Development of an Oncology Acute Care Risk Prediction Model. (2021)

• Record Type: Journal Article
• Authors: Csik, Valerie P.; Li, Michael; Binder, Adam F.; Handley, Nathan R.
• Is Part Of: JCO Clinical Cancer Informatics. Volume 5(2021)
• Publication Date: 2021
• Subjects: Machine learning -- Electronic health records -- Older adults -- End of life care -- Value based care -- Cancer prevention -- Professional development -- Oncology Care Model -- Hospital utilization -- aprepitant -- OCM; 616.994
• DOI: 10.1200/CCI.20.00146
• Languages: English
• ISSNs: 2473-4276

Abstract

Abstract : PURPOSE: Acute care utilization (ACU), including emergency department (ED) visits or hospital admissions, is common in patients with cancer and may be preventable. The Center for Medicare & Medicaid Services recently implemented OP-35, a measure in the Hospital Outpatient Quality Reporting Program focused on ED visits and inpatient admissions for 10 potentially preventable conditions that arise within 30 days of chemotherapy. This new measure exemplifies a growing focus on preventing unnecessary ACU. However, identifying patients at high risk of ACU remains a challenge. We developed a real-time clinical prediction model using a discrete point allocation system to assess risk for ACU in patients with active cancer. METHODS: We performed a retrospective cohort analysis of patients with active cancer from a large urban academic medical center. The primary outcome, ACU, was evaluated using a multivariate logistic regression model with backward variable selection. We used estimates from the multivariate logistic model to construct a risk index using a discrete point allocation system. RESULTS: Eight thousand two hundred forty-six patients were included in the analysis. ED utilization in the last 90 days, history of chronic obstructive pulmonary disease, congestive heart failure or renal failure, and low hemoglobin and low neutrophil count significantly increased risk for ACU. The model produced an overall C-statistic of 0.726. Patients defined as high risk (achieving a score of 2 or higher on the risk index) represented 10% of total patients and 46% of ACU. CONCLUSION: We developed an oncology acute care risk prediction model using a risk index–based scoring system, the REDUCE (Reducing ED Utilization in the Cancer Experience) score. Further efforts to evaluate the effectiveness of our model in predicting ACU are ongoing.