Generalized A
Anxiety Disorder in Women
Population-Based Twin Study
Kenneth S. Kendler, MD; Michael C. Neale, PhD; Ronald C. Kessler, Andrew C. Heath, DPhil; Lindon J. Eaves, PhD, DSc
is known about the role of familial and genetic facetiology of generalized anxiety disorder (GAD), a new disorder first proposed in DSM-III. We examine this question in 1033 female-female twin pairs from a population-based registry. Both members in each twin pair were "blindly" assessed by structured psychiatric interview. Our results suggest the following: (1) GAD is a moderately familial disorder; (2) the tendency for GAD to run in families seems to be due largely or entirely to genetic factors shared between relatives rather than to the effects of the familial environment; (3) the heritability of GAD, estimated at around 30%, is modest, with the remainder of the variance in liability resulting from environmental factors not shared by adult twins; (4) the heritability of GAD cannot be explained solely by the occurrence of GAD only during episodes of major depression or panic disorder; and (5) the etiologic role of genetic factors is probably similar in GAD with a 1- vs a 6-month minimum duration of illness. (Arch Gen Psychiatry. 1992;49:267-272) \s=b\ Little
tors in the
major change in the nosology of the anxiety disorders A appeared in DSM-UP when the diagnostic category articulated of neurosis, DSM-II,2 anxiety
as
in
was
split
into panic disorder, distinguished by paroxysmal epi¬ sodes of intense anxiety, and generalized anxiety disorder (GAD), characterized by sustained "free-floating" anxi¬ ety. Since the publication of DSM-III, panic disorder has been accepted rapidly as a valid psychiatric disorder3 that has high reliability,4 aggregates in families,5 and has a distinct pattern of response to treatment.6 By contrast, GAD has remained more controversial and its validity has been questioned.3-7 Symptomatic of the uncertainty regarding GAD has been the controversies over its duration and its place in the diagnostic hierarchy. In DSM-III, the minimum dura-
Accepted
for publication May 27, 1991. From the Departments of Psychiatry (Drs Kendler and Eaves) and Human Genetics (Drs Kendler, Neale, and Eaves), Medical College of Virginia/Virginia Commonwealth University, Richmond; the Institute for Social Research, University of Michigan, Ann Arbor (Dr Kessler); and the Department of Psychiatry, Washington University School of Medicine, St Louis, Mo (Dr Heath). Reprint requests to Department of Psychiatry, Medical College of Virginia, Box 710, Richmond, VA 23298-0710 (Dr Kendler).
PhD;
tion of GAD
was set at 1 month.1 Without strong empir¬ ical support, this was increased in DSM-III-R to 6 months.8 It remains unclear whether this change has confounded or improved the validity of GAD.9 In DSM-III,1 GAD could not be diagnosed when it was "due to another mental disorder." The hierarchy rules were narrowed in DSMIII-RS where GAD cannot be diagnosed when it occurs only during the course of a mood or psychotic disorder. Familial aggregation is an accepted validating criterion for psychiatric disorders.10 Although there are numerous family and twin studies of "anxiety neurosis,"11 we know of only four that have examined a GAD-like syndrome. Cloninger et al,12 in a "blind" family study, found no ex¬ cess of anxiety disorder in relatives of probands with a GAD-like syndrome termed "other anxiety neurosis." Noyes et al,13 in a nonblind family history study, found significantly higher rates of GAD in relatives of probands with GAD vs controls. Torgersen,14 in a sample of twins from Norway treated for psychiatric illness, found a con¬ cordance rate for GAD of 0% (0/12) in monozygotic (MZ) twins and 5% (1/20) in dizygotic (DZ) twins. Andrews et al,15 in a preliminary report from a volunteer sample of Australian twins, found no significant difference in con¬ cordance rates for DSM-ffl-defined GAD in MZ vs DZ
twins. In this report, we examine the etiologic role of genetic and environmental factors in GAD as assessed at personal interview in a large sample of female twins from the Vir¬ ginia Twin Registry. We examine whether these results change with the use of a 1- vs 6-month minimum duration of illness or of different diagnostic hierarchies.
PATIENTS AND METHODS The sample and our approach to the analysis of populationbased twin data have been detailed elsewhere.16 Briefly, femalefemale twin pairs from the population-based Virginia Twin Reg¬ istry were personally interviewed by social workers blind to the status of the cotwin. Zygosity was determined blindly by stan¬ dard questions,17 and, when necessary, DNA.18 Ninety-two percent of eligible twins were successfully inter¬ viewed, 89% in person and 11% by telephone. The diagnosis of GAD was made using an adaptation of an early version of the Structured Clinical Interview for DSM-III-R Diagnosis,19 which has been shown to diagnose GAD with high reliability.20 The introductory question was modified as follows: "Thinking back over your entire life, have you ever had a time for
photographs,
Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Massachusetts User on 09/25/2015
at least one month when you have been anxious, nervous or wor¬
ried more days than not?" The remaining questions were nearly identical to the current versions of the Structured Clinical Inter¬ view for DSM-III-R, except that no question inquired whether the anxiety or worry affected two or more life circumstances. The diagnosis of GAD was based on a blind diagnostic review by one of us (K.S.K.), an experienced psychiatric diagnostician, of the interview booklet, using DSM-III-R criteria.8 On initial re¬ view, however, the duration of illness and comorbidity were ig¬ nored. The longest duration of a GAD episode was recorded, as was the judgment of whether GAD always occurred only when the twin also met diagnostic criteria for major depression or panic disorder. Comorbidity, as defined in this report, is restricted to syndromes occurring contemporaneously. A twin with a diagno¬ sis of GAD and major depression that occurred at different times would not, by this definition, be considered comorbid. We herein report only a definite or probable diagnosis of GAD, meaning that on review, it was considered that these subjects met, with little uncertainty, the relevant diagnostic criteria. Interrater reliability was assessed in 53 jointly conducted interviews. Each of these interviews was blindly reviewed by one of us (K.S.K.), and the percent agreement (and statistic)21 for the diagnosis of GAD with a 1- and 6-month minimum du¬ ration of illness was, respectively, 91% (+.77±.10) and 96% (.73±. 19). To test the "equal environment assumption" for GAD, the similarity of childhood and adult environments in the twin pairs was assessed as outlined previously.16 While probandwise concordance is presented herein, our analysis focuses on the more efficient and appropriate statistic: the tetrachoric correlation (or correlation of liability).22,23 This correlation of the latent liability to GAD and its SE were calcu¬ lated directly from the 2x2 tables cross classifying the affection status of the first and second twin in each twin pair by the com¬ puter program PRELIS.24 Genetic models were fit to these corre¬ lations by the method of asymptotic weighted least squares us¬ ing the computer program LISREL.25"28 These models postulated four sources of variance in liability to GAD: (1) additive genetic effects (A); (2) dominance genetic effects (D); (3) family or "com¬ mon" environment (C); and (4) individual-specific environment (E). The proportion of total variance in liability due to additive and dominance genetic effects and common and individualspecific environment are termed, respectively, a2, d2, c2, and e2. In situations where a 2 difference test could not choose unam¬ biguously between different models, the best fitting model was chosen using the information criterion of Akaike,29 which reflects both the goodness of fit and parsimony of the model. For further details regarding the application of biometrie genetic models to twin data, see the article by Eaves et al.17 A tetrachoric correlation fit to a 2 x 2 table is a "perfect fit" and provides no test of the liability-threshold model. However, in testing whether different definitions of GAD represent different levels of "severity" on the same liability continuum, a polychoric correlation is calculated from the table, cross tabulating the diagnoses of the two twins. 2 goodness of fit test is then available for testing the liability threshold model. The mean (±SD) age at interview of MZ twins in our sample is 1.4 years younger than DZ twins (29.5±7.5 vs 30.9±7.5 years), which, because of our large sample size, is statistically significant (i 4.02, df= 2,056, =.0001). In presenting the lifetime preva¬ lence rates and probandwise concordance for GAD in MZ and DZ twins, we remove, by regression analysis, the effect of this small difference in mean age of the two zygosity groups. Age at interview is highly and positively correlated in our sample with lifetime prevalence for GAD. Because twins are perforce of the same age, this age effect will inflate twin resemblance equally in MZ and DZ twins. In our twin model fitting, therefore, we es¬ timate the impact of age on liability to GAD so that its effect can be separated from that of the genetic and environmental factors of interest.28 The impact of type of interview (phone vs face to face) and zy¬ gosity on the risk of GAD was analyzed by logistic regression,30 controlling for the effect of age. An additional variable (interview =
status of
cotwin) was included in these analyses as an index of
"cooperation effect." The impact of environmental similarity in childhood or adulthood was assessed by logistic regression em¬ ploying a "dummy" dependent variable coded zero if both twins were concordant for affection status and one if they were discor¬ a
dant.
RESULTS
Lifetime Prevalence Of the 2163 personally interviewed twins, 509 (23.5%) met cri¬
teria for GAD with a 1-month minimum duration of illness ("1month GAD") diagnosed without hierarchies at some point in their lives. Ekminating subjects who met criteria for 1-month GAD only when they also met criteria for panic disorder or de¬ pression reduced the prevalence estimates to 22.5% and 16.7%,
respectively.
Of the interviewed twins, 127 (5.9%) met criteria for lifetime GAD with a 6-month minimum duration of illness ("6-month GAD") diagnosed without hierarchies. This figure declined to 5.7% and 3.6% if subjects meeting criteria only during episodes of panic disorder and major depression, respectively, were eliminated.
Testing for Biases We will, in this section, examine biases using the criteria of 1-month GAD diagnosed without hierarchies, as the large sam¬ ple size gives us maximal power. In our sample, the probability of a lifetime diagnosis of 1-month GAD diagnosis is strongly and positively related to age at interview ( 2 37.80, df= 1, P
Anxiety Disorder in Women
Population-Based Twin Study
Kenneth S. Kendler, MD; Michael C. Neale, PhD; Ronald C. Kessler, Andrew C. Heath, DPhil; Lindon J. Eaves, PhD, DSc
is known about the role of familial and genetic facetiology of generalized anxiety disorder (GAD), a new disorder first proposed in DSM-III. We examine this question in 1033 female-female twin pairs from a population-based registry. Both members in each twin pair were "blindly" assessed by structured psychiatric interview. Our results suggest the following: (1) GAD is a moderately familial disorder; (2) the tendency for GAD to run in families seems to be due largely or entirely to genetic factors shared between relatives rather than to the effects of the familial environment; (3) the heritability of GAD, estimated at around 30%, is modest, with the remainder of the variance in liability resulting from environmental factors not shared by adult twins; (4) the heritability of GAD cannot be explained solely by the occurrence of GAD only during episodes of major depression or panic disorder; and (5) the etiologic role of genetic factors is probably similar in GAD with a 1- vs a 6-month minimum duration of illness. (Arch Gen Psychiatry. 1992;49:267-272) \s=b\ Little
tors in the
major change in the nosology of the anxiety disorders A appeared in DSM-UP when the diagnostic category articulated of neurosis, DSM-II,2 anxiety
as
in
was
split
into panic disorder, distinguished by paroxysmal epi¬ sodes of intense anxiety, and generalized anxiety disorder (GAD), characterized by sustained "free-floating" anxi¬ ety. Since the publication of DSM-III, panic disorder has been accepted rapidly as a valid psychiatric disorder3 that has high reliability,4 aggregates in families,5 and has a distinct pattern of response to treatment.6 By contrast, GAD has remained more controversial and its validity has been questioned.3-7 Symptomatic of the uncertainty regarding GAD has been the controversies over its duration and its place in the diagnostic hierarchy. In DSM-III, the minimum dura-
Accepted
for publication May 27, 1991. From the Departments of Psychiatry (Drs Kendler and Eaves) and Human Genetics (Drs Kendler, Neale, and Eaves), Medical College of Virginia/Virginia Commonwealth University, Richmond; the Institute for Social Research, University of Michigan, Ann Arbor (Dr Kessler); and the Department of Psychiatry, Washington University School of Medicine, St Louis, Mo (Dr Heath). Reprint requests to Department of Psychiatry, Medical College of Virginia, Box 710, Richmond, VA 23298-0710 (Dr Kendler).
PhD;
tion of GAD
was set at 1 month.1 Without strong empir¬ ical support, this was increased in DSM-III-R to 6 months.8 It remains unclear whether this change has confounded or improved the validity of GAD.9 In DSM-III,1 GAD could not be diagnosed when it was "due to another mental disorder." The hierarchy rules were narrowed in DSMIII-RS where GAD cannot be diagnosed when it occurs only during the course of a mood or psychotic disorder. Familial aggregation is an accepted validating criterion for psychiatric disorders.10 Although there are numerous family and twin studies of "anxiety neurosis,"11 we know of only four that have examined a GAD-like syndrome. Cloninger et al,12 in a "blind" family study, found no ex¬ cess of anxiety disorder in relatives of probands with a GAD-like syndrome termed "other anxiety neurosis." Noyes et al,13 in a nonblind family history study, found significantly higher rates of GAD in relatives of probands with GAD vs controls. Torgersen,14 in a sample of twins from Norway treated for psychiatric illness, found a con¬ cordance rate for GAD of 0% (0/12) in monozygotic (MZ) twins and 5% (1/20) in dizygotic (DZ) twins. Andrews et al,15 in a preliminary report from a volunteer sample of Australian twins, found no significant difference in con¬ cordance rates for DSM-ffl-defined GAD in MZ vs DZ
twins. In this report, we examine the etiologic role of genetic and environmental factors in GAD as assessed at personal interview in a large sample of female twins from the Vir¬ ginia Twin Registry. We examine whether these results change with the use of a 1- vs 6-month minimum duration of illness or of different diagnostic hierarchies.
PATIENTS AND METHODS The sample and our approach to the analysis of populationbased twin data have been detailed elsewhere.16 Briefly, femalefemale twin pairs from the population-based Virginia Twin Reg¬ istry were personally interviewed by social workers blind to the status of the cotwin. Zygosity was determined blindly by stan¬ dard questions,17 and, when necessary, DNA.18 Ninety-two percent of eligible twins were successfully inter¬ viewed, 89% in person and 11% by telephone. The diagnosis of GAD was made using an adaptation of an early version of the Structured Clinical Interview for DSM-III-R Diagnosis,19 which has been shown to diagnose GAD with high reliability.20 The introductory question was modified as follows: "Thinking back over your entire life, have you ever had a time for
photographs,
Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Massachusetts User on 09/25/2015
at least one month when you have been anxious, nervous or wor¬
ried more days than not?" The remaining questions were nearly identical to the current versions of the Structured Clinical Inter¬ view for DSM-III-R, except that no question inquired whether the anxiety or worry affected two or more life circumstances. The diagnosis of GAD was based on a blind diagnostic review by one of us (K.S.K.), an experienced psychiatric diagnostician, of the interview booklet, using DSM-III-R criteria.8 On initial re¬ view, however, the duration of illness and comorbidity were ig¬ nored. The longest duration of a GAD episode was recorded, as was the judgment of whether GAD always occurred only when the twin also met diagnostic criteria for major depression or panic disorder. Comorbidity, as defined in this report, is restricted to syndromes occurring contemporaneously. A twin with a diagno¬ sis of GAD and major depression that occurred at different times would not, by this definition, be considered comorbid. We herein report only a definite or probable diagnosis of GAD, meaning that on review, it was considered that these subjects met, with little uncertainty, the relevant diagnostic criteria. Interrater reliability was assessed in 53 jointly conducted interviews. Each of these interviews was blindly reviewed by one of us (K.S.K.), and the percent agreement (and statistic)21 for the diagnosis of GAD with a 1- and 6-month minimum du¬ ration of illness was, respectively, 91% (+.77±.10) and 96% (.73±. 19). To test the "equal environment assumption" for GAD, the similarity of childhood and adult environments in the twin pairs was assessed as outlined previously.16 While probandwise concordance is presented herein, our analysis focuses on the more efficient and appropriate statistic: the tetrachoric correlation (or correlation of liability).22,23 This correlation of the latent liability to GAD and its SE were calcu¬ lated directly from the 2x2 tables cross classifying the affection status of the first and second twin in each twin pair by the com¬ puter program PRELIS.24 Genetic models were fit to these corre¬ lations by the method of asymptotic weighted least squares us¬ ing the computer program LISREL.25"28 These models postulated four sources of variance in liability to GAD: (1) additive genetic effects (A); (2) dominance genetic effects (D); (3) family or "com¬ mon" environment (C); and (4) individual-specific environment (E). The proportion of total variance in liability due to additive and dominance genetic effects and common and individualspecific environment are termed, respectively, a2, d2, c2, and e2. In situations where a 2 difference test could not choose unam¬ biguously between different models, the best fitting model was chosen using the information criterion of Akaike,29 which reflects both the goodness of fit and parsimony of the model. For further details regarding the application of biometrie genetic models to twin data, see the article by Eaves et al.17 A tetrachoric correlation fit to a 2 x 2 table is a "perfect fit" and provides no test of the liability-threshold model. However, in testing whether different definitions of GAD represent different levels of "severity" on the same liability continuum, a polychoric correlation is calculated from the table, cross tabulating the diagnoses of the two twins. 2 goodness of fit test is then available for testing the liability threshold model. The mean (±SD) age at interview of MZ twins in our sample is 1.4 years younger than DZ twins (29.5±7.5 vs 30.9±7.5 years), which, because of our large sample size, is statistically significant (i 4.02, df= 2,056, =.0001). In presenting the lifetime preva¬ lence rates and probandwise concordance for GAD in MZ and DZ twins, we remove, by regression analysis, the effect of this small difference in mean age of the two zygosity groups. Age at interview is highly and positively correlated in our sample with lifetime prevalence for GAD. Because twins are perforce of the same age, this age effect will inflate twin resemblance equally in MZ and DZ twins. In our twin model fitting, therefore, we es¬ timate the impact of age on liability to GAD so that its effect can be separated from that of the genetic and environmental factors of interest.28 The impact of type of interview (phone vs face to face) and zy¬ gosity on the risk of GAD was analyzed by logistic regression,30 controlling for the effect of age. An additional variable (interview =
status of
cotwin) was included in these analyses as an index of
"cooperation effect." The impact of environmental similarity in childhood or adulthood was assessed by logistic regression em¬ ploying a "dummy" dependent variable coded zero if both twins were concordant for affection status and one if they were discor¬ a
dant.
RESULTS
Lifetime Prevalence Of the 2163 personally interviewed twins, 509 (23.5%) met cri¬
teria for GAD with a 1-month minimum duration of illness ("1month GAD") diagnosed without hierarchies at some point in their lives. Ekminating subjects who met criteria for 1-month GAD only when they also met criteria for panic disorder or de¬ pression reduced the prevalence estimates to 22.5% and 16.7%,
respectively.
Of the interviewed twins, 127 (5.9%) met criteria for lifetime GAD with a 6-month minimum duration of illness ("6-month GAD") diagnosed without hierarchies. This figure declined to 5.7% and 3.6% if subjects meeting criteria only during episodes of panic disorder and major depression, respectively, were eliminated.
Testing for Biases We will, in this section, examine biases using the criteria of 1-month GAD diagnosed without hierarchies, as the large sam¬ ple size gives us maximal power. In our sample, the probability of a lifetime diagnosis of 1-month GAD diagnosis is strongly and positively related to age at interview ( 2 37.80, df= 1, P
