Presentation (CTI)

OBJECTIVES: State-transition modelling (STM) offers an alternative to partitioned-survival modelling, employing structural relationships between endpoints rather than assuming independence. However, STM typically requires individual-patient-data (IPD), which are often unavailable. When relying on published aggregate data (e.g., progression-free survival [PFS] and overall survival [OS]), existing literature-based methods for approximating competing-risk transitions, involve detecting event simultaneity from published Kaplan-Meier (KM) survival curves. This study evaluates the performance of simultaneity-based event assignment methods to estimate pre-progression death rates using simulated trial data.
METHODS: Censored IPD for progression and death events were simulated for a hypothetical oncology trial. KM curves and at-risk tables were generated at 3-month intervals. Pseudo-IPD for PFS and OS were derived from hand-digitized XY coordinates of the KM curves, using the Guyot algorithm. Pre-progression deaths were assigned by applying predefined error margins, which defined the time window within which PFS and OS events were considered simultaneous. Estimated pre-progression death counts, and cumulative hazards were compared to the simulated data across different error margins ranging up to 0.1 months.
RESULTS: In the simulated trial data (N=430, median follow-up ~33 months), 5.6% of the patients experienced pre-progression death and 53% experienced disease progression. Narrow error margins (e.g., 0.01 months) misclassified ~75% of pre-progression deaths as progression events. Wider margins (e.g., 0.1 months) overestimated pre-progression deaths by more than 100%. A similar trend of event-misclassification was observed in other simulated datasets with differing error-margins.
CONCLUSIONS: Simultaneity-based event assignment methods appear to have notable limitations when approximating competing-risk events from published KM survival curves. Their performance is highly sensitive to competing-event prevalence, data-extraction precision, and chosen simultaneity error-margins. Misclassification of competing events can potentially distort health-economic analyses employing STM, leading to inaccurate projections and unreliable decision-making in policy or clinical practice. Further research is required to assess the performance of alternative multi-state modelling approaches for approximating competing-risk events.