Natural Language Processing to Predict Deep Brain Stimulation Outcomes in Parkinson's Disease

Area of Science:

• Neurology
• Computational Linguistics
• Biomedical Informatics

Background:

• The Unified Parkinson's Disease Rating Scale (UPDRS Part III) is crucial for assessing deep brain stimulation (DBS) efficacy in Parkinson disease (PD).
• Post-operative UPDRS Part III scores are frequently unavailable in routine clinical practice, hindering objective treatment response evaluation.
• Natural language processing (NLP) presents a potential method to extract valuable motor outcome data from unstructured clinical notes.

Purpose of the Study:

• To investigate the feasibility of using NLP to predict motor outcomes (UPDRS Part III scores) from clinical documentation in PD patients receiving DBS.
• To determine if NLP can predict treatment response, defined as improvement in UPDRS Part III scores, using pre-operative notes.

Main Methods:

• Utilized neurology notes and UPDRS Part III assessments from 25 PD patients pre- and post-operatively.
• Evaluated four NLP approaches: TF-IDF with linear models, BERT-base, ClinicalBERT, and Bio-ClinicalBERT embeddings.
• Employed leave-one-out cross-validation and assessed performance using mean absolute error, R², Spearman correlation, and AUC.

Main Results:

• All NLP methods successfully differentiated pre-operative from post-operative clinical notes.
• The TF-IDF model achieved moderate accuracy in predicting total UPDRS Part III scores and percentage change.
• Significant correlations were found between predicted and observed values for tremor, rigidity, and limb agility subscores; prediction error varied by PD phenotype.

Conclusions:

• NLP techniques show feasibility in extracting meaningful signals from routine DBS clinical documentation for motor outcome assessment.
• NLP-derived insights may serve as a valuable adjunct to formal UPDRS-III assessments, offering complementary data when formal scores are absent.
• These findings are exploratory and suggest NLP's potential role in enhancing the evaluation of DBS therapy for Parkinson disease.