The original (parent) cohort had phenotypic data regarding blood pressure collected at 21 time points spaced 2 years apart over 40 years while the offspring cohort was studied over 20 years at five time points every 4 years apart except for 8 years between the first two time points. To generate populations with maximal power, the parent and offspring cohorts were combined for this analysis. For each time point at which there were offspring data, phenotypic data from the closest date in the parent data set were used. This resulted in combining the 1971, 1979, 1983, 1987, and 1991 offspring data with the 1972, 1978, 1982, 1986, and 1988 parent data. The traits of interest were a diagnosis of hypertension (HTN) and systolic blood pressure (SBP). HTN was defined as systolic blood pressure ≥ 140 mm Hg, diastolic blood pressure ≥ 90 mm Hg, or use of medical therapy for hypertension. For those who were receiving antihypertensive medications, the SBP was assumed to be 10 mm Hg greater than the measured SBP as data suggest that this is the average reduction seen with medical therapy [3, 4]. This "correction factor" for SBP has also been found to usefully recover the genetic information in subjects receiving antihypertensive treatment [5, 6].
We used the Whittemore and Halpern NPL (nonparametric linkage) all statistic to test for allele sharing among all hypertensive individuals in a pedigree [7]. The Kong and Cox linear model was used to calculate a nonparametric LOD score [8]. This process was repeated for each of the five time points of data collection.
To test for allele sharing using SBP as the quantitative trait of interest, a variance components model was constructed for each of the five time points. The total phenotypic variance (conditional on the mean model) was based on a conventional covariate structure appropriate to the extended families present in the Framingham cohort. The model specified was:
σ2Total = σ2A + σ2CS + σ2C + σ2E.
In this model, σ2A represents additive genetic effects, σ2CS the effects of common sibling environment, σ2C the effects of a common family environment, and σ2E the residual variance (which is assumed to arise from nonfamilial factors). The narrow sense heritability (h2N) was calculated as σ2A / σ2Total. Age, gender, alcohol use (g/day), smoking (cigarettes/day), height, weight, and fasting glucose were included in the model as possible fixed effects. Linkage to the locus of interest was tested by comparing the likelihood of a model where the variance due to the locus of interest was constrained to zero versus an unrestricted model.
The linkage analyses of HTN were undertaken using the program MERLIN [9]; variance components analyses of SBP were undertaken in SOLAR [10]. Whole-genome, multi-point linkage analyses at 1-cM intervals of HTN and SBP were performed using 394 polymorphic markers on the 22 autosomes for each time point. Potential evidence of linkage was defined as LOD > 1.0 at ≥ 1 time point. Although this threshold does not meet Lander and Kruglyak's criteria for genome-wide significance [11], LOD > 1 is commonly used as a measure of promising evidence for linkage in the context of a complex disease [12].
The coefficient of variation (CV) of the heritability over the five time points was calculated to assess the reproducibility of this measure of additive genetic effect. To assess the reproducibility of linkage findings, Spearman correlations of LOD scores between each pair of time points were calculated over the entire genome as well as over the subset of loci where evidence for potential linkage was observed. In addition, the CV was calculated at each locus over the five time points to describe the variability in the LOD scores for linkage with each of the two traits. Mean and SD of the CVs at those loci where evidence for potential linkage was found are reported. The evidence for linkage was dichotomized at a threshold of LOD = 1.0 and Cohen's kappa statistic [13] was calculated between each pair of time points to provide another estimate of the reproducibility of results.