Genetic and environmental influences on relationship between anxiety sensitivity and anxiety subscales in children.

Anxiety sensitivity, a belief that symptoms of anxiety are harmful, has been proposed to influence development of panic disorder. Recent research suggests it may be a vulnerability factor for many anxiety subtypes. Moderate genetic influences have been implicated for both anxiety sensitivity and anxiety, however, little is known about the aetiology of the relationship between these traits in children. Self-reports of anxiety sensitivity and anxiety symptoms were collected from approximately 300 twin pairs at two time points. Partial correlations indicated that anxiety sensitivity at age 8 was broadly associated with most anxiety subtypes at age 10 (r=0.11-0.17, p<0.05). The associations were largely unidirectional, underpinned by stable genetic influences. Non-shared environment had unique influences on variables. Phenotypic results showed that anxiety sensitivity is a broad predictor of anxiety symptoms in childhood. Genetic results suggest that childhood is a developmental period characterised by genetic stability and time-specific environmental influences on anxiety-related traits.

Introduction

Anxiety disorders

Anxiety is one of the most prevalent psychiatric conditions amongst young people (Beesdo et al., 2010). About 10% experience anxiety by the age of 16 (Costello, Mustillo, Erkanli, Keeler, & Angold, 2003), with lifetime prevalence estimated at around 29% and mean onset age of 11 years (Kessler et al., 2005). Anxiety disorders have negative impact on child development, disturbing well-being and impairing academic performance and interpersonal interactions (Langley, Bergman, McCracken, & Piacentini, 2004; Van Ameringen, Mancini, & Farvolden, 2003). They are also reliable predictors of long-term mental health difficulties (Gregory et al., 2007; Otto et al., 2001). Anxiety is a broad term bringing together specific disorders, such as generalised anxiety disorder, panic disorder or phobias, that are characterised by excessive, persistent and impairing worry or fear (American Psychiatric Association, 2000). It is important to investigate developmental trajectories of each anxiety disorder in order to learn about the specific as well as shared aetiology.

Although anxiety disorders are characterised by homotypic continuity (prediction of disorder by the same disorder) and heterotypic continuity (prediction of disorder by another disorder), certain anxiety disorders seem to co-vary more than others (Gregory et al., 2007). Panic disorder and separation anxiety are thought to show such close developmental relationship, called the separation anxiety hypothesis (Klein, 1964; Silove, Manicavasagar, Curtis, & Blaszczynski, 1996). The two conditions share common physiological perturbations, such as somatic symptoms (Pine et al., 2005; Slattery et al., 2002). Separation anxiety in childhood has been associated with increased risk of panic disorder in adulthood (Klein, 1995; Silove, Manicavasagar, Vasey, & Dadds, 2001; Kossowsky et al., 2013), a longitudinal relationship which has been shown to be influenced by a shared genetic diathesis (Roberson-Nay, Eaves, Hettema, Kendler, & Silberg, 2012). However, the specificity of this developmental relationship is not clear, as some studies identified separation anxiety as a general risk factor for multiple adult anxiety and nonanxious disorders (Aschenbrand, Kendall, Webb, Safford, & Flannery-Schroeder, 2003; Kossowsky et al., 2013). Despite some evidence of clinical, developmental and biological similarity between separation anxiety and panic disorder, little is known about shared aetiology of these anxiety subtypes in childhood.

Anxiety sensitivity

Evidence from twin studies suggests moderate genetic and environmental influences on anxiety in childhood and across the lifespan, implicating a complex aetiology (Gregory & Eley, 2009). Another risk factor for anxiety might be biased cognitions. These are thought to play a role in both the emergence and maintenance of anxiety disorders (Clark, 1986; Ehlers, 1991). The biases can influence information processing at automatic information encoding stage (attentional biases), as well as at interpretational stage (interpretation and memory biases) (Muris & Field, 2008). Anxiety sensitivity represents one such bias: a tendency to perceive bodily cues related to experiencing anxiety as having threatening or dangerous consequences (Reiss, 1986). It is distinct from trait anxiety, which refers to the extent to which individual is fearful and prone to anxiety, while anxiety sensitivity is a fear of experiencing anxiety symptoms themselves (Taylor, 1996; Zinbarg, Brown, Barlow, & Rapee, 2001). Anxiety sensitivity is thought to be underscored by information processing abnormalities in the brain circuitry (Paulus & Stein, 2006) and variation in the trait is due to both genetic and environmental influences (Zavos, Gregory, & Eley, 2012). Anxiety sensitivity emerges in middle childhood (Reiss, Silverman, & Weems, 2001), a period characterised by a cognitive developmental stage of concrete operations and an overall cognitive maturation (Bibace & Walsh, 1981; Piaget, 1952), corresponding to acquisition of ability to consider physical symptoms in relation to anxiety from the age of 7 (Muris et al., 2008). Childhood anxiety sensitivity shows significant homotypic continuity, as well as predicts future anxiety symptoms when accounting for the current anxious state (Rabian, Embry, & MacIntyre, 1999; Weems, Hammond-Laurence, Silverman, & Ginsburg, 1998). Importantly, anxiety sensitivity and anxiety symptoms both emerge at similar age, making it an ideal time to investigate potential aetiological relationship of the two constructs.

Anxiety sensitivity – specific or broad risk factor?

Anxiety sensitivity was originally proposed as a specific risk factor for panic disorder. The presence of this cognitive bias in childhood has been found to predict panic attacks concurrently (Calamari et al., 2001; Mattis & Ollendick, 1997), as well as longitudinally in adulthood (Maller & Reiss, 1992; Schmidt et al., 2006). Several studies in adults have found that cognitive-behavioural therapy and pharmaceutical treatment targeted at panic reduce anxiety sensitivity, and this decline in cognitive bias was found to mediate the treatment (Simon et al., 2004; Smits, Powers, Cho, & Telch, 2004). Furthermore, one study found that children with good heart beat perception, which indicates good awareness of and attention to own body state, show the highest level of panic and somatic symptoms, but also heightened separation anxiety symptoms (Eley, Stirling, Ehlers, Gregory, & Clark, 2004). This could be due to a close developmental relationship between panic disorder and separation anxiety. Anxiety sensitivity, therefore, could be investigated as a specific risk factor not only for panic, but also for separation anxiety.

Other studies have shown a much broader relationship between anxiety sensitivity and anxiety subtypes (Schmidt et al., 2010; Taylor, 2003), suggesting that anxiety sensitivity might be a risk factor for a range of internalising symptoms (Plehn & Peterson, 2002). Two recent meta-analyses of adult studies support this view (Naragon-Gainey, 2010; Olatunji & Wolitzky-Taylor, 2009), finding that anxiety sensitivity was significantly related to all anxiety subtypes and depression. The associations were strongest between anxiety sensitivity and panic, general anxiety and post-traumatic stress disorder, suggesting some degree of specificity. A meta-analysis of studies of anxiety sensitivity in childhood (Noël & Francis, 2011) confirmed that anxiety sensitivity was associated with higher anxiety levels. Few studies have looked at the associations between anxiety sensitivity and specific anxiety subtypes in young people, but preliminary results based on 2 studies suggested a degree of specificity to panic symptoms.

The majority of studies that found association between anxiety sensitivity and anxiety are cross-sectional and are therefore not able to establish whether anxiety sensitivity predates anxiety symptoms, or is a consequence of anxiety. Interestingly, some longitudinal studies have directly addressed this question and suggest that the relationship might be bidirectional. For example, one study found a reciprocal longitudinal associations between anxiety sensitivity and both anxiety and depression in adolescence (Zavos, Rijsdijk, & Eley, 2012), while another found that the experience of panic and anxiety symptoms in adulthood lead to an increase in anxiety sensitivity (Schmidt, Lerew, & Joiner, 2000). This suggests that anxiety sensitivity increases subsequent anxiety, but also that symptoms of anxiety themselves increase levels of anxiety sensitivity. However, none of the studies have investigated these reciprocal processes in younger age groups, when both anxiety sensitivity and anxiety disorders emerge and when it might be possible to establish whether anxiety sensitivity predates anxiety symptoms.

Genetics of anxiety sensitivity and anxiety

Very little is known about the mechanisms underpinning the association between anxiety sensitivity and anxiety symptoms. To date, there are no multivariate twin studies investigating genetic and environmental relationship of these constructs in adults. In adolescence, anxiety sensitivity and anxiety were found to have high and significant genetic correlations (Zavos, Rijsdijk, Gregory, & Eley, 2010). This suggests that genetic factors are important in the concurrent association between anxiety sensitivity and anxiety in young people. In childhood, a very high genetic correlation has been reported between anxiety sensitivity and panic symptoms (r = 0.98; Eley, Gregory, Clark, & Ehlers, 2007), suggesting a substantial overlap of genetic influences on the two constructs. This is consistent with the pattern found in the adolescent sample, but longitudinal associations and specificity to other anxiety subtypes have not been addressed. In sum, the genetic and environmental influences underpinning the relationship between anxiety sensitivity and specific anxiety subtypes remain largely unknown.

Recently, twin studies have begun investigating developmental patterns of genetic and environmental effects in longitudinal study designs, in order to see how these influences operate over time (Ronald, 2011). Genetic influences on anxiety sensitivity have been found to be largely stable, with new genetic influences emerging late in adolescence (Zavos, Gregory, et al., 2012). Similarly, genetic stability in anxiety has been observed during childhood, with new genetic influences emerging in early and late adolescence, and in early adulthood (Kendler, Gardner, & Lichtenstein, 2008; Trzaskowski, Zavos, Haworth, Plomin, & Eley, 2011). Unlike genetic effects, environmental influences are more time-specific, possibly because non-shared environmental experiences such as stressful life events are transient (Kendler, Gardner, Annas, et al., 2008; Kendler, Gardner, & Lichtenstein, 2008; Lau & Eley, 2006; Trzaskowski et al., 2011). However, there is also evidence that idiosyncratic experiences may contribute to the continuity of anxiety (Kendler et al., 2011), suggesting some non-shared environment stability over time. Overall, very few studies have addressed these developmental questions and even fewer have explored genetic stability and change on the co-morbidity between two traits or disorders. To our knowledge, the stability of genetic and environmental influences on relationship between anxiety sensitivity and anxiety subtypes during childhood has not been investigated.

Aims

The current study aimed to investigate the developmental association between anxiety sensitivity and anxiety disorders when these problems first emerge in middle childhood (8–10 years old). Five anxiety subtypes were chosen – panic/somatic symptoms, general anxiety, separation anxiety, school anxiety and social phobia. Using a prospective study design, the phenotypic associations were first investigated to examine if anxiety sensitivity predicted future anxiety symptoms over and above any concurrent associations with other anxiety subscales. It was also hypothesised, based on previous literature, that the longitudinal relationship between anxiety sensitivity and anxiety subtypes would be bidirectional. Second, the genetic and environmental influences underpinning these longitudinal relationships were explored, controlling for the concurrent relationship between the variables. It was predicted that the multivariate genetic findings would mirror the phenotypic results. Finally, the degree of genetic and environmental continuity over time was investigated. It was expected that in the child sample environmental influences would be time-specific and genetic influences would be relatively stable over time.

Methods

Participants

The present analyses use data from the ECHO study (see Lau, Gregory, Goldwin, Pine, & Eley, 2007 for more details), a spin-off from a larger longitudinal sample of twins born in England and Wales during 1994–1996 (TEDS; Trouton, Spinath, & Plomin, 2002). In order to maximise power and include children with high emotional symptoms, the majority of twins (N = 247 pairs) were recruited due to one or both of them scoring within top 15% on child anxiety at age 7, as reported by parents. A smaller group of ‘control’ pairs were chosen, out of which none of the twins scored high on anxiety symptoms (N = 53 pairs). This selection ensured that the data represented a full range of scores on test measures. The sample characteristics at both waves are presented in Table 1. A total of 11 twin pairs (4%) were excluded because at least one of the twins had co-morbid diagnosis of neurological impairments, autistic spectrum disorders, severe receptive language impairments or persistent attentional difficulties. Zygosity was established using parent-report questionnaires. This method is estimated to be 90% accurate (Goldsmith, 1991). Where zygosity was ambiguous, DNA was collected from cheek swabs in order to assign zygosity using highly polymorphic markers of 99.9% accuracy (Price et al., 2000). The social-economic status (SES) of ECHO participants was somewhat higher than a population based sample, where for example 32% of parents were in education until 18 years or more (Meltzer, Gatward, Goodman, & Ford, 2000).

Table 1

Sample characteristics.

	Total tested	Top 15% score on anxiety/controls	Male/female	White/non-white ethnic background	Mother employed/unemployed	Father employed/unemployed	Mothers in education above age 18/below age 18	Fathers in education above age 18/below age 18	MZ/DZ/unknown	Mean age (range)	Total after exclusions
Twin pairsWave 1	300	247/53(82%)	130.5/169.5(43%)	256/44(85%)	215/85(72%)	269/31(90%)	157/143(52%)	175/125(58%)	100/199/1(33%)	8 years and 6 months (8 years 2 months–8 years 11 months)	289
Twin pairsWave 2	250	203/47(81%)	109/141(44%)	216/34 (86%)	186/64(74%)	225/25(90%)	140/110(56%)	144/106(58%)	83/167(33%)	10 years 1 month (9 years 7 months-10 years 10 months)	248 anxiety sensitivity; 249 anxiety

For both waves, parents provided written informed consent through the post prior to data collection. Data collection was conducted at the Institute of Psychiatry (King's College London, United Kingdom), apart from a small number of children who were visited in their homes. The study was granted ethical approval by the Maudsley Hospital Ethics Committee (London, United Kingdom).

In order to be able to generalise the results from this selected sample to the whole population, a weight was incorporated into all analyses. The weight controls for biases due to ascertainment – oversampling symptomatic children. It also controls for two response biases: higher SES of families from TEDS sample who agreed to participate in ECHO study as compared to the whole sample, and higher attrition rate in the families with mothers reporting higher levels of emotional problems and experiencing more negative life events. The weight used the ratio of the selection probability of high symptom families to that of nonsymptomatic families to control for bias associated with ascertainment across waves, and the inverse of the predicted probability of families remaining at Wave 2 to control for bias associated with attrition. In short, lower weights were assigned to individuals from categories over-represented in the sample, and higher weights to individuals from categories under-represented in the sample relative to the population distribution.

Measures

The questionnaires were administered on a laptop computer by a member of the research team at both waves. Items were read aloud if child had difficulty reading them.

Anxiety sensitivity

The Child Anxiety Sensitivity Index (CASI; Silverman, Fleisig, Rabian, & Peterson, 1991) was used to measure children's sensitivity to different symptoms of anxiety. Children were asked to rate on a 3 point Likert scale (1 = none, 2 = some, 3 = a lot) the 18 questionnaire items which included statements such as ‘Unusual feelings in my body scare me’. The construct validity of CASI is good, as suggested by the high correlations with fear scores in normative and clinical samples (Silverman et al., 1991). The internal consistency of the CASI measure is also very good (α = 0.87), with test–retest reliability ranges between 0.70 and 0.80 (Reiss, 1986; Silverman et al., 1991). In the current sample the internal consistency was comparable to published statistics: α = 0.80 at both waves.

Anxiety

The Screen for Child Anxiety Related Emotional Disorders (SCARED; Birmaher et al., 1999) was used to assess anxiety disorder symptoms. Children ranked, on a 3 point Likert scale (0 = almost never, 2 = often), how often in the last 3 months they experienced symptoms described by the 41 items of the questionnaire. The items can be summed up to create total anxiety score, but can also be used to create 5 DSM-IV-related anxiety subscales: panic/somatic, general anxiety, separation anxiety, social anxiety and school phobia. An example of a panic/somatic subscale item is ‘When I get frightened, I feel like passing out‘, while separation anxiety is measured by items such as ‘I worry that bad things will happen to my parents’. The psychometric properties of the SCARED have been extensively reviewed (Birmaher et al., 1999; Monga et al., 2000) and are very high, showing very good internal consistency (α = 0.90 for the total score, α = 0.78–0.87 for the subscales) and test–retest reliability ranging between 0.60 and 0.90. The SCARED also discriminates well between anxiety and other psychiatric disorders, including co-morbid conditions such as depression. In this sample the internal consistency ranged between α = 0.88–0.90 at both time points. For individual subscales, internal consistency was between α = 0.50 (school phobia) to α = 0.75 (panic/somatic scale) at wave 1, and α = 0.58 (social and school phobia) and α = 0.76 (panic/somatic scale) at wave 2.

Statistical analyses

Phenotypic analyses

Descriptive statistics were calculated using Stata (StataCorp., 2007). Variance, distribution and means were estimated for all variables. Associations between anxiety sensitivity and anxiety subtypes, both concurrent and longitudinal, were explored using full and partial correlations. For example, in order to investigate the longitudinal relationship between anxiety sensitivity at wave 1 and panic/somatic symptoms at wave 2, the scores on all anxiety scales at wave 1 were controlled. This allowed investigating the longitudinal links over and above the relationships with other variables at time 1 that might confound the longitudinal association due to high co-morbidity between anxiety symptoms. These descriptive analyses were performed on untransformed variables to allow for comparisons with other samples.

Genetics analyses

The twin design compares the degree of similarity between MZ (sharing 100% of the genes) and DZ (sharing on average 50% of their genes) twin pairs. These relative differences in within-pair correlations allow to disentangle the influences caused by additive genetics (A), common environment (C) and non-shared environment (E). For more details of quantitative genetic methods see Rijsdijk and Sham (2002).

Models were fitted using the OpenMx program (Boker et al., 2011) in R (www.R-project.org; TeamRDC, 2010), a structural equation modelling package for the analysis of genetically informative data that controls for non-independence of family members data. Sampling weights were incorporated into analyses, but did not influence the results in a manner that would alter the interpretation. As is standard in model fitting, the variables were regressed for age and sex, and the variables with skew greater than 1 were transformed to ensure normal distribution (anxiety sensitivity, panic/somatic and general anxiety time 2 were square transformed, school phobia time 2 was log transformed). Sex differences in genetic and environmental variance components were not examined owing to the relatively small sample size.

In order to maximise the sample, raw data was modelled. Saturated models were run for each set of variables. To assess model-fit to saturated model, a fit index, twice the negative loglikelihood (−2LL) of the data was calculated. The difference in this statistic between two models is distributed as chi-square, with the degrees of freedom being the difference in degrees of freedom between the two models. A p-value associated with chi-square was calculated to test for the significance of the discrepancy between the raw data outcomes and the expected parameters. The goodness-of-fit was also indexed by the Akaike information criterion (AIC). Information about the precision of parameter estimate was obtained by 95% likelihood-based confidence intervals (CIs).

The univariate analyses assessing the A, C and E influences on each variable were conducted at both waves. Next, three longitudinal trivariate Cholesky decomposition models were fitted (see Fig. 1). The model assumes three distinct sets of genetic and environmental influences (A1 to A3, C1 to C3 and E1 to E3) on each variable (though the paths a1 to a3, c1 to c3 and e1 to e3). A1, C1 and E1 are common factors influences on all variables; A2, C2 and E2 influence only the second and third variable; and A3, C3 and E3 are unique influences specific to the third variable only. The paths from each factor to the measured variable are denoted by lower case letters with subscripts denoting the number of the latent factor and the measured variable (for example A1 influence on second variable is denoted by a12). Total A, C and E effects on each individual measure can be obtained by summing all genetic and environmental paths to that measure. Although any ordering of the variables explains the variance–covariance matrix between variables equally well, the current variables are ordered in a way to investigate whether any longitudinal genetic or environmental relationships remained between anxiety sensitivity at time 1 and each of the anxiety subscales at wave 2 after accounting for scores on the same anxiety subscale at time 1. Unlike phenotypic analyses, which controlled for all concurrent variables at time 1, multivariate analyses controlled only for one concurrent variable at time 1 due to problems associated with interpreting Cholesky decompositions of more than 4 variables.

Fig. 1

Trivariate Cholesky decomposition. Note: A denotes additive genetic effects; C shared environmental effects; E non-shared environmental effects. Variance paths, which must be squared to estimate the proportion of variance accounted for, are represented by lowercase letters and followed by two numerals, e.g. a11, c22, e33.

There was an item on the panic/somatic scale that reflected anxiety sensitivity as much as panic/somatic subscale (‘I am afraid of having anxiety (or panic) attacks’), so all the analyses were repeated with a reduced version of the panic/somatic subscale that did not include this item. This exclusion did not make significant difference to the results and for this reason the results for the entire scale are presented.

Results

Phenotypic results

The descriptive statistics for both waves are presented in Table 2. There was a significant drop in scores on anxiety sensitivity (t(495) = 2.85, p = 0.001, d = 0.26) and all anxiety scales (t(497) = 6.65, p < 0.001, d = 0.60) from 8 to 10 years. As expected, there were significant sex differences in the scores. Females scored significantly higher than males on anxiety sensitivity time 1 (t(576) = 2.07, p = 0.02, d = 0.17) and on total anxiety scores at both times (time 1: t(576) = 3.20, p < 0.001, d = 0.27; time 2: t(496) = 2.33, p = 0.01, d = 0.21).

Table 2

Descriptive Statistics.

		Anxiety sensitivity	Total anxiety	Somatic/panic	General anxiety	Separation anxiety	Social phobia	School phobia
Mean (SD, range)	Time 1	31.31 (6.24, 18–52)	29.39 (12.63, 0–68)	7.15 (4.53, 0–22)	5.52 (3.51, 0–16)	7.46 (3.53, 0–16)	6.8 (2.96, 0–14)	2.45 (1.72, 0–8)
	Time 2	30.32 (5.51, 18–51)	25.17 (11.59, 1–71)	5.71 (3.93, 0–24)	5.08 (3.46, 0–17)	6.06 (3.24, 0–16)	6.27 (3.03, 0–14)	2.04 (1.59, 0–8)

Note: Results presented on untransformed variables for comparison with other published samples.

N for all variables at time 1 = 578 (289 twin pairs). At time 2, N = 496 (248 twin pairs) for anxiety sensitivity and N = 498 (249 twin pairs) for all anxiety subtypes.

The within-time and longitudinal correlations between variables were all significant at the p < 0.001 level, as presented in Table S.1. The more stringent longitudinal partial correlations, which controlled for all variables at time 1, are presented in Table 3. Anxiety sensitivity time 1 was found to be most strongly correlated with panic/somatic and separation anxiety subtypes at time 2 over and above associations with other time 1 variables (r = 0.17 and r = 0.16 respectively, both p < 0.001). Anxiety sensitivity was also significantly correlated with general and social anxiety subtypes at time 2 after controlling for the confounding associations (r = 0.11 both, p < 0.05). Anxiety sensitivity had no significant longitudinal relationship with school phobia subscale over and above associations with other variables at time 1. It is important to note that weaker relationship could be due to poorer internal consistency of school phobia measure at both waves. Post hoc comparisons revealed that the longitudinal partial correlation coefficients were not significantly different from each other.

Table 3

Longitudinal partial correlations between anxiety sensitivity and anxiety subtypes.

		Time 1
		Anxiety sensitivity	Panic/somatic	General anxiety	Separation anxiety	Social phobia	School phobia
Time 2	Anxiety Sensitivity	0.18***	0.05	0.10*	0.03	0.04	−0.06
	Somatic/panic	0.17***	0.15***	0.10*	−0.02	−0.01	0.05
	General anxiety	0.11*	0.01	0.21***	−0.03	0.04	0.01
	Separation anxiety	0.16***	−0.03	0.07	0.22***	−0.01	0.00
	Social phobia	0.11*	−0.05	0.05	0.05	0.20***	−0.01
	School phobia	0.03	0.03	0.01	0.05	−0.04	0.26***

Note: Analyses controlled for all other variables at time 1.

The significant correlations are presented in bold.

p < 0.05.

p < 0.001.

Table 3 shows that longitudinal associations were not bidirectional, as none of the anxiety subtypes at time 1 were significantly correlated to anxiety sensitivity at time 2, over and above concurrent associations, with the exception of general anxiety (r = 0.10, p < 0.05). There were significant longitudinal homotypic correlations of each anxiety subtype (r = 0.15–0.26, p < 0.001). Almost no longitudinal heterotypic correlations between the variables were evident, confirming that the SCARED scales are well designed to differentiate between anxiety subtypes.

Table S1

Within-time and longitudinal correlations between anxiety sensitivity and anxiety subscales.

Genetic results

The within-pair correlations for MZ and DZ pairs and the univariate model-fitting results are presented in Table S.2. They indicate small to moderate genetic effects on all variables (ranging from 5% on social phobia time 1 to 39% on social phobia time 2), non-significant shared environmental influences and large non-shared environmental influences (ranging from 59% on anxiety sensitivity time 2 to 95% on social phobia time 1).

The trivariate twin analyses focused on the longitudinal relationship between anxiety sensitivity and three anxiety subtypes: panic/somatic, general and separation anxiety. Panic/somatic and separation anxiety subtypes were selected because they had the strongest longitudinal phenotypic association with anxiety sensitivity. General anxiety was selected in addition, because it represents a broader anxiety phenotype to serve as a control variable. A saturated model was fitted to estimate variances, covariances, and means for the raw data to get a baseline index of fit. The trivariate Cholesky model was compared with the saturated model (χ2 = 8907.41, df = 1566) to determine the fit (χ2 = 8966.77, df = 1599, AIC = 5768.77, p = 0.003).

The trivariate Cholesky decompositions are shown in Fig. 2. Anxiety sensitivity time 1 and panic/somatic (Fig. 2a), separation (Fig. 2b) and general anxiety (Fig. 2c) at time 2 shared no genetic or non-shared environmental variance over and above that shared with the respective anxiety subtypes time 1. Shared environmental influences were non-significant. The results indicate that these significant longitudinal associations reflect shared genetic influences on anxiety sensitivity and anxiety subtypes at time 1. Furthermore, for all three anxiety subtypes, no new genetic influences on anxiety subtypes emerged at time 2. New non-shared environmental effects were evident at time 2 (67% of variance for panic and general anxiety, 61% for separation anxiety). There were also independent non-shared environmental influences on anxiety sensitivity time 1 over and above the influences shared concurrently with anxiety subtypes (54% of variance for panic, 55% for separation and general anxieties). Results from the correlated factors solution are given in Table S.3, in addition to analyses between anxiety sensitivity and total anxiety.

Fig. 2

The trivariate Cholesky decomposition for anxiety sensitivity time 1 and panic/somatic (Fig. 1a), separation anxiety (Fig. 1b) and general anxiety (Fig. 1c) at both times. Note: All non-significant paths are presented with the dashed line. The paths denote the proportion of the variance accounted for. The square root of the values must be taken to estimate the variance paths. All shared environmental paths were non-significant and for this reason they are not presented on the figures. However, due to small sample size, C has not been formally dropped from the model. 95% confidence intervals (CIs) are presented in brackets. CIs above or below 0 indicate significant correlations. Non-overlapping CIs mean significant difference between the values.

Table S2

Univarite model-fitting results for anxiety sensitivity, total anxiety and anxiety subtypes.

Table S3

Correlated factors solution results complementing Cholesky decompositions. Genetic and non-shared environmental correlations between anxiety sensitivity at time 1 and panic/somatic, separation, general and total anxiety scores at both waves.

Discussion

Summary

The current study found that anxiety sensitivity at age 8 is a significant predictor of anxiety symptoms at age 10, even when controlling for anxiety symptoms at age 8. Specifically, anxiety sensitivity most strongly predicted the development of somatic/panic symptoms and separation anxiety and to a lesser extent development of general anxiety and social phobia (although differences between the correlation strengths did not reach significance). School phobia was the only anxiety subtype that did not have significant longitudinal associations with anxiety sensitivity when controlling for concurrent associations. These longitudinal associations were not bidirectional – anxiety at age 8 did not predict anxiety sensitivity at age 10 – with the exception of general anxiety which weakly predicted onset of anxiety sensitivity over and above the concurrent associations.

Twin analyses revealed that genetic influences, at age 8, were largely shared between anxiety sensitivity and panic as well as separation and general anxiety. These genes also accounted for the longitudinal associations between anxiety sensitivity and subsequent anxiety symptoms. In other words, the longitudinal associations were not due to any new genetic influences emerging within the middle-childhood period. Conversely, significant new non-shared environmental effects on anxiety emerged at age 10. There were also independent non-shared environmental influences on anxiety sensitivity over and above the influences shared concurrently with anxiety subtypes. Lastly, as expected, shared environmental influences did not seem to play a role in the relationship between anxiety sensitivity and anxiety subtypes.

Implications

One of the criticisms of anxiety sensitivity is that this construct may not be distinct from measuring anxiety, yet the current study found significant longitudinal associations between anxiety sensitivity and subsequent anxiety over and above concurrent anxiety symptoms. Therefore, current results add to the evidence that anxiety sensitivity in middle childhood has incremental validity above and beyond measure of anxiety (Taylor, 1996). The phenotypic findings also support and strengthen the prevailing view in the literature that anxiety sensitivity is a broad anxiety predictor, with a degree of specificity to panic disorder and separation anxiety. This is supported by several meta-analyses of adult samples which have shown that anxiety sensitivity is broadly associated with all anxiety scales as well as depression. These studies, however, also show a degree of specificity to panic, general anxiety and post-traumatic stress disorder (Naragon-Gainey, 2010; Olatunji & Wolitzky-Taylor, 2009). Few studies have explored this question in child samples, and the current study extended these conclusions to the youngest age group to date.

To our knowledge the current study is the first to focus on the association between anxiety sensitivity and separation anxiety. Anxiety sensitivity has been extensively investigated as a potential specific vulnerability trait for panic disorder, yet the specificity to separation anxiety, an anxiety subscale hypothesised to be most closely related to panic (Silove et al., 1996), has not been investigated. Interestingly, current study found the association between anxiety sensitivity and separation anxiety to be as strong as that with panic disorder. This could be due to the common physiological symptoms and/or the shared developmental trajectory between separation anxiety and panic (Kossowsky et al., 2013; Pine et al., 2005). The results suggest that anxiety sensitivity might act as a shared risk factor for these problems. It is especially interesting in the context of the phenotypic correlations, which show that in the current sample separation anxiety does not predict subsequent panic symptoms over and above concurrent associations with all anxiety subscales and anxiety sensitivity. Therefore, separation anxiety hypothesis (Klein, 1964) is not supported in our sample, but anxiety sensitivity could be implicated in the developmental relationship between separation anxiety and panic across longer time spans.

Another novel phenotypic finding is that the longitudinal associations between anxiety sensitivity and anxiety subtypes in childhood were unidirectional. It suggests that as internalising problems emerge in childhood, cognitive biases predate anxiety symptoms, indicating that anxiety sensitivity is a developmental risk factor for anxiety, but not vice versa. These results contrast with the previous studies of older samples that found reciprocal relationships between cognitive biases and mood disorder symptoms (Schmidt et al., 2000; Zavos, Rijsdijk, et al., 2012). This may be because once the symptoms are established, they might in turn affect the cognitive biases, resulting in a vicious circle that maintains high internalising problems. The current study, being the first to investigate these processes in child sample, provides preliminary support for initial direction of influence between anxiety sensitivity and anxiety.

From the quantitative genetics perspective, to our knowledge this study is the first to show that the relationship between anxiety sensitivity and anxiety subtypes as early as at 8 years of age is underpinned by very high and significant genetic correlations. Results for panic and general anxiety suggested a complete genetic overlap with anxiety sensitivity, both concurrently and longitudinally, while separation anxiety also shared a high and significant proportion of its genetic influences with anxiety sensitivity. These results are in agreement with the only other evidence present in the literature, which found that anxiety sensitivity and anxiety symptoms as a whole share most of their genetic influences concurrently in adolescence (Zavos et al., 2010).

The study also furthers current understanding of age-to-age changes and continuity in genetic and environmental influences on anxiety-related disorders. The results showed that the longitudinal relationships between anxiety sensitivity and anxiety subscales are due to stable genetic influences. There were no changes in genetic influences at age 10. On the other hand, non-shared environmental influences did not contribute to the longitudinal relationship between anxiety sensitivity and anxiety symptoms, acting instead in a time and measure-specific fashion. The genetic findings provide support for the generalist genes hypothesis (Eley, 1997; Kovas & Plomin, 2006), which proposes that a shared set of genes makes individuals vulnerable to a wide range of disorder-related phenotypes. In contrast to broad genetic effects, non-shared environmental influences are generally time specific and as a result they are responsible for the discontinuity between the traits (Trzaskowski et al., 2011; Zavos, Gregory, et al., 2012). Although it has recently been suggested that accumulating environmental experiences might also contribute to continuity of internalising disorders, including anxiety (Kendler et al., 2011), such environmental continuity has not been observed in the current sample. This discrepancy might be due to the young age of the participants, as it might be too early to see such cumulative effects in childhood.

Limitations and future directions

The current study has several limitations. First, the analyses were conducted on questionnaire data, which may fail to capture some of the complexities of the investigated symptoms. It has been shown, however, that prepubescent children are able to reliably report on their mood, and as internalising problems are sometimes difficult to observe, this method may be more reliable than data from parental reports (Michael & Merrell, 1998). It would be very informative if future research incorporated clinical and multiple-observers-based diagnoses of anxiety disorders.

Secondly, the sample consists of twins and the quantitative genetics methodology carries several concerns, including equal environment assumption, assumption that gene–environment correlations and interactions are minimal for the trait and that twins do not differ from general population. For detailed discussion of these limitations, see Plomin, DeFries, McClearn, and McGuffin (2008). Overall, these limitations are likely to only have small effects in different directions. As such, the estimates of genetic and environmental influences should be taken only as indicative rather than absolute.

Lastly, the current twin study, despite a large sample size for phenotypic analysis, had little power for quantitative analyses to detect common environmental effects and dominant genetic effects. Considering that shared environment is thought to play an important role in childhood psychopathology (Eley, 2011), it would be important to replicate these results in a bigger sample. However, based on the MZ and DZ correlations, shared-environmental influences were not expected to emerge, and some of the recent studies found no evidence for shared environmental influences on anxiety in middle childhood (Kendler, Gardner, & Lichtenstein, 2008; Ogliari et al., 2010). The study was also underpowered to investigate sex differences in the analyses. However, results of a recent study suggest that genetic influences on symptoms of anxiety in males and females are very similar in childhood (Kendler, Gardner, & Lichtenstein, 2008), so this should not be too significant a limitation.

Final conclusions

Despite caveats, the current study provides novel evidence for the theoretical understanding of cognitive risk factors involved in development of anxiety. It addressed questions not previously explored in child samples and found that anxiety sensitivity is a broad predictor of anxious traits, with relative specificity to panic and separation anxiety subtypes, and that this relationship is unidirectional. The current study also found that this longitudinal association is underpinned by shared additive genetic influences and that no new genetic influences emerge to influence this relationship within the 2 years middle childhood period. Furthermore, non-shared environment was found to act as unique influence on these variables, suggesting that idiosyncratic experiences can trigger anxiety symptoms over and above the genetically initiated cognitive biases. Further research is required to see if these developmental associations change over longer periods as children reach adolescence and adulthood, but the current multivariate approach adds to the understanding of aetiology of developmental trajectories of cognitive biases and affective disorders as early as childhood.