Proof of Concept of Natural Language Processing in Predicting Patient-Reported Outcomes in Spinal Cord Stimulation

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Research

Background:

• Patient-reported outcome measures (PROMs) are crucial for evaluating treatment efficacy.
• Spinal cord stimulation (SCS) is used for chronic pain management.
• Predicting treatment response in SCS patients is essential for personalized care.

Purpose of the Study:

• To analyze the efficacy of Natural Language Processing (NLP) models in predicting improvements in patient-reported outcome measures (PROMs) following spinal cord stimulation.
• To compare the performance of TF-IDF, BERT-base, and Bio-ClinicalBERT models in predicting PROM changes.

Main Methods:

• Examined PROMs and clinical notes from 20 SCS patients preoperatively and one year postoperatively.
• Utilized Numeric Rating Scale (NRS), Patient Global Impression of Change (PGIC), Oswestry Disability Index (ODI), Beck Depression Inventory (BDI), McGill Pain Questionnaire (MPQ), and Pain Catastrophizing Scale (PCS).
• Applied Term-frequency-inverse document frequency (TF-IDF), BERT-base, and Bio-ClinicalBERT text embedding models for analysis.

Main Results:

• TF-IDF demonstrated robust prediction for MPQ, PCS, NRS, ODI, and BDI, with strong correlations (ρ up to -0.97).
• TF-IDF achieved perfect classification (AUC 1.0) for binary responders based on Minimal Clinically Important Difference (MCID) for PGIC, MPQ, and PCS.
• BERT-base and Bio-ClinicalBERT showed variable performance, with higher accuracies for most PROMs excluding PGIC and ODI.

Conclusions:

• NLP models, particularly TF-IDF, can approximate the order of improvement in PROMs for SCS patients.
• Accurate prediction of individual patient responders remains an ongoing challenge.
• Advancements in machine learning algorithms are expected to enhance the clinical utility of these predictive models.