4.3 Review

Dynamic Prediction of Patient Outcomes in the Intensive Care Unit: A Scoping Review of the State-of-the-Art

Journal

JOURNAL OF INTENSIVE CARE MEDICINE
Volume 38, Issue 7, Pages 575-591

Publisher

SAGE PUBLICATIONS INC
DOI: 10.1177/08850666231166349

Keywords

predictive modeling; dynamic prediction; patient outcomes; critical care; intensive care unit

Ask authors/readers for more resources

This scoping review examines the current state-of-the-art dynamic prediction models for patient outcomes in the ICU. The review found that there is significant potential in developing dynamic prediction models to aid decision-making in real-time. However, most of the models focus on predicting mortality, and there is a need for further models to predict other serious complications.
Introduction Intensive care units (ICUs) are high-pressure, complex, technology-intensive medical environments where patient physiological data are generated continuously. Due to the complexity of interpreting multiple signals at speed, there are substantial opportunities and significant potential benefits in providing ICU staff with additional decision support and predictive modeling tools that can support and aid decision-making in real-time. This scoping review aims to synthesize the state-of-the-art dynamic prediction models of patient outcomes developed for use in the ICU. We define dynamic models as those where predictions are regularly computed and updated over time in response to updated physiological signals. Methods Studies describing the development of predictive models for use in the ICU were searched, using PubMed. The studies were screened as per Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines, and the data regarding predicted outcomes, methods used to develop the predictive models, preprocessing the data and dealing with missing values, and performance measures were extracted and analyzed. Results A total of n = 36 studies were included for synthesis in our review. The included studies focused on the prediction of various outcomes, including mortality (n = 17), sepsis-related complications (n = 12), cardiovascular complications (n = 5), and other complications (respiratory, renal complications, and bleeding, n = 5). The most common classification methods include logistic regression, random forest, support vector machine, and neural networks. Conclusion The included studies demonstrated that there is a strong interest in developing dynamic prediction models for various ICU patient outcomes. Most models reported focus on mortality. As such, the development of further models focusing on a range of other serious and well-defined complications-such as acute kidney injury-would be beneficial. Furthermore, studies should improve the reporting of key aspects of model development challenges.

Authors

I am an author on this paper
Click your name to claim this paper and add it to your profile.

Reviews

Primary Rating

4.3
Not enough ratings

Secondary Ratings

Novelty
-
Significance
-
Scientific rigor
-
Rate this paper

Recommended

No Data Available
No Data Available