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

Objectives The Unified Parkinson's Disease Rating Scale (UPDRS Part III), is often not obtained post-operatively in routine clinical practice, limiting objective assessment of treatment response following deep brain stimulation (DBS). Natural language processing (NLP) offers a potential approach to use narrative clinical documentation to predict motor outcomes and evaluate therapeutic responsiveness when formal post-operative scoring is unavailable. Methods Neurology notes and UPDRS Part III assessments were obtained from 25 patients with Parkinson disease (PD) pre- and post-operatively. DBS response was defined as percent improvement in UPDRS Part III from the pre-operative OFF-medication state to the post-operative DBS/medication ON assessment. We assessed whether UPDRS Part III scores could be predicted from clinic notes and whether improvement could be predicted from pre-operative notes. Four NLP approaches were evaluated: TF-IDF with linear models, BERT-base, ClinicalBERT, and Bio-ClinicalBERT embeddings. Models were assessed using leave-one-out cross-validation, with performance evaluated using mean absolute error, coefficient of determination (R²), Spearman correlation, and area under the ROC curve. Results All approaches differentiated pre-operative notes from post-operative notes. The TF-IDF model demonstrated moderate accuracy at predicting UPDRS Part III total scores and percentage change. Several individual motor subscores demonstrated significant correlations between predicted and observed values, including tremor, rigidity, and limb agility measures. Tremor-dominant and rigidity-bradykinesia phenotypes were associated with lower prediction error, whereas the PIGD phenotype showed increased error when predicting postoperative motor improvement. Conclusions These findings support the feasibility of using NLP to detect signal in routine DBS clinical documentation, but they should be interpreted as exploratory. NLP-derived outputs may complement, but should not replace, formal UPDRS-III assessment.