Computing short-interval transition matrices of a discrete-time Markov chain from partially observed data

Area of Science:

• Biostatistics
• Mathematical Modeling
• Epidemiology

Background:

• Markov chains are widely used for modeling chronic disease progression through severity states.
• Estimating short-interval transition matrices from cohort data observed at longer intervals is challenging.
• Traditional matrix decomposition can yield invalid matrices with negative or complex entries.

Purpose of the Study:

• To present a novel regularization-based method for computing valid short-interval transition matrices.
• To address the limitations of existing matrix decomposition techniques in chronic disease modeling.
• To evaluate the performance of the proposed method on diverse matrix structures and a real-life HIV model.

Main Methods:

• A regularization technique is applied to compute short-interval transition matrices.
• The method operates row-wise on invalid matrices to minimize a distance measure.
• The approach was tested on various matrix structures and sizes.

Main Results:

• The proposed regularization method successfully computes valid short-interval transition matrices.
• The technique overcomes the issue of negative or complex entries arising from standard decomposition.
• Performance was evaluated using a real-world HIV transition model.

Conclusions:

• Regularization techniques offer a robust solution for deriving valid short-interval transition matrices.
• This method enhances the accuracy of Markov chain models for chronic disease progression.
• The approach is effective for epidemiological modeling, as demonstrated with HIV data.