External validation of AI-based scoring systems in the ICU: a systematic review and meta-analysis
2024

External validation of AI-based scoring systems in the ICU

Sample size: 572 publication 10 minutes Evidence: moderate

Author Information

Author(s): Rockenschaub Patrick, Akay Ela Marie, Carlisle Benjamin Gregory, Hilbert Adam, Wendland Joshua, Meyer-Eschenbach Falk, Näher Anatol-Fiete, Frey Dietmar, Madai Vince Istvan

Primary Institution: Charité - Universitätsmedizin Berlin

Hypothesis

How frequently is external validation performed for machine learning-based risk scores in ICU settings, and how does their performance change in external data?

Conclusion

External validation of machine learning-based scoring systems in the ICU is increasing but remains uncommon, with performance generally lower in external data.

Supporting Evidence

  • 14.7% of studies were externally validated, increasing to 23.9% by 2023.
  • On average, AUROC was reduced by -0.037 in external data.
  • 49.5% of validated studies showed a performance reduction of more than 0.05.

Takeaway

This study looked at how often hospitals check if their AI tools for predicting patient problems work well in different places, and found that they often don't work as well outside the original hospital.

Methodology

Systematic review and meta-analysis of studies using machine learning to predict deterioration in ICU patients, assessing external validation and performance changes.

Potential Biases

Potential overfitting due to reliance on specific datasets and differences in patient populations.

Limitations

The study primarily relied on a few datasets for external validation, which may not represent the broader ICU population.

Participant Demographics

Studies included adult ICU patients from various hospitals, primarily using US data.

Statistical Information

P-Value

p<0.001

Confidence Interval

95% CI -0.052 to -0.027

Statistical Significance

p<0.001

Digital Object Identifier (DOI)

10.1186/s12911-024-02830-7

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication