ClinicNet: machine learning for personalized clinical order set recommendations.

Area of Science:

• * Clinical Informatics
• * Machine Learning in Healthcare
• * Electronic Health Records (EHR)

Background:

• * Existing clinical decision support tools often rely on static order sets, which may not reflect individual clinician practices or evolving patient needs.
• * Accurate prediction of clinical orders and order set usage is crucial for streamlining workflows and improving patient care.

Purpose of the Study:

• * To evaluate the efficacy of neural networks trained on EHR data in predicting individual clinical orders and institutional order set usage.
• * To compare the predictive accuracy of the proposed neural network model (ClinicNet) against existing decision support tools and baseline methods.

Main Methods:

• * A feed-forward neural network (ClinicNet) and logistic regression were trained on 57,624 patients' EHR data from 2008-2014.
• * The models were applied to predict both individual clinical items and the usage of institutional order set templates.

Main Results:

• * ClinicNet demonstrated superior performance in predicting individual clinical orders (precision=0.32, recall=0.47) compared to institutional order sets (precision=0.15, recall=0.46).
• * ClinicNet achieved higher average precision (0.31) in predicting clinician usage of order sets than frequency baselines (0.20) and logistic regression (0.12).

Conclusions:

• * Machine learning, specifically ClinicNet, can predict clinical decision-making patterns with greater accuracy than static order sets, streamlining workflows.
• * While effective, this approach should complement, not replace, manually authored content for purposeful care pathway design.
• * Combining top-down (order sets) and bottom-up (EHR data) approaches via ML offers advanced clinical decision support.