Supervised machine learning to predict non-home discharge following surgical treatment of pelvic fractures

Abstract

Background: Decision-tree-based machine learning (ML) algorithms such as random forest (RF) are useful for their ability to predict outcomes and rank variables according to their utility in the decision-making process. This study utilizes RF to identify important predictors of discharge to facility following surgical stabilization of pelvis fractures, a traumatic injury that often precludes mortality and diminished quality of life. Methods: The American College of Surgeons national surgical quality improvement program (ACS-NSQIP) database was queried for patients aged 16 to 70 undergoing surgical fixation of pelvis fractures between 2008 and 2018. Outcome of interest was discharge home versus to facility. RF was trained with surgical variables, comorbidities, and other patient factors and tasked with predicting discharge location. Permutation feature importance (PFI) was then generated to identify important variables. Results: Out of 492 patients, 184 patients were discharged to facility, and 308 patients were discharged home. RF identified age, American Society of Anesthesiologists (ASA) classification, and preoperative hematocrit as top predictors for discharge to facility. Patients being discharged home were younger, had lower ASA scores, and had higher preoperative hematocrit. Conclusions: RF identified age, ASA classification, and preoperative hematocrit as top predictors for discharge destination following pelvic surgery. Knowledge of the impact of these variables can inform preoperative planning for both patients and their care team, while highlighting the opportunity to address preoperative hematocrit to both reduce cost and improve quality of care.