Heterogenous effect of automated alerts on mortality

Area of Science:

• Nephrology
• Medical Informatics
• Clinical Decision Support

Background:

• Electronic alerts aim to improve patient outcomes by flagging potential health issues like acute kidney injury (AKI).
• The effectiveness of these alerts can vary significantly among different patient populations.
• Understanding this heterogeneity is crucial for optimizing alert systems.

Purpose of the Study:

• To investigate the diverse treatment effects of electronic alerts for acute kidney injury (AKI).
• To identify patient subgroups that benefit from or are potentially harmed by AKI alerts.
• To explore how alert-triggered provider actions influence patient mortality.

Main Methods:

• Secondary analysis of individual patient data from three randomized controlled trials.
• Utilized machine learning to predict individualized alert effects on 14-day all-cause mortality.
• Performed internal and external validation, including a meta-analysis of individual patient data.

Main Results:

• Patients predicted to benefit from alerts showed a lower risk of death compared to those predicted to be harmed.
• Machine learning identified reduced mortality with alerts in patients with higher blood pressure and lower predicted risk.
• Increased mortality was observed with alerts in non-urban and non-teaching hospitals, with varied provider responses.

Conclusions:

• Significant heterogeneity exists in the impact of AKI alerts on patient mortality.
• Tailoring alert delivery based on predicted benefit may reduce harm and improve clinical outcomes.
• Individualized alerts hold potential for reducing all-cause mortality, warranting prospective trials.