Diagnostic models for fever of unknown origin based on ¹⁸F-FDG PET/CT: a prospective study in China

Fever of unknown origin (FUO) was originally described by Petersdorf and Beeson as a temperature repeatedly higher than 38.3 °C for at least 3 weeks with no ascertainable cause after a week of hospitalization and investigation [1]. FUO remains a substantial challenge for clinicians due to its inclusion of approximately 200 potential causes in four categories: infections, noninfectious inflammatory diseases (NIIDs), malignancies, and miscellaneous causes [2‐6]. The pathophysiology, treatment, and prognosis of these categories differ markedly [7]. Early identification of the underlying category is important to optimize treatment strategies for patients with FUO [8‐10]. Currently, ¹⁸F-fluorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG PET/CT) is considered one of the most promising diagnostic tools for FUO [11‐29].

Previous studies have mainly been small case series (between 24 and 376 cases) and retrospective studies [11‐21]. The application of PET/CT in etiological classification of FUO was not sufficient. Although PET/CT can provide some diagnostic clues, it is difficult to directly provide a definitive etiological diagnosis of FUO. Therefore, we conducted a study in a Chinese population to explore the PET/CT characteristics for different causes of FUO and further evaluate the clinical significance of PET/CT in the diagnosis of FUO. Specifically, the purpose of this study was to identify the most suitable model for diagnosing patients with FUO based on PET/CT imaging.

This study aims to analyze the PET/CT characteristics of different causes of FUO and identify independent predictors to develop a suitable diagnostic model for distinguishing between these causes.

Due to the complexity and diversity of FUO-related diseases, the multiplicity of etiologies, the atypical symptoms, and the small number of diagnostic clues, diagnosing the intrinsic etiologies of FUO remains a challenge for contemporary clinicians, requiring extensive time-consuming physical and laboratory tests and even diagnostic treatment. In this study, infection (223/524, 42.6%) accounted for the largest proportion of cases. A total of 23.1% (121/524) of patients had malignancies, 20.8% (109/524) had NIIDs, and 9.4% (49/524) received no diagnosis. This etiological distribution is in concordance with a Chinese survey involving 1641 patients with FUO [34]. Early identification of the underlying cause and implementation of appropriate empirical treatment are effective strategies for patients with FUO.

The high sensitivity and wide visual field of PET/CT are helpful in detecting lesions that are clinically unknown or undetectable with conventional imaging techniques and in obtaining multisystem observations [25, 26]. FDG uptake is an indicator of increased intracellular glucose metabolism, and this molecule can be absorbed not only by malignant cells, but also by cells involved in infection and inflammation [13, 14, 18, 19, 25]. The presence of abnormal uptake can help guide further investigations, leading to a definitive diagnosis. The absence of abnormal absorption reasonably ensures that the possible underlying conditions are not present, thus avoiding unnecessary additional testing. Therefore, in recent years, the diagnostic role of PET/CT in FUO has been widely recognized, especially in oncology [11‐19]. SUVmax can be obtained by routine scanning, which is easy to obtain and reproduce and directly reflects the highest activity of the lesion, so it is widely used in clinical practice. Consistent with previous studies, the uptake of FDG in neoplastic diseases was significantly higher than that in other diseases. SUVmax had moderate diagnostic performance for malignancy with an AUC of 0.79 and a low diagnostic performance for nonneoplastic diseases, which is not satisfactory for clinicians. Although the lymph nodes and bone marrow were the most common sites of the highest enhancement in all etiologies, other hyperenhancement sites varied among etiologies. In particular, nasopharyngeal hypermetabolism was common in tumors; the final diagnostic model confirmed that nasopharyngeal metabolic characteristics were helpful for the diagnosis of tumors.

Previous studies divided positive PET/CT results into focal FDG uptake and nonspecific abnormal uptake (NAU). NAU manifests as spleen and bone marrow diffuse high uptake and multiple reactive hyperplasia lymph nodes with high uptake and symmetrical distribution. NAU is common and has been considered a negative or a minor factor in diagnosis in some studies [12‐15, 17, 18]. However, there were significant differences in the metabolic characteristics of the spleen, bone marrow, and lymph nodes among different etiologies of FUO. In general, infection manifested a low incidence of increased FDG uptake and a low corresponding SUVmax, malignancy was characterized by a high incidence of increased FDG avidity and a high corresponding SUVmax, and NIID showed a high incidence of increased FDG uptake (except in the central lymph nodes) but a low corresponding SUVmax. Nevertheless, these parameters performed finitely when used separately in differentiating the cause of FUO into infection, malignancy, and NIID, and their diagnostic efficiency need to be further improved.

Preliminary studies have demonstrated that clinical parameters, such as rash, weight loss, SF, and LDH, can be combined with imaging findings to obtain a more confident diagnosis [15, 35]. Some researchers have built diagnostic models for patients with FUO, but only for particular causes or patients [9, 10, 15]. Efstathiou et al. focused on the distinction between infectious and noninfectious, and XU et al. emphasized the discrimination of bacterial infections. Wang’s research was also based on PET/CT and clinical features, but the diagnostic formula was mainly aimed at NAU patients, which was not simple and feasible in practice. Unlike previous studies that focused only on blood markers at admission, this study suggested that dynamic changes in these parameters were more conducive to a differential diagnosis, further confirming the importance of dynamic observation in the diagnosis of FUO. The indexes used to construct the models were consistent with the clinical characteristics of the corresponding diseases. There are some common clinical indicators used to identify certain etiology, such as IGRA for infection, low platelet count and splenomegaly for cancer, rash, ANA and ANCA for NIID, but they are not sufficiently sensitive and specific when used alone. Theoretically, the combined application can further improve the diagnostic performance, but there is no consensus on how to combine them. Through a large sample population in the real world, the favorable factors conducive to the etiological classification of FUO were screened out, and the weight of each index was further analyzed to obtain the formula. The AUCs of the infection, malignancy, and NIID diagnostic models were 0.89, 0.94, and 0.95, respectively, showing good discrimination ability. Similar results were obtained for the validation cohort, which verified the stability of the models. Especially for infectious diseases and NIID, the diagnostic model combined with other clinical indicators significantly improves the diagnostic efficacy of PET/CT. When PET/CT cannot provide a clear etiological diagnosis, the diagnostic models can be used to calculate the probability of each classification, rapidly and accurately distinguishing the etiologies of FUO: infectious or noninfectious, malignant or nonmalignant, and typical NIID or non-NIID.

The strengths of our study include its prospective setting, the large sample size divided into prediction and validation cohorts, the combination of clinical parameters and PET/CT imaging, and the creation of diagnostic models in a real-life clinical setting. The models were established according to PET/CT characteristics, which further improved the diagnostic efficiency of this imaging modality in FUO. The models performed satisfactorily in differentiating FUO cases, especially for malignancy and NIID. When the etiology of FUO is unclear or the clinical features are not obvious, the models can help clinicians achieve a more accurate FUO diagnosis, avoid unnecessary invasive procedures and unreasonable use of antibiotics, and reduce the rate of missed diagnosis and delayed diagnosis.

There are some limitations to this study. The single-center design likely limits the robustness of our findings. Multicenter and larger sample studies are needed to further confirm these results. Due to the wide variety of infectious diseases and obvious differences among subgroups, the sensitivity and specificity of the infection model are not adequately high. In later stages, we will further subdivide the types of infectious diseases and build more sophisticated diagnostic models. Since SUVmax is closely related to acquisition time, PET scanner, and patient metabolism and may represent only a small pixel, the use of SUVmax in this study may have limitations. More metabolic indicators, such as SUVmean or SUVpeack, can be considered in the later stage.

¹⁸F-FDG PET/CT has a certain level of sensitivity and accuracy in the diagnosis of FUO, which can be further improved by combining clinical parameters. Diagnostic models based on PET/CT show excellent performance in the diagnosis of FUO and can be used as a reliable tool to discriminate the cause of the condition. According to the current study results on FUO patients, it is important to further investigate the application of PET/CT in diagnosing nonneoplastic diseases.