Longitudinal familial analysis of blood pressure involving parametric (co)variance functions
© Soler and Blangero; licensee BioMed Central Ltd 2003
Published: 31 December 2003
For analyzing longitudinal familial data we adopted a log-linear form to incorporate heterogeneity in genetic variance components over the time, and additionally a serial correlation term in the genetic effects at different levels of ages. Due to the availability of multiple measures on the same individual, we permitted environmental correlations that may change across time.
Systolic blood pressure from family members from the first and second cohort was used in the current analysis. Measures of subjects receiving hypertension treatment were set as censored values and they were corrected. An initial check of the variance and covariance functions proposed for analyzing longitudinal familial data, using empirical semi-variogram plots, indicated that the observed trait dispersion pattern follows the assumptions adopted.
The corrections for censored phenotypes based on ordinary linear models may be an appropriate simple model to correct the data, ensuring that the original variability in the data was retained. In addition, empirical semi-variogram plots are useful for diagnosis of the (co)variance model adopted.
Longitudinal designs in family studies represent additional opportunities to model temporal variation in genetic and environmental factors influencing quantitative traits. For certain phenotypes, like anthropometric measures and blood pressure, more insight into its continuous physiological variation may be provided for adopting phenotype (co)variance components as a function of time, rather than scalar components. In this article the term (co)variance is been used to refer both to variance and covariance components.
The classical approach to the genetic analysis of longitudinal traits, under the variance component framework, considers an unstructured covariance matrix  for modeling the correlations on the sequence of measurements within an individual. Without making assumptions about the (co)variance, terms the model is not flexible for incomplete profiles, and it is not clear how to define the heritability measures to incorporate the longitudinal features. One approach is to use structured (co)variance patterns, assuming, for instance, that the different variance components change across time with ages, according to a parametric function, and allowing the autocorrelation process within the measurements of the individuals. In this regard, gains in precision are obtained by reducing the number of parameters involved in the analysis.
To absorb the dependence in the sequential measurements within an individual, environmental correlations were added in the model. Data from family members from the first and second cohort that were re-examined 21 and 5 times, respectively, during the longitudinal phase of the study, were used in the current analysis. The precision of the estimates obtained from these analyses was compared. For subjects receiving hypertension treatment, the recorded systolic blood pressure was considered as censored value, and to accommodate it in the analysis, we proceeded with corrections using a nonparametric algorithm to adjust the censored phenotypes, as considered by Levy et al. . Only systolic blood pressure measures taken when subjects were aged 18 years or more were supposed to be informative for the analysis.
Adjusted right-censored phenotypes
For subjects receiving hypertension treatment, the recorded systolic blood pressure was considered as a right-censored value, since one knows that it is less than what the untreated value would be. To accommodate the censoring process in the analysis we addressed corrections on the censored phenotypes through the nonparametric algorithm used by Levy et al. . Separate adjustments were conducted according to sex and age groups (<35, 35 to 44, 45 to 54, 55 to 64, 65 to 74, and 75 years). In this phase of the analysis, for adjustment of the censored measurements, ordinary linear regression models were adopted to investigate the relationship between age and systolic blood pressure, despite multiple observations being available in the same individuals. Because assumptions about phenotype (co)variance structure can affect the fitted values, it is useful to have simple methods for correction of the censored responses. The residuals obtained were ordered to generate a sample from a discrete reference distribution. Censored phenotypes were adjusted by conditional expectations of the untreated residuals given residual values of equal or greater magnitude. Once untreated observations constituted the vast majority of the data, this replacement process should not reduce the original variability of the data .
Longitudinal familial variance component model
To define the adopted model, let Y j (t) denote a measurement on the j th individual at the time t. The cross-sectional familial variance component approach  can be extended for longitudinal data. The polygenic model was defined as:
where μ is the overall mean, X jt is a covariates vector (describing, for instance, sex scores, linear and quadratic polynomial coefficients associated with levels of age, and possible interactions) at occasion t for the j th individual, β is a parameters vector associated with the fixed effects, g jt and e jt are uncorrelated random effects due to polygenic and environmental sources of variation, respectively. For relatives i and j, observed at t and s occasions, respectively, we assume:
where 2φ ij is the kinship coefficient between relatives i and j.
Following Aitkin , who considered a log-linear form for modeling the heterogeneity of the variance, we employed the polygenic variance function , where t denotes age. The parameter γ accommodates longitudinal changes in the trait heritability, and represents the interaction effect between the polygene and age. Since polygenic correlations between the expressions of the trait at different ages usually depends on their time range, we adopted a monotone decreasing function ρ(t,s)=exp(-λ|t-s|), as proposed by Diggle  and Huggins et al. . Stating that ρ(t,s) is different from 1 suggests that different polygenes influence the trait in different ages. Also, modeling the correlation structure in continuous time admits unbalanced sequences of measurements on the different individuals.
To estimate the parameters involved in the specification of the (co)variance components under multivariate normal distribution and involving longitudinal and familial dependence patterns, we used a version of the EM (expectation maximization) algorithm, as described by Iturria and Blangero  and implemented into the software SOLAR. Hypothesis tests concerning the (co)variance parameters involved were conducted through likelihood ratio statistics.
Trait average to untreated (SBP), hypertension treated (SBP_Hy) and adjusted (SBP*_Hy) values
Residual < 0
Residual > 0
In this work, we presented a methodology to analyze longitudinal familial data set, considering the adoption of parametric functions for modeling genetic and environmental (co)variance components involved. The approach is applied for trials of systolic blood pressure measurements. Considering that the hypertension treatment generates a right-censored response, we proceeded with corrections on the values observed under treatment. The corrections were based on information from residuals obtained through ordinary linear models, an appropriate simple model useful to estimation in this phase of the analysis. Because the vast majority of the data are untreated measures, such replacement ensures the original variability of the data, which will be modeled in terms of genetic and environmental (co)variance components. An initial check of the proposed parametric (co)variance functions, using empirical semi-variogram plots, indicated that the systolic blood pressure dispersion pattern follows assumptions as heterogeneity of variance and autocorrelation process. Having done these preliminary analyses, we are now involved in the estimation and hypothesis tests related to the genetic parameters of the proposed model.
This work was supported by NIH grant MH59490 and by a postdoctoral fellowship from Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP), São Paulo, SP, Brazil, for JMPS.
- de Andrade M, Guéguen R, Visvikis S, Sass C, Siest G, Amos CI: Extention of variance components approach to incorporate temporal trends and longitudinal pedigree data analysis. Genet Epidemiol. 2002, 22: 221-232. 10.1002/gepi.01118.View ArticlePubMedGoogle Scholar
- Levy D, DeStefano AL, Larson MG, O'Donnell CJ, Lifton RP, Gavras H, Cupples LA, Myers RH: Evidence for a gene influencing blood pressure on chromosome 17. Genome scan linkage results for longitudinal blood pressure phenotypes in subjects from the Framingham Heart Study. Hypertension. 2000, 36: 477-483.View ArticlePubMedGoogle Scholar
- Murray GD, Findlay JG: Correcting for bias caused by drop-outs in hypertension trials. Stat Med. 1988, 7: 941-946.View ArticlePubMedGoogle Scholar
- Almasy L, Blangero J: Multipoint quantitative trait linkage analysis in general pedigrees. Am J Hum Genet. 1998, 62: 1198-1211. 10.1086/301844.PubMed CentralView ArticlePubMedGoogle Scholar
- Aitkin M: Modelling variance heterogeneity in normal regression using GLIM. Appl Stat. 1987, 36: 332-339. 10.2307/2347792.View ArticleGoogle Scholar
- Diggle PJ: An approach to the analysis of repeated measurements. Biometrics. 1988, 44: 959-971. 10.2307/2531727.View ArticlePubMedGoogle Scholar
- Huggins RM, Hoang NH, Loesch DZ: Analysis of longitudinal data from twins. Genet Epidemiol. 2000, 19: 345-353. 10.1002/1098-2272(200012)19:4<345::AID-GEPI6>3.0.CO;2-S.View ArticlePubMedGoogle Scholar
- Iturria SJ, Blangero J: An EM algorithm for obtaining maximum likelihood estimates in the multi-phenotype variance components linkage model. Ann Hum Genet. 2000, 64: 349-369. 10.1046/j.1469-1809.2000.6440349.x.View ArticlePubMedGoogle Scholar
- Laird NM, Christ D, Ware JH: Longitudinal studies with continuous responses. Stat Meth Med Res. 1992, 1: 225-247.View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.