Constructing training set using distance between learnt graphical models of time series data on patient physiology,...

Area of Science:

• Computational biology
• Medical informatics
• Network science

Background:

• Early prediction of disease susceptibility is crucial for timely medical intervention.
• Veno-occlusive disease (VOD) is a serious complication following bone marrow transplants, where survival is linked to early detection.
• Existing methods for predicting VOD risk have limitations in accuracy and scope.

Purpose of the Study:

• To develop and validate a novel VOD-score for predicting disease susceptibility in patients undergoing bone marrow transplantation.
• To establish a reliable method for quantifying individual patient risk for VOD.
• To facilitate early intervention strategies by providing a quantitative measure of VOD risk.

Main Methods:

• Utilized time series data on patient physiology from pre-transplant to post-transplant.
• Defined VOD-score based on graph theory, specifically the distance between Soft Random Geometric Graphs derived from patient data and a reference.
• Employed vector-variate Gaussian Processes and Markov Chain Monte Carlo inference for learning and prediction.

Main Results:

• Successfully learned a VOD-score from retrospective cohort data, correlating with disease susceptibility.
• Developed a predictive model linking the VOD-score to pre-transplant patient parameters.
• Demonstrated that the VOD-score calculation is robust and independent of data length.

Conclusions:

• The developed VOD-score offers a reliable method for predicting VOD risk post-bone marrow transplant.
• This approach enables personalized risk assessment and supports timely clinical decision-making.
• The findings pave the way for improved patient outcomes through early intervention in VOD.