THE IMPACT OF DIFFERENT TIME-VARYING NETWORK META-ANALYSIS METHODS TO ASSESS TALAZOPARIB + ENZALUTAMIDE VERSUS OLAPARIB + ABIRATERONE IN METASTATIC CASTRATION-RESISTANT PROSTATE CANCER
Author(s)
Melissa Kirker, PharmD1, Neo Su, MPH, MS, PharmD2, Jack Said, MSc3, Adam Gough, MSc4, Emma Tyas, PhD5, Hannah Kilvert, MSc6, Atharva Phadke, MSc7, Amy Barber, MSc5, Yang Meng, PhD8.
1Pfizer, Washington, DC, USA, 2Pfizer, Houston, TX, USA, 3Pfizer, Manchester, United Kingdom, 4Lumanity, Boston, MA, USA, 5Lumanity, Sheffield, United Kingdom, 6Lumanity, SHEFFIELD, United Kingdom, 7Lumanity, Utrecht, Netherlands, 8Lumanity, Las Vegas, NV, USA.
1Pfizer, Washington, DC, USA, 2Pfizer, Houston, TX, USA, 3Pfizer, Manchester, United Kingdom, 4Lumanity, Boston, MA, USA, 5Lumanity, Sheffield, United Kingdom, 6Lumanity, SHEFFIELD, United Kingdom, 7Lumanity, Utrecht, Netherlands, 8Lumanity, Las Vegas, NV, USA.
OBJECTIVES: The Phase III TALAPRO-2 trial showed that talazoparib plus enzalutamide (TALA+ENZA) improved overall survival (OS) and radiographic progression-free survival (rPFS) versus ENZA alone in metastatic castration-resistant prostate cancer (mCRPC). For health technology assessment, indirect comparisons with key comparators, such as olaparib + abiraterone (OLA+ABI), were needed. Due to concerns about non-proportional hazards in the evidence network, we applied time-varying network meta-analysis (NMA) methods to estimate the relative OS and rPFS of TALA+ENZA versus key comparators, to inform the cost-effectiveness analysis in mCRPC.
METHODS: Three time-varying NMA methods were used: fractional polynomial NMA (FP-NMA), restricted cubic spline NMA (RCS-NMA), and parametric NMA (pNMA). Comparator survival was estimated by applying time-varying HRs to the TALA+ENZA reference curve. Average survival time was calculated and compared across treatments.
RESULTS: FP and RCS-NMA models did not converge for OS, and pNMA models produced HRs with wide credible intervals (CrIs) that were uninformative for decision making, so OS results are not presented. Convergence was improved for rPFS, likely because data were more mature. Preferred models, selected by statistical and visual fit, included: first order p=-1 for FP-NMA, log-normal for pNMA, and two-knot for RCS-NMA. Results based on the FP-NMA and pNMA models indicated that TALA+ENZA provides longer average rPFS compared with OLA+ABI, with 95% CrIs excluding 0. Conversely, results based on the RCS-NMA showed 95% CrI including 0, indicating no evidence of a difference between treatments. Time-varying HRs and their statistical significance are challenging to interpret, so differences in mean survival time were used to provide a single estimate of relative efficacy. Model choice substantially influenced the mean survival for all treatments.
CONCLUSIONS: TALA+ENZA demonstrated greater rPFS compared with ENZA and OLA+ABI, regardless of the time-varying NMA method. The choice of time-varying NMA method can have a meaningful impact on survival estimates.
METHODS: Three time-varying NMA methods were used: fractional polynomial NMA (FP-NMA), restricted cubic spline NMA (RCS-NMA), and parametric NMA (pNMA). Comparator survival was estimated by applying time-varying HRs to the TALA+ENZA reference curve. Average survival time was calculated and compared across treatments.
RESULTS: FP and RCS-NMA models did not converge for OS, and pNMA models produced HRs with wide credible intervals (CrIs) that were uninformative for decision making, so OS results are not presented. Convergence was improved for rPFS, likely because data were more mature. Preferred models, selected by statistical and visual fit, included: first order p=-1 for FP-NMA, log-normal for pNMA, and two-knot for RCS-NMA. Results based on the FP-NMA and pNMA models indicated that TALA+ENZA provides longer average rPFS compared with OLA+ABI, with 95% CrIs excluding 0. Conversely, results based on the RCS-NMA showed 95% CrI including 0, indicating no evidence of a difference between treatments. Time-varying HRs and their statistical significance are challenging to interpret, so differences in mean survival time were used to provide a single estimate of relative efficacy. Model choice substantially influenced the mean survival for all treatments.
CONCLUSIONS: TALA+ENZA demonstrated greater rPFS compared with ENZA and OLA+ABI, regardless of the time-varying NMA method. The choice of time-varying NMA method can have a meaningful impact on survival estimates.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR16
Topic
Methodological & Statistical Research
Disease
SDC: Oncology