Risk prediction model for acute care use among patients with advanced cancer on clinical trials.

Abstract

11074 Background: Reducing acute care use is an important strategy for improving value. Patients with advanced cancer are at high risk for unplanned Emergency Room visits (ER_visit) and hospital stays (HS). Identifying patients at highest risk could inform personalized risk-based intervention strategies and resource allocation. Methods: The data were from the SWOG Cancer Research Network. We linked data from trials in advanced cancer to Medicare claims data. The primary endpoint was the occurrence of any HS or ER_visit within 1 year after enrollment. Patients were required to have had continuous Medicare parts A and B claims from registration to 12 months after registration or death to detect all utilization outcomes. We examined sociodemographic (age, sex, race, ethnicity, insurance status), geographic (rural or urban, area-level deprivation), clinical (prognostic risk, performance status), treatment characteristics (line of therapy, therapy type), and individual comorbidity factors to establish a predictive risk model. Twenty-six factors were evaluated. A training/test approach was used. First, a 60% random sample training set of patients was generated. Best subset selection was used to identify candidate variables that minimized mean squared error using 5-fold cross validation repeated 10 times. Candidate variables were examined in multivariable logistic regression, with factors with p<.05 retained. A risk model was built by summing adverse factors and creating high vs low- risk groups by splitting at the median and into quartiles. The derived model is reported. Results: Among N=1397 total patients from 6 trials (lung, 3; prostate, 2; pancreas, 1), 839 (60%) comprised the training set. In these, 32.9% were 75+ years, 21.3% were female, and 7.7% were Black. The overall proportion of patients with >1 hospitalization/ER visit was 67.5%. In the training set, adverse risk factors were first line treatment (vs. subsequent), high (vs. low) prognostic risk, coronary artery disease (yes vs. no), hypertension (yes vs. no), and liver disease (yes vs. no). Patients with >2 factors (high risk; n=487, 58.0%) vs. 0/1 risk factor (low risk; n=352, 42.0%), were more likely to experience hospitalization/ER visit (80.3% vs. 49.7%, p<.001), corresponding to a >4-fold increase in risk (OR=4.12, 95% CI, 3.03-5.59, p<.001). Quartile-level proportions were 38.8%, 55.4%, 76.8%, and 84.0%, respectively, with an eightfold increased risk for those in the highest vs. lowest quartiles (Q4 vs. Q1, OR=8.25, 95% CI, 4.98-13.65, p<.001). Validation of the derived risk model in an independent set of 558 patients (40%) is planned. Conclusions: A limited set of 5 comorbid conditions, clinical and treatment variables predicted a 4-fold increased risk of HS and ER_visits in cancer patients with advanced disease. Personalized targeted interventions aimed at preventing acute care use could decrease the cost and improve the quality of cancer care.