Clinically validated graphical approaches identify hepatosplenic multimorbidity in individuals at risk of schistosomiasis

The global burden of multimorbidity is increasing yet poorly understood, owing to insufficient methods for modelling complex systems of conditions. In particular, hepatosplenic multimorbidity has been inadequately investigated. From 17 January to 16 February 2023, we examined 3186 individuals aged 5-92 years from 52 villages across Uganda within the SchistoTrack Cohort. Point-of-care B-mode ultrasound was used to assess 45 hepatosplenic conditions within the context of schistosomiasis (Schistosoma mansoni). Three graph learning methods for representing hepatosplenic multimorbidity were compared. Thresholds for including graph edges were found using graph kernels and tested with graph neural networks to assess predictive utility for unobserved conditions. Clinical validity was assessed by identifying medically relevant condition inter-dependencies for portal hypertension. 54.65% (1741/3186) of individuals were multimorbid with two or more hepatosplenic conditions. Thresholds were 50.16 and 64.46% for graphical lasso and signed distance correlation, respectively, but could not be inferred for co-occurrence. Co-occurrence graphs were clinically uninformative with low predictive capacity. Graph learning algorithms with statistical assumptions, e.g. graphical lasso, enabled accurate and clinically valid multimorbidity representations. Severe conditions related to portal hypertension were predicted with high sensitivity and specificity. This work presents a generalizable framework for understanding multimorbidity to enable more accurate diagnoses of complex diseases.