Dietary energy density and adiposity: Employing bias adjustments in a meta-analysis of prospective studies

BMC Public Health

Table 2 Correlation coefficients of studies calculated from p-values according to the principal results extracted

Study (source for extracted results)	Extracted results		Re-calculated results
	N	Model & results	p-value	z (SE)	r
Butte et al [12](Table 2) ^a	548	GEE: β = 0.23; SE = 0.35; p = 0.5		0.029 (0.043)	0.029
Deierlein et al [13](Table 2) ^b	1231	Quartile mean differences from Q₁: Q₂ 0.49; CI - 0.4,1.37; Q₃ 1.13; CI 0.24,2.01; Q₄ 1.08; CI 0.20,1.97	0.046	0.057 (0.029)	0.057
Iqbal et al [14](Table 4) ^c	1762	LM;β- 13.49; SE = 36.46; p = 0.711		-0.009 (0.024)	-0.009
Johnson et al [15](Table 3, Model 4)	584	GLM; OR = 1.36; CI 1.09,1.69	0.006	0.114 (0.042)	0.114
McCaffrey et al [16](Table 6, Model 2)	48	GLM; OR = 2.16; CI 1.099, 4.25; p = 0.026		0.332 (0.149)	0.320
Savage et al [17](Text) ^d	183	Tertile means: T₁ 2.5, SD = 6.8; T₂ 4.8, SD = 9.2; T₃ 6.4, SD = 6.5	0.067	0.137 (0.075)	0.136

β = regression coefficient; CI = confidence interval; GEE = generalized estimated equations; GLM = generalized linear model; LM = linear regression model; n = sample size; OR = odds ratio; p = p-value; r = correlation; SD = standard deviation; SE = standard error; T = tertile; Q = quintile; z = Fisher transformed correlation.
^aThe p-value was derived from the GEE regression coefficient. The GEE model assumed an exchangeable correlation structure within families. An intra-family correlation coefficient of c = 0.26 is reported. The design effect is 1.46 = 1+(m-1)*c = 1+(879/319-1)*0.26, where m is the average number of people within each family, and thus the effective sample size is 548 (= 796/1.46). It is assumed that the average cluster size of 879/319 is correct for the 798 individuals included in the analysis. ^bThe regression slope and p-value were obtained from a weighted regression using analytical weights in Stata. A complete balance across quartiles has been assumed and the mean differences in DED were calculated from the published table. The p-value and sample size were used to calculate the implied correlation.^c The paper reports results for males and females separately; groups have been merged. ^dP-value was re-estimated using weighted regression.

ISSN: 1471-2458