ACCURACY OF LIFE YEAR GAIN PREDICTIONS FOR NIVOLUMAB MONOTHERAPY IN THE LONG TERM: AN ANALYSIS ACROSS FOUR INDICATIONS
Author(s)
Porteous A1, van Hest N2, Curteis T3, Wickstead R1
1Costello Medical, London, UK, 2Costello Medical, Cambridge, CAM, UK, 3Costello Medical, Cambridge, UK
OBJECTIVES Survival profiles for immunotherapies such as nivolumab may exhibit a plateau or a delayed effect, leading to complex long-term hazard functions. We aimed to retrospectively analyze the accuracy of overall survival (OS) extrapolations of interim data cuts in predicting realized long-term life years (LYs), to ascertain whether certain survival models might better predict long-term survival with nivolumab across indications. METHODS Standard parametric models, spline models (1–2 knots; normal, odds and hazard) and mixture cure models (MCMs) were fitted to digitized published OS data for nivolumab from successive interim data cuts of CheckMate 057, 017, 067 and 141. Models were tested for statistical fit using Akaike and Bayesian information criteria. Cumulative LYs were estimated for each model over a time horizon corresponding to the longest duration of published OS data and compared to realized LYs over this period. RESULTS Considering standard parametric models, statistical fit to interim data cuts appeared to correlate with accuracy of LY predictions. The LogNormal and LogLogistic models provided best statistical fit and most accurate LY predictions, on average, with a mean absolute percentage difference between predicted and realized LYs across indications of 2.0% and 1.6%, respectively. Statistical fit to interim data cuts was not necessarily an indicator of LY prediction accuracy for spline models and MCMs. For interim data cuts with limited follow-up, MCMs were poor predictors of long-term survival, substantially overestimating LYs. However, these models became better predictors at successive data cuts with longer follow-up, suggesting a minimum follow-up requirement for MCMs to accurately predict long-term survival. CONCLUSIONS Models reflecting non-monotonic hazards were consistently associated with better statistical fit and more accurate predictions of long-term survival for nivolumab across indications than alternative standard parametric models. MCMs may also be appropriate to model long-term survival for nivolumab, but only if data with sufficient follow-up are available.
Conference/Value in Health Info
2020-05, ISPOR 2020, Orlando, FL, USA
Value in Health, Volume 23, Issue 5, S1 (May 2020)
Code
PCN4
Topic
Clinical Outcomes
Topic Subcategory
Relating Intermediate to Long-term Outcomes
Disease
Oncology