Predictive Modeling to Uncover Parkinson’s Disease Characteristics That Delay Diagnosis

Abstract

Background: People with Parkinson's disease (PwPD) present with a variety of motor and non-motor symptoms, and a more biological definition of PD is poised to expand the diagnostic spectrum beyond the stereotypical "elderly male with tremor". This heterogeneity can potentially pose a challenge for an accurate and early diagnosis. Objectives: To determine whether demographic or clinical characteristics systematically affect the time till diagnosis, by modeling large-scale longitudinal data. Methods: Using longitudinal data from three large PD cohorts and a latent time joint mixed-effects model (LTJMM), we aligned the disease courses of individual PwPD and estimated whether individual PD diagnosis was early or late compared to the average time of PD diagnosis in each cohort. Initial clinical manifestations at the typical time of PD diagnosis were estimated using mixed-effects models. Results: We included 1,124 PwPD in our analysis. Several clinical and demographic factors were associated with a later-than-average diagnosis of PD: higher age, tremor dominance, rapid progression, anxiety, autonomic symptoms, depression, fatigue, pain, sleep problems, and in general more non-motor symptoms. In contrast, postural and gait disturbance was associated with an earlier-than-average PD diagnosis. Sex, family history of PD and predominantly affected side did not impact the time of PD diagnosis. Conclusions: Using statistical modeling, we were able to study initial clinical characteristics of PwPD even in the absence of directly observable clinical data at the time when PD is diagnosed typically. Our findings are consistent with a biological definition of PD that includes patients who present initially with non-motor symptoms.