Text

Objectives and questions: Primary spinal infections (PSIs) are an increasing global health concern associated with substantial morbidity/mortality. Management remains heterogeneous due to the limited scope/reproducibility of existing classification systems. To address these limitations, the AO Spine Knowledge Forum Trauma and Infection (T&I) developed a novel comprehensive classification system incorporating anatomic location, morphologic pattern, neurologic status, and systemic modifiers. This study aimed to perform internal and international validation of the AO Spine Primary Infection Classification System by assessing accuracy, interobserver reliability, and intraobserver reliability.

Material and methods: De-identified imaging studies and standardized clinical vignettes from 19 cases of primary spinal infection were independently classified in two assessment rounds using the AO Spine Primary Infection Classification System. An internal validation cohort included AO Spine Knowledge Forum T&I experts, while an international cohort included AO Spine surgeons from diverse geographic regions. An expert panel consensus served as the reference standard. Accuracy was defined as exact agreement with the reference standard at the infection-type level (A/B/C) and across the full 12-subcategory classification (A0-A3, B0-B3, C0-C3). The reliability of Interobserver and intraobserver observations was assessed using Fleiss’ kappa statistics.

Results: A total of 570 expert ratings were generated across two rounds, with 228 surgeons participating and 150 completing both rounds. In the internal validation cohort, case-level accuracy exceeded 85% across both rounds, with near perfect interobserver reliability for infection type (κ ≈ 0.90) and substantial reliability for the full classification (κ ≈ 0.70). Intraobserver reliability in this cohort was excellent (mean κ ≈ 0.90). In the international cohort, case-level accuracy ranged from 82.6% to 87.7%. Interobserver reliability was almost perfect for infection type (κ = 0.90–0.91) and substantial for the full classification (κ = 0.68–0.72), with lower agreement observed for cervical canal–involvement subtypes (C2-C3; κ ≈ 0.47-0.54). Intraobserver reliability among surgeons completing both rounds remained excellent (mean κ = 0.90; median 1.00).

Discussion and conclusions: The AO Spine Primary Infection Classification System demonstrated high accuracy and strong interobserver/intraobserver reliability across both expert panel and international surgeon cohorts. Our findings support its validity and generalizability as a standardized framework for primary spinal infections. Targeted educational strategies may further improve consistency for cervical canal-involvement subtypes, and future studies should assess its impact on clinical decision-making and outcomes.