Friday, September 28, 2018 10:30 AM–12:00 PM abstracts: deformity: technical factors: 167. Preoperative prediction of cost and catastrophic cost in adult spine deformity surgery: feasibility analysis of predictive analytics to establish 90-day bundled payments

Abstract

BACKGROUND CONTEXT Adult spinal deformity (ASD) surgery is responsible for 6% of all United States (US) healthcare costs. There has been increasing interest in cost prediction for bundled payment models and risk sharing initiatives. Catastrophic cost (CC) outliers are typically excluded in bundled payment settings making their preoperative identification critical. PURPOSE The objective of this study is to offer a comprehensive framework using predictive analytics to preoperatively determine direct cost and catastrophic cost outliers for ASD surgery. STUDY DESIGN/SETTING Direct cost modeling based on prospective multicenter ASD data. PATIENT SAMPLE Prospectively collected multicenter ASD patients. OUTCOME MEASURES Actual, direct costs incurred to the hospital for ASD surgery. METHODS We performed regression models (generalized linear regression and random forest) for direct costs and classification models (random forest) for CC for ASD surgery. The goal of the regression models was to explain the determinants of direct costs by means of patient, surgical and contextual factors. The goal of the CC models was to predict which patients would have a direct CC (>$100,000). RESULTS A total of 210 ASD patients (83% women, 45% revision surgery) from four sites located in four geographic US areas were included. Cost data obtained were actual, direct costs incurred to the hospital. Average index cost per patient was $75,772. 14.8% of patients had a cost above the $100,000 threshold. Direct cost could be predicted preoperatively using random forest models with an accuracy of 72.1%. Out of the total variance explained, 22.6% was site and surgeon fixed-effects. The top predictors of cost in order were: surgeon, number of levels fused, interbody fusion and site. Catastrophic cost was predicted with a 90.4% accuracy and 87.7% AUC. In our sample alone, if we reduced the number of CC occurrence by one-third the associated savings would be $452,181. Reducing the number of CC occurrence by one-third across the US would produce savings of at least $80 million per year. CONCLUSIONS Bundled payment models and risk sharing initiatives have been proposed as means of controlling ASD surgery cost, but these approaches require accurate cost prediction. This study demonstrates that direct cost in ASD surgery can be reliably predicted in a preoperative setting and that CC outliers can be predicted preoperatively with a 90% accuracy. The high degree of cost variance explained by factors such as site and surgeon suggest potential efficiency gains offered by standardization in patient selection and treatment strategies.