Prediction model for etiology of fever of unknown origin in children.

Abstract

Diagnosing fever of unknown origin (FUO) in children remains challenging, particularly in differentiating between infections, autoimmune diseases, and malignancies. We aimed to develop and validate a prediction model for determining the etiology of pediatric FUO. We retrospectively reviewed medical records of children aged 1-18 years with FUO lasting ≥ 7 days from 2007 to 2023. Clinical and laboratory data were collected. The study was conducted in two phases: (1) model development (development cohort) and (2) internal validation (validation cohort). Multinomial logistic regression and predictive margin analyses were used to construct the model, with performance assessed by the area under the Receiver Operating Characteristic curve (AUC). In the development cohort (n = 240, median age: 6.4 years, IQR 3.4-11.6), FUO was attributed to infections (32.5%), autoimmune diseases (34.2%), and malignancies (33.3%). Using infections as a reference, arthritis (OR = 32.8, 95%CI 6.5-166.4) and fever > 30 days (OR = 10.3, 95%CI 2.9-35.4) were predictors of autoimmune diseases; while splenomegaly (OR = 5.2, 95%CI 1.8-15.6), lymphadenopathy (OR = 4.2, 95%CI 1.6-11.2), severe anemia (OR = 9.2, 95%CI 2.3-36.9), thrombocytopenia (OR = 10.0, 95%CI 3.3-30.1), and fever > 30 days (OR = 19.4, 95%CI 5.1-73.8) were predictors of malignancies. Coughing was inversely associated with both autoimmune (OR = 0.1, 95%CI 0.1-0.4) and malignancies (OR = 0.1, 95%CI 0.04-0.4). A computerized prediction model was constructed using these parameters. The validation cohort (n = 78) demonstrated good discrimination for infection (AUC = 0.82), autoimmune (AUC = 0.88), and malignancies (AUC = 0.83).Conclusions: A prediction model has been developed and validated to assist pediatricians in differentiating the causes of FUO. It demonstrates good performance and supports data-driven decision-making in pediatric FUO.