| 摘要: |
| ［摘要］ 目的 分析妇科恶性肿瘤患者院内感染的病原学分布特征，并构建主要致病菌大肠埃希菌感染的预测模型，为临床抗感染治疗和风险评估提供依据。方法 回顾性分析2020年5月至2022年1月安徽省肿瘤医院妇科肿瘤术后1周内及化疗住院期间发生感染的146例患者的临床资料。分析感染病原菌分布特征，分析主要致病菌大肠埃希菌耐药情况及感染的影响因素并构建预测模型。结果 共检出180株病原菌，中段尿液样本占比最高（65.56%，118/180）。革兰阴性杆菌占80.56%（145/180），其中大肠埃希菌占革兰阴性杆菌的70.34%（102/145），占总病原菌的56.67%（102/180）。药物敏感性试验结果显示，大肠埃希菌对头孢曲松、环丙沙星、复方新诺明及左氧氟沙星的耐药率均超过55%，对亚胺培南、厄他培南、哌拉西林/他唑巴坦、呋喃妥因及阿米卡星的耐药率均低于10%。在全部样本中，单因素logistic回归分析显示大肠埃希菌感染仅与中段尿液样本类型显著相关（P=0.004），故后续仅纳入中段尿液样本进行分析。在中段尿液样本中，多因素logistic回归分析结果表明，大肠埃希菌感染可能与肿瘤类型、白细胞计数（WBC）、脂蛋白a（Lpa）、尿亚硝酸盐（UNIT）及尿液细菌数量相关（P<0.1）。基于上述5个变量，采用多种机器学习方法构建大肠埃希菌感染评估模型。结果显示，决策树算法和逻辑回归算法模型表现稳定，决策树算法模型训练集与测试集曲线下面积（AUC）分别为0.838和0.818，逻辑回归算法模型训练集与测试集AUC分别为0.775和0.737。两种模型的校准曲线及决策曲线分析（DCA）均显示出优良的性能。结论 妇科恶性肿瘤患者院内感染以泌尿系统革兰阴性杆菌为主，大肠埃希菌为优势菌。基于肿瘤类型、WBC、Lpa、UNIT及尿液细菌数量构建的决策树算法模型和逻辑回归算法模型对中段尿液样本中该菌感染识别性能优异，有助于早期识别高危患者，指导精准用药，对减少耐药及改善抗感染结局具有重要价值。 |
| 关键词: 妇科恶性肿瘤 机器学习 病原菌 感染预测模型 院内感染 |
| DOI：10.3969/j.issn.1674-3806.2026.07.04 |
| 分类号: |
| 基金项目:国家自然科学基金项目（编号：82404092）；常州市卫生健康委重大科技项目（编号：ZD202314） |
|
| A machine-learning framework for predicting nosocomial Escherichia coli infection in patients with gynecological malignant tumors |
|
Shan Wulin1,2, Peng Wenju3, Xu Xinxin4, Kan Jinsong1,2, Zhang Jiayun1,2, Chen Jiming5
|
|
1.Department of Laboratory Medicine, the West District of the First Affiliated Hospital of University of Science and Technology of China(Anhui Provincial Hospital), Hefei 230031, China; 2.Department of Laboratory Medicine, Anhui Provincial Cancer Hospital, Hefei 230031, China; 3.Department of Gynecological Tumor Surgery, the West District of the First Affiliated Hospital of University of Science and Technology of China(Anhui Provincial Hospital), Hefei 230031, China; 4.Department of Medical Laboratory Science, Anhui Institute of Medicine, Hefei 230601, China; 5.Department of Gynecology, the Third Affiliated Hospital of Nanjing Medical University(the Second People′s Hospital of Changzhou), Changzhou 213000, China
|
| Abstract: |
| ［Abstract］ Objective To analyze the pathogenic distribution characteristics of nosocomial infections in patients with gynecological malignant tumors and to construct a predictive model for infections caused by the predominant causative agent, Escherichia coli, thereby providing a basis for clinical anti-infection treatment and risk assessment. Methods A retrospective analysis was conducted on the clinical data of 146 patients who developed infections within 1 week after gynecological tumor surgery and during hospitalization for chemotherapy in Anhui Provincial Cancer Hospital from May 2020 to January 2022. The distribution characteristics of the pathogenic microorganisms causing infections, the drug resistance of the predominant causative agent, Escherichia coli, and the influencing factors of the infections were analyzed, and a predictive model was constructed. Results A total of 180 strains of pathogenic microorganisms were isolated, with mid-stream urine samples accounting for the highest proportion(65.56%, 118/180). Gram-negative bacilli accounted for 80.56%(145/180) of all the isolates, and Escherichia coli accounted for 70.34%(102/145) of the Gram-negative bacilli and 56.67%(102/180) of the total pathogenic microorganisms. Antimicrobial susceptibility testing results showed that Escherichia coli exhibited resistance rates to ceftriaxone, ciprofloxacin, trimethoprim/sulfamethoxazole and levofloxacin exceeding 55%, while the resistance rates to imipenem, ertapenem, piperacillin/tazobactam, nitrofurantoin and amikacin were less than 10%. Across all the sample types, univariate logistic regression analysis revealed that Escherichia coli infection was only significantly associated with mid-stream urine samples(P=0.004). Therefore, the subsequent analyses were restricted to the mid-stream urine samples. In the mid-stream urine samples, multivariate logistic regression analysis showed that Escherichia coli infection might be associated with tumor type, white blood cell count(WBC), lipoprotein a(Lpa), urinary nitrite(UNIT), and bacterial count in urine(P<0.1). Based on the above 5 variables, several machine learning methods were adopted to construct assessment models for Escherichia coli infection. The results showed that the decision tree and logistic regression models demonstrated stable performance. For the decision tree model, area under the curve(AUC) was 0.838 in the training set and 0.818 in the test set. For the logistic regression model, AUC was 0.775 in the training set and 0.737 in the test set. Furthermore, the calibration curves and decision curve analysis(DCA) in both models demonstrated excellent performance. Conclusion Nosocomial infections in patients with gynecological malignant tumors are predominantly caused by Gram-negative bacilli in the urinary tract, with Escherichia coli being the dominant pathogen. The decision tree and logistic regression models built on tumor type, WBC, Lpa, UNIT and bacterial count in urine show excellent performance in identifying Escherichia coli infection in mid-stream urine samples. The two models facilitate early identification of high risk patients and guide targeted antimicrobial therapy, which are of great value in reducing antibiotic resistance and improving anti-infection outcomes. |
| Key words: gynecological malignant tumors machine learning pathogenic microorganisms predictive model for infections nosocomial infections |
