Secondary vs. Primary Spinal Infection in Early Clinical Assessment: A Parsimonious, Leakage-Resistant Modelling...

Area of Science:

• Infectious Diseases
• Nephrology
• Endocrinology
• Spine Surgery
• Biostatistics

Background:

• Spinal infections are diverse, necessitating etiological classification for effective diagnosis and treatment.
• Comorbidities like chronic kidney disease (CKD) and diabetes mellitus (DM) may influence the etiology of spinal infections.
• Multicenter data is crucial for understanding the complex interplay between comorbidities and spinal infection types.

Purpose of the Study:

• To evaluate etiological patterns in a multicenter cohort of spinal infections.
• To investigate the association between CKD, DM, and primary versus secondary spinal infection etiologies.
• To contribute novel insights into the role of comorbidity profiles in spinal infection classification.

Main Methods:

• Employed a ridge-penalized logistic regression (L2) model with nested cross-validation for robust probability estimation.
• Optimized model parameters (penalty parameter C) using log-loss minimization to prevent data leakage.
• Conducted a supplementary Firth-penalized analysis for validation, using CKD/DM-negative patients as a reference group.

Main Results:

• The model effectively discriminated between spinal infection probabilistic profiles (OOF AUC 0.762).
• Diabetes mellitus (DM) patients showed a high probability of secondary spinal infections (84.4%), contrasting with CKD patients' higher likelihood of primary infections (84.6%).
• No cases with both CKD and DM were observed; analyses were confined to relevant patient strata.

Conclusions:

• Chronic kidney disease (CKD) is predominantly associated with primary spinal infection etiology.
• Diabetes mellitus (DM) is more frequently linked to secondary spinal infection etiologies.
• Comorbidity profiles significantly influence the etiological classification of spinal infections, warranting further investigation in larger cohorts.