Using recurrent time-to-event models with multinomial outcomes to generate toxicity profiles.

Area of Science:

• Clinical pharmacology
• Biostatistics
• Oncology

Background:

• Clinical studies often report adverse event frequencies but lack temporal data.
• Understanding the timing of adverse events is crucial for managing long-term therapies and assessing new treatments.
• Current methods do not fully capture the dynamic nature of treatment-related toxicities.

Purpose of the Study:

• To develop and illustrate a method for generating 'toxicity profiles' that detail the temporal patterns of adverse events.
• To enable simultaneous assessment of event risks over time and provide cumulative probabilities for different adverse event types.
• To aid in disease management and the evaluation of novel therapeutic agents.

Main Methods:

• An adaptation of the ordinal time-to-event model was used.
• A two-step process involved multinomial logistic regression for response probabilities.
• These were combined with recurrent time-to-event hazard estimates to produce cumulative event probabilities.

Main Results:

• The developed method generates detailed toxicity profiles for adverse events.
• Cumulative event probabilities for each adverse event category were calculated.
• Toxicity profiles were illustrated for three adverse events in advanced breast cancer patients receiving two treatment regimens.

Conclusions:

• The proposed method provides a robust tool for analyzing the temporal aspects of adverse events.
• Toxicity profiles can enhance the assessment of therapeutic interventions, particularly long-term treatments.
• This approach supports better clinical decision-making and evaluation of treatment benefits versus costs.