Equitable Machine Learning for Hypoglycaemia Risk Management.

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Healthcare
• Patient Safety

Background:

• Hypoglycaemia poses a significant risk to hospitalized patients.
• Current detection methods may be insufficient for timely intervention.
• Machine learning offers potential for proactive risk identification.

Purpose of the Study:

• To develop and validate a machine learning model for predicting high-risk hypoglycaemia events in hospitalized patients.
• To improve the early detection and management of hypoglycaemia.
• To ensure equitable performance across different patient subgroups.

Main Methods:

• Development of a machine learning model using data from an Australian regional health district.
• Training the model on historical patient data to identify risk factors for hypoglycaemia.
• Conducting subgroup analyses based on gender and Indigenous status to assess model fairness.

Main Results:

• The machine learning model demonstrated strong predictive performance with an Area Under the Curve (AUC) of 0.837.
• Subgroup analysis indicated that the model did not disadvantage specific population groups.
• The model's reliance on objective data reduced bias associated with practice patterns.

Conclusions:

• Machine learning models can effectively identify high-risk patient cohorts for hypoglycaemia.
• Careful equity analysis is crucial for unbiased ML deployment in healthcare.
• Automated detection and mitigation strategies hold promise for reducing preventable errors and improving patient safety.