Wednesday, September 26, 2018 7:35 AM–9:00 AM ePosters: P81. Successful creation of deployable preoperative predictive risk calculators for individual patient event-free survivorship for major complications, hospital readmissions and unplanned surgery following adult spinal deformity (ASD) surgery

Abstract

BACKGROUND CONTEXT Few predictive models allow for proper patient selection, adjustment of invasiveness and patient frailty optimization to predict and reduce postoperative major complications (MC), hospital readmissions (READMIT), and unplanned surgery (UNPLAN). PURPOSE The objective of this project is to create accurate predictive models for the occurrence and timing of MC, READMIT, and UNPLAN following ASD surgery. STUDY DESIGN/SETTING Retrospective analysis of two independent prospective, multicenter ASD databases with identical fixed data fields. PATIENT SAMPLE Data from 1,018 ASD surgically treated patients (57 surgeons, 24 sites, and 5 countries) were used to build MC, READM and UNPLAN risk calculating models with proved successful model fit. OUTCOME MEASURES Postoperative major complications (MC), hospital readmissions (READMIT), unplanned surgery (UNPLAN). METHODS Surgical ASD patients with >2yFU were identified. Patient demographic, radiographic, operative, baseline PROMs, and complications data were analyzed to build event free survival curves for MC, READMIT and UNPLAN, and to create predictive models by means of a random survival forest with 80/20 train or test sets. 101 variables were used. Missing value imputation was performed with the missForest package. RESULTS A total of 1,018 ASD patients treated surgically before September-2014 (77.7% women, 55.5 mean age, 10.7 levels fused segments, 55.5% pelvic fixation, 21.2% 3CO) by 57 surgeons at 24 sites in 5 countries (2 continents), with 2,047.9 observation-years, were included in the analysis. Missing value imputation was 14.59%. C-statistic value (70.6%) proved successful model fit. Models demonstrate that 87.9% of patients are MC-free at 10days postop, 78.5% at 90days and 63% at 2years. Surgical invasiveness (LIV-pelvic fixation, length of fusion, prior surgery), age, magnitude of sagittal deformity, patient frailty (walking and lifting capacity) and blood loss most strongly predict MC. Surgeon and site most strongly predict READMIT and UNPLAN. Curves show a continued survivorship decrease for event free MC, READMIT and UNPLAN beyond >2yFU. CONCLUSIONS Risk calculating models for event-free MC, READMIT and UNPLAN following ASD surgery demonstrate that patient-related factors, >1/3 of which are modifiable, account for 55% of the MC predictive model weight. Surgeon and site represent 4% for MC, but are most relevant for READMIT and UNPLAN.