Introduction:
Sexually transmitted infections (STIs) are a major public health issue, and prompt diagnosis is crucial for controlling infections. Multiplex PCR-based nucleic acid amplification tests (NAATs) are the gold standard for diagnosis, especially for identifying pathogens involved in gonococcal and nongonococcal urethritis. However, their frequent use in low-risk or asymptomatic men increases laboratory workload and healthcare expenses.

Material and methods:
This study employed machine learning algorithms to develop and validate a rule-out model designed to reduce unnecessary NAAT requests. Data from 2020 to 2024 (n = 455) were used for model training, and data from 2025 (n = 75) served as an external temporal test set. Model thresholds were optimized to ensure high sensitivity (≥ 95%) and a high negative predictive value (NPV ≥ 90%), emphasizing patient safety and reducing the risk of false negatives. The selected rule-out threshold (0.079) represented the lowest predicted-probability cut-point achieving these safety criteria and was applied to the independent test cohort. Data were collected from our institutional database. Five supervised models – XGBoost, logistic regression (LR), support vector machine (SVM), k-nearest neighbors (KNN), and random forest – were built and assessed.

Results:
XGBoost achieved the best rule-out performance (AUC = 0.86; sensitivity = 0.98; NPV = 0.91) while safely deferring NAAT testing in 44.2% of patients. Urinary WBC, leukocyte esterase, and nitrite were the strongest positive predictors, whereas absence of inflammation and asymptomatic presentation favored test deferral.

Conclusions:
This model can assist clinicians in identifying male patients likely to test negative on NAATs, supporting diagnostic stewardship through cost-effective, data-driven decision-making.