68. PROMIS physical function and pain interference are independent predictors of unplanned readmission after instrumented lumbar spinal arthrodesis

BACKGROUND CONTEXT Risk stratification is essential for informed decision-making and patient counseling. The PROMIS physical function computer adaptive test (PFCAT) and pain interference (PI) scales directly evaluate physical function and limitation due to pain with low question burden and may be clinically relevant predictors of adverse outcomes in spine patients. PURPOSE To evaluate whether preoperative PROMIS PFCAT or PI scores are important predictors of unplanned readmission in adults undergoing lumbar arthrodesis. STUDY DESIGN/SETTING Retrospective analysis of a spine patient registry at a Level I academic center. PATIENT SAMPLE A total of 1,782 patients who underwent lumbar arthrodesis between 2015 and 2020 were included. OUTCOME MEASURES Primary outcome measures was unplanned hospital readmission. METHODS All cases with unplanned hospital readmission were adjudicated by manual chart review. We performed random forest prediction modeling to identify variables with the greatest predictive importance to unplanned hospital readmission risk using the mean decrease in Gini index. RESULTS Mean patient age was 62.8 (12.8) years. Mean BMI was 30.2 (6.4) kg/m2. Mean preoperative PFCAT was 33.8 (6.6) and mean PI was 66.5 (6.5). Preop CCI mean was 2.0 (2.5). Of 1,782 patients included, 146 patients (8.19%) had unplanned readmissions after surgery. In both model PFCAT and model pain interference, BMI had the greatest importance (68.9, 63.3, respectively). In model PFCAT, PFCAT had the second highest predictive importance (61.3), while in model pain interference PROMIS PI had the second greatest predictive importance (50.1). In both models, age (49.5, 46.7, respectively) and CCI (32.0, 28.7) were the third and fourth most important predictive variables for unplanned readmission, respectively. CONCLUSIONS Random forest modeling is a more flexible modeling technique than linear regression techniques, and therefore, can identify under-explored risk factors. Our analysis using random forest modeling identified BMI as the most important predictive variable for unplanned readmission, with PROMIS PFCAT and PROMIS PI demonstrating the next greatest predictive importance in respective models. Both models also included age and CCI as predictive variables of lesser importance. While traditionally used to assess post-treatment outcomes, PROMIS PFCAT and PI are stronger independent predictors of postoperative readmission than age or CCI. Further research to develop risk-stratification tools incorporating PROMIS measures and random forest models is warranted. FDA DEVICE/DRUG STATUS This abstract does not discuss or include any applicable devices or drugs.