A causal network framework to identify disease-to-disease determinants in multimorbidity

Background

Understanding how diseases interact to shape health outcomes is a central challenge in multimorbidity research. Beyond identifying risk factors for mortality, it is crucial to detect diseases that act as determinants of other diseases, structuring cascading health processes.

Methods

We developed an integrated computational framework combining Model-X Knockoffs for false discovery rate (FDR) controlled selection, Bayesian network learning for causal structure identification, and recurrent determinant analysis. The approach includes prevalence filtering, automated model selection, and centrality measures to identify structurally influential diseases. The framework was applied to a dataset of 4,000 patients with 80 potential disease indicators and mortality outcomes.

Results

After FDR-controlled selection (q = 0.20), 26 diseases were identified as significant determinants. The causal network revealed that several diseases not only directly influenced mortality but also acted as determinants of other diseases, forming interconnected pathways of multimorbidity. High betweenness centrality nodes emerged as key bridging diseases linking otherwise separate clusters, indicating their role in propagating health risk. Recurrent determinant analysis confirmed higher-order structures, identifying diseases that systematically influence multiple other conditions. The model showed strong predictive performance (AUC up to 0.89) and good calibration.

Conclusions

This framework advances multimorbidity research by identifying diseases as causal determinants within interconnected systems, rather than isolated risk factors. By capturing disease-to-disease relationships and cascading effects, it provides a novel and interpretable approach for understanding complex health dynamics and informing public health and clinical strategies.

Key messages

• Diseases act as causal determinants of other diseases, shaping cascading multimorbidity pathways beyond isolated risk factors.

• An integrated AI-driven framework identifies disease-to-disease relationships, revealing key bridge conditions for targeted public health interventions.