Assessing the Accuracy of a Semi-Markov Model (SMM) with Split Local and Distant Recurrence Health States in Predicting Overall Survival (OS) for Muscle Invasive Urothelial Carcinoma (MIUC)
Author(s)
Kurt M1, Mamtani R2, Jain R3, Srinivasan S4, Broughton E5, Teitsson S6
1Bristol Myers Squibb, Lawrenceville, NJ, USA, 2University of Pennsylvania, Philadelphia, PA, USA, 3Parexel International, Bengaluru, India, 4University of North Carolina at Chapel Hill, Chapel Hill, NC, USA, 5An employee at the time the study was conducted; Bristol Myers Squibb, Lawrenceville, NJ, USA, 6Bristol Myers Squibb, Uxbridge, UK
OBJECTIVES: OS is the gold standard for demonstrating clinical benefit in oncology trials and its absence poses challenges in cost-effectiveness assessments. We examined the accuracy of a four-state SMM in predicting medium- (10-year) and long-term (30-year) OS using published data derived from the control arm of the EORTC-30994 study of adjuvant chemotherapy in MIUC.
METHODS: Disease evolution after radical cystectomy was captured by four mutually-exclusive health states as disease-free (DF), local recurrent (LR), distant-recurrent (DR) and death. Reconstructed DF survival (DFS) data was extrapolated using parametric- and spline-based survival models. Transitions from DF state were estimated via extrapolated DFS assuming constant relative frequency of first recurrence events over five years after which DF patients were considered functionally cured. Transitions from LR state were estimated by jointly utilizing the median survival reported from treatment initiation for LR patients and survival from DR state which was informed by the distribution of deferred chemotherapy treatments and indirect comparisons of OS from pivotal trials in first-line systemic therapy for MIUC. Model-predicted weekly OS-rates and restricted mean survival times (RMSTs) were compared with their counterparts obtained from reported and extrapolated OS.
RESULTS: Spline-based models provided the best statistical fit for both DFS and OS extrapolations. The estimated 10-year RMSTs from the reported OS and model-predicted OS were 5.23 and 5.19 years, respectively. Over 10-years, model-predicted OS laid within the 95% confidence interval of the reported OS 98.2% of the time. The estimated 30-year RMSTs from the extrapolated OS and model-predicted OS were 9.38 and 8.43 years, respectively. Model-predicted OS had an average absolute gap of 0.02 and 0.035 versus the reported OS and extrapolated OS over 10- and 30-years, respectively.
CONCLUSIONS: For MIUC patients receiving no adjuvant chemotherapy after radical cystectomy, OS projections from our SMM were accurate in the medium term and more conservative in the long-term.
Conference/Value in Health Info
Value in Health, Volume 25, Issue 6, S1 (June 2022)
Code
MSR24
Topic
Methodological & Statistical Research, Study Approaches
Topic Subcategory
Decision Modeling & Simulation
Disease
Oncology, Urinary/Kidney Disorders