Validation of a Parkinson Disease Predictive Model in a Population-Based Study

Figure 2

Hazard ratio of PD based on the predicted probability of PD at baseline (U.S. Medicare). Age-only model: predicted by age (2 linear splines with a knot at age 85). Basic model: predicted by age (2 linear splines with a knot at age 85), sex, race/ethnicity (7 categories), probability of ever/never smoking (continuous), constipation (ICD-9 564, 564.0, 564.00, 564.01, 564.02, and 564.09), REM sleep behavior disorder (ICD-9 327.42), and anosmia/hyposmia (included in ICD-9 781.1 as smell and taste disturbances), with or without the total number of unique ICD-9 diagnosis codes (continuous) as a measure of overall use of medical care. Full model: predicted by age, sex, race/ethnicity, smoking, total number of unique ICD-9 diagnosis codes, all as modeled above, and 536 diagnosis or procedure codes including codes for constipation and REM sleep behavior as above. Abbreviations: CI = confidence interval; ICD-9 = International Classification of Diseases, Version 9; HR = hazard ratio; PD = Parkinson disease. The predicted probability of PD as of baseline for the full model and both basic models was positively associated with the hazard ratio (HR) for PD , with the highest HR occurring for the top decile of predictive probabilities from the full model (HR = 13.5, 95% CI 10.6–17.3). The predicted probability of PD for the full model departed from the three simpler models in the top 2-3 deciles of predicted probabilities when these probabilities were assessed in terms of the magnitude of their association with time to PD diagnosis.