A Machine Learning Framework for Predicting Nosocomial Escherichia coli Infections in Cervical Cancer

This study aimed to characterize the etiological profile of nosocomial infections in cervical cancer patients and to develop a machine learning-based prediction model for infections caused by the predominant pathogen, Escherichia coli, to support clinical decision-making in anti-infective therapy and risk stratification.

We conducted a retrospective analysis of clinical data from 118 cervical cancer patients to evaluate the distribution and antimicrobial resistance patterns of infectious pathogens. Predictive factors for Escherichia coli infection were identified, and a corresponding prediction model was developed. All the statistical analyses were carried out via R software (version 4.3.2) and iResearch (version 2.9.2).

A total of 151 pathogenic isolates were obtained, with the highest prevalence detected in mid-stream urine samples (69.54%, 105/151). Gram-negative bacteria constituted 76.82% (116/151) of the isolates, among which Escherichia coli was the most frequently identified species (50.33%, 76/151). Antimicrobial susceptibility testing revealed resistance rates exceeding 55% to ceftriaxone, ciprofloxacin, trimethoprim-sulfamethoxazole, and levofloxacin among Escherichia coli isolates, whereas high susceptibility was retained to carbapenems, piperacillin-tazobactam, and amikacin. Logistic regression analysis revealed that Escherichia coli infection was positively associated with earlier clinical stage, absence of anemia, and mid-stream urine sample type. Within the urinary infection subgroup, positive urinary nitrite was also correlated with increased infection risk. Feature selection utilizing multiple approaches informed the construction of the prediction model. Logistic regression and svm_cross_validation exhibited stable performance in the full sample analysis. Restricting the analysis to mid-stream urine samples substantially improved model performance. The svm-based model yielded AUC values of 0.81 and 0.89 in the training and test sets, respectively, and the logistic model achieved AUCs of 0.87 and 0.90, respectively.

Nosocomial infections in cervical cancer patients are caused primarily by gram-negative bacilli within the urinary tract, with Escherichia coli representing the most prevalent pathogen. The machine learning model, which incorporates readily available clinical parameters such as disease stage, anemia status, and urinalysis results, demonstrated robust discriminatory performance in predicting Escherichia coli infection in mid-stream urine samples. This tool offers a practical approach for risk identification and guides a more targeted empiric therapy, holding promise for improving treatment outcomes in patients with cervical cancer.