials, more frequently with response endpoints (60%). In only 7 (37%) of the AML trials, the BOE endpoint in the USPI was a prespecified endpoint in the trial in addition to the efficacy population sizes and the point estimates of the efficacy endpoints being the same in the USPI and trial publication; of these, 5 used OS as the endpoint and 3 used a response endpoint. We found a substantial discordance in clinical trial efficacy results reported in the USPI compared to the clinical trial publications, especially when the BOE endpoint was response rather than TTE. While some of the differences may be due to use of data sets with different cut-off dates, the high degree of discordance in results for response endpoints raises questions about the clarity of the response criteria in the study protocols. For the clinical trials in this analysis, the differences noted were not likely clinically meaningful since they were used as the basis of the approvals, but a limitation of this study is that it does not include applications that were not approved, since those study reports are not typically publicly available. How the differences in reported efficacy results impact the integrity of future trial designs based on these results remains to be determined.