SURG-32. A NOVEL MACHINE LEARNING APPROACH FOR MODELING THE PREDICTIVE VALUE OF COMORBIDITY INDICES IN PITUITARY SURGERY

Abstract Although pituitary adenomas (PAs) are common intracranial tumors, literature evaluating the utility of comorbidity indices for predicting perioperative complications in patients undergoing surgery remains limited, thereby hindering development complex models that aim to identify high-risk patient populations. Accordingly, we utilized comparative modeling strategies evaluate predictive validity various and combinations thereof key outcomes. The Nationwide Readmissions Database was used who underwent transsphenoidal tumor operations (n=19,653) 2016-2017. Patient frailty assessed using Johns Hopkins Adjusted Clinical Groups (JHACG). Charlson Comorbidity Index (CCI) Elixhauser (ECI) were calculated each patient. Five sets generalized linear mixed-effects developed, 1) Frailty, 2) CCI, 3) ECI, 4) Frailty+CCI, or 5) Frailty+ECI as primary predictor. Complications investigated included inpatient mortality, non-routine discharge (e.g., locations other than home) , length stay (LOS) within top quartile, cost one-year readmission rates. Postoperative mortality occurred 73 (0.4%), one year reported 2,994 (15.2%),and 2,176 (11.1%) patients. mean adjusted all-payer procedure $25,553.85±$26,518.91 (Top Quartile: $28,261.20) LOS 4.8 days±7.4 days 5.0 days). model frailty+ECI predictor consistently outperformed models, with statistically significant p-values determined by comparing their AUCs, most complications. For prediction however (AUC:0.831) not better ECI alone (AUC:0.831;p=0.95). (AUC:0.617) (AUC:0.606;p=0.10) Frailty+CCI (AUC:0.610;p=0.29). Knowledge gained from these may help neurosurgeons requiring additional clinical attention specific resource utilization prior surgical planning.