OBJECTIVES

: No methods exist on how real-world (RW) outcomes for new therapies in a pre-approval setting could be predicted. We aimed to develop a methodological framework for this purpose.

METHODS

: Our study was based on Pembrolizumab in patients with advanced-non-small-cell-lung-cancer (aNSCLC). Using patient-level data from the KEYNOTE-024 trial (generated by applying a STATA reconstruction algorithm/digitizer software approach based on the published Kaplan-Meier curve) and real-world (RW) data in the target-label population in a German claims dataset, we stepwise predicted overall survival (OS) by; (1) adjusting trial patient characteristics using a propensity-score-weighting method, (2) adjusting for expected differences in drug dosage/adherence, (3) adjusting for other RW effects, and (4) predicting long-term OS by fitting the simulated KM curve in parametric models. We validated our approach by comparing the predicted OS with the RW-OS of aNSCLC patients treated with Pembrolizumab.

RESULTS

: In KEYNOTE-04, 154/151 aNSCLC patients received Pembrolizumab/chemotherapy (median OS was not reached in either group). After step 1 (based on 206 RW-patients), simulated OS of the Pembrolizumab decreased (median OS Pembrolizumab 13). Dosage effects adjustment was not done as the respective phase-Ib study (KEYNOTE-001) did not identify a dosage-outcome-relationship.

Comparing OS for chemotherapy based on KEYNOTE-024 and RW chemotherapy OS adjusted for the clinical trial patient characteristic, a HR of 1.255 was identified and applied to the Pembrolizumab curve (finally predicted median OS: 10 months). Based on survival curve fitting, this corresponded with a 12‑/24-months OS of 45%/20%.

In comparison to median OS (8.5 months) of 32 aNSCLC patients who received Pembrolizumab in our RW database, our predicted OS was a much better proxy than original trial results.

CONCLUSIONS

: Our approach might provide a framework for the pre-approval prediction of RW-outcomes of new therapies.