Multimorbidity network patterns improve the explanation of inpatient outcomes beyond conventional comorbidity assessment: a retrospective cohort study across 32 tertiary hospitals

Multimorbidity, the co-occurrence of multiple chronic conditions, is increasingly prevalent among middle-aged and older hospitalized populations. While traditional comorbidity indices such as the age-adjusted Charlson Comorbidity Index (ACCI) offer predictive value, they overlook the structural relationships among diseases. This study aims to evaluate the incremental predictive value of multimorbidity network patterns beyond the ACCI in predicting key inpatient outcomes. We conducted a retrospective multicenter cohort study using inpatient electronic health records from 32 tertiary hospitals in Shanxi Province, China, covering the period 2018–2022. A total of 2,042,063 hospital admissions with multimorbidity (≥2 chronic conditions, restricted to physical chronic conditions) were included. The multimorbidity network was constructed using the Louvain modularity algorithm to detect disease communities. Three nested regression models were fitted for three outcomes—length of stay (LOS), total hospitalization cost, and 30-day readmission risk—adjusting for demographic factors, ACCI, and multimorbidity patterns. Model fit was assessed using Akaike information criterion(AIC) and Bayesian information criterion(BIC), and population-level attribution for each pattern was evaluated using counterfactual predictions. The multimorbidity network (304 nodes, 2513 edges) revealed hypertension, type 2 diabetes, and stroke as central hub diseases. Nine disease communities were identified. Multimorbidity patterns significantly improved model fit compared to ACCI alone, with reductions in AIC/BIC. Community 6 (endocrine-immune-renal diseases) had the highest readmission risk (odds ratio [OR] = 1.591) and was associated with lower costs (mean ratio = 0.696), while Community 8 (cardiovascular-arrhythmic diseases) was associated with shorter LOS(rate ratio = 0.691) and lower readmission risk (OR = 0.478). ACCI contributed 25.21% to readmission risk, while multimorbidity patterns showed substantial variability in risk contributions. Multimorbidity network patterns provide additional explanatory value over ACCI in forecasting hospitalization outcomes. This approach improves risk stratification and could inform more tailored clinical management and resource allocation.