Improving Pain Assessment Using Vital Signs and Pain Medication for Patients With Sickle Cell Disease: Retrospective...

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Computational Medicine

Background:

• Sickle cell disease (SCD) is a prevalent inherited blood disorder causing severe, unpredictable pain episodes.
• Accurate assessment of SCD pain intensity is challenging for clinicians.
• Pain episodes are a primary reason for emergency department visits in SCD patients.

Purpose of the Study:

• To develop deep feature representations of subjective pain trajectories using objective physiological signals.
• To leverage electronic health records for enhanced pain assessment in SCD.
• To build machine learning models for predicting pain intensity in SCD patients.

Main Methods:

• Utilized electronic health record data from 496 participants over 5 years.
• Extracted vital signs, self-reported pain scores (0-10), and medication data (type, status, dosage).
• Employed variational autoencoders for representation learning and random forest classifiers for pain prediction.

Main Results:

• Achieved average accuracies of 82.8% (2-point) to 47.4% (11-point) in predicting pain scores.
• Deep representational learning models outperformed models using raw data across all pain scales.
• Inclusion of medication data alongside vital signs improved model performance.

Conclusions:

• Medication information significantly enhances subjective pain prediction models compared to vital signs alone.
• Data-driven pain score estimation offers valuable supplementary information for clinicians.
• This approach shows promise for improving pain management in sickle cell disease.