Presentation (CTI)

OBJECTIVES: PROs are routinely employed to generate time to deterioration (TTD) endpoints. A recent literature review demonstrated that when clinical terminal events (CTEs; OS/PFS) demonstrate efficacy, PRO-based TTD endpoints fail 72% of the time. In contrast to CTEs, TTD data contain terminal and recurrent events: participants deteriorate, recover (with or without intervention) and deteriorate again, generating a time series of deterioration. The existing TTD framework has two flaws:
1. Forcing a single-event model onto recurrent event data, eg: TTFD analyzes time to first deterioration ignoring subsequent deteriorations; and TUDD ignores events preceding the event captured in the last record.
2. Existing frameworks cannot evaluate how CTEs and TTD are related. This research presents a recurrent-event joint survival model (JM) analyzing the entire time series of deterioration events and directly estimating how recurrent PRO-based deterioration predicts CTEs.
METHODS: Applied examples will be presented from 4 oncology trials. Methods and results characterize one: NCT01124786 was a Phase 2 trial evaluating efficacy of gemcitabine-5-elaidate (GE) vs. gemcitabine (G) for Metastatic Pancreatic Adenocarcinoma.
Recurrent Brief Pain Inventory deterioration (R-BPI-D) events were analyzed to Cycle 25 via frailty Cox models. The primary terminal endpoint, OS to Cycle 25, was analyzed via the cox proportional hazard model. The JM estimated all effects predicting risk of OS from R-BPI-D risk, using JMBayes2; R version 4.3.2.
RESULTS: Of 299 analysis-eligible participants, 141 (42.99%), 137 (41.77%), 18 (6%), 2 (0.67%) and 1 (0.33%) experienced zero (censored), one, two, three, or eight BPI-D events, respectively.
Analysis demonstrated that each additional R-BPI-D event strongly predicted risk of death (ln(HR) = 8.26, p = 0.00001) and was representative of findings in all trials examined across varying cancers and PROs.
CONCLUSIONS: This presentation will use evidence from 4 oncology trials illustrating these two limitations of the existing PRO-based TTD framework and how the proposed approach resolves them.