Table 2.

Section 2: Factors associated with risk for possible or probable AD considering all data (left columns) and considering only data collected at research clinic study visits (right columns), along with bootstrap results (far right column)^*

	All data		Clinic-only data		Bootstrap results
	Hazard ratio	p value	Hazard ratio	p value	p value
Sex
Male	1	Reference	1	Reference
Female	1.07	0.52	1.16	0.46	0.50
Age	1.12	<0.001	1.14	<0.001	0.040
Education	Omnibus: p<0.001		Omnibus: p=0.32
≤12 years	1	Reference	1	Reference
13–16 years	0.79	0.015	0.84	0.38	0.61
17+ years	0.57	<0.001	0.64	0.15	0.44
Race
Other	1	Reference	1	Reference
White	0.83	0.26	1.00	0.99	0.35
Medical conditions
Diabetes	1.16	0.40	1.76	0.09	0.039
Hypertension	0.85	0.13	0.90	0.63	0.62
Heart	1.03	0.81	1.36	0.24	0.09
Cerebrovascular disease	1.29	0.13	2.42	0.002	0.006
APOE genotype
0 ε4 alleles	1	Reference	1	Reference
1 or more ε4 alleles	1.66	<0.001	2.28	<0.001	0.008
Comorbidity	Omnibus: p=0.39		Omnibus: p=0.55
RxRisk = 1	1	Reference	1	Reference
RxRisk = 2	0.92	0.54	0.99	0.96	0.66
RxRisk = 3	0.96	0.78	0.69	0.20	0.034
RxRisk = 4	1.18	0.26	0.81	0.54	0.044

^*Bootstrap results give the probability that hazard ratios this different would be observed by chance alone. Each bootstrap sample is generated by randomly drawing individuals from the “all data” dataset with replacement until the same sample size as the “clinic only” dataset is reached. We then ran the same Cox model with that sample and kept track of its hazard ratio. We repeated this procedure 10,000 times, and ranked the hazard ratios. We compared the observed hazard ratio for the “clinic only” dataset to the bootstrapped hazard ratios to obtain the bootstrapped p values.