背景与目的

不明原因发热（Fever of unknown origin，FUO）是临床常见的疑难病症，其诊断与管理面临重大挑战。目前，经典型FUO的未确诊率仍高达51%，且缺乏对出院时不同诊断结局患者特征的系统分析及未确诊患者大规模长期随访数据。此外，基于血常规衍生的新型炎症标志物作为重要的辅助指标，在经典型FUO鉴别诊断中的价值尚未明确。本研究旨在分析经典型FUO不同诊断结局的影响因素及预后特征，并构建经典型FUO中感染性疾病早期风险预测模型，为经典型FUO的分层管理与精准诊疗提供临床决策依据。

方法

第一部分 经典型FUO诊断结局及预后分析：一项单中心回顾性队列研究

纳入2015年12月至2018年11月北京协和医院感染内科以经典型FUO为表现的住院患者，分析人口学及临床特征与出院时不同诊断结局的关系，并对未确诊（临床诊断和诊断不明）患者进行长期随访。

第二部分 经典型FUO中感染性疾病早期风险预测模型的建立

基于明确诊断的感染性疾病、非感染性疾病（非感染性炎症性疾病和肿瘤性疾病）患者数据，对比基线特征和13种炎症标志物的差异，筛选关键指标并构建逻辑回归模型。通过Hosmer-Lemeshow检验和受试者工作特征（Receiver Operating Characteristic，ROC）曲线评估模型效能。

结果

第一部分 经典型FUO诊断结局及预后分析：一项单中心回顾性队列研究

1. 共纳入739例患者，中位年龄50岁[四分位数间距（Interquartile Range，IQR） (32-61）]，华北地区占比最高（56.3%）。城镇职工医保（48.3%）和新农合（32.9%）为主要医疗付款方式。

2. 明确诊断率36.8%（272例），临床诊断率48.8%（361例），诊断不明率14.3%（106例）。明确诊断组患者重症监护室（Intensive Care Unit，ICU）入住率（3.7% vs 0.3%，P=0.0013）和住院费用[ 2.2（1.4，3.7）vs 1.7（1.1，2.9），P<0.001]显著高于临床诊断组。不同诊断结局组在性别、年龄、地区、文化程度、婚姻状况、医疗付款方式、吸烟史、饮酒史、基础合并症、住院时间以及行侵入性操作比例方面未见显著差异（P>0.05）。

3. 经典型FUO明确病因诊断患者中，感染性疾病最常见（56.6%），其次为非感染性炎症性疾病（19.9%）、肿瘤性疾病（9.9%）和其他疾病（13.6%）。细菌感染占47.1%；13例为多重耐药细菌感染，其中1例耐碳青霉烯类鲍曼不动杆菌（Carbapenem-Resistant Acinetobacter baumannii，CRAB）肺炎。其他疾病组女性患者比例显著高于感染性疾病组（64.9% vs 45.5%，P=0.034）和肿瘤性疾病组（64.9% vs 29.6%，P=0.005），且肿瘤性疾病组男性患者比例显著高于非感染性炎症性疾病组（70.4% vs 42.6%，P=0.018）。其他疾病组患者年龄显著低于感染性疾病[35(24, 59) vs52(33, 62)，P=0.004]、非感染性炎症性疾病[35(24, 59) vs52(34, 64)，P=0.010]和肿瘤性疾病组[35(24, 59) vs 54(39, 67)，P=0.002]。感染性疾病组合并心脏病患者比例显著高于非感染性炎症性疾病组（15.6% vs 3.7%，P=0.029）。与其他三种病因组相比，肿瘤性疾病患者住院费用和侵入性操作比例最高（P均<0.05），并且吸烟与肿瘤性疾病之间存在弱强度关系（Crammer’s V=0.166，P=0.021）。

4. 中位随访时间1836天（IQR 213-2624），总体病死率为3.3%（13例）。289例（73.0%）患者被确诊，其中与出院时临床疑似诊断一致的占93.0%（253/272），以病毒感染（44.7%）、结核病（20.9%）、成人stills病（16.2%）为主。31.8%（34/107）自行缓解，36.4%（39/107）经短期抗炎治疗缓解。

5. 诊断性抗结核治疗有效率79.1%（53/67），其中肺外结核占71.7%。治疗成功组患者入院时病程更短（P=0.028），T-SPOT.TB阳性率更高（P=0.036）。

第二部分 经典型FUO中感染性疾病早期风险预测模型的建立

1. C反应蛋白（C-reactive Protein，CRP）、血沉（Erythrocyte Sedimentation Rate，ESR）、血清铁蛋白（Serum Ferritin，SF）、CLR [CRP /淋巴细胞计数（lymphocyte count，Lym）]、及RAR[红细胞分布宽度（Red Blood Cell Distribution Width，RDW]/白蛋白（Albumin，ALB）]与感染性疾病发生率存在显著线性相关（P<0.05）。

2. 既往有心脏病史、CRP及RDW为感染性疾病的独立危险因素（模型1）。在排除变量之间的共线性后，最终建立了8种基于临床特征和（或）炎症标志物的用于早期识别感染性疾病风险模型。ROC曲线下面积（Area Under The ROC Curve，AUC）为0.608-0.755。基于临床特征联合多种炎症标志物构建的模型1诊断效能最好[AUC 0.755（95%可信区间（confidence interval，CI）：0.678-0.830）]，灵敏度和特异度分别为58.8%和80.6%。

结  论

第一部分 经典型FUO诊断结局及预后分析：一项单中心回顾性队列研究

    经典型FUO诊断结局与疾病严重程度、复杂性及临床决策具有相关性。感染性疾病仍是主要病因，既往合并心脏病患者需重点筛查。以经典型FUO为表现的肺炎需警惕CRAB感染。年轻女性患者病因谱更广泛。肿瘤性疾病加剧了医疗负担。经全面评估未确诊的患者，大多数预后良好，应避免过度医疗。诊断性治疗是重要的临床诊治策略。

第二部分 经典型FUO中感染性疾病早期风险预测模型的建立

多种炎症标志物可有效区分感染性疾病和非感染性疾病。基于三个变量的简单工具可辅助鉴别感染性疾病，避免滥用抗生素，减少多重耐药细菌感染风险。

Background and Objectives

Fever of unknown origin (FUO) represents a challenging clinical condition with significant diagnostic and management difficulties. The undiagnosed rate of classic FUO remains at 51%, with a lack of systematic analysis on patients characteristics across different diagnostic outcome at discharge, as well as large-scale long-term follow-up data for undiagnosed cases. Furthermore, the diagnostic value of complete blood count-derived inflammatory markers as significant auxiliary indicators in classic FUO remains unclear. This study aimed to analyze influencing factors associated with different diagnostic outcome and prognostic characteristics in patients with classic FUO, and develop an early risk prediction model for infectious etiologies, thereby providing clinical evidence for stratified management and precision medicine in FUO.

Methods

Part I: Diagnositic Outcomes and Prognostic Analysis in Patients with Classic FUO: a Single-Center Retrospective Cohort Study
Patients admitted into the department of infectious diseases at Peking Union Medical College Hospital, who were diagnosed with classic FUO, between December 2015 and November 2018, were included in this study. We systematically analyzed demographic and clinical characteristics in relation to diagnostic outcome. And long-term follow-up was conducted for undiagnosed cases (both clinically diagnosed and unresolved cases).

Part II: Development of an Early Risk Prediction Model for Infectious Etiologies in Classic FUO

Patients with definitive infectious, non-infectious inflammatory, and neoplastic diseases were enrolled. The differences in baseline characteristics and thirteen inflammatory markers were compared, aiming to screen key predictors and perform logistic regression models. The Model efficacy was assessed by Hosmer-Lemeshow test and receiver operating characteristic (ROC) curve.

Results

Part I: Diagnostic Outcomes and Prognostic Analysis in patients with Classic FUO: a Single-Center Retrospective Cohort Study

1. A total of 739 patients were enrolled, with a median age of 50 years (IQR 32-61). The highest proportion in patients with North China was found, accounting for 56.3%. The Urban Employee Basic Medical Insurance and the new Rural Cooperative Medical Scheme Insurance were the primary medical payment methods, accounting for 48.3% and 32.9%, respectively.

2. 36.8% (272 cases) achieved definitive diagnosis, while 48.8% (361 cases) received clinical diagnosis and 14.3% (106 cases) remained unresolved cases, respectively. Bacterial infections accounted for 47.1%. There were thirteen cases on multidrug-resistant bacterial infections, including one patients with Carbapenem-Resistant Acinetobacter baumannii (CRAB) pneumonia. Compared with clinical diagnostic group, the patients in definitive diagnositic group had the significantly higher ICU admission rate (3.7% vs 0.3%, P=0.0013), and hospitalization costs [ 2.2( IQR 1.4–3.7) vs 1.7( IQR 1.1–2.9)，P<0.001] were similarly seen. No significant differences were seen between the different diagnostic outcome groups in terms of gender, age, region, education level, marital status, medical payment method, smoking history, drinking history, underlying comorbidities, length of hospitalization, and proportion of invasive procedures performed (P>0.05).

3. Infectious diseases were the most common etiology among definitive patients in classic FUO (56.6%) , followed by non-infectious inflammatory diseases (19.9%), neoplasms diseases (9.9%), and other diseases (13.6%). A significantly greater proportion of female patients was observed in the other disease group compared to both the infectious disease group (64.9% vs 45.5%, P=0.034) and the neoplastic disease group (64.9% vs. 29.6%, P=0.005). In contrast, male patients were more prevalent in the neoplastic disease group than in the noninfectious inflammatory disease group (70.4% vs 42.6%, P=0.018). Patients in the other disease group were significantly younger than those in the infectious disease [35 years (IQR 24–59) vs 52 years (IQR 33–62) , P=0.004], non-infectious inflammatory disease [35 years (IQR 24–59) vs 52 years (IQR 34–64) , P=0.010], and neoplastic disease groups [35 years (IQR 24–59) vs 54 years (39–67), P=0.002].The proportion of underlying comorbid heart disease patients in infectious diseases were significantly higher than the non-infectious inflammatory group (15.6% vs 3.7%, P=0.029). In comparison to the remaining three etiological categories, Patients with neoplasm diseases had the highest hospitalization costs and invasive procedure rates (both P<0.05). A statistically significant but weak association was observed between smoking and neoplasms (Crammerr's V=0.166, P=0.021).

4. The median duration of follow-up was 1836 days (IQR 213–2624), the overall mortality rate was 3.3% (13 cases). Among definitive cases (289 cases, 73.0%), 93.0% (253/272) were consistent with clinical suspected diagnoses at the time of discharge, predominantly vivus infection, tuberculosis and adult-onset still's disease, accounting for 44.7%, 20.9% and 16.2%, respectively. 31.8% (34/107) experienced spontaneous remission, while 36.4% (39/107) improved with short-term anti-inflammatory therapy.  

5. Diagnostic anti-tuberculosis treatment achieved a 79.1% efficacy rate (53/67), with extrapulmonary TB accounting for 71.7%. Patients in successful treatment group had a shorter duration of illness at admission (P=0.028), and a higher T-SPOT.TB positivity (P=0.036).

Part II: Development of an Early Risk Prediction Model for Infectious Etiologies in Classic FUO

1. There was a significant linear correlation (P<0.05) between CRP, ESR, SF, CLR (CRP/Lym), and RAR (RDW/ALB) and infectious diseases (P<0.05).

2. Underlying comorbid heart disease, CRP, and RDW were independent risk factors for infectious disease, to construct Model 1. After addressing multicollinearity among variables, we ultimately developed eight risk prediction models incorporating clinical features and/or inflammatory markers for early identification of infectious causes in FUO. The models demonstrated modest discriminative ability, with area under the ROC curve (AUC) values ranging from 0.608 to 0.755. The Model 1 demonstrated the optimal diagnostic performance, achieving an AUC of 0.755 (95% CI: 0.678–0.830), with a sensitivity of 58.8% at 80.6% specificity.

Conclusions

Part I: Diagnostic Outcomes and Prognostic Analysis in Patients with Classic FUO: a Single-Center Retrospective Cohort Study
Diagnostic outcomes in classic fever of unknown origin (FUO) significantly correlate with disease severity, complexity, and therapeutic decision-making processes. Infections predominate in FUO patients, especially in previous cardiology, while Neoplastic diseases incur a significant burden on healthcare systems. CRAB infection needs to be considered for pneumonia with classic FUO. Younger females have a wide spectrum of diseases. The majority of patients with undefinitive diagnosis subgroup generally have favorable prognosis after fully systemic evaluation, and overmedication should be avoided. The empirical treatment remains a key clinical diagnostic and therapitic strategy.

Part II: Development of an Early Risk Prediction Model for Infectious Etiologies in Classic FUO
Multiple inflammatory biomarkers demonstrate significant diagnostic utility in distinguishing infectious from noninfectious etiologies. A simple tool based on three variables enables differentiation of infectious diseases in FUO, avoiding misuse of antibiotics, and reducing the risk of infection with multidrug resistant bacteria.