Acute care events after outpatient systemic therapy among patients at elevated risk for acute care utilization.

Abstract

e23246 Background: A subset of acute care events (ACE) within 30 days of outpatientsystemic therapy is potentially preventable and is tracked under CMS qualitymeasure OP-35. It remains unclear whether baseline data alone can predict thesepreventable events (PACE30). We evaluated prediction of PACE30 and examined itsdistribution within the broader population at risk for any ACE30. Methods: We studied 12,231 patients receiving initial outpatient systemic therapy(cytotoxic, immunotherapy, targeted, and endocrine regimens) at a large cancer center(2012–2021). ACE30 and PACE30 were identified using CMS OP-35–alignedstandardized definitions. Baseline predictors included demographics, disease andtreatment characteristics, laboratory values, vital signs, and pre-treatment healthcareutilization. We developed standalone models to predict PACE30. For context, weapplied a previously developed 56-variable L1-regularized logistic regression model(internally developed and benchmarked against published PROACCT models) to stratifyACE30 risk. We assessed PACE30 enrichment among ACE30 high-risk patients andtested whether two-stage (conditional) modeling within high-risk strata improvedPACE30 discrimination. Model performance was assessed via AUC, positive predictivevalue (PPV), and risk enrichment at fixed capacity thresholds. Results: ACE30 occurred in 22.9% of patients, and PACE30 occurred in 4.5% (~20%of all ACE30). Standalone baseline models showed modest discrimination for PACE30(AUC 0.62, PPV 0.10) compared with stronger performance for ACE30 (AUC 0.68, PPV0.47). Conditional modeling restricted to the top 20% of ACE30 risk did not meaningfullyenhance PACE30 discrimination or stratification. However, the top 20% ACE30 riskgroup exhibited nearly twofold higher PACE30 rates (8.2% vs 4.5% overall),demonstrating substantial enrichment of preventable events among patients withelevated baseline clinical instability and acute care risk. Conclusions: Baseline pre-treatment variables offer limited direct predictive power todistinguish preventable from non-preventable acute care events after systemic therapyinitiation, even within high-risk subgroups. Nevertheless, preventable events clusterdisproportionately in patients identified as high-risk for any acute care utilization viabaseline stratification. Preventability appears driven predominantly by downstream post-systemic therapy care delivery processes rather than pre-treatment patientcharacteristics. Efforts to reduce avoidable acute care should prioritize targeted post-systemic therapy interventions—including proactive symptom monitoring, rapid access pathways, and enhanced care coordination—focused on patients identified as high-riskby baseline models, rather than relying solely on direct baseline prediction ofpreventability.