Cognitive Symptoms in Cross-Sectional Parkinson Disease Cohort Evaluated by Human-in-the-Loop Machine Learning and Natural Language Processing.

Background and objectives: Cognitive impairment is experienced by up to 80% of people with Parkinson disease (PD). Little is known regarding the subjective experience and frequency of bothersome cognitive problems across the range of disease duration as expressed directly in patients' own words. We describe the types and frequency of bothersome cognitive symptoms reported verbatim by patients with PD.

Methods: Through the online Fox Insight study and the Parkinson Disease Patient Report of Problems, we asked patients with PD to self-report by keyboard entry up to five most bothersome problems and how these problems affect their functioning. Human-in-the-loop curation, natural language processing, and machine learning were used to categorize responses into 8 cognitive symptoms: memory, concentration/attention, cognitive slowing, language/word finding, mental alertness/awareness, visuospatial abilities, executive abilities/working memory, and cognitive impairment not otherwise specified. Associations between cognitive symptoms and demographic and disease-related variables were examined in our cross-sectional cohort using multivariate logistic regression.

Results: Among 25,192 participants (55% men) of median age 67 years and 3 years since diagnosis (YSD), 8,001 (32%) reported a cognitive symptom at baseline. The 3 most frequently reported symptoms were memory (13%), language/word finding (12%), and concentration/attention (9%). Depression was significantly associated with bothersome cognitive problems in all domains except visuospatial abilities. Predictors of reporting any cognitive symptom in PD were depression (adjusted OR 1.5), increasing MDS-UPDRS Part II score (OR 1.4 per 10-point increment), higher education (OR 1.2 per year), and YSD 1, 2, 6-7, and 8-9 vs 0 YSD. Among individuals with at least one cognitive symptom, posterior cortical-related cognitive symptoms (i.e., visuospatial, memory, and language) were reported by 17% (n = 4325), frontostriatal-related symptoms (i.e., executive abilities, concentration/attention) by 7% (n = 1,827), and both by 14.2% (n = 1,020). Odds of reporting posterior cortical symptoms vs frontostriatal symptoms increased with age and MDS-UPDRS part II score, but not depression.

Discussion: Nearly one-third of participants with PD, even early in the disease course, report cognitive symptoms as among their most bothersome problems. Online verbatim reporting analyzed by human-in-the-loop curation, natural language processing, and machine learning is feasible on a large scale and allows a detailed examination of the nature and distribution of cognitive symptoms in PD.