Clinical Phenotypes of Oral and Maxillofacial Space Infection: Derivation, External Assessment, and Prognostic Evaluation

Background/Objectives: Oral and maxillofacial space infection (OMSI) has heterogeneous clinical presentations. We aimed to derive candidate clinical phenotypes at hospital presentation and assess their external structural reproducibility and prognostic value beyond anatomical severity. Methods: This two-center retrospective study included 287 patients in the development cohort and 51 in the external cohort. Consensus k-means clustering was performed using 24 presentation variables across 20 imputed datasets, followed by frozen external phenotype assignment. Firth’s bias-reduced logistic regression was used to fit the traditional severity score (TSS), phenotype-only, and combined models for the recorded primary composite outcome of intensive care unit admission, tracheostomy, septic shock, or descending mediastinal infection. Phenotype assignments remained fixed during bootstrap internal validation. Results: Four candidate phenotypes showed different inflammatory, physiological, and metabolic profiles. The febrile and glucose-dominant phenotypes both had elevated inflammatory markers but differed in temperature and glucose levels. A history of diabetes was recorded in 29 of 31 patients in the glucose-dominant phenotype. The primary composite outcome occurred in 35/287 (12.2%) and 22/51 (43.1%) patients in the development and external cohorts, respectively. Independent external clustering showed limited agreement with the frozen assignments (adjusted Rand index = 0.320). Bootstrap optimism-corrected areas under the receiver operating characteristic curve (AUCs) were 0.798, 0.631, and 0.799 for the TSS, phenotype-only, and combined models, respectively. The corresponding external AUCs were 0.926, 0.665, and 0.891. All three frozen models underestimated absolute risk in the external cohort (observed-to-expected ratios: 4.425, 3.620, and 4.424, respectively). Conclusions: The candidate phenotypes described overlapping inflammatory, physiological, and metabolic patterns at presentation. Independent external clustering showed limited structural agreement, and adding phenotype membership to TSS did not consistently improve prognostic discrimination. Their role in clinical decision-making requires further evaluation.