Table 2.

Multivariate associations between the 22 serum ions/metabolites selected in the metabolomics signature and the odds of subsequent cognitive decline, in the 3C Bordeaux cohort (n = 418)

Selection rank	Ion	Metabolite name	Identification level	OR
1	M497.2383_T10.11	Atractyligenin glucuronide	2	0.72
2	M144.1018_T0.96	Proline betaine	1	1.56
3	M195.0876_T8.09	Caffeine	1	1.75
4	M251.1278_T13.25	CMPFP	2	0.74
5	M129.0658_T0.90	–	4	1.49
6	M160.1331_T0.97	–	4	1.83
7	M271.2056_T12.99	–	4	0.69
8	M197.1284_T7.95	Cyclo(prolyl‐valyl)	1	0.90
9	M372.3108_T13.12	Myristoylcarnitine	1	1.01
10	M626.3536_T11.69	GDCA	1	1.12
11	M383.1161_T0.87	Glucose	1	1.24
12	M114.0660_T0.88	Creatinine	1	1.41
13	M189.1597_T0.79	N‐trimethyl‐Lysine	1	1.09
14	M159.0276_T1.15	–	4	1.40
15	M211.1441_T8.78	Cyclo(leucyl‐prolyl)	1	0.68
16	M287.6256_T14.09	LysoPC(18:3)	2	0.76
17	M363.2166_T10.80	Cortisol	1	0.74
18	M330.2639_T11.58	Undecanoylcarnitine/4,8 dimethylnonanoylcarnitine	3	0.83
19	M245.0768_T1.15	–	4	1.29
20	M256.6796_T14.27	–	4	0.85
21	M175.1189_T0.81	L‐Arginine	1	1.29
22	M344.2795_T12.23	Lauroylcarnitine	1	1.32

Ions/metabolites are ordered by decreasing frequency of selection across bootstraps and referred to by mass‐to‐charge‐ratio (M) and retention time (T). Odds Ratios (ORs) for cognitive decline were estimated using a conditional logistic regression conditioned on matching variables (age, gender and educational level) and adjusted for BMI and number of medications regularly consumed. ORs are for 1SD‐increment of metabolite intensity. Confidence intervals are not valid in post‐selection inference and hence were not estimated. Level of identification is assigned as: 1, identification validated with standards; 2, putative identification by comparison with databases or literature; 3, putative identification of a chemical class; 4, unknown. CMPFP, 3‐carboxy‐4‐methyl‐5‐pentyl‐2‐furanpropionic acid; GDCA, glycodeoxycholic acid‐3‐glucuronide; LysoPC(18:3), 1‐linolenoyl‐sn‐glycero‐3‐phosphocholine.