Multifactor machine learning models for predicting urinary tract infections: a pilot study.

Area of Science:

• Immunology and Microbiology
• Biochemistry
• Artificial Intelligence in Healthcare

Background:

• Vitamin D is crucial for immune function and calcium-phosphate homeostasis.
• Vitamin D deficiency is linked to immune dysregulation, increasing risks of infections like UTIs.
• Urinary 25-hydroxyvitamin D (25(OH)D) levels are explored as a potential biomarker for UTI risk.

Purpose of the Study:

• To investigate the association between urinary 25(OH)D levels and UTI occurrence.
• To develop and evaluate machine learning models for UTI prediction using demographic and biochemical data.

Main Methods:

• Analysis of a cohort of 358 subjects with collected demographic, biochemical, and microbiological data.
• Development of 12 machine learning models utilizing urinary 25(OH)D, urine pH, age, and gender.
• Models were evaluated using metrics including accuracy, sensitivity, specificity, PPV, NPV, AUC-ROC, and F1 score.

Main Results:

• Significant differences in urinary 25(OH)D levels were observed between individuals with positive and negative urine cultures (p < 0.001).
• Machine learning models achieved accuracies ranging from 64% to 87%, with a stacking model reaching 88% accuracy.
• The best-performing stacking model demonstrated 83% sensitivity, 94% specificity, and an AUC-ROC of 0.93.

Conclusions:

• Lower urinary 25(OH)D levels are significantly associated with increased UTI occurrence.
• Machine learning models show promise as adjunct tools for UTI diagnosis, complementing traditional methods.
• Further validation and prospective studies are needed to assess the clinical impact of these predictive models.