A novel calibration framework for survival analysis when a binary covariate is measured at sparse time points

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Evaluating time-dependent treatments, such as aspirin initiation, and their association with survival outcomes is crucial in clinical and cohort studies.
• Colorectal cancer (CRC) survival studies often examine aspirin use, but intermittent measurements complicate accurate association estimation.
• Existing methods for analyzing time-dependent binary exposures can introduce substantial bias.

Purpose of the Study:

• To develop and present a class of calibration models for estimating the association between time-dependent binary exposures and survival outcomes.
• To address the challenges posed by intermittent measurements of time-dependent covariates in survival data analysis.
• To implement and validate these novel methods using data from aspirin initiation and colorectal cancer (CRC) survival.

Main Methods:

• Development of non-parametric, semiparametric, and parametric calibration models for the distribution of time-dependent covariate status changes.
• Incorporation of calibration model estimates into the proportional hazards partial likelihood for survival analysis.
• Introduction of a risk-set calibration approach for situations with strong associations between the binary covariate and survival.

Main Results:

• The proposed calibration models provide less biased estimates of the association between time-dependent aspirin use and colorectal cancer survival.
• The developed methods effectively handle intermittent covariate measurements, a common issue in longitudinal studies.
• The risk-set calibration approach offers improved utility in scenarios with strong exposure-survival associations.

Conclusions:

• The novel calibration modeling framework offers a robust approach to analyzing time-dependent treatments and survival outcomes, particularly in cancer research.
• These methods enhance the accuracy of survival analyses by mitigating bias arising from irregularly measured exposure data.
• The risk-set calibration provides a valuable alternative for specific clinical scenarios, strengthening the toolkit for epidemiological research.