USING EXTENDED COX MODELS WITH TIME-VARYING COVARIATES TO EXAMINE EFFECT OF VARIABLE OPIOID EXPOSURE IN CANCER RECURRENCE IN DISEASE-FREE BREAST CANCER SURVIVORS
Author(s)
Albert Truong, PharmD1, Vasco M. Pontinha, MA, MSc, PhD2;
1Virginia Commonwealth University, Richmond, VA, USA, 2Virginia Commonwealth University School of Pharmacy, Richmond, VA, USA
1Virginia Commonwealth University, Richmond, VA, USA, 2Virginia Commonwealth University School of Pharmacy, Richmond, VA, USA
OBJECTIVES: Opioid exposure during disease-free survivorship may affect cancer recurrence in time-dependent ways. While survival models are well suited to study time-to-recurrence, the Cox proportional hazards model assumes constant exposure effects, an assumption that may not hold for long-term opioid use. We used real-world evidence to test it with a cohort of disease-free breast cancer survivors. Additionally, we examined how risk estimates using standard and extended Cox models influence Cox proportional model assumptions.
METHODS: Using MarketScan, we identified (n=99,794) disease-free breast cancer survivors and followed patients from the end of treatment until recurrence or censoring. Opioid exposure was measured using a continuous exposure score and quartile-based dose groups. Time-to-recurrence was analyzed using Cox models adjusted for sociodemographic and clinical covariates. Proportional hazards were tested using Schoenfeld’s residuals and alternative models used extended Cox models, assuming time-varying coefficients.
RESULTS: In standard Cox proportional models, higher opioid exposure appeared protective against recurrence, with hazard ratios ranging from 0.37-0.87 across exposure quartiles and 0.995 per 1-unit increase in exposure score. However, Schoenfeld residual tests indicated violations of the proportional hazards assumption for all opioid exposure measures (p<0.001). In extended Cox models accounting for time-varying effects, continuous opioid exposure showed a lower hazard at the disease-free index (baseline HR:0.95; 95% CI 0.95-0.96), but the hazard increased over time (time-varying HRs:1.01; 95% CI 1.006-1.008). Similarly, quartile-based extended Cox models showed strong early protective associations (baseline HRs:0.003-0.073) with significantly time-varying increases in hazard (time-varying HRs:1.59-2.25), indicating attenuation and reversal of the early protective effect with longer follow-up.
CONCLUSIONS: Opioid exposure has a time-dependent association with breast cancer recurrence in disease-free survivors. Standard Cox models that assume proportional hazards yielded biased estimates, whereas extended Cox models revealed a shift from early protection to increasing risk over longer follow-up. Accounting for time-varying effects is essential for valid inference in cancer survivorship research.
METHODS: Using MarketScan, we identified (n=99,794) disease-free breast cancer survivors and followed patients from the end of treatment until recurrence or censoring. Opioid exposure was measured using a continuous exposure score and quartile-based dose groups. Time-to-recurrence was analyzed using Cox models adjusted for sociodemographic and clinical covariates. Proportional hazards were tested using Schoenfeld’s residuals and alternative models used extended Cox models, assuming time-varying coefficients.
RESULTS: In standard Cox proportional models, higher opioid exposure appeared protective against recurrence, with hazard ratios ranging from 0.37-0.87 across exposure quartiles and 0.995 per 1-unit increase in exposure score. However, Schoenfeld residual tests indicated violations of the proportional hazards assumption for all opioid exposure measures (p<0.001). In extended Cox models accounting for time-varying effects, continuous opioid exposure showed a lower hazard at the disease-free index (baseline HR:0.95; 95% CI 0.95-0.96), but the hazard increased over time (time-varying HRs:1.01; 95% CI 1.006-1.008). Similarly, quartile-based extended Cox models showed strong early protective associations (baseline HRs:0.003-0.073) with significantly time-varying increases in hazard (time-varying HRs:1.59-2.25), indicating attenuation and reversal of the early protective effect with longer follow-up.
CONCLUSIONS: Opioid exposure has a time-dependent association with breast cancer recurrence in disease-free survivors. Standard Cox models that assume proportional hazards yielded biased estimates, whereas extended Cox models revealed a shift from early protection to increasing risk over longer follow-up. Accounting for time-varying effects is essential for valid inference in cancer survivorship research.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR59
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference
Disease
SDC: Injury & Trauma, SDC: Oncology