Development and validation of 18F-FDG PET/CT-based multivariable clinical prediction models for etiological types of fever of unknown origin.

Background: The etiology of fever of unknown origin (FUO) is complex, and early diagnosis of FUO is important for determining the method of treatment, but there are no uniform diagnostic criteria. This study aimed to evaluate the value of 18F-fluorodeoxyglucose positron emission tomography/computed tomography (18F-FDG-PET/CT) in differentiating the etiological types of FUO and to develop a predictive model. Methods: A total of 456 FUO patients with baseline 18F-FDG-PET/CT data were enrolled. The etiological categories were classified into neoplastic diseases (NDs) and nonneoplastic diseases (NNDs) and then classified into infectious diseases (IDs) and noninfectious inflammatory diseases (NIIDs). Continuous variables were compared with the Mann-Whitney U test, and categorical variables were compared with the Chi-squared test. Features were screened via the above method, and multivariate logistic regression (Bonferroni correction) was subsequently applied to construct two models: a clinical model (laboratory data only) and a combined model (laboratory + PET/CT data), with corresponding nomograms generated. The model performance was assessed by calculating the area under the receiver operating characteristic curve (AUC), which was compared via the Delong test. Results: To differentiate between NDs and NNDs, five PET/CT variables [asymmetrical distribution of hypermetabolic lymph nodes, maximum standardized uptake value (SUVmax) in the bone marrow, SUVmax in the spleen, splenomegaly, and focal liver lesion], and three laboratory variables [white blood cell (WBC) count, platelet count (PLT) and lactate dehydrogenase (LDH)] (all P<0.05) were selected to construct the models. To differentiate between IDs and NIIDs, three PET/CT variables (short diameter of the lymph node, hypermetabolic spleen, and focal spleen lesion), one laboratory variable (WBC count) and age (all P<0.05) were selected to construct the models. In the validation group, the combined model (AUC =0.742) did not demonstrate significant improvement over the clinical model-either in differentiating NDs from NNDs (AUCcombined =0.742 vs. AUCclinical =0.680; P=0.080) or in distinguishing ID from NIID (AUCcombined =0.737 vs. AUCclinical =0.678; P=0.111). Conclusions: This study demonstrated the diagnostic value of 18F-FDG PET/CT in differentiating etiological types of FUO. However, the addition of 18F-FDG PET/CT did not significantly enhance the ability of clinical models to distinguish between NDs and NNDs or between IDs and NIIDs. Further validation in future studies is warranted.