Natural Language Processing of Referral Letters for Machine Learning-Based Triaging of Patients With Low Back Pain to...

Area of Science:

• Clinical Informatics
• Artificial Intelligence in Healthcare
• Musculoskeletal Disorders

Background:

• Current decision support systems (DSSs) for low back pain (LBP) lack sufficient accuracy for clinical use.
• Existing DSSs primarily rely on patient-reported outcomes.
• Electronic health records (EHRs) offer additional qualitative data, such as referral reasons and patient goals, for DSS improvement.

Purpose of the Study:

• To evaluate the added value of qualitative data from EHRs and referral letters to DSS accuracy for LBP.
• To enhance the triage accuracy of DSSs by incorporating referral letter information.

Main Methods:

• A retrospective study analyzed data from 1608 Dutch LBP patients.
• Natural Language Processing (NLP) extracted referral reasons and patient goals from referral letters.
• Machine learning models (SVM, k-NN, MLP) were trained with and without referral letter data to predict treatment paths.

Main Results:

• Incorporating referral reasons from letters improved triage accuracy by up to 19.5% for specific interventions (anesthesiology, rehabilitation).
• Confusion matrices validated the enhanced predictive accuracy of the models.
• Specific referral reasons significantly impacted the model's ability to suggest optimal LBP treatments.

Conclusions:

• Enriching DSSs with NLP-extracted referral letter content enhances model accuracy for LBP treatment recommendations.
• Despite improvements, overall DSS model accuracies remain insufficient for direct clinical application.
• Further research is needed to optimize DSSs for effective clinical implementation in LBP management.