Using multiple timescale models for the evaluation of a time-dependent treatment

Area of Science:

• Biostatistics
• Survival Analysis
• Clinical Epidemiology

Background:

• Kaplan-Meier estimator is widely used but yields biased results for time-dependent interventions.
• Accurate survival analysis requires accounting for interventions that occur after baseline.

Purpose of the Study:

• Extend multiple timescale models for time-dependent interventions.
• Estimate treatment effects (hazard ratios) using flexible modeling.
• Develop a valid prediction tool for patients changing treatment.
• Create appropriate graphical representations of survival with time-dependent treatment changes.

Main Methods:

• Utilized a multiple timescale model with two timescales to account for treatment changes.
• Applied the model to compare chemotherapy versus transplant in high-risk acute lymphoblastic leukemia.
• Proposed an alternative approach to survival estimation, improving upon the traditional landmark approach.

Main Results:

• The proposed model offers advantages over the traditional landmark approach.
• It effectively uses all available data for survival estimation from remission.
• Explicitly models hazard changes due to interventions like transplantation.

Conclusions:

• Multiple timescale models provide a more accurate method for survival analysis with time-dependent interventions.
• This approach enhances prognostic predictions and graphical representations in complex clinical scenarios.
• The method is particularly beneficial for analyzing treatment changes in pediatric leukemia survivors.