Prediction of delirium using data mining: A systematic review

Area of Science:

• Medical Informatics
• Data Science
• Clinical Prediction Models

Background:

• Delirium is a serious condition with significant morbidity, mortality, and economic impact.
• The increasing volume and velocity of health data, termed "big data", offers potential for predictive modeling.
• Predictive models for delirium incidence in inpatient settings are crucial for timely intervention.

Purpose of the Study:

• To systematically review and meta-analyze the utility of big data analytics in predicting inpatient delirium incidence.
• To evaluate the performance of various machine learning algorithms for delirium prediction.
• To identify key variables and statistical methods employed in big data-driven delirium prediction.

Main Methods:

• Systematic review and meta-analysis of randomized and observational studies.
• Searched multiple databases (Medline, Embase, etc.) using MeSH terms related to big data and delirium.
• Included six retrospective observational studies (n=178,091) with quality assessment using the CHARMs checklist.

Main Results:

• Random forest was the most studied and best-performing method, with Area Under the Receiver Operating Curve (AUROC) from 0.78 to 0.91.
• Sensitivity ranged from 0.59 to 0.81, and specificity ranged from 0.73 to 0.92.
• Support vector machines and artificial neural networks were also explored, but with generally lower performance than random forest.

Conclusions:

• Machine learning techniques, particularly random forest, demonstrate significant potential for predicting delirium incidence in hospitalized patients.
• Big data analytics can be effectively utilized to develop robust predictive models for delirium.
• Further research into prospective studies is warranted to validate these findings and implement clinical decision support tools.