Abstract
PURPOSE: Acute care use (ACU) is more costly and prolonged for oncology patients and often leads to treatment disruptions and worsened outcomes. Reducing ACU requires understanding risk factors and proactively identifying at-risk patients. This study addresses research gaps by developing predictive models to assess all-cause acute care use (A-ACU) versus preventable acute care use (P-ACU) and rural-specific barriers. PATIENTS AND METHODS: We conducted a retrospective cohort study of adult oncology patients who received intravenous cancer treatment between October 2021 and April 2024 within a rural midwestern regional cancer network. We used predictor and outcome data from electronic medical records and insurance claims. We defined P-ACU using the Centers for Medicare & Medicaid Services' OP-35 criteria and classified A-ACU as any emergency department visit or hospitalization, regardless of reason. We trained LASSO and Random Forest models on 80% of the cohort to predict 30-, 90-, and 180-day risk of P-ACU and A-ACU after regimen initiation. RESULTS: Among 2,922 patients, 45.3% experienced A-ACU and 10.3% had P-ACU within 180 days of chemotherapy regimen initiation. Key predictors included number of previous inpatient stays and comorbidities. Insurance type and age were more influential in predicting P-ACU, whereas laboratory values (albumin, sodium, and neutrophil-to-lymphocyte ratio) were more important in A-ACU models. Nearly all LASSO and Random Forest models showed strong performance (mean area under the receiver operating characteristic curve = 0.73, mean F1 score = 0.79). CONCLUSION: Our models effectively identify patients at high risk for ACU using routinely collected data and validate known risk factors in a large rural oncology population. Future work should integrate these tools into practice and address rural-specific challenges to reduce ACU during chemotherapy.