Table 2.

Main association results. Effect corresponds to OR (odds ratio) for logistic regression on AD status not adjusted by age (LogReg), exp(β) for linear regression on AD-age score (LinReg), and HR (hazard ratio) for multivariate Cox regression on age-at-onset (CoxReg). Correlation between these measures is high for suggestive associations as shown on Figure S11. P p value, m model subversion. Subversion codes are (1) adjusted for sex and 10 first principal components of population structure and (2) additionally adjusted for APOE ε2 and APOE ε4 alleles. Two types of weighted AD-age score were used with (A) corresponding to a linear effect of age between 60 and 100 and (B) accounting for the changes in AD prevalence slope in this age range [8]

SNP (hg19) / gene	Discovery									Replication
	LogReg			LinReg			CoxReg			LogReg			LinReg			CoxReg
	OR	P	m	exp(β)	P	m	HR	P	m	OR	P	m	exp(β)	P	m	HR	P	m
1:200959302:G:A / KIF21B	0.87	2.10−4	2	0.90	5.10−6	B2	0.89	5.10−6	2	0.96	0.13	1	0.96	0.01	B1	0.96	0.02	2
1:216270469:G:A / USH2A	9.12	4.10−3	2	6.76	4.10−8	B1	4.07	8.10−9	2	1.58	0.14	1	1.70	0.04	A1	1.33	0.12	1
2:26332640:T:C / RAB10	17.4	0.06	1	10.46	2.10−15	B1	4.92	5.10−7	1	4.50	0.05	1	5.03	2.10−3	B1	2.69	6.10−4	1
6:41129252:C:T / TREM2	4.83	3.10−10	1	3.22	2.10−27	A1	2.58	1.10−23	1	2.32	2.10−9	1	2.69	1.10−14	A1	1.95	2.10−18	2
7:99971313:T:C / PILRA	0.88	2.10−5	1	0.87	6.10−8	A2	0.90	9.10−7	2	0.92	6.10−5	1	0.90	2.10− 7	B1	0.93	5.10−7	1
11:59945745:T:C / MS4A6A	0.88	9.10−6	1	0.91	1.10−6	B1	0.92	1.10−5	1	0.89	1.10−8	1	0.89	3.10−12	A1	0.93	2.10−8	1
14:93022240:G:T / RIN3	16.3	7.10−3	2	6.54	6.10−11	A2	3.46	4.10−7	2	1.95	0.04	2	1.69	0.02	A2	1.59	0.01	2
16:29998200:A:G / TAOK2	1.12	6.10−5	1	1.08	3.10−7	A1	1.09	6.10−6	2	1.04	0.07	2	1.05	1.10−3	B1	1.05	4.10−4	2
17:44828931:G:A / NSF/MAPT/KANSL1	0.85	5.10−6	2	0.89	5.10−8	B2	0.89	7.10−7	2	0.97	0.20	2	0.97	0.06	B2	0.98	0.18	2
19:1047507:AGGAGCAG:A / ABCA7	3.36	1.10−4	2	2.18	3.10−7	A1	1.94	1.10−6	1	1.36	0.12	1	1.33	0.07	B2	1.22	0.13	2