What is the clinical evidence from this study?

Study design: Cohort. Population: Fluid overload in critically ill adults (n=391). Intervention: Machine learning models (XGBoost) vs. Traditional logistic regression. Primary outcome: Prediction of fluid overload (positive fluid balance ≥ 10% of admission body weight) at 48-72 hours after ICU admission (AUROC).

June 19, 2023Open Access

Machine learning vs. traditional regression analysis for fluid overload prediction in the ICU

ASAndrea SikoraUniversity of Colorado Anschutz Medical Campus TZTianyi ZhangJiangsu University DMDavid J. MurphyChildren's Hospital of Eastern Ontario

Key Result

Machine learning models, specifically XGBoost (AUROC 0.78), performed similarly to traditional logistic regression (AUROC 0.70) in predicting fluid overload in critically ill adults.

Study Design

Type

Cohort (n=391)

Multicenter

Structured PICO

Do machine learning models improve the prediction of fluid overload compared to traditional regression techniques in adult ICU patients?

Population

391 adult patients admitted to an intensive care unit for at least 72 hours with available fluid balance data.

Exposure

Machine learning models (Random Forest, Support Vector Machine, XGBoost) for predicting fluid overload

Comparator

Traditional logistic regression models (stepwise backward elimination)

Outcome

Presence of fluid overload (defined as a positive fluid balance ≥10% of admission body weight) at 48-72 hours (day 3) after ICU admission

Machine learning and traditional logistic regression models perform similarly in predicting fluid overload in the ICU, identifying baseline severity of illness and medication regimen complexity as key predictors.

Main Result

Absolute Event Rate: 0.78% vs 0.7%

Limitations

Patient sample may have been too small to demonstrate superiority of the machine learning approaches compared to traditional regression.
No validation in a separate, external dataset was undertaken.
Bias may exist due to which patients had fluid balance data available.
Other predictors for fluid overload not included in the models may exist.
Relying on prediction data derived in the first 24 hours did not fully capture the dynamic nature of critical illness over the entire three-day ICU period.

Abstract

ABSTRACT Background Fluid overload, while common in the ICU and associated with serious sequelae, is hard to predict and may be influenced by ICU medication use. Machine learning (ML) approaches may offer advantages over traditional regression techniques to predict it. We compared the ability of traditional regression techniques and different ML-based modeling approaches to identify clinically meaningful fluid overload predictors. Methods This was a retrospective, observational cohort study of adult patients admitted to an ICU ≥ 72 hours between 10/1/2015 and 10/31/2020 with available fluid balance data. Models to predict fluid overload (a positive fluid balance ≥10% of the admission body weight) in the 48-72 hours after ICU admission were created. Potential patient and medication fluid overload predictor variables (n=28) were collected at either baseline or 24 hours after ICU admission. The optimal traditional logistic regression model was created using backward selection. Supervised, classification-based ML models were trained and optimized, including a meta-modeling approach. Area under the receiver operating characteristic (AUROC), positive predictive value (PPV), and negative predictive value (NPV) were compared between the traditional and ML fluid prediction models. Results A total of 49 of the 391 (12.5%) patients developed fluid overload. Among the ML models, the XGBoost model had the highest performance (AUROC 0.78, PPV 0.27, NPV 0.94) for fluid overload prediction. The XGBoost model performed similarly to the final traditional logistic regression model (AUROC 0.70; PPV 0.20, NPV 0.94). Feature importance analysis revealed severity of illness scores and medication-related data were the most important predictors of fluid overload. Conclusion In the context of our study, ML and traditional models appear to perform similarly to predict fluid overload in the ICU. Baseline severity of illness and ICU medication regimen complexity are important predictors of fluid overload.

Ask AI

Helpful

Bookmark

View Full Paper