TOPLINE

A predictive risk model identifies adolescent and young adult (AYA) patients with cancer at high risk for acute care events (ACEs) during early survivorship, with nearly half of patients experiencing at least one hospitalization or emergency department visit between 2 and 5 years after diagnosis.

METHODOLOGY

• More than 85% of the 90,000 AYAs diagnosed with cancer annually in the US survive at least 5 years after diagnosis, leading to a rapidly growing population of AYA survivors who may experience lifelong adverse health effects distinct from peers without a cancer history. Prior risk models in oncology have largely focused on patients receiving active treatment, with survivorship models concentrating on secondary cancers or late effects from chemotherapy, leaving a critical gap in risk prediction tailored to the AYA population during early survivorship transitions.
• Researchers utilized the University of North Carolina Lineberger Cancer Information and Population Health Resource, linking North Carolina Cancer Registry records with health insurance claims data from private insurance plans and Medicaid in North Carolina. The study included 7393 AYA patients diagnosed with cancer between 2006 and 2018 at ages 15-39 years, with an average follow-up of 1.9 years during the observation period from 2 to 5 years after diagnosis. 
• Study participants were randomly assigned to development (70%, n = 5176) and validation (30%, n = 2217) cohorts, with continuous insurance enrollment required from 18 to 24 months after diagnosis to define baseline clinical exposures. 
• Outcome measures included any ACE, defined as either a hospitalization or an emergency department visit occurring between 2 and 5 years after diagnosis, excluding obstetric delivery-related hospitalizations. 
• Analysis involved developing three multivariable logistic regression models using stepwise inclusion of predictive variables, with performance evaluated using sensitivity, specificity, positive predictive value (PPV), and area under the curve (AUC), defining high risk as the top 20% of predicted probability scores.

TAKEAWAY

• In the development cohort, 2522 of 5176 patients (49%) experienced at least one ACE during the observation period, with factors significantly associated with ACEs including age at diagnosis, race, insurance type, cancer stage, treatment type, Charlson comorbidity score, mood disorder, tobacco use, and prior acute care use (P < .001 for most variables). 
• Model 3, the final selected model, achieved an AUC of 0.76, specificity of 0.95, sensitivity of 0.32, and PPV of 0.84 in the validation cohort when defining high risk as the top 20% of predicted probability scores. 
• Patients with ACEs were more frequently female (68% vs 64%), Black (30% vs 20%), and publicly insured via Medicaid (69% vs 43%) and had higher rates of tobacco use disorder (10% vs 3%) and mood disorders (26% vs 16%) than those without ACEs. 
• In a sensitivity analysis excluding 1757 patients (24%) with claims indicating active treatment during the observation period, model performance remained stable with an AUC of 0.77, specificity of 0.95, sensitivity of 0.33, and PPV of 0.85, with 71% of patients experiencing ACEs having no claims for active treatment. 

IN PRACTICE

“This is the first validated risk model to predict ACEs in AYA patients with cancer during the early posttreatment period. The model, designed for seamless electronic health record integration, enables early identification of high-risk patients, presenting opportunities for targeted interventions to reduce acute care use,” the authors of the study wrote.

SOURCE

The study was led by Ryanne C. Buckley, MD, The University of North Carolina School of Medicine, Chapel Hill, and Andrew B. Smitherman, MD, MS, Lineberger Comprehensive Cancer Center, The University of North Carolina at Chapel Hill. It was published online on August 25 in JCO Oncology Practice.

LIMITATIONS

The study was limited by its reliance on insurance claims data, which excluded uninsured individuals and did not capture ACEs occurring during periods without coverage, potentially affecting generalizability to populations with higher acute care use. Recurrent or refractory disease could not be reliably identified using claims data alone, limiting the ability to measure the impact of disease progression on model performance. The model was developed using data from a single state-based cancer registry in North Carolina, which may affect applicability in states with different healthcare delivery systems. The baseline clinical time window was defined as 18-24 months after cancer diagnosis, and variables may vary if this period were modified. The non-disease-specific approach, while enhancing clinical applicability in large healthcare systems, may introduce clinical heterogeneity and reduce predictive precision compared with disease-specific models.

DISCLOSURES

This study received support from a University of North Carolina Lineberger Comprehensive Cancer Center Developmental Award, which is funded in part by the Cancer Center Core Support Grant P30 CA016086. Work on this study also received funding from the University Cancer Research Fund via the state of North Carolina. Buckley disclosed no relevant conflicts of interest. Jacob N. Stein, MD, MDH, disclosed receiving consulting or advisory fees from SpringWorks Therapeutics and Deciphera, honoraria from GO2 for Lung Cancer, and research funding from Lilly. Additional disclosures are noted in the original article.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.