Using Regression Discontinuity in Time Design for Real-World Comparative Effectiveness: A Case Study for the Second-Line Use of Pembrolizumab in Advanced Non-Small Cell Lung Cancer
Author(s)
Nai-Chia Chen, MS1, Antal T. Zemplenyi, MSc, PhD1, Blythe Adamson, MPH, PhD2, Cindy L. OBryant, PharmD1, Robert B. McQueen, BA, MA, PhD1, Kelly Anderson, MPP, PhD1;
1University of Colorado Anschutz Medical Campus, Aurora, CO, USA, 2Flatiron Health, New York, NY, USA
1University of Colorado Anschutz Medical Campus, Aurora, CO, USA, 2Flatiron Health, New York, NY, USA
Presentation Documents
OBJECTIVES: Real-world evidence offers promising opportunities for comparative effectiveness research but can be subject to unobserved confounders tied to treatment selection and patient outcomes. Regression discontinuity in time (RDiT) is a quasi-experimental framework adapting regression discontinuity where time is the running variable. The study aimed to model and compare changes in treatment probabilities over time and their impact on real-world comparative effectiveness outcomes versus trial efficacy when comparing immunotherapies to chemotherapy, using a case-study of advanced non-small-cell lung cancer (aNSCLC).
METHODS: A retrospective cohort of the second-line use of pembrolizumab against docetaxel for aNSCLC data was extracted. This study used the nationwide Flatiron Health electronic health record (EHR)-derived de-identified database. Both the RDiT or time-stratified inverse probabilities treatment weighting (time-stratified IPTW) methods were used to estimate treatment probabilities over time. Outcome measures derived from estimated treatment probabilities included median survival and hazard ratios (HR) with 95% confidence intervals. Individual patient-level data from the targeted trial with 5-year follow-up were reconstructed for performance assessments.
RESULTS: A total of 1,975 patients from 2011 to 2023 were included in our cohort, with 1,014 patients receiving pembrolizumab monotherapy and 961 patients receiving docetaxel monotherapy after failing platinum-based chemotherapy. Overall, pembrolizumab had better survival outcomes with respect to different specifications of baseline hazard functions. Adjusted median overall survival and HRs between pembrolizumab and docetaxel were 11.5 months vs 6.9 months (HR 0.65, 95% CI: 0.48, 0.88) in RDiT method, and 13.6 months vs 7.3 months (HR 0.52, 95% CI: 0.42, 0.65) in time-stratified IPTW method, compared to 11.8 months vs 8.4 months in the targeted trial (HR 0.70, 95% CI: 0.61, 0.80).
CONCLUSIONS: In the presence of unobserved confounders, in this case study, RDiT estimated smaller differences in absolute survival gains to trial-based efficacy as compared to propensity score adjustments.
METHODS: A retrospective cohort of the second-line use of pembrolizumab against docetaxel for aNSCLC data was extracted. This study used the nationwide Flatiron Health electronic health record (EHR)-derived de-identified database. Both the RDiT or time-stratified inverse probabilities treatment weighting (time-stratified IPTW) methods were used to estimate treatment probabilities over time. Outcome measures derived from estimated treatment probabilities included median survival and hazard ratios (HR) with 95% confidence intervals. Individual patient-level data from the targeted trial with 5-year follow-up were reconstructed for performance assessments.
RESULTS: A total of 1,975 patients from 2011 to 2023 were included in our cohort, with 1,014 patients receiving pembrolizumab monotherapy and 961 patients receiving docetaxel monotherapy after failing platinum-based chemotherapy. Overall, pembrolizumab had better survival outcomes with respect to different specifications of baseline hazard functions. Adjusted median overall survival and HRs between pembrolizumab and docetaxel were 11.5 months vs 6.9 months (HR 0.65, 95% CI: 0.48, 0.88) in RDiT method, and 13.6 months vs 7.3 months (HR 0.52, 95% CI: 0.42, 0.65) in time-stratified IPTW method, compared to 11.8 months vs 8.4 months in the targeted trial (HR 0.70, 95% CI: 0.61, 0.80).
CONCLUSIONS: In the presence of unobserved confounders, in this case study, RDiT estimated smaller differences in absolute survival gains to trial-based efficacy as compared to propensity score adjustments.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR87
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Oncology