Bayesian networks to identify potential high-risk multimorbidity and intervention clusters in inpatients: an...

Area of Science:

• Healthcare Analytics
• Clinical Informatics
• Patient Outcomes Research

Background:

• Routine hospital data offers a rich source for understanding patient care.
• Identifying patterns in multimorbidity and interventions is crucial for improving patient outcomes.
• Existing methods may not fully capture complex relationships in large datasets.

Purpose of the Study:

• To explore multimorbidity and intervention clusters associated with high risks for in-hospital mortality and unplanned readmissions.
• To apply data-driven analytical methods to routine hospital data.
• To identify potential targets for quality improvement programs.

Main Methods:

• An explorative, historical cohort study using routine healthcare data from a Swiss tertiary care center (2012-2017).
• Inclusion of all inpatients aged 18 years and older.
• Mutual information analysis to identify multimorbidity and intervention clusters.
• Multi-layer Bayesian networks to analyze associations with in-hospital death and unplanned readmissions.

Main Results:

• Analysis of 190,837 inpatient cases revealed 7994 diagnoses and 6639 interventions.
• Identified 32 multimorbidity clusters and 24 intervention clusters using mutual information analysis.
• Bayesian network analysis linked several clusters to in-hospital mortality and/or unplanned readmissions.

Conclusions:

• Bayesian network analysis is a valuable tool for mining large healthcare databases.
• Identified clusters can inform intervention targets for quality improvement initiatives.
• Further research with causal models is needed to substantiate observed associations.