1016 - Disease Site-Specific Risk Stratification for Acute Care Utilization In Radiation Oncology: A Large-Scale Competing Risk Analysis
Presenter(s)
L. Pandey, M. J. Lane, K. Chang, A. J. Chang, and R. R. Savjani; Department of Radiation Oncology, University of California, Los Angeles, Los Angeles, CA
Purpose/Objective(s): Acute care utilization during radiation treatment varies by cancer type. While predictive models exist, comparative risk stratification across disease sites within the same population accounting for death as competing risk remains unexplored. We hypothesized that disease site independently predicts 90-day emergency department (ED) and inpatient (IP) admission risk from CT simulation after adjusting for demographics, disease state, and treatment.
Materials/Methods: This retrospective cohort included 24,234 patients undergoing 39,335 CT simulations (2013-2026) at one institution using EMRs. We categorized 994 diagnoses into 27 disease sites and 7 disease states. Primary endpoints were time to first ED visit and IP admission within 90 days of first simulation, with 77.5% of patients having confirmed 90-day survival or mortality data. Cause-specific Cox proportional hazards models identified predictors, genitourinary (most common, 24%) as reference. Death within 90 days (893 patients, 3.8%) was evaluated as competing risk using Fine-Gray hazard models. Proportional hazards violations underwent time-stratified analysis (0-30, 31-60, 61-90 days). Model discrimination was assessed via concordance index.
Results: Median age was 66 years; 51% male, 61% White. Within 90 days, 2,918 (12.4%) had ED visits (median 30 days) and 3,382 (14.4%) had IP admissions (median 25 days).
On multivariable Cox regression (concordance 0.815 IP, 0.747 ED), IP risk varied 12-fold across disease sites. Hematologic malignancies had highest risk (HR 7.57; 95% CI, 6.64-8.63; p<0.001), with 24.5% population-attributable fraction (PAF) despite 4.9% prevalence. CNS (HR 2.58; CI, 2.20-3.03; PAF 10.7%), gynecologic (HR 2.23; CI, 1.92-2.59; PAF 7.6%), and soft tissue (HR 3.23; CI, 2.74-3.80; PAF 6.5%) also elevated risk, while breast cancer was protective (HR 0.64; CI, 0.53-0.77; PAF 5.5%). Metastatic disease (HR 1.33, PAF 3.2%), recent surgery (HR 2.87, PAF 29.6%), chemotherapy (HR 2.00, PAF 19.8%) were strongest treatment-related predictors. Time-stratified analysis revealed gynecologic IP risk peaked early (HR 2.56, 0-30 days) then attenuated (HR 1.04, 61-90 days), while chemotherapy effect amplified (HR 1.63 to 2.67). ICU admission (2.8%) was predicted by CNS (HR 3.16), hematologic (HR 3.38), and endocrine (HR 3.75) with death in 50.5% of ICU candidates versus 11.4% of IP candidates. Fine-Gray models showed <0.5% average HR difference from Cox, validating this approach.Conclusion: IP risk varies 12-fold across cancer types with timing-specific intervention opportunities. Hematologic malignancies drive 24.5% of admissions from 5% of patients, warranting supportive care. Gynecologic patients warrant intensive weeks 0-4 surveillance (HR 2.56) before risk normalizes. Chemotherapy toxicity escalates progressively (HR 1.6× to 2.7×). These disease-specific and time-sensitive profiles enable targeted surveillance and preemptive multidisciplinary care coordination at simulation.