Survival

Kaplan–Meier analysis was used to estimate the survival functions, and log-rank test to determine the difference of survival curve between groups. A P-value of <0.05 was considered statistically significant. All these survival analyses were conducted in software R (Version 3.3.3, The R Foundation, Vienna, Austria) with survival package (Version 2.41-3; survival/index.html">https://cran.r-project.org/web/packages/survival/index.html).

To determine PD subtype differences in clinical milestones (DBS, dementia, mortality), multivariate Cox proportional hazards regression models using the longitudinal follow‐up data were conducted, with censoring based on last date of contact. SURVIVAL and events were calculated as such for each milestone: for DBS, events were defined as date of DBS and SURVIVAL time was calculated as time since baseline visit to most recent contact; for dementia, events were defined as the date when a participant received a CDR score ≥1, and SURVIVAL time was calculated as time since baseline visit to most recent CDR; and for mortality, events were defined as date of death, and SURVIVAL time was calculated as time since baseline to most recent contact. As follow‐up visits may only occur every three years, date of last contact (for the DBS and mortality analyses) was based on most recent contact from either a study visit, study contact, or clinical visit.
LCA was conducted using MPlus (Muthen & Muthen, Los Angeles CA). The longitudinal SURVIVAL analyses (Cox proportional hazards regression) were conducted in R Version 3.5.2, SURVIVAL and SURVMINER packages (R Foundation, Vienna Austria). Additional analyses were conducted with PASW Version 25 (IBM, Chicago, IL). All tests were 2‐tailed and P < 0.05 defined statistical significance.

To compare baseline characteristics, we used the chi-squared test and the Kruskal–Wallis test. Continuous and categorical variables were presented as median values with interquartile ranges and numbers with percentages, respectively. The all-cause mortality risks were estimated by multivariate Cox regression analysis adjusted for variables including age, sex, body mass index (BMI), smoking status, drinking status, serum glucose, systolic blood pressure and total cholesterol. To allow the modeling of smooth non-linear effects, we used penalized splines implemented in the spline function of the R package ‘survival’ (R Foundation for Statistical Computing, Vienna, Austria). We performed the statistical analysis using SAS version 9.4 (SAS Institute, Cary, NC, USA) with a two-sided P-value <0.05 as the criterion for statistical significance.