Harmonizing behavioral outcomes across studies, raters, and countries: application to the genetic analysis of aggression in the ACTION Consortium.

BACKGROUND: Aggression in children has genetic and environmental causes. Studies of aggression can pool existing datasets to include more complex models of social effects. Such analyses require large datasets with harmonized outcome measures. Here, we made use of a reference panel for phenotype data to harmonize multiple aggression measures in school-aged children to jointly analyze data from five large twin cohorts. METHODS: Individual level aggression data on 86,559 children (42,468 twin pairs) were available in five European twin cohorts measured by different instruments. A phenotypic reference panel was collected which enabled a model-based phenotype harmonization approach. A bi-factor integration model in the integrative data analysis framework was developed to model aggression across studies while adjusting for rater, age, and sex. Finally, harmonized aggression scores were analyzed to estimate contributions of genes, environment, and social interaction to aggression. The large sample size allowed adequate power to test for sibling interaction effects, with unique dynamics permitted for opposite-sex twins. RESULTS: The best-fitting model found a high level of overall heritability of aggression (~60%). Different heritability rates of aggression across sex were marginally significant, with heritability estimates in boys of ~64% and ~58% in girls. Sibling interaction effects were only significant in the opposite-sex twin pairs: the interaction effect of males on their female co-twin differed from the effect of females on their male co-twin. An aggressive female had a positive effect on male co-twin aggression, whereas more aggression in males had a negative influence on a female co-twin. CONCLUSIONS: Opposite-sex twins displayed unique social dynamics of aggressive behaviors in a joint analysis of a large, multinational dataset. The integrative data analysis framework, applied in combination with a reference panel, has the potential to elucidate broad, generalizable results in the investigation of common psychological traits in children.

Data Integration Methods for Phenotype Harmonization in Multi-Cohort Genome-Wide Association Studies With Behavioral Outcomes.

Parallel meta-analysis is a popular approach for increasing the power to detect genetic effects in genome-wide association studies across multiple cohorts. Consortia studying the genetics of behavioral phenotypes are oftentimes faced with systematic differences in phenotype measurement across cohorts, introducing heterogeneity into the meta-analysis and reducing statistical power. This study investigated integrative data analysis (IDA) as an approach for jointly modeling the phenotype across multiple datasets. We put forth a bi-factor integration model (BFIM) that provides a single common phenotype score and accounts for sources of study-specific variability in the phenotype. In order to capitalize on this modeling strategy, a phenotype reference panel was utilized as a supplemental sample with complete data on all behavioral measures. A simulation study showed that a mega-analysis of genetic variant effects in a BFIM were more powerful than meta-analysis of genetic effects on a cohort-specific sum score of items. Saving the factor scores from the BFIM and using those as the outcome in meta-analysis was also more powerful than the sum score in most simulation conditions, but a small degree of bias was introduced by this approach. The reference panel was necessary to realize these power gains. An empirical demonstration used the BFIM to harmonize aggression scores in 9-year old children across the Netherlands Twin Register and the Child and Adolescent Twin Study in Sweden, providing a template for application of the BFIM to a range of different phenotypes. A supplemental data collection in the Netherlands Twin Register served as a reference panel for phenotype modeling across both cohorts. Our results indicate that model-based harmonization for the study of complex traits is a useful step within genetic consortia.

Genetic and environmental contributions to the development of childhood aggression.

Longitudinal data from a large sample of twins participating in the Netherlands Twin Register (n = 42,827, age range 3-16) were analyzed to investigate the genetic and environmental contributions to childhood aggression. Genetic auto-regressive (simplex) models were used to assess whether the same genes are involved or whether new genes come into play as children grow up. The authors compared 2 different simplex models to disentangle potentially changing behavioral expressions from changes in genetic and environmental effects. One model provided estimates of genetic and environmental effects at the level of individual aggression questionnaire items, and the other model assessed the effects at the level of an aggression sum score computed from the individual items. The results from both models provided evidence for largely stable genetic effects throughout childhood. The results also highlighted the differential heritability of the different indicators of aggression measured with the Childhood Behavior Checklist, with destruction of property showing a very high genetic component during early childhood and fighting behaviors being more heritable in early adolescence. (PsycINFO Database Record

Sum Scores in Twin Growth Curve Models: Practicality Versus Bias.

To study behavioral or psychiatric phenotypes, multiple indices of the behavior or disorder are often collected that are thought to best reflect the phenotype. Combining these items into a single score (e.g. a sum score) is a simple and practical approach for modeling such data, but this simplicity can come at a cost in longitudinal studies, where the relevance of individual items often changes as a function of age. Such changes violate the assumptions of longitudinal measurement invariance (MI), and this violation has the potential to obfuscate the interpretation of the results of latent growth models fit to sum scores. The objectives of this study are (1) to investigate the extent to which violations of longitudinal MI lead to bias in parameter estimates of the average growth curve trajectory, and (2) whether absence of MI affects estimates of the heritability of these growth curve parameters. To this end, we analytically derive the bias in the estimated means and variances of the latent growth factors fit to sum scores when the assumption of longitudinal MI is violated. This bias is further quantified via Monte Carlo simulation, and is illustrated in an empirical analysis of aggression in children aged 3-12 years. These analyses show that measurement non-invariance across age can indeed bias growth curve mean and variance estimates, and our quantification of this bias permits researchers to weigh the costs of using a simple sum score in longitudinal studies. Simulation results indicate that the genetic variance decomposition of growth factors is, however, not biased due to measurement non-invariance across age, provided the phenotype is measurement invariant across birth-order and zygosity in twins.

Disentangling Heterogeneity of Childhood Disruptive Behavior Problems Into Dimensions and Subgroups.

OBJECTIVE: Irritable and oppositional behaviors are increasingly considered as distinct dimensions of oppositional defiant disorder. However, few studies have explored this multidimensionality across the broader spectrum of disruptive behavior problems (DBPs). This study examined the presence of dimensions and distinct subgroups of childhood DBPs, and the cross-sectional and longitudinal associations between these dimensions. METHOD: Using factor mixture models (FMMs), the presence of dimensions and subgroups of DBPs was assessed in the Generation R Study at ages 6 (n = 6,209) and 10 (n = 4,724) years. Replications were performed in two population-based cohorts (Netherlands Twin Registry, n = 4,402, and Swedish Twin Study of Child and Adolescent Development, n = 1,089) and a clinical sample (n = 1,933). We used cross-lagged modeling in the Generation R Study to assess cross-sectional and longitudinal associations between dimensions. DBPs were assessed using mother-reported responses to the Child Behavior Checklist. RESULTS: Empirically obtained dimensions of DBPs were oppositional behavior (age 6 years), disobedient behavior, rule-breaking behavior (age 10 years), physical aggression, and irritability (both ages). FMMs suggested that one-class solutions had the best model fit for all dimensions in all three population-based cohorts. Similar results were obtained in the clinical sample. All three dimensions, including irritability, predicted subsequent physical aggression (range, 0.08-0.16). CONCLUSION: This study showed that childhood DBPs should be regarded as a multidimensional phenotype rather than comprising distinct subgroups. Incorporating multidimensionality will improve diagnostic accuracy and refine treatment. Future studies need to address the biological validity of the DBP dimensions observed in this study; herein lies an important opportunity for neuroimaging and genetic measures.

Assessing Model Selection Uncertainty Using a Bootstrap Approach: An update.

Model comparisons in the behavioral sciences often aim at selecting the model that best describes the structure in the population. Model selection is usually based on fit indices such as AIC or BIC, and inference is done based on the selected best-fitting model. This practice does not account for the possibility that due to sampling variability, a different model might be selected as the preferred model in a new sample from the same population. A previous study illustrated a bootstrap approach to gauge this model selection uncertainty using two empirical examples. The current study consists of a series of simulations to assess the utility of the proposed bootstrap approach in multi-group and mixture model comparisons. These simulations show that bootstrap selection rates can provide additional information over and above simply relying on the size of AIC and BIC differences in a given sample.

Using a multivariate model to assess the interactive effects of demographics and lifestyle on the hematological profile.

AIM: To assess the extent to which a multivariate approach to modeling interrelated hematological indices provides more informative results than the traditional approach of modeling each index separately. MATERIALS & METHODS: The effects of demographics and lifestyle on ten hematological indices collected from a Dutch population-based sample (n = 3278) were studied, jointly using multivariate distance matrix regression and separately using linear regression. RESULTS: The multivariate approach highlighted the main effects of all predictors and several interactions; the traditional approach highlighted only main effects. CONCLUSION: The multivariate approach provides more power than traditional methods to detect effects on interrelated biomarkers, suggesting that its use in future research may help identify subgroups that benefit from different treatment or prevention measures.

Fitting latent variable mixture models.

Latent variable mixture models (LVMMs) are models for multivariate observed data from a potentially heterogeneous population. The responses on the observed variables are thought to be driven by one or more latent continuous factors (e.g. severity of a disorder) and/or latent categorical variables (e.g., subtypes of a disorder). Decomposing the observed covariances in the data into the effects of categorical group membership and the effects of continuous trait differences is not trivial, and requires the consideration of a number of different aspects of LVMMs. The first part of this paper provides the theoretical background of LVMMs and emphasizes their exploratory character, outlines the general framework together with assumptions and necessary constraints, highlights the difference between models with and without covariates, and discusses the interrelation between the number of classes and the complexity of the within-class model as well as the relevance of measurement invariance. The second part provides a growth mixture modeling example with simulated data and covers several practical issues when fitting LVMMs.

Joint developmental trajectories of internalizing and externalizing disorders between childhood and adolescence.

This study sought to identify trajectories of DSM-IV based internalizing (INT) and externalizing (EXT) problem scores across childhood and adolescence and to provide insight into the comorbidity by modeling the co-occurrence of INT and EXT trajectories. INT and EXT were measured repeatedly between age 7 and age 15 years in over 7,000 children and analyzed using growth mixture models. Five trajectories were identified for both INT and EXT, including very low, low, decreasing, and increasing trajectories. In addition, an adolescent onset trajectory was identified for INT and a stable high trajectory was identified for EXT. Multinomial regression showed that similar EXT and INT trajectories were associated. However, the adolescent onset INT trajectory was independent of high EXT trajectories, and persisting EXT was mainly associated with decreasing INT. Sex and early life environmental risk factors predicted EXT and, to a lesser extent, INT trajectories. The association between trajectories indicates the need to consider comorbidity when a child presents with INT or EXT disorders, particularly when symptoms start early. This is less necessary when INT symptoms start at adolescence. Future studies should investigate the etiology of co-occurring INT and EXT and the specific treatment needs of these severely affected children.

Extending multivariate distance matrix regression with an effect size measure and the asymptotic null distribution of the test statistic.

Person-centered methods are useful for studying individual differences in terms of (dis)similarities between response profiles on multivariate outcomes. Multivariate distance matrix regression (MDMR) tests the significance of associations of response profile (dis)similarities and a set of predictors using permutation tests. This paper extends MDMR by deriving and empirically validating the asymptotic null distribution of its test statistic, and by proposing an effect size for individual outcome variables, which is shown to recover true associations. These extensions alleviate the computational burden of permutation tests currently used in MDMR and render more informative results, thus making MDMR accessible to new research domains.

Finding structure in data using multivariate tree boosting.

Technology and collaboration enable dramatic increases in the size of psychological and psychiatric data collections, but finding structure in these large data sets with many collected variables is challenging. Decision tree ensembles such as random forests (Strobl, Malley, & Tutz, 2009) are a useful tool for finding structure, but are difficult to interpret with multiple outcome variables which are often of interest in psychology. To find and interpret structure in data sets with multiple outcomes and many predictors (possibly exceeding the sample size), we introduce a multivariate extension to a decision tree ensemble method called gradient boosted regression trees (Friedman, 2001). Our extension, multivariate tree boosting, is a method for nonparametric regression that is useful for identifying important predictors, detecting predictors with nonlinear effects and interactions without specification of such effects, and for identifying predictors that cause 2 or more outcome variables to covary. We provide the R package "mvtboost" to estimate, tune, and interpret the resulting model, which extends the implementation of univariate boosting in the R package "gbm" (Ridgeway, 2015) to continuous, multivariate outcomes. To illustrate the approach, we analyze predictors of psychological well-being (Ryff & Keyes, 1995). Simulations verify that our approach identifies predictors with nonlinear effects and achieves high prediction accuracy, exceeding or matching the performance of (penalized) multivariate multiple regression and multivariate decision trees over a wide range of conditions. (PsycINFO Database Record

Inference Based on the Best-Fitting Model can Contribute to the Replication Crisis: Assessing Model Selection Uncertainty Using a Bootstrap Approach.

Inference and conclusions drawn from model fitting analyses are commonly based on a single "best-fitting" model. If model selection and inference are carried out using the same data model selection uncertainty is ignored. We illustrate the Type I error inflation that can result from using the same data for model selection and inference, and we then propose a simple bootstrap based approach to quantify model selection uncertainty in terms of model selection rates. A selection rate can be interpreted as an estimate of the replication probability of a fitted model. The benefits of bootstrapping model selection uncertainty is demonstrated in a growth mixture analyses of data from the National Longitudinal Study of Youth, and a 2-group measurement invariance analysis of the Holzinger-Swineford data.

A powerful phenotype for gene-finding studies derived from trajectory analyses of symptoms of anxiety and depression between age seven and 18.

To investigate the utility of longitudinal data in genetic analyses of symptoms of anxiety and depression, we assessed individual differences between age 7 and 18 using growth mixture models, and investigated the genetic and non-genetic factors contributing to the trajectories. Mothers of 7,706 girl and 7,418 boy twins from the Netherlands Twin Register rated the anxious depression scale (SxAnxDep) of the Child Behavior Check List (CBCL) at age 7, 10, and 12 years. Two thousand seven hundred and six girl and 1,856 boy twins completed the Youth Self Report (YSR) at age 14, 16, and 18. While individual trajectories varied considerably, these differences were largely idiosyncratic and could not be grouped into separate latent classes with class-specific average growth curves. The intercept, which reflects the individuals' baseline level of SxAnxDep across time, explained 55-58% of the total phenotypic variance. The slope factor, which captures a common average trend over time, did not explain variance in the phenotype. This finding also underlines the high level of idiosyncrasy of trajectories that lack a common longitudinal structure. The analyses of twin data showed that the random intercept factor of SxAnxDep during childhood and during adolescence is considerably more heritable than the observations at any single age, namely between 60% and 84%. One explanation is that different factors contribute to the level of symptoms of anxiety and depression at any given time point, including temporary events and emotions. When considering baseline stability, these temporary influences average out, with the result of a more reliable and more heritable phenotype. © 2015 Wiley Periodicals, Inc.

Population heterogeneity of trait anger and differential associations of trait anger facets with borderline personality features, neuroticism, depression, Attention Deficit Hyperactivity Disorder (ADHD), and alcohol problems.

Anger is an emotion consisting of feelings of variable intensity ranging from mild irritation to intense fury. High levels of trait anger are associated with a range of psychiatric, interpersonal, and health problems. The objectives of this study were to explore heterogeneity of anger as measured by the Spielberger Trait Anger Scale (STAS), and to assess the association of the different anger facets with a selection of psychiatric disorders covering externalizing and internalizing problems, personality disorders, and substance use. Factor mixture models differentiated between a high and low scoring class (28% vs. 72%), and between three factors (anger-temperament, anger-reaction, and immediacy of an anger response). Whereas all psychiatric scales correlated significantly with the STAS total score, regressing the three STAS factors on psychiatric behaviors model showed a more detailed pattern. Only borderline affect instability and depression were significantly associated with all three factors in both classes whereas other problem behaviors were associated only with 1 or 2 of the factors. Alcohol problems were associated with immediacy only in the high scoring class, indicating a non-linear relation in the total sample. Taking into account these more specific associations is likely to be beneficial when investigating differential treatment strategies.

Genetic analyses benefit from using less heterogeneous phenotypes: an illustration with the hospital anxiety and depression scale (HADS).

Phenotypic heterogeneity of depression has been cited as one of the causes of the limited success to detect genetic variants in genome-wide studies. The 7-item Hospital Anxiety and Depression Scale (HADS-D) was developed to detect depression in individuals with physical health problems. An initial psychometric analysis showed that a short version ("HADS-4") is less heterogeneous and hence more reliable than the full scale, and correlates equally strong with a DSM-oriented depression scale. We compared the HADS-D and the HADS-4 to assess the benefits of using less heterogeneous phenotype measures in genetic analyses. We compared HADS-D and HADS-4 in three separate analyses: (1) twin- and family-based heritability estimation, (2) SNP-based heritability estimation using the software GCTA, and (3) a genome-wide association study (GWAS). The twin study resulted in heritability estimates between 18% and 25%, with additive genetic variance being the largest component. There was also evidence for assortative mating and a dominance component of genetic variance, with HADS-4 having slightly lower estimates of assortment. Importantly, when estimating heritability from SNPs, the HADS-D did not show a significant genetic variance component, while for the HADS-4, a statistically significant amount of heritability was estimated. Moreover, the HADS-4 had substantially more SNPs with small P-values in the GWAS analysis than did the HADS-D. Our results underline the benefits of using more homogeneous phenotypes in psychiatric genetic analyses. Homogeneity can be increased by focusing on core symptoms of disorders, thus reducing the noise in aggregate phenotypes caused by substantially different symptom profiles.

Childhood ODD and ADHD Behavior: The Effect of Classroom Sharing, Gender, Teacher Gender and Their Interactions.

One criterion for a diagnostic and statistical manual of mental disorders (DSM-IV) diagnosis of attention deficit hyperactivity disorder (ADHD) and oppositional defiant disorder (ODD) is that symptoms are present in at least two settings, and often teacher ratings are taken into account. The short Conners' Teacher Rating Scales-Revised (CTRS-R) is a widely used standardized instrument measuring ODD and ADHD behavior in a school setting. In the current study CTRS-R data were available for 7, 9 and 12-year-old twins from the Netherlands Twin Register. Measurement invariance (MI) across student gender and teacher gender was established for three of the four scales (Oppositional Behavior, Hyperactivity and ADHD Index) of the CTRS-R. The fourth scale (ATT) showed an unacceptable model fit even without constraints on the data and revision of this scale is recommended. Gene-environment (GxE) interaction models revealed that heritability was larger for children sharing a classroom. There were some gender differences in the heritability of ODD and ADHD behavior and there was a moderating effect of teacher's gender at some of the ages. Taken together, this indicates that there was evidence for GxE interaction for classroom sharing, gender of the student and gender of the teacher.

Heritability of liver enzyme levels estimated from genome-wide SNP data.

Variation in the liver enzyme levels in humans is moderately heritable, as indicated by twin-family studies. At present, genome-wide association studies have traced <2% of the variance back to genome-wide significant single-nucleotide polymorphisms (SNPs). We estimated the SNP-based heritability of levels of three liver enzymes (gamma-glutamyl transferase (GGT); alanine aminotransferase (ALT); and aspartate aminotransferase (AST)) using genome-wide SNP data in a sample of 5421 unrelated Dutch individuals. Two estimation methods for SNP-based heritability were compared, one based on the distant genetic relatedness among all subjects as summarized in a Genetic Relatedness Matrix (GRM), and the other one based on density estimation (DE). The DE method was also applied to meta-analysis results on GGT and ALT. GRM-derived SNP-based heritability estimates were significant for GGT (16%) and AST (11%), but not for ALT (6%). DE estimates in the same sample varied as a function of pruning and were around 23% for all liver enzymes. Application of the DE approach to meta-analysis results for GGT and ALT gave SNP-based heritability estimates of 6 and 3%. The significant results in the Dutch sample indicate that genome-wide SNP platforms contain substantial information regarding the underlying genetic variation in the liver enzyme levels. A major part of this genetic variation remains however undetected. SNP-based heritability estimates, based on meta-analysis results, may point at substantial heterogeneity among cohorts contributing to the meta-analysis. This type of analysis may provide useful information to guide future gene searches.

Epi2Loc: an R package to investigate two-locus epistatic models.

Epistasis is a growing area of research in genome-wide studies, but the differences between alternative definitions of epistasis remain a source of confusion for many researchers. One problem is that models for epistasis are presented in a number of formats, some of which have difficult-to-interpret parameters. In addition, the relation between the different models is rarely explained. Existing software for testing epistatic interactions between single-nucleotide polymorphisms (SNPs) does not provide the flexibility to compare the available model parameterizations. For that reason we have developed an R package for investigating epistatic and penetrance models, Epi2Loc, to aid users who wish to easily compare, interpret, and utilize models for two-locus epistatic interactions. Epi2Loc facilitates research on SNP-SNP interactions by allowing the R user to easily convert between common parametric forms for two-locus interactions, generate data for simulation studies, and perform power analyses for the selected model with a continuous or dichotomous phenotype. The usefulness of the package for model interpretation and power analysis is illustrated using data on rheumatoid arthritis.