Hospital-acquired infections surveillance: The machine-learning algorithm mirrors National Healthcare Safety Network...

Area of Science:

• Infectious Diseases
• Medical Informatics
• Artificial Intelligence

Background:

• Hospital-acquired infections (HAIs) surveillance is critical for infection control.
• Machine learning (ML) shows promise for enhancing HAI surveillance.
• This study compares manual surveillance with a ML-based semiautomated method.

Purpose of the Study:

• To compare the efficacy of manual HAI surveillance with a supervised, semiautomated ML method.
• To explore infection types and important features identified by the ML model.
• To assess the potential of ML in improving HAI detection rates.

Main Methods:

• A semiautomated ML random forest algorithm was implemented in a Brazilian hospital from July to December 2021.
• Inpatient records were reviewed manually and by the ML method.
• A panel of experts validated the ML-identified HAIs.

Main Results:

• The semiautomated method identified 4.7% of patients with HAIs compared to 2.9% with manual surveillance.
• The ML method detected 77 additional respiratory infections, accounting for 93.9% of newly identified HAIs.
• The ML model utilized 447 features, with over 50% aligning with NHSN criteria for HAI classification.

Conclusions:

• The ML algorithm augmented human capacity for HAI classification, identifying more cases than traditional methods.
• The inclusion of numerous features, including NHSN criteria, enhances the model's performance.
• Documented ML algorithm performance can support the integration of AI into clinical and epidemiological practices for improved HAI surveillance.