Secondary vs. Primary Spinal Infection in Early Clinical Assessment: A Parsimonious, Leakage-Resistant Modelling Approach with Internal Validation: A Multicenter Retrospective Study

TL;DR

This study uses a statistical model to show how chronic kidney disease and diabetes are linked to different types of spinal infections, helping guide clinical decisions.

Contribution

A novel leakage-resistant modeling approach is introduced to associate comorbidities with primary or secondary spinal infection etiologies.

Findings

Diabetes mellitus is strongly associated with secondary spinal infections.

Chronic kidney disease is predominantly linked to primary spinal infections.

The model effectively discriminates infection profiles with an AUC of 0.762.

Abstract

Background and Objectives: Spinal infections represent a heterogeneous group of diseases where primary or secondary etiological classification is fundamental for diagnosis and clinical decision-making. The aim is to present multicenter data evaluating etiological patterns associated with comorbidity. This study investigated the etiological distribution of spinal infections in a multicenter cohort and examined the relationships between chronic kidney disease (CKD) and diabetes mellitus (DM) and primary and secondary spinal infection etiologies, which emerged in the study and are thought to contribute to the literature. Materials and Methods: For this early-phase exploratory modelling study, a ridge-penalized logistic regression (L2) model was trained using repeated nested cross-validation (outer 5-fold stratified CV ×10 repetitions; inner 5-fold CV) to generate out-of-fold (OOF) probabilities. The penalty parameter (C) was optimized by minimizing log-loss. All preprocessing was performed within the CV pipeline to prevent data leakage. A supplementary Firth-penalized analysis was conducted as a plausibility check, using the CKD0/DM0 group as reference. Results: The model demonstrated effective discrimination between spinal infection probabilistic profiles (OOF AUC 0.762; conditional OOF bootstrap 95% CI 0.608–0.885). A contrasting probabilistic profile concordance was observed: DM-only patients had a high likelihood of secondary infection (observed secondary risk 93.3%; mean OOF estimated probability 84.4%), compared to a higher likelihood of primary infection in CKD-only patients (observed secondary risk 15.4%, which translates to a primary risk of 84.6%; mean OOF estimated probability 21.8%). Calibration was near-ideal (intercept 0.069; slope 1.028). Decision curve analysis showed a clear utility between the thresholds of 0.15 and 0.84. There were no CKD+DM+ cases (n = 0); analyses were restricted to supported strata. Conclusions: In this multicenter analysis of spine infections, CKD was predominantly related to primary spine infection etiology, whereas DM was more frequently related to secondary spine infections. These findings emphasize the potential role of comorbidity profiles in etiologic classification and need to be confirmed in larger multicenter cohorts.

Figures