Impact of palliative care and predictive modeling with clinical features in hospitalized lung cancer patients: A National Inpatient Sample analysis (2018-2022).

313 Background: Early palliative care (PC) integration is crucial for improving outcomes in lung cancer, but its utilization varies. This study assessed PC's impact on hospital outcomes for deceased lung cancer patients and developed machine learning (ML) models, enhanced with extensive clinical features, to predict PC use and quantify potential missed opportunities. Methods: A retrospective cohort study using national inpatient data (2018-2021 for training, N = 734,212 total (unweighted), representing an approximate 20% stratified sample of U.S. hospital discharges; 2022 for validation, N = 181,636 total (unweighted)). Deceased lung cancer patients (# of PC Pts w/ LC = 17,058, # of Pts w/ LC without PC = 10,894 in 2018-2021 cohort) were compared on median total hospital charges (TOTCHG), length of stay (LOS) (Mann-Whitney U), and mean clinical characteristics (engineered from ICD-10 codes, organ failure/metastasis/symptom). ML models (Logistic Regression, Random Forest, Gradient Boosting) using demographic, administrative, and clinical features predicted PC utilization, validated internally (75/25 split) and externally. The best model by PR-AUC (Precision-Recall Area Under Curve) was calibrated and used to identify deceased non-PC patients with high predicted PC probability (> 0.5). Results: For deceased lung cancer patients (2018-2021 NIS sample), PC was associated with significantly lower median TOTCHG ($48,908 vs. $64,882, p < 0.0001) and shorter median LOS (4.0 vs. 5.0 days, p < 0.0001). PC recipients had higher mean metastasis counts (0.72 vs. 0.63). Gradient Boosting performed best for PC prediction (Internal Test: ROC-AUC 0.786, PR-AUC 0.433; External Validation: ROC-AUC 0.780, PR-AUC 0.451), though with low sensitivity (~0.11-0.14). Using this calibrated model, an estimated 6,540 deceased non-PC lung cancer patients nationally (1,308 in sample x ~5) in the 2018-2021 cohort represented potential missed PC opportunities (predicted probability > 0.5). This extrapolates to a projected potential national reduction in charges of approximately $20.9 million in a representative population and 6,540 hospital days. For 2022, an estimated 1,805 national cases (361 in sample x ~5) were identified, with a projected charge reduction of $28.8 million. Conclusions: Palliative care is associated with significantly reduced hospital resource utilization for deceased lung cancer patients nationally. Clinically enhanced ML models robustly predict PC use. Identifying and engaging potential "missed opportunities" for PC, projected to over 6,500 lung cancer patients annually in the 2018-2021 period, could yield substantial national cost savings and optimize hospital stays, supporting improved end-of-life care.