DSaaS

DSaaS applies machine learning to predict the risk of multidrug-resistant urinary tract infections (MDR UTIs) in hospitalized patients.

Key Features:

• Machine Learning Integration: Incorporates machine learning algorithms including Catboost, Support Vector Machine (SVM), and Neural Networks to build predictive models.
• Model Validation and Evaluation: Validates models using supervised regression and classification and reports performance metrics: accuracy (ACC), AUC-ROC, AUC-PRC, F1 score, sensitivity (SEN), specificity, and Matthews correlation coefficient (MCC).
• Real-world Clinical Dataset: Applied to a dataset of 1,486 hospitalized patients with nosocomial urinary tract infections from the Clinical Pathology Operative Unit of the Principe di Piemonte Hospital in Senigallia, Italy, using predictors sex, age, age class, ward, and time period.
• Predictive Performance: Demonstrated superior performance of Catboost over SVM and Neural Networks, with reported metrics: MCC 0.909, SEN 0.904, F1 0.809, AUC-PRC 0.853, AUC-ROC 0.739, and ACC 0.717.
• Planned Computational Enhancements: Expansion to unsupervised machine learning, streaming data analysis, distributed computation, and big data storage and management.

Scientific Applications:

• MDR UTI Risk Stratification: Enables identification of hospitalized patients at high risk of acquiring multidrug-resistant urinary tract infections.
• Support for Targeted Interventions and Antibiotic Stewardship: Provides predictive analytics to inform targeted clinical interventions and optimize antibiotic use.
• Analysis of Nosocomial UTI Datasets: Facilitates analysis of hospitalized patient datasets for nosocomial urinary tract infections to evaluate predictors of resistance.

Methodology:

Model training and validation using supervised regression and classification algorithms; evaluation with ACC, AUC-ROC, AUC-PRC, F1, sensitivity, specificity, and MCC; algorithms applied include Catboost, Support Vector Machine, and Neural Networks; analysis performed on a dataset of 1,486 hospitalized patients with nosocomial UTIs from the Clinical Pathology Operative Unit of the Principe di Piemonte Hospital (Senigallia, Italy), using predictors sex, age, age class, ward, and time period.