Clinical prediction for surgical versus nonsurgical interventions in patients with vertebral osteomyelitis and discitis.

Abstract

Vertebral osteomyelitis and discitis (VOD), an infection of intervertebral discs, often requires spine surgical intervention and timely management to prevent adverse outcomes. Our study aims to develop a machine learning (ML) model to predict the indication for surgical intervention (during the same hospital stay) versus nonsurgical management in patients with VOD. This retrospective study included adult patients (≥18 years) with VOD (ICD-10 diagnosis codes M46.2,3,4,5) treated at a single institution between 01/01/2015 and 12/31/2019. The primary outcome studied was surgery. Candidate predictors were age, sex, race, Elixhauser comorbidity index, first-recorded lab values, first-recorded vital signs, and admit diagnosis. After splitting the dataset, XGBoost, logistic regression, and K-neighbor classifier algorithms were trained and tested for model development. A total of 1,111 patients were included in this study, among which 30% (n=339) of patients underwent surgical intervention. Age and sex did not significantly differ between the two groups; however, race did significantly differ (P<0.0001), with the surgical group having a higher percentage of white patients. The top ten model features for the best-performing model (XGBoost) were as follows (in descending order of importance): admit diagnosis of fever, negative culture, Staphylococcus aureus culture, partial pressure of arterial oxygen to fractional inspired oxygen ratio (PaO2:FiO2), admit diagnosis of intraspinal abscess and granuloma, admit diagnosis of sepsis, race, troponin I, acid-fast bacillus culture, and alveolar-arterial gradient (A-a gradient). XGBoost model metrics were as follows: accuracy =0.7534, sensitivity =0.7436, specificity =0.7586, and area under the curve (AUC) =0.8210. The XGBoost model reliably predicts the indication for surgical intervention based on several readily available patient demographic information and clinical features. The interpretability of a supervised ML model provides robust insight into patient outcomes. Furthermore, it paves the way for the development of an efficient hospital resource allocation instrument, designed to guide clinical suggestions.