Clustering based on adherence data

Area of Science:

• Biostatistics
• Medical Informatics
• Public Health

Background:

• Patient adherence to medical treatments is crucial for effective management and research.
• Current methods often rely on averages, potentially oversimplifying complex adherence behaviors.
• Missing data due to patient outcomes is a common challenge in long-term studies.

Purpose of the Study:

• To propose a novel stochastic process model for analyzing patient adherence.
• To utilize Markov chains for a more dynamic and comprehensive description of adherence behavior.
• To enable patient clustering and prediction based on adherence patterns.

Main Methods:

• Adapting stochastic process theory to model patient adherence measures.
• Analyzing repeated adherence measures as a Markov chain with finite states.
• Estimating transition probabilities to characterize patient adherence dynamics.
• Clustering patients based on their estimated transition probabilities.

Main Results:

• The Markov chain approach provides a robust framework for analyzing adherence data.
• Patient clustering based on transition probabilities reveals distinct adherence patterns.
• The method is illustrated effectively using data from the DART trial.

Conclusions:

• Modeling adherence as a Markov chain offers a powerful alternative to traditional averaging methods.
• This approach facilitates a deeper understanding of adherence variability and patient stratification.
• The findings have implications for identifying adherence predictors and forecasting future health events.