We compared an NIMH repository population that was screened using an online self-report to a population that was directly interviewed by clinically trained assessors, and who passed best-estimate procedures for a diagnosis. Our report demonstrates that (1) subjects without any self-reported symptoms in the screened sample were indistinguishable from the clinically interviewed control group on either neuroticism or extraversion; (2) subjects in the screened sample who met CIDI-SF criteria for anxiety disorders had lower neuroticism and higher extraversion scores than subjects who had been formally interviewed and assigned the same diagnosis, but conversely had higher neuroticism and lower extraversion scores than subjects with no symptoms (3) the difference in neuroticism and extraversion scores between subjects with and without anxiety (that is, the ‘effect size’) was greater in the clinically interviewed population. These patterns were invariant to the subject’s age, sex and race, comorbid depression or anxiety as well as to the diagnostic criteria employed.
These findings suggest that the subset of the repository sample with no discernable symptoms on the CIDI-SF may be representative (on assessed measures) of healthy controls who have never been mentally ill and have no family psychiatric history. This is an important ascertainment, as the NIMH subjects have never themselves been directly interviewed. Subjects who meet CIDI-SF criteria for anxiety disorders, however, if included in control groups (particularly when used in studies of anxiety or other disorders that co-segregate with it) could confound the studies, as we found them to deviate significantly on neuroticism and extraversion from non-ill controls. In fact, further inspection of the CIDI-SF revealed that for each covered diagnostic module depression, generalized anxiety, specific phobia, agoraphobia, obsessive-compulsive disorder and alcohol/drug dependence), subjects who either endorsed gate questions or met diagnostic criteria had significantly higher neuroticism and lower extraversion than the non-ill controls. Excluding individuals with any psychiatric symptoms from control groups, regardless of the disorder group addressed, may therefore be the most conservative approach.
Imposing such criteria would substantially reduce the available pool, which coupled with other restrictions of genetic studies (for example, matching) may necessitate trade-offs between having a larger sample versus greater homogeneity. In the case of replication studies, an increasingly vital component of psychiatric genetic research,28
the sample must also be further matched to the original for comparability. Finally, investigators should note that a significant number of subjects in the NIMH repository sample were young at the time of the screen. Because the bulk of psychiatric disorders tend to have their first onset in adolescence or early adulthood, for example,29–31
persons under 30 might optimally be excluded from non-ill control groups to minimize the likelihood of including subjects who have not passed through the age of risk and who might later develop the illness.
Our findings showed that subjects meeting CIDI-SF criteria for anxiety introduce heterogeneity into control groups. At the same time, our data also suggest that such subjects should ideally not be used to generate anxiety disorder groups, as they may represent a less severe phenotype than those meeting our rigorous interview criteria, which could possibly lead to diminished effect sizes (note the smaller difference between anxiety and control trait scores in the screened, as compared to the interviewed sample, in and ). The lower scores in the screened, vis-a-vis
the interviewed, samples were not attributable to differences in diagnostic instruments, and likely reflect intrinsic variation among the two populations (such as differences in age of onset, frequency/severity of attacks and so on). Because these variables were not assessed in the NIMH sample, we cannot examine this argument further. The lack of family history and onset information in the screened sample however bears note, as earlier age of onset has been found to be predictive of greater familial aggregation,16,17
and forms of disorders that aggregate within families may be genetically distinct from those that do not.32
The unavailability of such information may limit the ability to construct reliably homogenous and heritable disorder groups from the screened sample.
Finally, it is important that the findings of this report be interpreted cautiously. Because only two traits were assessed (there were no other comparable measures across the two samples), the findings of equivalence among the two control groups cannot be generalized, as it is possible that the populations differed substantially on other unmeasured characteristics. Furthermore, neuroticism and extraversion, despite evidence for heritability,33–35
are not genetic markers, and our findings do not address genetic
homogeneity. Third, because of the design of the interviewed sample, comparisons across disorders were restricted to panic and social anxiety. For other disorder groups with different comparative reliabilities for clinically assessed versus self-reported data, comparability may quite vary. Despite these limitations, no other study to our knowledge has directly compared the screened repository sample to a fully characterized interviewed population. The repository sample is being used and requested with increasing frequency, especially now that whole genome data are available, and this report highlights comparability across two important traits associated with psychiatric functioning.