Health Services/Policy
Length of stay risk adjustment for patients with acute pancreatitis using a large U.S. cohort Tahmina Begum* Tahmina Begum Begum Begum Begum Begum Department of Epidemiology & Community Health, University of North Carolina at Charlotte, USA
Background: Acute pancreatitis (AP) is a common gastrointestinal cause of hospitalization. Gaps remain in population-based risk adjustments leveraging large, commonly available datasets across U.S. hospital systems, often ignoring non-clinical factors, competing in-hospital mortality risks, and discharges against medical advice (AMA). We examine patient and facility factors associated with hospital length of stay (LOS), and hypothesize that commonly available variables beyond clinical factors can enhance risk-adjustments of LOS among AP patients, adjusting for AMA and in-hospital mortality.
Methods: A retrospective cohort study of 63,970 hospitalized patients was used with AP principal diagnosis discharged in 2024 over 784 U.S. facilities, using administrative claims data. The outcome was time to clinical discharge, with in-hospital mortality as competing risk and AMA discharge a censoring event. Two separate facility-clustered Fine-Gray models were used to estimate LOS, one with Elixhauser index (EI) and another using individual comorbidities, adjusting for patient and facility variables. C-statistics were estimated within an 80:20 derivation-validation cohort split.
Results: Patient and facility factors were associated with LOS. For example, women (hazard ratio; HR=0.978), count of secondary diagnoses (HR=0.941), and procedures (HR=0.851) were associated with longer LOS (p<0.03), while smaller facilities (≤250 beds) were associated with shorter LOS (HR≥1.07; p<0.05). The individual comorbidity-based model demonstrated substantial HR differences across comorbidities vs. HRs implied from the EI-based model. Both models demonstrated moderate out-of-sample performance (C≥0.67).
Conclusions: Risk-adjustment of national administrative data can identify additional non-clinical determinants of AP LOS, with implications of informing interventions to optimize care and refining processes for more equitable payments, particularly for hospitals caring for underserved communities.