Preparing for future pandemics: Automated intensive care electronic health record data extraction to accelerate clinical insights.

Summary

In 1,515 intubated COVID-19 ICU patients, automatically extracted EHR data replicated prior findings that Harris-Benedict dead-space fraction rises over time and is consistently higher in non-survivors. This supports automated extraction as a credible, scalable alternative to manual abstraction for critical care research and preparedness.

Key Findings

• Automated EHR extraction replicated prior manual-abstraction findings in 1,515 intubated patients.
• Harris-Benedict dead-space fraction increased over time and remained higher in non-survivors at each time point.
• Demonstrated feasibility and credibility of automated extraction for multicenter ICU research during a pandemic.

Clinical Implications

Hospitals can adopt automated EHR extraction to monitor dead-space indices and other ICU metrics in near-real time, informing prognostication and resource allocation during surges.

Limitations

• Retrospective observational design with potential unmeasured confounding
• Reliance on estimates (e.g., HB dead-space) and potential site-level data heterogeneity

Future Directions

Prospective validation of automated pipelines, broader phenotype extraction (e.g., ventilator dyssynchrony), and integration with predictive modeling for early warning systems.