Considerable literature relates developmental perturbation in stress regulation to later diagnosis of anxiety or major depression (1, 2). However, little is known about stress regulation in children with anxiety disorders. In a laboratory paradigm that exposed anxious and normal children to 5% CO₂, patients with anxiety disorders experiencing CO₂-induced panic symptoms exhibited elevations in cortisol levels before and after CO₂ exposure relative to patients and healthy subjects not experiencing panic symptoms. Cortisol levels did not increase further during CO₂ exposure, even in subjects who developed panic (3). These results, consistent with data in adults (4), suggest a relationship in juveniles between anticipatory anxiety and hypothalamic-pituitary-adrenal (HPA) axis function.

With a family-based design, the present study addressed whether parental panic disorder relates to HPA activity in offspring during exposure to CO₂ inhalation. The study also examined whether the previously noted positive relationship between anxiety response to CO₂ and HPA activity could be replicated in a nonclinical group that was studied at home.

The methodological details of the study have appeared elsewhere (5). We recruited 142 offspring from 93 families, ages 9—19; 131 (92%) provided cortisol samples. The subjects included offspring of parents with panic disorder, major depression, and neither condition. The participants were diagnosed with the Structured Clinical Interview for DSM-IV (SCID) (for the parents) and the Parent as Respondent Informant Schedule (6) (for the offspring). All of the subjects provided consent or assent.

CO₂ inhalation was conducted in the subjects’ homes. The subjects breathed through a face mask. Response to CO₂ was assessed with the Acute Panic Inventory (3, 5) at three time points: before the procedure (baseline), immediately before CO₂ inhalation (threat), and during CO₂ inhalation. A technician rated the occurrence of panic attacks by applying standard criteria (3); 14 of 131 (11%) met the criteria.

Cortisol samples, obtained from saliva immediately before and after CO₂ inhalation, were analyzed as described previously (3). Because the study was home based, it was not possible to standardize the time. The procedures began within 1 hour of four times: 10:00 a.m., 12:00 noon, 2:00 p.m., and 4:00 p.m.

To maintain consistency with our previous report (3), cortisol levels measured before and during CO₂ administration were treated as the dependent measures. Predictors included a within-subject repeated-measure or pre-to-postsample effect as well as time of day coded from 1 to 4 to indicate procedure initiation time, Acute Panic Inventory score, CO₂-induced panic attack, anxiety disorder in offspring, and panic disorder and/or major depression in parents. We used the SAS mixed-model procedure (7) to examine whether cortisol levels before and after CO₂ inhalation were predicted by Acute Panic Inventory scores and by panic attacks, with control for time of day and offspring or parent diagnosis. This approach allows modeling of variability due to familial factors resulting from the inclusion of multiple offspring from the same family and of variability due to repeated measures resulting from repeated cortisol samplings. These methods account for the nonindependence of data deriving from related subjects. Predictor variables in the multivariate models included fixed effects (time of day, parent with major depression, parent with panic disorder, offspring anxiety diagnosis, Acute Panic Inventory score, and repeated measure or sample) and random effects (parent and offspring). The autoregressive covariance structure was used, and denominator degrees of freedom for F or t tests on fixed and random effects were estimated by using Satterthwaite approximation, with Bonferroni correction for post hoc tests (significance was two-tailed at alpha=0.05).

The table presented online (Data Supplement 1 at http://ajp.psychiatryonline.org) shows data for offspring dichotomized by the presence or absence of CO₂-induced panic (i.e., demographic characteristics, Acute Panic Inventory ratings, diagnostic data, and offspring cortisol levels). A previous report described associations among parent diagnoses, offspring diagnoses, and offspring Acute Panic Inventory responses to CO₂(5). The prior report noted that offspring anxiety diagnosis but not parent diagnosis predicted offspring Acute Panic Inventory scores during CO₂ inhalation.

The mean of the two cortisol values was significantly higher in offspring who did versus those who did not experience CO₂-induced panic (F=5.2, df=1, 154, p<0.05) with control for the time of day, parents with panic disorder, parents with major depression, and offspring with an anxiety diagnosis—none of which predicted cortisol level. Because the model entered these variables as covariates, the noted between-group difference was independent of time of day and parent or offspring diagnosis. Adjusted cortisol values in the entire sample were significantly higher before (mean=2.0 mg/dl, SD=0.2) than after (mean=1.8 mg/dl, SD=0.2) CO₂ inhalation (F=6.0, df=1, 154, p<0.05). Cortisol values were similar in subjects with (mean=1.9 mg/d, SD=0.2) and without (mean=1.9 mg/dl, SD=0.2) parental panic disorder. No interactions with repeated measure or sample emerged for any predictor.

A second multivariate model was fit, including as predictors the two continuous Acute Panic Inventory ratings but not the categorical designation of "panic attack." This analysis found a positive association between mean cortisol level and Acute Panic Inventory scores recorded during (F=7.2, df=1, 153, p<0.01) but not before CO₂ inhalation (F=0.7, df=1, 153, p=0.42). As in the first model, time of day was controlled, but it only marginally predicted cortisol level (F=2.1, df=1, 153, p=0.10). This analysis also controlled for parental major depression, parental panic disorder, and offspring anxiety diagnosis—none of which predicted cortisol level. Again, no interactions with repeated measure or trial emerged. Thus, the relationship between mean cortisol level and Acute Panic Inventory score during CO₂ exposure was independent of time of day and parent or offspring diagnosis. Acute Panic Inventory scores at baseline showed no association with cortisol level.

Parental panic disorder, parental major depression, and offspring anxiety diagnosis were not associated with offspring cortisol level. However, consistent with prior findings (3), cortisol level was elevated in juveniles susceptible to CO₂-induced panic attacks.

Novel environments are known to increase cortisol levels. The previous CO₂ study (5) was conducted in the laboratory—a novel setting—whereas the current study occurred in the home. Consequently, the current findings cannot be explained by an interaction between the novelty of the laboratory context and the stressor. Current and previous findings document consistent associations in juveniles between elevations in cortisol levels and anxiety response to a standard stressor. Such findings are notable, given inconsistencies in previous research on developmental psychobiology of the human HPA axis (1). Current data suggest that translational research on the HPA axis in children may benefit from the use of standard stress paradigms that produce clinically relevant changes in affect and physiology.

Study limitations should be noted. First, findings are based on at-risk offspring recruited through parents seeking treatment for panic disorder or major depression, a sampling strategy vulnerable to referral biases. Second, cortisol exhibits diurnal variations, but it was not possible to standardize the procedural timing. Of importance, analyses used multivariate models that controlled for diurnal variability. Similarly, accurate assessment of baseline HPA activity requires acquisition of multiple cortisol samples over multiple days. The study examined stress-related but not baseline cortisol activity; because of habituation, intermediate pre-CO₂ cortisol value is expected to be elevated over a baseline value. Prestress cortisol level is most relevant to our study hypotheses. Because both limitations are more likely to increase type II than type I errors, they are unlikely to account for the positive findings obtained.