Table 3.

A) Claims, EMR Prediction accuracy at first gait or tremor, first tremor, and first gait. B) Analysis of advance prediction time at various FPR thresholds for first gait or tremor Deep Neural Network Model

A)
	Claims					EMR
	Demographics			Validation AUROC (95% Confidence)		Demographics			Validation AUROC (95% Confidence)
	Cohort Size	Percent Progressing to PD	Average days to PD (STD)	Deep Neural Network	Logistic Regression	Cohort Size	Percent Progressing to PD	Average days to PD (STD)	Logistic Regression
First Gait or Tremor	8475	2.43	469 (493)	0.874 (0.869–0.879)	0.803 (0.791–0.816)	1349	3.08	548 (517)	0.804 (0.792–0.816)
First Gait Only	3925	1.37	575 (521)	0.769 (0.759–0.780)	0.791 (0.772–0.809)	694	2.23	606 (530)	0.714 (0.679–0.750)
First Tremor Only	4550	6.69	377 (447)	0.698 (0.679–0.718)	0.697 (0.674–0.719)	681	5.24	479 (490)	0.757 (0.730–0.784)
B)
False Positive Rate Threshold	Observed False Negative Rate	Mean Days Accelerated Diagnosis (STD)
0.90	0.00	377 (399)
0.80	0.01	375 (397)
0.70	0.03	369 (395)
0.60	0.04	368 (396)
0.50	0.07	360 (390)
0.40	0.12	348 (384)
0.30	0.18	339 (376)
0.20	0.26	334 (372)
0.10	0.33	322 (371)
0.01	0.44	303 (369)