Clinical prediction for surgical versus nonsurgical interventions in patients with vertebral osteomyelitis and discitis
Jennifer Lee,Miguel A Ruiz-Cardozo,Rujvee P Patel,Saad Javeed,Raj Swaroop Lavadi,Catherine Newsom-Stewart,Anton Alyakin,Camilo A Molina,Nitin Agarwal,Wilson Z Ray,Michele Santacatterina,Brenton H Pennicooke
DOI: https://doi.org/10.21037/jss-23-111
2024-06-21
Abstract:Background: 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. Methods: 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. Results: 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. Conclusions: 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.