Analytical Approaches to Estimate Medication Persistence From Electronic Health Record Data: A Study of Tyrosine Kinase Inhibitors in Patients With Epidermal Growth Factor Receptor-Positive Advanced Non-Small Cell Lung Cancer
Author(s)
Xinye Li, ScM, Brooke Jarret, PhD, MSPH, Brendan T. Kerr, MS, Nathaniel Wade, PharmD, BCOP, Patrick Ward, PhD, MPH, Anthony Proli, PharmD, BCOP.
Flatiron Health, New York, NY, USA.
Flatiron Health, New York, NY, USA.
Presentation Documents
OBJECTIVES: Despite the increased use of electronic health record (EHR) data in oncology research, studies estimating medication persistence using EHR data alone remain limited. In the biotech industry, persistence is often assessed using time-to-event (TTE) and non-TTE approaches. This study compared approaches for estimating the persistence of epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKI), including erlotinib, gefitinib, dacomitinib, afatinib, osimertinib, and lazertinib, among patients with EGFR-positive advanced non-small cell lung cancer (advNSCLC).
METHODS: This retrospective study used the nationwide Flatiron Health EHR-derived deidentified database with a data cutoff of November 30, 2024. Patients with advNSCLC and a positive EGFR result 60 days before or 30 days after initiating any EGFR TKI were included. In the TTE approach, persistence was estimated with the time from EGFR TKI initiation until death or the earliest subsequent episode of EGFR TKI followed by more than 60 days of EGFR TKI-free patient activity, whichever occurred first. Patients were censored at their last confirmed activity or data cutoff. In the non-TTE approach, persistence was defined as the proportion of patients remaining on EGFR TKIs at different time points, among those still under follow-up at their last confirmed activity date.
RESULTS: Among 4851 patients, persistence (95% CI) of EGFR TKIs with the TTE approach vs non-TTE approach was 69% (67%-70%) vs 82% (82%-83%) at 6 months, 50% (49%-52%) vs 70% (69%-71%) at 12 months, and 37% (35%-38%) vs 61% (60%-62%) 18 months.
CONCLUSIONS: The non-TTE approach estimated higher EGFR TKI persistence than the TTE approach at all timepoints. The TTE approach accounts for censoring and estimates cumulative persistence, whereas the non-TTE approach provides a point-in-time snapshot. The TTE approach may also provide insights into other real-world outcomes, such as real-world treatment duration. Future work could enhance this approach by incorporating competing risks.
METHODS: This retrospective study used the nationwide Flatiron Health EHR-derived deidentified database with a data cutoff of November 30, 2024. Patients with advNSCLC and a positive EGFR result 60 days before or 30 days after initiating any EGFR TKI were included. In the TTE approach, persistence was estimated with the time from EGFR TKI initiation until death or the earliest subsequent episode of EGFR TKI followed by more than 60 days of EGFR TKI-free patient activity, whichever occurred first. Patients were censored at their last confirmed activity or data cutoff. In the non-TTE approach, persistence was defined as the proportion of patients remaining on EGFR TKIs at different time points, among those still under follow-up at their last confirmed activity date.
RESULTS: Among 4851 patients, persistence (95% CI) of EGFR TKIs with the TTE approach vs non-TTE approach was 69% (67%-70%) vs 82% (82%-83%) at 6 months, 50% (49%-52%) vs 70% (69%-71%) at 12 months, and 37% (35%-38%) vs 61% (60%-62%) 18 months.
CONCLUSIONS: The non-TTE approach estimated higher EGFR TKI persistence than the TTE approach at all timepoints. The TTE approach accounts for censoring and estimates cumulative persistence, whereas the non-TTE approach provides a point-in-time snapshot. The TTE approach may also provide insights into other real-world outcomes, such as real-world treatment duration. Future work could enhance this approach by incorporating competing risks.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR157
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
SDC: Oncology