Comorbidity and multimorbidity prediction of major chronic diseases using machine learning and network analytics

The prevalence of chronic disease comorbidity and multimorbidity is a significant health issue worldwide. In many cases, for individuals, the occurrence of one chronic disease leads to the development of one or more other chronic conditions. This exerts a significant burden on healthcare systems globally. Disease comorbidity is defined as the simultaneous occurrence of more than one disease. And a person having more than two comorbidities is referred to as multimorbid. This study followed a machine learning and network analytics-based approach to predict major chronic disease comorbidity and multimorbidity. In doing so, this study first extracted patient networks from the research dataset. In such networks, nodes represent patients and edges between two nodes indicate that the underlying two patients had at least one common disease. This study also considered other relevant features from patients' health trajectories. Out of the five machine learning models considered in this study (Logistic regression, k-nearest neighbours, Nave Bayes, Random Forest and Extreme Gradient Boosting) and two deep learning models (Multilayer perceptrons and Convolutional neural networks), Extreme Gradient Boosting showed the highest accuracy (95.05%), followed by the Convolutional neural networks (91.67%). The attribute of the number of episodes from the patient trajectory had been found as the most important feature, followed by the patient network attribute of transitivity. Other relevant results (feature correlation, variable clustering, confusion matrix and kernel density estimation) were also reported and discussed. The findings and insights of this study can help healthcare stakeholders and policymakers mitigate the negative impact of disease comorbidity and multimorbidity.