Improved ability of biological and previous caries multimarkers to predict caries disease as revealed by multivariate PLS modelling

Table 1 Ability of biological and previous caries variables to predict caries

		Caries Y matrices^a
		Baseline		Increment		Incidence		Progression
X matrices^a	[n variables]^b	R²	Q²	R²	Q²	R²	Q²	R²	Q²
Primary models (all VIPs)
Plaque, diet, saliva	[88]	0.597	0.260^M1	0.619	0.084	0.534	0.056	0.632	0.034
Plaque, diet, saliva + previous caries	[88 + 18]			0.567	0.211^M2	0.406	0.164^M3	0.365	0.162^M4
Secondary models (VIPs > 1.0)
Plaque markers	[6,4,6,4]	0.353	0.289	0.218	0.142	0.181	0.097	0.286	0.226
Diet markers	[7,9,9,0]	0.155	0.113	0.303	0.153	0.170	0.083	nt^c	nt^c
Saliva markers	[5,5,7,2]	0.129	0.034	ns^d	ns^d	ns^d	ns^d	nt^c	nt
Multimarkers (n = 6-39)
Plaque, diet, saliva	[18,18,22,6]	0.528	0.404	0.377	0.224	0.351	0.204	0.311	0.212^e
Previous caries	[ ,17,17,16]			0.382	0.323	0.287	0.247	0.494	0.421
Plaque, diet, saliva + previous caries	[ ,35,39,22]			0.662	0.354	0.418	0.318	0.511	0.386
Five marker panel^f	[5]	0.430	0.275
Single lactobacilli marker^g	[1]	0.340	0.270

^a The primary models harbor all variables. The secondary models harbor influential (VIP > 1.0) markers from the corresponding primary models M1 to M4. All models had ≤ 3 significant components except for the model with a single saliva lactobacilli marker. The R² (explanatory) and Q² (predictive) values gives the performance of each PLS model (see Methods).
^b Numbers of variables or markers in each separate PLS model.
^c nt = not tested
^d ns = no significant component.
^e No available diet variable with VIP > 1.0.
^f Saliva lactobacilli, plaque mutans streptococci, diet retinol and zinc, saliva calcium with high VIP-values.
^g Saliva lactobacilli.

ISSN: 1472-6831