Defining and analysis of multimorbidity pattern of diseases using Markov random field approach: a comparative analysis

Abstract

Aim: Multi-morbidity remains poorly understood due to the multifactorial complexity of this phenomenon and the lack of a standardized methodology for building and analysing Multimorbidity network. A comparative analysis of methods of modeling Multimorbidity network in literature may help to understand the pros and cons of these methods, then to facilitate a consensus about a standardized methodology. We propose to study two approaches for building Multimorbidity network focusing in their technical specificities.

Subject and Methods: We propose to model Multimorbidity using Ising Model, a Markov Random field based approach, and to compare its performance to the approach consisting in building a network of co-occurence using pairwise association strength estimated by Multimorbidity Coefficient. Besides, we illustrate how to use network science techniques to extract structural knowledge from Multimorbidity network.

Results: The results show that the Ising model is able to detect a similar structural patern as the approach of computing Multimorbidity coefficient for all paires of diseases. An evaluation of the stability and precision of the obtained comorbidity network has proved its reliability.

Conclusion: Defining methods and algorithms of detecting Multimorbidity network in formal language may help interdisciplinary cooperative research. Ising Model is a machine learning based on a probabilistic formalism capable of detecting the same pattern as traditional approaches in Multimorbidity research literature. Understanding how diseases co-occur at the same time will help physicians to reason on multimorbidity burden as a complex system rather than reasoning on diseases as single and isolated entities.