 |
|
Genetic and Environmental Contributions to the Child Behavior Checklist Obsessive-Compulsive Scale
A Cross-cultural Twin Study
James J. Hudziak, MD;
C. E. M. van Beijsterveldt, PhD;
Robert R. Althoff, MD, PhD;
Catherine Stanger, PhD;
David C. Rettew, MD;
Elliot C. Nelson, MD;
Richard D. Todd, MD, PhD;
Meike Bartels, PhD;
Dorret I. Boomsma, PhD
Arch Gen Psychiatry. 2004;61:608-616.
ABSTRACT
| |
Context We have reported elsewhere on the development of an 8-item Obsessive-Compulsive Scale (OCS) contained in the Child Behavior Checklist (CBCL) to identify children who meet criteria for DSM-IV obsessive-compulsive disorder. Twin studies of obsessive-compulsive disorder have indicated a significant genetic component to its expression.
Objective To determine the relative contributions of genetic and environmental influences on childhood obsessive-compulsive behavior using the CBCL OCS in twin samples.
Design The CBCL data were received by survey of twins in the Netherlands Twin Registry (NTR) and the Missouri Twin Study (USA/MOTWIN).
Setting General community twin samples.
Participants Participants were 4246 twin pairs aged 7 years, 2841 aged 10 years, and 1562 aged 12 years (who also participated in the study at 7 and 10 years of age) from the NTR and 1461 mixed-age twin pairs (average age, approximately 9 years) from the USA/MOTWIN.
Main Outcome Measures Model fitting to test for genetic and environmental influences, sex differences, and sibling interaction/rater contrast effects on the CBCL OCS.
Results In each case, the best-fitting model was one that indicated significant additive genetic influences (range, 45%-58%; 95% confidence interval [CI], 45%-61%), and unique environmental influences (range, 42%-55%; 95% CI, 39%-55%), with shared environmental influences in the NTR sample aged 12 years (16%). Sex differences were seen in the mixed-age USA/MOTWIN model, but not in the NTR samples. No evidence of dominance, sibling interaction, or rater-contrast effects was seen. These data were relatively consistent across age and cultures.
Conclusions The CBCL OCS is influenced by genetic factors (approximately 55%) and unique environmental factors (approximately 45%) in the younger sample, with common environmental influences only at 12 years of age. These effects do not vary with differences in sex or sibling interaction/rater contrast effects. Our data reveal higher genetic influences for obsessive-compulsive behavior and do not demonstrate genetic differences across sex.
INTRODUCTION
Obsessive-compulsive disorder (OCD) is common in children and adults. It is associated with serious impairment and, in many cases, has a lifelong course. Studies on prevalence indicate that the lifetime rates of OCD in adolescents range from 1.9% to 4.1%,1-6 with a higher prevalence in girls than in boys.7 The wide variability in prevalence rates may be due to differences in the populations sampled and the methods used for assessment. Nonetheless, the prevalence of OCD in children and adolescents may be higher than originally thought. Other reports8 (including unpublished data March 2003: J.J.H., R.R.A., C.S., C.E.M.v B., E.C.N., G. L. Hanna, MD, D.I.B., and R.D.T.) have described our development of an 8- item Obsessive-Compulsive Scale (OCS) contained in the Child Behavior Checklist (CBCL) that is useful in identifying children aged 7 to 12 years who meet criteria for DSM-IV OCD (Table 1). Using a summed score of 5 (the borderline clinical range, >95th percentile based on the CBCL normative sample), the CBCL OCS demonstrated high sensitivity (92%), moderate specificity (67%), high negative predictive value (90%), and moderate positive predictive value (73%) in subjects who had been diagnosed as having DSM-IV OCD by board-certified psychiatrists. Using a higher 98% clinical range, which corresponds to a score of 6, the sensitivity of the test was 79%; specificity was 78.1%; positive predictive value was 77.8%; and negative predictive value was 79.4%. In 4 large twin samples (1 in the United States and 3 in the Netherlands), the percentage of participants with CBCL OCS scores in the borderline clinical range was 2.4% to 4.3%, with 1.4% to 3.8% in the clinical range (Table 2).
|
|
|
|
Table 1. Items Used for the OCS
|
|
|
|
|
|
|
Table 2. Percentage of Whole Sample Meeting CBCL OCS Cut Points
|
|
|
The purpose of the present study was to extend our previous work and determine the contributions of environment, genes, sex, and age to the expression of CBCL OCS and, by extension, childhood OCD.
PREVIOUS GENETIC STUDIES OF OCD
The estimation of genetic and environmental influences on a disorder typically includes twin, family, adoption, and molecular genetic approaches. Tsuang et al9 argued that although our molecular genetic techniques have advanced to the point that identifying genetic variants that contribute to the development of a phenotype is now a trivial laboratory exercise, the development of phenotypic identification strategies that refine diagnoses for molecular genetic investigations remain problematic. This is especially true in childhood OCD, where relatively few genetic studies have been performed. Family and twin studies of OCD are rare relative to the studies of other psychiatric disorders. Of the OCD studies, many have used small samples that were generally derived from highly comorbid clinical populations, with a resultant reduction in the generalizability of the results.10 Despite these limitations, the belief that OCD is influenced by genetic factors is widely held; in fact, initial reports on the genetic contributions to OCD are more than 60 years old.11
FAMILY STUDIES
Family studies to date have revealed conflicting results. The percentage of affected first-degree relatives of patients with OCD has ranged from being indistinguishable from control subjects12 to 30%,13 with several estimates between.14-18 Many OCD studies include patients with other disorders such as Tourette syndrome and other tic disorders. These studies have reported rates of OCD as high as 26% among first-degree relatives of patients with Tourette syndrome, a familial condition known to have significant genetic influences.19-20 A recent meta-analysis21 demonstrated an unadjusted aggregate risk of 8.2% in first-degree relatives of patients with OCD vs 2% in relatives of controls. In summary, published family studies support the contention that OCD, alone or comorbid, is a condition that is influenced by genetic factors.
TWIN STUDIES
Twin studies have also demonstrated genetic influences on OCD. Monozygotic (MZ) twins have been shown to have a concordance rate for OCD as high as 70% to 80%, compared with 22% to 47% among dizygotic (DZ) twins.22-23 Heritability estimates have been calculated in the range of 26% to 33%.10, 24 However, Hettema et al21 were unable to find any twin studies of adequate size without ascertainment bias to meet their criterion for inclusion, although they noted that published twin studies had found consistent evidence of genetic contributions to obsessions and compulsions.
MOLECULAR GENETIC STUDIES
Most molecular genetic studies of OCD have focused on the monoamine pathway genes. Numerous molecular genetic studies have aimed to determine the relative contributions of different candidate genes to the pathophysiology of OCD. The catechol O-methyltransferase (COMT) gene,25-28 the serotonin 2 receptor (5HT2B) gene,29-30 the serotonin transporter gene,31-32 the serotonin 2 receptor (5HT2A) gene,33-34 and the 7 repeat of the dopamine D4 receptor (DRD4) gene,35 among others, have all been implicated in OCD or OCD-like phenotypes, some perhaps in a sex-specific36 or a population-specific27, 37 fashion. Other reports have shown the contrary.38-40 In each of these studies, the authors considered the results preliminary and called for studies on much larger populations and more refined samples to clearly understand the contribution of individual gene variations to the etiology of OCD.
SUMMARY
In aggregate, family, twin, and molecular genetics studies support the premise that OCD or features of OCD are influenced by genetic factors. Most of these studies did not control for sex, age, referral bias, or comorbidity. Thus, confusion remains about the best way to conceptualize and refine the OCD phenotype for genetic analysis.
AIMS OF THE PRESENT STUDY
There were multiple aims of this study. The first was to determine the genetic and environmental contributions to CBCL OCS scores. The second was to determine whether evidence of sex-genetic interactions existed. The third was to determine whether the age of the child contributed to the genetic/environmental influences on CBCL OCS scores by analyzing these data in samples of twins aged 7, 10, and 12 years and a mixed sample of twins aged 8 to 12 years. Finally, the study assessed cultural differences by determining whether the genetic/environmental contributions differed by country (the Netherlands vs the United States).
METHODS
SUBJECTS AND PROCEDURE
The Netherlands Twin Registry
The study was part of an ongoing twin-family study of health-related characteristics, personality, and behavior in the Netherlands. The subjects were all part of the Netherlands Twin Registry (NTR).41 At present, the NTR has data on more than 25 000 twin pairs from ages 3 to 30 years. For this study, we assessed a sample of Dutch twin pairs whose parents reported on their behavior when they were 7, 10, and 12 years of age (NTR-7, NTR-10, and NTR-12 samples, respectively). The total sample with CBCL data available in the NTR at these age groups at the time of the study included 4484, 2905, and 1664 twin pairs aged 7, 10, and 12 years, respectively. Of these, 238, 64, and 102 twin pairs, respectively, were excluded owing to missing data, leaving a total of 4246, 2841, and 1562 twin pairs available for analysis at ages 7, 10, and 12 years, respectively. The socioeconomic status of the parents of the twins was somewhat higher than the level in the general Dutch population.42
The assessment procedures at ages 7, 10, and 12 years have been described elsewhere.41 Parents received a CBCL by mail. Parents who did not return the forms within 2 months received a reminder. Those who did not respond after 4 months were called by the NTR research assistant. This procedure resulted in an average continued participation rate of 80% from 7 to 10 and 10 to 12 years of age.
Zygosity were determined by means of DNA analyses of blood group polymorphisms for 634 same-sex twin pairs. For the remaining twins, zygosity was determined by questionnaire items about physical similarity and frequency of confusion of the twins by family and strangers. The classification of zygosity was based on a discriminant analysis, relating the questionnaire items to zygosity based on blood/DNA typing. According to this analysis, the zygosity was correctly classified by questionnaire in nearly 95% of the cases.43
Missouri Twin Study Sample
To apply this screening tool to a large sample of twins with different ages and mixed ethnicity, we selected twins from an ongoing project, the Missouri Twin Study (USA/MOTWIN). This sample has been described previously.44 Briefly, an attempt was made to contact parents of all twins born in Missouri from 1975 to 1991 to invite them to participate. They were paid $5 for completing survey materials. At the time that the genetic analyses were performed, data on 1461 of 1565 twin pairs who were sent CBCLs were used. One hundred four pairs were excluded because 1 or more of the 8 items of the CBCL OCS were missing. The CBCL OCS scores from each of these remaining twin pairs were computed.
Zygosity was determined by means of questionnaire items as for the NTR samples, with assignment based on a latent class approach. In a comparison of genotypic determination of 121 twin pairs, only 1 pair was misassigned using this method.45
Table 3 provides a description of the numbers of twin pairs by sex and zygosity for the 4 samples. There was a small but statistically significant difference in age, with girls slightly younger; however, there was no statistical age difference by zygosity group. Although the USA/MOTWIN sample was ethnically mixed (85% European American, 13.4% African American, and <1% Hispanic, Asian, or Native American), the NTR samples consisted primarily of children of European descent. The NTR-10 sample was a subset of the NTR-7 sample studied 3 years later, and the NTR-12 sample was a subset of the NTR-10 sample.
|
|
|
|
Table 3. Numbers and Ages of Twin Pairs
|
|
|
MEASURES
The CBCL is a widely used questionnaire for parents to respond to 118 problem behaviors exhibited by their child during the previous 6 months. The parent responds along a 3-point scale with the code of 0 if the item is not true of the child, 1 for sometimes true, and 2 for often true. The characteristics and psychometric stability of the CBCL have been well established.46 The analyses performed herein used the 1991 version of the CBCL, but the same items can be scored on the more recent 2001 version.47
The CBCL OCS was developed using factor analysis on 11 CBCL items that were thought to likely predict OCD.8 Using a 1-factor model, 8 items were retained and were shown to have good internal consistency (Cronbach = .84). Those items retained are shown in Table 1, along with their CBCL item number. A numerical value for the CBCL OCS is created by adding the scores on these 8 items (0, 1, or 2 for each), thus limiting the scale to a range of 0 to 16. The CBCL OCS was tested to determine prevalence, specificity, and sensitivity.9
DATA ANALYSES
Differences in Means
Means, variances, and twin correlations were calculated using Mx.48 Differences in mean scores and variances between sex and zygosity were tested by means of likelihood ratio c2 tests. These tests were performed taking into account the dependency that exists between scores of the twins. Because the CBCL OCS score was not normally distributed, the data were square root transformed to approximate a normal distribution.
Models
Genetic and environmental influences on CBCL OCS scores were computed using structural equation modeling. The relative contributions of genetic and environmental factors to individual differences in CBCL OCS scores can be inferred from the different level of genetic relatedness of MZ and DZ twins.49 Figure 1 summarizes the fundamental univariate genetic model that underlies our analyses. The variance may be due to additive genetic factors (A), common or shared environment factors (C), or nonshared environment effects (E). We also tested for dominance genetic effects (D), which correlate at 1.0 in MZ twins and 0.25 in DZ twins. Estimating D and C at the same time is not possible in a design using only MZ and DZ twins reared together. Using D instead of C in the models did not contribute to a better fit; thus, D was not examined further. The genetic factors are correlated at 1.0 in MZ twins, as they are genetically identical. For DZ twins, the additive genetic factors are correlated at 0.5, because DZ twins share on average half of their genes. The environment shared by a twin pair is assumed not to depend on zygosity, and thus shared environmental factors correlate at 1.0 in both MZ and DZ twins. The E term is by definition uncorrelated. All uncorrelated error is also absorbed in the E term. The parameters a, c, and e are loadings of the observed phenotype on the latent factors A, C, and E and indicate the degree of relations between the latent factors and the observed phenotype. The proportion of the variance accounted for by genetic and environmental influences is calculated by squaring the parameters a, c, and e and dividing them by the total variance (a2+c2+e2). In addition, in the univariate model, the effects of sibling interaction (path s) are also considered. Sibling interaction reflects the effect of the behavior of one twin on the behavior of the other twin. The interaction effect may also be due to bias in parental reports when parents rate their children's behavior in comparison with each other. Whether the sibling interaction effects are a function of rater contrast or of real sibling interaction cannot be tested with the current data, but would need information from more than 1 informant. The AE model with sibling interaction was tested and did not lead to a better fit than the AE model without sibling interaction, and it was not examined further.
|
|
|
|
Sex-genetic model. Additive genetic factors are correlated at 0.5 in dizygotic (DZ) twins, because DZ twins share on average half of their genes, and at 1.0 in monozygotic (MZ) twins. The parameters a, c, and e are loadings of the observed phenotype (P) on the latent additive genetic factors (A), common or shared environment factors (C), and nonshared environment effects (E) and indicate the degree of relations between the latent factors and the observed P. The proportion of the variance accounted for by genetic and environmental influences is calculated by squaring the parameters a, c, and e and dividing them by the total variance (a2+c2+e2). In addition, in the univariate model, the effects of sibling interaction (path s) are also considered. Genotypexsex interaction effects (illustrated for the case of unlike-sex sibling pairs) may take the form of sex differences in the magnitude of genetic or environmental influences (paths am, cm, and em vs af, cf, and ef) or the form of an additional genetic or common environmental influence on only 1 of the unlike-sex sibling pairs (paths a' and c'). OCS indicates Obsessive-Compulsive Scale.
|
|
|
Figure 1 extends the latent variable component of the model by allowing for genotypexsex interaction effects (illustrated for the case of unlike-sex sibling pairs). This may take the form of sex differences in the magnitude of genetic or environmental influences (paths am, cm, and em vs af, cf, and ef) or the form of an additional genetic or common environmental influence on only 1 of the unlike-sex sibling pairs (paths a' and c'). These analyses allowed us to test for sex differences on CBCL OCS scores.
Model Fitting
To estimate the genetic and environmental contributions, the data for twins 1 and 2 were summarized into 2 x 2 covariance matrices, computed by PRELIS scientific software.50 All model fitting was performed with Mx,48 a statistical software package designed for conducting genetic analyses with an approach that is standard in structural equation modeling.51 The basic model tested was an ACE model. The significance of the A and C factors was tested by dropping each of these variance components one at a time and using the c2 difference test. The c2 statistic was computed by subtracting the c2 statistic for the full model from that for a reduced model. The degrees of freedom for this test are equal to the differences between the degrees of freedom for the full and the reduced model. If the c2 statistic is significant, this means that the variance component makes a significant contribution to the fit of the full model, because removing it significantly worsens the fit of the model. In addition, the Akaike information criterion, a goodness-of-fit index that considers the rule of parsimony, was calculated. A smaller Akaike information criterion indicates a better fit. We also computed likelihood-based 95% confidence intervals (CIs) for each parameter.48, 52 More technical details of genetic model-fitting analyses are reviewed elsewhere.49
RESULTS
The square root–transformed mean CBCL OCS scores and variances across sex and zygosity are presented in Table 4. Raw CBCL OCS scores were quite similar across age and country. The homogeneity of the variance across sex was tested with Mx48 and revealed significant sex differences for CBCL OCS scores in all 3 samples. In the USA/MOTWIN sample, boys had higher scores; in both Dutch samples, girls had higher scores. The variance and covariance matrices for all zygosity groups are given in Table 5. There were no differences in the means, variances, and covariances across the 5 zygosity groups for the USA/MOTWIN sample and the NTR-10 or NTR-12 samples. In the NTR-7 sample, the means and the variances of the DZ twins were larger than for MZ twins ( c28 = 21.18). As shown in Table 5, although significant, these differences were very small.
|
|
|
|
Table 4. Estimated Square Root–Transformed CBCL OCS Scores
|
|
|
|
|
|
|
Table 5. Observed Variance-Covariance Matrix for the Square Root–Transformed CBCL OCS Scores*
|
|
|
The twin correlations for the CBCL OCS score are shown in Table 6. Analysis of twin correlations yielded evidence of the influence of genetic and environmental factors. In all 3 samples, MZ correlations were larger than DZ correlations, indicating the influences of genes. The MZ and DZ correlations were not different across sex, with 1 exception. The female DZ correlation was lower than the male DZ correlation in all samples, but only significantly so in the USA/MOTWIN sample. In the remaining samples, the magnitude of genetic and environmental effects was equal across sex. In addition, the DZ opposite-sex correlations equaled the same-sex male correlations in all 3 samples, suggesting that the same genes and environmental influences play a role for boys and girls.
|
|
|
|
Table 6. Twin Correlations for Transformed CBCL OCS Syndrome
|
|
|
A summary of the model-fitting results is given in Table 7. The c2 statistic indicates the goodness of fit, and smaller c2 statistics indicate better agreement of the observed data with the model. First, we computed a model for each sample that allowed the variance components to differ between boys and girls. In the second set of models, A, C, and E parameters were constrained to be equal across sex. These constrained models were compared with the unconstrained models, and the best-fitting models were selected. These results showed no deterioration in fit in the 2 Dutch samples when the parameters were constrained to be the same across sex. In the USA/MOTWIN sample, this resulted in a worsening of the fit (c23 = 16.85). This sex difference was probably due to differences in total variance between boys and girls seen in this sample only. Next, the significance of the C factor was tested by dropping it from the models and calculating change in the goodness of fit. Dropping C from the model did not lead to a deterioration of the fit in any of the samples except for the NTR-12, meaning that the AE was the best model for all 3 of the other samples, with an ACE model being the best fit for the NTR-12 sample only.
|
|
|
|
Table 7. Model-Fitting Results for Square Root–Transformed CBCL OCS Scores*
|
|
|
SUMMARY OF MODEL FITTING
The best-fitting model for 3 samples included A and E contributions (Table 7), with C contributions evident in the NTR-12 sample. There were no sex effects in the Dutch samples, although there were minor sex effects in the USA/MOTWIN sample, likely due to underlying differences in the means and variances of the CBCL OCS score in this mixed-aged sample. Across age groups and cultures, the additive genetic influence of the CBCL OCS varied from 45% to 58% (95% CI, 45%-61%). The E factors (which conspire to make members of a twin pair different) ranged from 42% to 55% (95% CI, 39%-55%). In the NTR-12 only, the magnitude of the shared environmental influences was about 16%.
COMMENT
With these results, we extend our previous work on the CBCL OCS by revealing that scores on this proposed scale for assessing childhood OCD are highly heritable and influenced by additive genetic and unique environmental factors in younger children, with common environmental influences appearing to play a role beginning at 12 years of age. The magnitude and type of the genetic and environmental influences were surprisingly stable across age, sex, and culture within the younger group, but may differ as the children enter puberty, given the differences reported herein at 12 years of age.
The magnitude of the genetic contribution for each sample is larger than previous estimates,10, 24 which fell in the range of 26% to 33%. There are several reasons for this difference. First, restriction of range introduced by examining genetic effects on OCD in clinical samples may have attenuated previous estimated genetic contributions. Clinical cases are often more severe, with higher rates of comorbid illness that may be influenced by environmental fa |
