Preventing sepsis; how can artificial intelligence inform the clinical decision-making process? A systematic review

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Predictive Analytics

Background:

• Sepsis is a critical, life-threatening condition with high mortality rates.
• Artificial intelligence (AI) offers potential for early sepsis detection and clinical decision support.
• Identifying optimal predictors for AI models is crucial for effective sepsis prediction.

Purpose of the Study:

• To systematically review and identify the optimal set of predictors for training machine learning algorithms.
• To predict the likelihood of infection and subsequent sepsis in adult patients.
• To inform clinical decision-making and resource allocation in critical care.

Main Methods:

• Systematic literature review across Medline, CINAHL, and Embase databases.
• Inclusion of quantitative primary research studies on sepsis prediction in adults.
• Analysis of predictors used in machine learning algorithms for sepsis forecasting.

Main Results:

• Seventeen articles met inclusion criteria, identifying 194 unique predictors.
• Common predictors include age, vital signs, medical history (e.g., cardiovascular disease, cancer), and lab values (e.g., lactate, white blood cell count).
• AI models demonstrated an average sensitivity of 75.7% and specificity of 63.1%.

Conclusions:

• The selection of predictors significantly impacts the performance and timeframe of AI-driven sepsis prediction models.
• AI-powered sepsis prediction can help target resources to high-risk patients.
• Future research should aim to enhance the sensitivity and specificity of these predictive algorithms.