Can Artificial Intelligence Optimize the Early Diagnosis of Invasive Candidiasis? A Systematic Review and...

Area of Science:

• Medical Informatics
• Infectious Diseases
• Machine Learning

Background:

• Early diagnosis of invasive candidiasis is difficult in immunocompromised patients.
• Artificial intelligence (AI) offers potential for improved clinical decision-making.

Purpose of the Study:

• To systematically review and meta-analyze AI-based predictive models for early identification of invasive *Candida* infections.
• To assess the diagnostic accuracy and limitations of current AI models.

Main Methods:

• PROSPERO-registered systematic review and meta-analysis of AI models for invasive candidiasis.
• Searched multiple databases for studies on hospitalized immunocompromised patients.
• Bivariate random-effects meta-analysis for candidemia prediction models.

Main Results:

• Eight studies met inclusion criteria; models used retrospective data with varied predictors.
• Pooled sensitivity for candidemia prediction was 81.3%, specificity was 81.6%.
• High negative predictive values, modest positive predictive values; moderate to high risk of bias; low certainty of evidence.

Conclusions:

• AI models demonstrate potential for early candidemia identification with moderate accuracy.
• These models may serve as decision-support tools.
• Multicenter prospective validation is essential before routine clinical adoption.