Estimation of Caffeine Regimens: A Machine Learning Approach for Enhanced Clinical Decision Making at a Neonatal...

Area of Science:

• Neonatal Medicine
• Computational Biology
• Pharmacology

Background:

• Optimal drug dosage is critical in neonatal intensive care units (NICUs), especially for preterm neonates with apnea of prematurity.
• Inaccurate dosing of caffeine, a common treatment, can have life-or-death consequences.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the adequacy and therapeutic concentration of caffeine in preterm neonates.
• To improve clinical decision-making for caffeine administration in the NICU.

Main Methods:

• Utilized optimized Support Vector Machine (SVM), decision trees (Bagging, Boosting, Random Forest), optimized Multi Layer Perceptron (MLP), and Deep Learning models.
• Evaluated models using 100 clinical caffeine cases from a NICU.
• Investigated the impact of the Score for Neonatal Acute Physiology I (SNAP I) as an input feature.

Main Results:

• A deep belief network (DBN) achieved an Area Under the Curve (AUC) of 0.91, outperforming other models in assessing caffeine effectiveness.
• An optimized MLP, using SNAP I, also demonstrated high accuracy in predicting therapeutic caffeine concentrations.
• SNAP I was identified as a critical input variable for enhancing prediction model performance.

Conclusions:

• Machine learning, particularly DBN and optimized MLP with SNAP I, offers a valuable approach for decision support systems in the NICUs.
• These models can optimize the administration of caffeine, ensuring efficacy and adequacy before administration to sensitive neonates.
• The findings suggest a more precise and effective method for managing apnea of prematurity in preterm infants.