A Twin Study of the Etiology of Comorbidity: Attention-deficit Hyperactivity Disorder and Dyslexia JEFFREY W. GILGER, PH.D., BRUCE F. PENNINGTON, PH.D.,
AND
JOHN C. DEFRIES, PH.D.
Abstract. Monozygotic and dizygotic twin pairs, in which at least one member of each pair is reading disabled (RD), were assessed for attention-deficit hyperactivity disorder (ADHD). Within pair cross-concordances of the RD and ADHD qualitative diagnoses for monozygotic twins were larger than for dizygotic twins, although not significantly so (p < 0.10). Thus, the data suggest that RD and ADHD may be primarily genetically independent. However, trends in the data and subtype analyses suggest that in some cases RD and ADHD may occur together because of a shared genetic etiology and that a genetically mediated comorbid subtype may exist. J. Am. Acad. Child Adolesc. Psychiatry, 1992, 31, 2:343-348. Key Words: comorbidity, dyslexia, attention-deficit hyperactivity disorder, genetics, twins.
Comorbidity or the association of two disease states in an individual frequently occurs for psychiatric disorders. Familiar examples include the association between attention deficit hyperactivity disorder (ADHD) and learning disabilities, and depression with other psychiatric disturbances such as alcoholism (American Psychiatric Association, 1987). In addition to causing problems for diagnosis and treatment, comorbidity raises interesting questions pertaining to the etiology and mutual interdependence of the comorbid disorders. The comorbidity of ADHD and reading disabilities (RD) has been well documented (Cantwell and Baker, 1991; Cantwell and Satterfield, 1978; Holobrow and Berry, 1986; McGee et aI., 1987; Shaywitz and Shaywitz, 1988). It is possible that in some cases RD and ADHD may cooccur because of a common etiology, either genetic or environmental. A genetic contribution to both RD and ADHD, each as separate disorders, has in fact been demonstrated (Rutter et aI., 1990), but, at this time, no genetic analyses have been reported that examine the etiology of RD-ADHD comorbidity. One method to assess the genetic contribution to the comorbidity observed for any two qualitative characters is to examine the pattern of cosegregation of the two disorders in families (e.g., Pauls et aI., 1986). Basically, relatives of probands can be examined for evidence that the two diseases occur together more frequently than is expected by chance. Should the cosegregation rates differ significantly from exAccepted July 17, 1991. Drs. Gilger and Pennington are from the University of Denver. Dr. DeFries is from the University of Colorado. During preparation of this article, Dr. Gilger was supported in part by grant MH15442 through the University of Colorado. Dr. Pennington was supported by a NIMH RSDA (MH00419), project grant (MH38820), and grants from the March ofDimes (12-135) and the Orton Dyslexia Society. The twin research reported in this paper was supported in part by NICHD Grants HD 11681 and HD 27802. We thank the staffandfamilies ofthe Colorado school districts that participated in the twin study. Reprint requests to Dr. Gilger, University of Denver, Department of Psychology, 2155 S. Race Street, Denver, CO, 80208. 0890-8567/92/3102-Q343$03.00/O©1992 by the American Academy of Child and Adolescent Psychiatry. J. Am. Acad. Child Adolesc. Psychiatry, 31:2, March 1992
pectation, it is concluded that their association is due to the effects of some common etiological mechanism. Pauls and colleagues (1986) used this method to rule out a genetic basis to the comorbidity found for ADHD and Gilles de la Tourette's syndrome. The purpose of this paper is to provide the first genetic cosegregation analysis of ADHD-RD comorbidity. Although cosegregation analysis of family data is a useful technique that is currently being applied to psychiatric disorders, an underutilized and potentially more powerful method to identify the genetic and environmental components of comorbidity is the study of twins (Gilger, in press). Unlike family studies, which confound genetic and environmental influences, twin data facilitate a separate analysis of genetic and environmental effects. Therefore, in this paper monozygotic (MZ) and dizygotic (DZ) twin cross-concordances for RD and ADHD were compared. Because MZs are genetically identical, whereas DZs share only half of their segregating genes on average, a significantly larger MZ than DZ cross-concordance would provide evidence for a common genetic etiology for the observed comorbidity. Method
Subjects
Subjects in the present study are the subset of twin pairs participating in the ongoing Colorado Reading Project (CRP) (DeFries, 1985) for which complete behavioral information was available. Twin pairs in the CRP were identified from 27 cooperating school districts within a 150 mile radius of Denver, Colorado. After a pair was identified, parental permission was sought to examine the twins' school records for evidence of reading problems (e.g., low standardized test scores, referral to a reading therapist, or referrals from classroom teachers or school psychologists). For the reading disabled group, twin pairs in which at least one member of the pair had a positive school history of reading problems was identified. A sample of matched control twin pairs, where neither member of the pair had a history of reading problems, was also ascertained. All subjects received an extensive battery of psychometric tests, including subtests from the Peabody Individual Achievement Test (Dunn and Markwardt, 1970) and the Wechsler Intelligence Scales for 343
GILGER ET AL.
Children-Revised (Wechsler, 1974) or the Wechsler Adult Intelligence Scale -Revised (Wechsler, 1981). Questionnaire data covering various aspects of the twins' behavior, health, and interests was also gathered. Standard criteria were used to exclude a twin from participating in the study, including the presence of neurological or emotional problems or uncorrected visual or auditory acuity deficits, which may be responsible for their learning problems. All twins came from English-speaking, middle-class homes and range in age from 8 to 20 years. Zygosity of the twin pairs was determined using a modified version of the Nichols and Bilbro (1966) questionnaire, which has a reported accuracy of 95%. In doubtful cases, zygosity was confirmed by analysis of blood samples. After identification and testing, twins were diagnosed as RD by a discriminant function (DeFries and Gillis, 1991) that uses weighted scores from the Reading Recognition, Reading Comprehension, and Spelling subtests of the Peabody Individual Achievement Test (Dunn and Markwardt, 1970). The weights used were estimated from an earlier analysis of data from an independent sample of 140 RD and 140 non-RD children. Seventy-seven percent of those pairs having at least one member with a positive school history for reading problems also had at least one member that could be identified as RD via the discriminant score method. After the inception of the CRP, resources became available to recontact the families and obtain detailed information on subjects' medical history for various disorders (e.g., allergies, auto-immune diseases, asthma, etc.). Information was also obtained as to the presence of ADHD symptomology in the twins by means of parental responses on the Diagnostic Interview for Children and Adolescents (mCA; Herjanic et aI., 1982). The overall response rates for the medical history and DICA surveys exceeded 75% of the recontacted sample. Parental report DICA data on 81 MZ and 52 same-sex DZ pairs of the RD sample of twins are currently available. Because of the relatively low rates of RD and ADHD in the control twins, only those twin pairs originally ascertained for the reading disabled group will be used in the analyses for this paper. Opposite-sex DZ twins will not be included in the analyses in case there are contrast effects in RD or ADHD because of gender. Measures
In accordance with standard procedures, a total score was computed for the DICA attention deficit disorder scale (Herjanic et aI., 1982). The reliability of the mCA scale is reported as 0.82 (Weiner et aI., 1987). The mCA has been successfully used in a variety of clinical and research settings for identifying children and adolescents with ADHD symptomology (Biederman et aI., 1990; Dykman and Ackerman, 1991). The DICA attention deficit disorder parental interview scale consists of 21 questions requiring a "yes" response if the child manifests the behavior in question. For example, a parent is asked: "When your child is in school, does hel she have trouble sitting in his/her seat for a long time?" Because the overlap between the items of this DICA scale and the DSM-IlI-R criteria for ADHD is very substantial, it
344
seems appropriate to use the term ADHD in this paper when diagnosing subjects on the basis of the mCA. When the criteria of the mCA and DSM-III-R are compared, 12 of 14 DSM-IIl-R items are included in the mCA, and only one DICA item is not covered by DSM-IlI-R. These three noncongruent items were Numbers 11 and 13 from the DSMIII-R (interrupts, intrudes, and loses things), and Number 20 on the DICA survey (restless sleeper). Parents responded to all 21 questions of the mCA, regardless of whether or not an item was answered positively or negatively. A composite score is then computed by counting the number of positive DICA responses out of 21 questions. Thus, higher scores reflect more severe attentional deficits. As in previous research (e.g., Felton et aI., 1987), a subject is diagnosed as an ADHD proband in the present study if he or she has a DICA composite score of 6 or greater, and the behavior problems have existed since at least the second grade (or age 7). Analysis
There are a variety of methods for analyzing bi-or multivariate twin data, such as log-linear regression and various complex modeling procedures (e.g., Hannah et aI., 1983; Hopper et aI., 1990; DeFries and Fulker, 1986). However, because of their assumptions, these techniques may not be appropriate for samples selected to represent an extreme group, such as the reading disabled sample of the CRP twins. Moreover, large sample sizes are required to obtain stable and reliable parameter estimates from these procedures. Because the sample size is relatively small and because it represents a selected sample, in this paper, a simpler crossconcordance analysis will be used to assess the genetic and environmental etiologies of the comorbidity between RD and ADHD. Two kinds of analyses were performed, one that tested for a common etiology in the entire twin sample, and another that tested for an etiological subtype. In the first analysis, all MZ and DZ pairs were compared on the degree of probandwise cross-concordance observed for RD and ADHD. Specifically, the method is to select twin pairs where at least one member is RD and examine the rates of ADHD in the cotwins. A significantly larger MZ than DZ cross-concordance rate would suggest a common genetic etiology to the RD-ADHD comorbidity found in this sample. For these analyses, probandwise concordance rates are more appropriate than pair-wise, given the manner in which the CRP twins were ascertained. For the methods of calculation of probandwise rates, the reader is referred to DeFries and Gillis (1991). For the second analysis, the possibility of etiological heterogeneity for RD-ADHD comorbidity was considered by examining MZ and DZ concordances for an RD-ADHD subtype. In a two disorder model, such as RD-ADHD, there are four possible ways a person may be affected. He or she may have both disorders (++), just RD and not ADHD (+-), just ADHD and not RD (-+), or neither disorder (- -). It is possible that these subtypes are etiologically distinct. The best chance of providing support for a genetically mediated subtype of RD-ADHD comorbidity would be to J. Am. Acad. Child Adolesc. Psychiatry, 3 J: 2, March J992
TWIN STUDY OF ETIOLOGY OF COMORBIDITY
select a subset of MZ and DZ twin pairs where at least one member of the pair has both RD and ADHD (++). A comparison can then be made of the frequency of MZ and DZ cotwins also possessing the ++ condition, as opposed to the other three cotwin subtypes possible (+-, -+, - -). Again, a higher MZ than DZ concordance rate for the ++ subtype is suggestive of a genetic etiology to the cooccurrence of RD-ADHD in some individuals. Although the ++ subtype analysis seems straightforward, the results are confounded by the fact that RD and ADHD are both heritable disorders. Thus, the probability of finding larger MZ than DZ twin pair concordances for the ++ subtype is artificially elevated because pairs of twins have been selected a priori where at least one member has two heritable disorders. Some form of correction must therefore be applied to account for this artifact. Although there are a variety of ways to derive such a correction, the one used here assumes equal RD and ADHD base rates and a simple genetic model of RD-ADHD comorbidity. The simplest genetic model for the comorbid subtype is one where a single gene "causes" both RD and ADHD. (Another indistinguishable model at this level of analysis is the situation where two separate, but tightly linked, genes "cause" the comorbid subtype.) The task of this analysis is to determine whether the observed MZ-DZ concordances for the ++ subtype significantly exceed the MZ-DZ concordance rates for this subtype that would be expected by chance alone. Assuming that the four possible subtypes (+ - , -+, ++, - -) may be etiologically heterogeneous and given the available data, the best guess as to the expected chance magnitude of MZ and DZ concordances for the genetic ++ subtype are the rates observed for MZ and DZ pairs that have been ascertained through a ++ proband. That is, it is assumed that pairs have been ascertained that have the "cormorbidity" gene, although it may not be fully penetrant in all cases. The best guess of the expected concordance rates for RD and ADHD when the disorders are "caused" by independent genetic factors are the rates observed for twin pairs ascertained through probands having just one of the disorders but not both. By taking the cross-product of the these latter' 'independent" RD and ADHD concordances for MZs and DZs, an estimate of the concordance rates one would expect to see for the ++ subtype if RD and ADHD were cooccurring more often in MZs than DZs just because of their separate heritabilities and not because of the effects of a single gene could be obtained. If the observed MZ-DZ difference in concordances for the ++ subtype significantly exceeds these expectations, then the data support the conclusion of a separate genetic etiology for the ++ subtype. In fact, given that the statistical corrections were designed to be conservative, such a result would provide strong evidence for a genetically based comorbid subtype. Results RD and ADHD concordance rates. When there is a genetic basis to comorbidity, it is reasonable to expect that both disorders are significantly heritable. Whereas it is possible that two disorders observed in the general population would not show substantial heritability, even though their J. Am. Acad. Child Adolesc. Psychiatry, 31:2, March 1992
TABLE 1.
Number of MZ and DZ Pairs of Particular Cotwin and Proband Subtypes
Cotwin Proband MZ twins ++ +DZ twins ++ +-
++
+-
-+
18 7
7 21
2 11
5 = 32 0=39
3 10
10 7
10 8
1 = 24 1= 26
Note: See text for explanation of table frequencies. Affected with RD and ADHD (++), affected with RD only (+-), affected with ADHD only (- +), and not affected with either disorder (- -).
cooccurrence may be mediated by genetic factors, such situations are probably rare. Substantial heritability was in fact exhibited in this sample for both RD and ADHD considered separately. (All analyses were conducted using the discriminant function method as well as an IQ discrepant method of RD diagnosis. Because the results of both methods were essentially parallel for all analyses, only those for the discriminant method are presented here.) In this sample of MZ and DZ twin pairs, where at least one member was RD, the probandwise concordances for RD were 84% and 66%, respectively (z = 1.80; p < 0.05). When pairs were reselected for at least one member possessing ADHD, the probandwise rates for ADHD were 81 % and 29% for MZs and DZs, respectively (z = 4.44; p < 0.001). These results indicate that genetic factors playa significant role in the etiology of each of these two disorders. The heritabilities for RD and ADHD estimated from continuous data also suggest a significant amount of genetic variance. Estimates of h2g (a measure of the extent to which the deficits in reading or attention-related behaviors are heritable) obtained from continuous data were 0.50 and 0.98 for RD and ADHD, respectively (DeFries and Gillis, 1990; Gillis et al., submitted manuscript). It is worth noting that the rates of RD and ADHD were not confounded with zygosity. In twin pairs where at least one member was reading disabled, 39% of the subjects in each of the MZ and DZ samples were diagnosed as ADHD by the mCA. The rates of subjects diagnosed as having both RD and ADHD are also similar for MZs (31 %) and DZs (26%). General cross-concordances for RD and ADHD. To assess cross-concordances for MZ and DZ pairs, cotwins of RD probands were examined for the presence of ADHD. Resulting probandwise cross-concordance rates for MZ and DZ pairs were 44% and 30%, respectively (z = 1.38; p < 0.10, one-tailed). Because this MZ-DZ difference in crossconcordance approaches statistical significance, heritable influences may contribute to RD-ADHD comorbidity to some extent. Moreover, it is possible that this genetic covariance is substantial in a subset of individuals in the CRP twin sample. Subtype analysis of RD-ADHD comorbidity. As noted in the Analysis section, the examination of subtypes is important in situations where etiological heterogeneity may 345
GILGER ET AL.
underlie phenotypic comorbidity. Because the major focus of this paper is the coheritability of RD and ADHD, the differential MZ-DZ concordances for the ++ subtype is of primary interest. Table 1 presents the relevant data. The data in Table 1 are organized such that rows represent the two types of probands possible (i.e., where at least one member of each pair is RD). Although all twin pairs were originally selected for inclusion in the CRP such that at least one member had a history of RD, in a few cases, some pairs did not have RD members by way of the discriminant diagnostic scheme. These pairs have been excluded from the data in Table 1. The columns of Table 1 represent the number of pairs falling into each of the four possible RD-ADHD subtypes (++, +-, -+, and --). Thus, the first cell in Table 1 for MZs reflects the number of pairs (18) where both members were ++. Similarly, Cell 2 shows the number of pairs (7), where the cotwins of ++ probands were of the +- subtype. As can be seen in Table 1, primarily for the MZs, there is a strong tendency toward concordance for subtype. This suggests that subtype heterogeneity may have a heritable basis. In the case of ++ and +- probands, it is clear that cotwin subtype is associated with proband subtype. Specifically, the proportion of MZ pairs where both members were ++ is 0.56 (probandwise concordance rate = 0.72), whereas the percentage of pairs with +- cotwins given a ++ proband is only 0.22. These rates can be compared with the rates observed in +- proband pairs. The proportion of MZ pairs concordant for the +- subtype is 0.54 (probandwise concordance rate = 0.79), whereas the percentage of pairs with ++ cotwins in +- proband pairs is 0.18. Thus, the data suggest the presence of subtype heterogeneity in regards to the association between RD and ADHD. That is, the probability of observing a particular comorbid phenotype (e.g., ++, +-, -+, or - -) in a cotwin appears to be influenced by the comorbid phenotype through which the proband was selected (e.g., ++ or +-), such that similar proband--cotwin phenotypes are more commonly observed. Thus, subtypes of RD may differ etiologically and phenotypically depending on whether the disorder occurs in isolation or in conjunction with ADHD. Because there is evidence for heterogeneity, it is reasonable to test whether the ++ phenotypic subtype reflects a genetically comorbid condition. There were 32 MZ and 24 DZ pairs with at least one member possessing both RD and ADHD. Eighteen MZ pairs and three DZ pairs were concordant for this subtype, yielding probandwise concordances of 72% for MZs and 22% for DZs (z = 3.70, p < 0.001). As was explained in the Analysis section, a correction must be applied before drawing the conclusion that the significantly greater MZ concordance rate is indicative of a genetically mediated comorbid subtype. This was done by taking the cross-products ofthe RD and ADHD concordance rates and using these cross-products as an estimate of the rates to be expected if the comorbidity observed is not due to a common genetic mechanism. Before taking the crossproducts, the RD and ADHD concordance rates were first adjusted by removing the ++ concordant pairs from the analysis. Given the conservativeness of this correction, 346
should a significant MZ-DZ concordance difference remain, strong evidence for an RD-ADHD comorbid subtype of genetic etiology would be provided. Removing the 18 MZ and three DZ ++ concordant pairs and recalculating concordance rates yielded probandwise values for RD of 0.76 and 0.63 for MZs and DZs, respectively. Rates for ADHD were 0.57 for MZs and 0.17 for DZs. Thus, assuming etiological heterogeneity, the expected ++ probandwise concordance rate for MZs, if the two disorders are not coheritable, is 0.43. For DZs, this expected rate is 0.11. A test of the MZ-DZ ++ concordance difference, after correcting for these expected rates yielded nearly statistically significant results (z = 1.33; P < 0.10, one-tailed). Discussion Twin data have been presented that examine the etiology ofRD-ADHD comorbidity, using qualitative diagnoses. Using conventional levels of statistical significance, the data suggest that RD and ADHD may cooccur because of reasons other than a common genetic etiology. However, patterns in the data suggest that in some cases the RD-ADHD comorbidity observed in children may be due, in part, to heritable influences. Although the results of this study suggest that RD and ADHD may not be purely genetically isomorphic disorders, this conclusion must be tempered for the following reasons. First, these data are preliminary in the sense that they are based on a relatively small sample, albeit the largest one reported on this topic to date. Larger sample sizes will facilitate more complex analyses that may yield additional data relevant to the relationship between RD and ADHD. That the general cross-concordance and the ++ subtype analyses yielded results approaching conventional significance levels suggest that there may be a subset of individuals with both RD and ADHD because of a common genetic etiology. Second, the method of diagnosing ADHD solely on the basis of the DICA may be subject to criticism because it relies on single versus multiple informants, and it is not confirmed by objective measures of attention and other ADHD symptoms. Again, these criticisms can be empirically examined using more extensive interview and objective measures of ADHD, which are currently being obtained. Finally, the subtype analyses are preliminary. There may be additional subtypes within the comorbid subtype, and only in some may RD and ADHD co-occur because of a common genetic etiology. As the sample size increases, more detailed subtype analyses can be conducted. The twin method listed in this paper can also be applied to the study of comorbidity found for other psychiatric and medical disorders (e.g., Gilger, in press; Gilger et aI., in press). When working with these types of data, it is important to consider alternative explanations for comorbidity other than a common genetic etiology, especially when a significantly greater MZ than DZ cross-concordance is found. Because comorbidity is so important in psychiatry, a more detailed discussion of the relevance of behavioral genetic studies to this issue is given below. It is noteworthy that this discussion not only applies to twin data but to family data as well. J. Am. Acad. Child Adolesc. Psychiatry, 3 J :2, March 1992
TWIN STUDY OF ETIOLOGY OF COMORBIDITY
Other than the mechanism of common genetic etiology, there are three additional alternative explanations for comorbidity: 1) RD leads to ADHD as a secondary symptom; 2) ADHD leads to RD as a secondary symptom; or 3) the observed association is an artifact. Again, it is likely that comorbid cases are etiologically heterogeneous; where in some, ADHD is secondary to RD; in others, RD and ADHD share a common etiology; and, in still others, the observed cooccurrence of ADHD and RD is due to chance. This having been said, these three additional explanatory mechanisms can be described as follows. Explanations 1 and 2 would suggest that having one disorder is a necessary (but it need not be a sufficient) condition for the other disorder. There is in fact evidence that at least some cases of ADHD may represent secondary symptoms ofRD (e.g., Cantwell and Baker, 1991; Shaywitz and Shaywitz, 1988). For example, the prevalence of comorbid ADHD and RD depends on the method of subject ascertainment. If subjects are selected on the basis of having RD, about 33% are also ADHD, but when subjects are first selected for ADHD, approximately 11 % are RD as well (Shaywitz and Shaywitz, 1988). This pattern of association suggests that RD may somehow facilitate the expression of ADHD symptoms, rather than ADHD being instrumental in the development of RD. Additional support for Explanation 1 and the conclusion that RD-ADHD comorbidity reflects little or no genetic covariance was found in two other studies conducted in Colorado. One study was a multiple regression analysis of the CRP twins, where the heritability of ADHD was assessed in terms of a continuous dimension, rather than as a dichotomous diagnosis (Gillis et aI., in press). Partialling out the variance resulting from IQ and/or reading ability-disability in these analyses had essentially no effect on the heritability estimates for ADHD. These analyses in a sense "controlled" for shared genetic variance between RD and ADHD. Finding little subsequent change in the heritability of ADHD suggests that there is minimal genetic covariance between RD and ADHD in this twin sample. In another study where data based on a separate nontwin sample was examined, additional evidence favoring the secondary symptom hypothesis (i.e., RD leads to ADHD) was found using a neuropsychological approach (Penningtonet aI., submitted manuscript). The third explanatory mechanism suggests that the observed RD-ADHD association may be artifactual because of sample selection methods (e.g., Berkson's bias, Berkson, 1946) or perhaps definitional overlap. Selection biases are particularly prevalent in clinically ascertained samples, where there is a greater tendency for the most severely affected children to seek care. Thus, the high rates of RD-ADHD comorbidity often reported may simply reflect artificially inflated statistical estimates of this association, rather than some causal (Explanations I and 2) or etiological (genetic) interrelatedness of RD and ADHD. Definitional and diagnostic overlap or ambiguity may also contribute to the observed rates of RD-ADHD comorbidity. For example, the symptoms of both RD and ADHD were once considered part of the minimal brain damage syndrome (Strauss and Lehtinen, 1947). Moreover, such symptoms as inattenJ. Am. Acad. Child Adolesc. Psychiatry, 31:2, March 1992
tiveness in the classroom and poor school performance are still considered by some to be hallmarks of both RD and ADHD. To the degree that RD and ADHD share diagnostic criteria, comorbidity may be inflated irrespective of whether or not the two disorders share a common neurological substrate, are causally related, or have a common etiology. Whereas each of these possible mechanisms responsible for comorbidity is distinct, they can lead to very similar patterns of results in twin cross-concordance or family cosegregation studies. If, for example, ADHD were secondary to RD in some percentage of RD cases in the population, a larger MZ than DZ RD-ADHD cross-concordance rate is expected, similar to the rate expected if the comorbidity were genetic in origin. Given that RD is heritable, more MZ than DZ cotwins will have RD, and, given that RD is expected to lead to ADHD in an equal number of MZ and DZ twins, more MZ than DZ cotwins will also exhibit ADHD. Thus, at first glance, the twin data would falsely suggest a genetic etiology to comorbidity when none may exist. The causal basis for a greater MZ than DZ cross-concordance could be elucidated by a more detailed examination of the cotwin data. If the MZ-DZ difference in cross-concordance was due to secondary factors, then a large proportion of ADHD cotwins should also have the primary disorder of RD. On the other hand, if the comorbidity and RD-ADHD cross-concordance was due to genetic factors, one would expect cotwins to exhibit ADHD even when they do not have RD (e.g., ADHD is in a sense a variable expression of the RD gene). Of course, the distinction between genetic and secondary factors in comorbidity becomes cloudy as the association between the primary disorder and the secondary disorder increases. In those cases where the primary disorder has directly led to the secondary disorder, two disorders in a sense share the same genetic basis, although one disorder may appear later in development than the other. It is also noteworthy that the potential confusion between secondary and common etiological effects does not exist when examining a comorbid subtype as was done in the second set of analyses presented in this paper. The subtype test is a robust yet conservative measure of whether a genetically based comorbid subtype exists in the twin sample. In the case of Explanation 3, where comorbidity is artifactual, there is little reason to expect to observe differences in the MZ and DZ rates. Artifact effects are probably best identified after the other possible mechanisms have been dismissed. Another way to test an artifact hypothesis in twin data or family data ascertained through a clinic or special referral is to compare the comorbidity rates in probands (or the person through which the pair or family was identified) with the comorbidity rates observed in their cotwins or relatives. Artifacts of sampling would be suggested by higher comorbidity rates in the probands than in the cotwins or relatives. In the case of twins, such a comparison is best performed on the DZ pairs. Because MZs are genetically identical, probands and cotwins may be very similar even if a selection bias is operating. In spite of the intricacies involved, the results of this study begin to address why RD and ADHD cooccur in children and suggest a method by which other comorbid disorders in 347
GILGER ET AL.
psychiatry may be studied. Such studies would help clarify the reliability, validity, and usefulness of combining primary and secondary diagnoses such as RD and ADHD in research and clinical practice. For example, it is clinically important to recognize that RD-ADHD comorbidity may not simply be a complex phenotype of a single biological or neurological mechanism in all individuals. This suggests that interventions may need to be directed at the various symptoms and causes of each "disorder"; treating comorbid RD and ADHD the same as ADHD by itself (e.g., with Ritalin®) may be ineffective, because many of these children may need tutoring for their primary RD. References American Psychiatric Association (1987), Diagnostic and statistical manual of mental disorders (3rd edition-Revised). Washington, D.C.: APA Press. Biederman, J., Faraone, S. V., Keenan, K, Knee, D. & Tsuang, M. T. (1990), Family-genetic and psychosocial risk factors in DSM-I1I attention deficit disorder. J. Am. Acad. Child Adolesc. Psychiatry, 29:526-533. Berkson, J. (1946), Limitations of the application of fourfold table analysis to hospital data. Biometrics, 2:47-51. Cantwell, D. P. & Satterfield, J. (1978), The prevalence of academic underachievement in hyperactive children. J. of Pediatr. Psychol., 3:168-171. - - Baker, L. (1991), Association between attention deficit-hyperactivity disorder and learning disorders. Journal ofLearning Disabilities, 24:88-95. DeFries, J. C. (1985), Colorado Reading Project. In: Biobehavioral Measures of Dyslexia, eds. D. B. Gray & J. F. Kavanagh. Parkton, MD: York Press, pp. 107-122. - - Fulker, D. W. (Eds.) (1986), Special issue: multivariate behavioral genetics and development. In: Behav. Genet., 16: 1-235. - - Gillis, J. J. (1991). Etiology of reading deficits in learning disabilities: quantitative genetic analysis. In: Advances in the Neuropsychology of Learning Disabilities: Issues, Methods, & Practices, eds. J. E. Obrzut & G. W. Hynd. Orlando, FL: Academic Press pp. 29-47. Dunn, L. M. & Markwardt, F. c., Jr. (1970), Manualfor the Peabody Individual Achievement Test. Circle Pine, MN: American Guidance Service. Dykman, R. A. & Ackerman, P. T. (1991), Attention deficit disorder and specific reading disability: Separate but often overlapping disorders. Journal of Learning Disabilities, 24:96-103. Felton, R. H., Wood, F. B., Brown, I. B., Campbell, J. & Harter, M.
348
R. (1987), Separate verbal memory and naming deficits in attention deficit disorder and reading disability. Brain and Language, 31:171-184. Gilger, J. W. (in press), A comment on twin analyses of comorbidity in psychiatry. Am. J. Psychiatry. - - Pennington, B. F., Green, P., Smith, S. D. & Smith, S. A. (1992), Dyslexia immune disorders, and left-handedness: twin and family studies of their relations. Neuropsychologia. (in press). Gillis, J. J., Gilger, J. W., Pennington, B. F. & DeFries, J. C. (1992), Attention-deficit hyperactivity disorder in reading disabled twins: evidence for a genetic etiology. J. Abnorm. Child Psycho/. (in press). Hannah, M. c., Hopper, J. L. & Mathews, J. D. (1983), Twin concordance for a binary trait: I. Statistical methods illustrated with data on drinking status. Acta Genet. Med. Gemello/., 32:127-137. Herjanic, B., Campbell, 1. & Reich, W. (1982), Development of a structured psychiatric interview for children: agreement between child and parent on individual symptoms. J. Abnorm. Child Psycho/., 10:307-324. Holobrow, P. & Berry, P. (1986), A multinational, cross-cultural perspective on hyperactivity. Am. J. Orthopsych. Assoc., 56:320-322. Hopper, J. L., Hannah, M. C., Macaskill, G. T. & Mathews, J. D. (1990), Twin concordance for a binary trait: III. A bivariate analysis of hay fever and asthma. Genet. Epidemio/., 7:277-289. McGee, R., Williams, S. & Silva, P. A. (1987), A comparison of girls and boys with teacher-identified problems of attention. J. Am. Acad. Child Adolesc. Psychiatry, 26:711-714. Nichols, R. C. & Bilbro, W. C. (1966), The diagnosis of twin zygosity. Acta Genet. Med. Gemello/., 16:265-275. Pauls, D. L., Hurst, C. R., Kruger, S. D., Leckman, S. D., Kidd, K & Cohen, D. J. (1986), Gilles de al Tourette's Syndrome and attention deficit disorder with hyperactivity: evidence against a genetic relationship. Arch. Gen. Psychiatry, 43:1177-1179. Rutter, M., Macdonald, H., Couteur, A. L., Harrington, R., Bolton, P. & Bailey, A. (1990), Genetic factors in child psychiatric disorders. II: Empirical findings. J. Child Psycho/. Psychiatry, 31 :39-83. Shaywitz, S. E. & Shaywitz, B. E. (1988), Attention deficit disorder: current perspectives. In: Learning Disabilities: Proceedings of the National Conference, eds, J. F. Kavanaugh & T. J. Tress. Parkton, MD: York Press, pp. 369-523. Strauss, A. & Lehtinen, L. (1947), Psychopathology and Education in the Brain Injured Child. New York: Grune & Stratton. Weiner, Z., Reich, W., Herjanic, B., Jung, K G. & Amado, H. (1987), Reliability, validity, and parent-ehild agreement studies of the Diagnostic Interview for Children and Adolescents (DICA). J. Am. Acad. Child Adolesc. Psychiatry, 26:649-653. Wechsler, D. (1974), Examiner's Manual: Wechsler Intelligence Scale for Children-Revised. New York: The Psychological Corporation. - - (1981), Examiner's Manual: Wechsler Intelligence Adult ScaleRevised. New York: The Psychological Corporation.
J. Am. Acad. Child Adolesc. Psychiatry, 3 I :2, March 1992
AND
JOHN C. DEFRIES, PH.D.
Abstract. Monozygotic and dizygotic twin pairs, in which at least one member of each pair is reading disabled (RD), were assessed for attention-deficit hyperactivity disorder (ADHD). Within pair cross-concordances of the RD and ADHD qualitative diagnoses for monozygotic twins were larger than for dizygotic twins, although not significantly so (p < 0.10). Thus, the data suggest that RD and ADHD may be primarily genetically independent. However, trends in the data and subtype analyses suggest that in some cases RD and ADHD may occur together because of a shared genetic etiology and that a genetically mediated comorbid subtype may exist. J. Am. Acad. Child Adolesc. Psychiatry, 1992, 31, 2:343-348. Key Words: comorbidity, dyslexia, attention-deficit hyperactivity disorder, genetics, twins.
Comorbidity or the association of two disease states in an individual frequently occurs for psychiatric disorders. Familiar examples include the association between attention deficit hyperactivity disorder (ADHD) and learning disabilities, and depression with other psychiatric disturbances such as alcoholism (American Psychiatric Association, 1987). In addition to causing problems for diagnosis and treatment, comorbidity raises interesting questions pertaining to the etiology and mutual interdependence of the comorbid disorders. The comorbidity of ADHD and reading disabilities (RD) has been well documented (Cantwell and Baker, 1991; Cantwell and Satterfield, 1978; Holobrow and Berry, 1986; McGee et aI., 1987; Shaywitz and Shaywitz, 1988). It is possible that in some cases RD and ADHD may cooccur because of a common etiology, either genetic or environmental. A genetic contribution to both RD and ADHD, each as separate disorders, has in fact been demonstrated (Rutter et aI., 1990), but, at this time, no genetic analyses have been reported that examine the etiology of RD-ADHD comorbidity. One method to assess the genetic contribution to the comorbidity observed for any two qualitative characters is to examine the pattern of cosegregation of the two disorders in families (e.g., Pauls et aI., 1986). Basically, relatives of probands can be examined for evidence that the two diseases occur together more frequently than is expected by chance. Should the cosegregation rates differ significantly from exAccepted July 17, 1991. Drs. Gilger and Pennington are from the University of Denver. Dr. DeFries is from the University of Colorado. During preparation of this article, Dr. Gilger was supported in part by grant MH15442 through the University of Colorado. Dr. Pennington was supported by a NIMH RSDA (MH00419), project grant (MH38820), and grants from the March ofDimes (12-135) and the Orton Dyslexia Society. The twin research reported in this paper was supported in part by NICHD Grants HD 11681 and HD 27802. We thank the staffandfamilies ofthe Colorado school districts that participated in the twin study. Reprint requests to Dr. Gilger, University of Denver, Department of Psychology, 2155 S. Race Street, Denver, CO, 80208. 0890-8567/92/3102-Q343$03.00/O©1992 by the American Academy of Child and Adolescent Psychiatry. J. Am. Acad. Child Adolesc. Psychiatry, 31:2, March 1992
pectation, it is concluded that their association is due to the effects of some common etiological mechanism. Pauls and colleagues (1986) used this method to rule out a genetic basis to the comorbidity found for ADHD and Gilles de la Tourette's syndrome. The purpose of this paper is to provide the first genetic cosegregation analysis of ADHD-RD comorbidity. Although cosegregation analysis of family data is a useful technique that is currently being applied to psychiatric disorders, an underutilized and potentially more powerful method to identify the genetic and environmental components of comorbidity is the study of twins (Gilger, in press). Unlike family studies, which confound genetic and environmental influences, twin data facilitate a separate analysis of genetic and environmental effects. Therefore, in this paper monozygotic (MZ) and dizygotic (DZ) twin cross-concordances for RD and ADHD were compared. Because MZs are genetically identical, whereas DZs share only half of their segregating genes on average, a significantly larger MZ than DZ cross-concordance would provide evidence for a common genetic etiology for the observed comorbidity. Method
Subjects
Subjects in the present study are the subset of twin pairs participating in the ongoing Colorado Reading Project (CRP) (DeFries, 1985) for which complete behavioral information was available. Twin pairs in the CRP were identified from 27 cooperating school districts within a 150 mile radius of Denver, Colorado. After a pair was identified, parental permission was sought to examine the twins' school records for evidence of reading problems (e.g., low standardized test scores, referral to a reading therapist, or referrals from classroom teachers or school psychologists). For the reading disabled group, twin pairs in which at least one member of the pair had a positive school history of reading problems was identified. A sample of matched control twin pairs, where neither member of the pair had a history of reading problems, was also ascertained. All subjects received an extensive battery of psychometric tests, including subtests from the Peabody Individual Achievement Test (Dunn and Markwardt, 1970) and the Wechsler Intelligence Scales for 343
GILGER ET AL.
Children-Revised (Wechsler, 1974) or the Wechsler Adult Intelligence Scale -Revised (Wechsler, 1981). Questionnaire data covering various aspects of the twins' behavior, health, and interests was also gathered. Standard criteria were used to exclude a twin from participating in the study, including the presence of neurological or emotional problems or uncorrected visual or auditory acuity deficits, which may be responsible for their learning problems. All twins came from English-speaking, middle-class homes and range in age from 8 to 20 years. Zygosity of the twin pairs was determined using a modified version of the Nichols and Bilbro (1966) questionnaire, which has a reported accuracy of 95%. In doubtful cases, zygosity was confirmed by analysis of blood samples. After identification and testing, twins were diagnosed as RD by a discriminant function (DeFries and Gillis, 1991) that uses weighted scores from the Reading Recognition, Reading Comprehension, and Spelling subtests of the Peabody Individual Achievement Test (Dunn and Markwardt, 1970). The weights used were estimated from an earlier analysis of data from an independent sample of 140 RD and 140 non-RD children. Seventy-seven percent of those pairs having at least one member with a positive school history for reading problems also had at least one member that could be identified as RD via the discriminant score method. After the inception of the CRP, resources became available to recontact the families and obtain detailed information on subjects' medical history for various disorders (e.g., allergies, auto-immune diseases, asthma, etc.). Information was also obtained as to the presence of ADHD symptomology in the twins by means of parental responses on the Diagnostic Interview for Children and Adolescents (mCA; Herjanic et aI., 1982). The overall response rates for the medical history and DICA surveys exceeded 75% of the recontacted sample. Parental report DICA data on 81 MZ and 52 same-sex DZ pairs of the RD sample of twins are currently available. Because of the relatively low rates of RD and ADHD in the control twins, only those twin pairs originally ascertained for the reading disabled group will be used in the analyses for this paper. Opposite-sex DZ twins will not be included in the analyses in case there are contrast effects in RD or ADHD because of gender. Measures
In accordance with standard procedures, a total score was computed for the DICA attention deficit disorder scale (Herjanic et aI., 1982). The reliability of the mCA scale is reported as 0.82 (Weiner et aI., 1987). The mCA has been successfully used in a variety of clinical and research settings for identifying children and adolescents with ADHD symptomology (Biederman et aI., 1990; Dykman and Ackerman, 1991). The DICA attention deficit disorder parental interview scale consists of 21 questions requiring a "yes" response if the child manifests the behavior in question. For example, a parent is asked: "When your child is in school, does hel she have trouble sitting in his/her seat for a long time?" Because the overlap between the items of this DICA scale and the DSM-IlI-R criteria for ADHD is very substantial, it
344
seems appropriate to use the term ADHD in this paper when diagnosing subjects on the basis of the mCA. When the criteria of the mCA and DSM-III-R are compared, 12 of 14 DSM-IIl-R items are included in the mCA, and only one DICA item is not covered by DSM-IlI-R. These three noncongruent items were Numbers 11 and 13 from the DSMIII-R (interrupts, intrudes, and loses things), and Number 20 on the DICA survey (restless sleeper). Parents responded to all 21 questions of the mCA, regardless of whether or not an item was answered positively or negatively. A composite score is then computed by counting the number of positive DICA responses out of 21 questions. Thus, higher scores reflect more severe attentional deficits. As in previous research (e.g., Felton et aI., 1987), a subject is diagnosed as an ADHD proband in the present study if he or she has a DICA composite score of 6 or greater, and the behavior problems have existed since at least the second grade (or age 7). Analysis
There are a variety of methods for analyzing bi-or multivariate twin data, such as log-linear regression and various complex modeling procedures (e.g., Hannah et aI., 1983; Hopper et aI., 1990; DeFries and Fulker, 1986). However, because of their assumptions, these techniques may not be appropriate for samples selected to represent an extreme group, such as the reading disabled sample of the CRP twins. Moreover, large sample sizes are required to obtain stable and reliable parameter estimates from these procedures. Because the sample size is relatively small and because it represents a selected sample, in this paper, a simpler crossconcordance analysis will be used to assess the genetic and environmental etiologies of the comorbidity between RD and ADHD. Two kinds of analyses were performed, one that tested for a common etiology in the entire twin sample, and another that tested for an etiological subtype. In the first analysis, all MZ and DZ pairs were compared on the degree of probandwise cross-concordance observed for RD and ADHD. Specifically, the method is to select twin pairs where at least one member is RD and examine the rates of ADHD in the cotwins. A significantly larger MZ than DZ cross-concordance rate would suggest a common genetic etiology to the RD-ADHD comorbidity found in this sample. For these analyses, probandwise concordance rates are more appropriate than pair-wise, given the manner in which the CRP twins were ascertained. For the methods of calculation of probandwise rates, the reader is referred to DeFries and Gillis (1991). For the second analysis, the possibility of etiological heterogeneity for RD-ADHD comorbidity was considered by examining MZ and DZ concordances for an RD-ADHD subtype. In a two disorder model, such as RD-ADHD, there are four possible ways a person may be affected. He or she may have both disorders (++), just RD and not ADHD (+-), just ADHD and not RD (-+), or neither disorder (- -). It is possible that these subtypes are etiologically distinct. The best chance of providing support for a genetically mediated subtype of RD-ADHD comorbidity would be to J. Am. Acad. Child Adolesc. Psychiatry, 3 J: 2, March J992
TWIN STUDY OF ETIOLOGY OF COMORBIDITY
select a subset of MZ and DZ twin pairs where at least one member of the pair has both RD and ADHD (++). A comparison can then be made of the frequency of MZ and DZ cotwins also possessing the ++ condition, as opposed to the other three cotwin subtypes possible (+-, -+, - -). Again, a higher MZ than DZ concordance rate for the ++ subtype is suggestive of a genetic etiology to the cooccurrence of RD-ADHD in some individuals. Although the ++ subtype analysis seems straightforward, the results are confounded by the fact that RD and ADHD are both heritable disorders. Thus, the probability of finding larger MZ than DZ twin pair concordances for the ++ subtype is artificially elevated because pairs of twins have been selected a priori where at least one member has two heritable disorders. Some form of correction must therefore be applied to account for this artifact. Although there are a variety of ways to derive such a correction, the one used here assumes equal RD and ADHD base rates and a simple genetic model of RD-ADHD comorbidity. The simplest genetic model for the comorbid subtype is one where a single gene "causes" both RD and ADHD. (Another indistinguishable model at this level of analysis is the situation where two separate, but tightly linked, genes "cause" the comorbid subtype.) The task of this analysis is to determine whether the observed MZ-DZ concordances for the ++ subtype significantly exceed the MZ-DZ concordance rates for this subtype that would be expected by chance alone. Assuming that the four possible subtypes (+ - , -+, ++, - -) may be etiologically heterogeneous and given the available data, the best guess as to the expected chance magnitude of MZ and DZ concordances for the genetic ++ subtype are the rates observed for MZ and DZ pairs that have been ascertained through a ++ proband. That is, it is assumed that pairs have been ascertained that have the "cormorbidity" gene, although it may not be fully penetrant in all cases. The best guess of the expected concordance rates for RD and ADHD when the disorders are "caused" by independent genetic factors are the rates observed for twin pairs ascertained through probands having just one of the disorders but not both. By taking the cross-product of the these latter' 'independent" RD and ADHD concordances for MZs and DZs, an estimate of the concordance rates one would expect to see for the ++ subtype if RD and ADHD were cooccurring more often in MZs than DZs just because of their separate heritabilities and not because of the effects of a single gene could be obtained. If the observed MZ-DZ difference in concordances for the ++ subtype significantly exceeds these expectations, then the data support the conclusion of a separate genetic etiology for the ++ subtype. In fact, given that the statistical corrections were designed to be conservative, such a result would provide strong evidence for a genetically based comorbid subtype. Results RD and ADHD concordance rates. When there is a genetic basis to comorbidity, it is reasonable to expect that both disorders are significantly heritable. Whereas it is possible that two disorders observed in the general population would not show substantial heritability, even though their J. Am. Acad. Child Adolesc. Psychiatry, 31:2, March 1992
TABLE 1.
Number of MZ and DZ Pairs of Particular Cotwin and Proband Subtypes
Cotwin Proband MZ twins ++ +DZ twins ++ +-
++
+-
-+
18 7
7 21
2 11
5 = 32 0=39
3 10
10 7
10 8
1 = 24 1= 26
Note: See text for explanation of table frequencies. Affected with RD and ADHD (++), affected with RD only (+-), affected with ADHD only (- +), and not affected with either disorder (- -).
cooccurrence may be mediated by genetic factors, such situations are probably rare. Substantial heritability was in fact exhibited in this sample for both RD and ADHD considered separately. (All analyses were conducted using the discriminant function method as well as an IQ discrepant method of RD diagnosis. Because the results of both methods were essentially parallel for all analyses, only those for the discriminant method are presented here.) In this sample of MZ and DZ twin pairs, where at least one member was RD, the probandwise concordances for RD were 84% and 66%, respectively (z = 1.80; p < 0.05). When pairs were reselected for at least one member possessing ADHD, the probandwise rates for ADHD were 81 % and 29% for MZs and DZs, respectively (z = 4.44; p < 0.001). These results indicate that genetic factors playa significant role in the etiology of each of these two disorders. The heritabilities for RD and ADHD estimated from continuous data also suggest a significant amount of genetic variance. Estimates of h2g (a measure of the extent to which the deficits in reading or attention-related behaviors are heritable) obtained from continuous data were 0.50 and 0.98 for RD and ADHD, respectively (DeFries and Gillis, 1990; Gillis et al., submitted manuscript). It is worth noting that the rates of RD and ADHD were not confounded with zygosity. In twin pairs where at least one member was reading disabled, 39% of the subjects in each of the MZ and DZ samples were diagnosed as ADHD by the mCA. The rates of subjects diagnosed as having both RD and ADHD are also similar for MZs (31 %) and DZs (26%). General cross-concordances for RD and ADHD. To assess cross-concordances for MZ and DZ pairs, cotwins of RD probands were examined for the presence of ADHD. Resulting probandwise cross-concordance rates for MZ and DZ pairs were 44% and 30%, respectively (z = 1.38; p < 0.10, one-tailed). Because this MZ-DZ difference in crossconcordance approaches statistical significance, heritable influences may contribute to RD-ADHD comorbidity to some extent. Moreover, it is possible that this genetic covariance is substantial in a subset of individuals in the CRP twin sample. Subtype analysis of RD-ADHD comorbidity. As noted in the Analysis section, the examination of subtypes is important in situations where etiological heterogeneity may 345
GILGER ET AL.
underlie phenotypic comorbidity. Because the major focus of this paper is the coheritability of RD and ADHD, the differential MZ-DZ concordances for the ++ subtype is of primary interest. Table 1 presents the relevant data. The data in Table 1 are organized such that rows represent the two types of probands possible (i.e., where at least one member of each pair is RD). Although all twin pairs were originally selected for inclusion in the CRP such that at least one member had a history of RD, in a few cases, some pairs did not have RD members by way of the discriminant diagnostic scheme. These pairs have been excluded from the data in Table 1. The columns of Table 1 represent the number of pairs falling into each of the four possible RD-ADHD subtypes (++, +-, -+, and --). Thus, the first cell in Table 1 for MZs reflects the number of pairs (18) where both members were ++. Similarly, Cell 2 shows the number of pairs (7), where the cotwins of ++ probands were of the +- subtype. As can be seen in Table 1, primarily for the MZs, there is a strong tendency toward concordance for subtype. This suggests that subtype heterogeneity may have a heritable basis. In the case of ++ and +- probands, it is clear that cotwin subtype is associated with proband subtype. Specifically, the proportion of MZ pairs where both members were ++ is 0.56 (probandwise concordance rate = 0.72), whereas the percentage of pairs with +- cotwins given a ++ proband is only 0.22. These rates can be compared with the rates observed in +- proband pairs. The proportion of MZ pairs concordant for the +- subtype is 0.54 (probandwise concordance rate = 0.79), whereas the percentage of pairs with ++ cotwins in +- proband pairs is 0.18. Thus, the data suggest the presence of subtype heterogeneity in regards to the association between RD and ADHD. That is, the probability of observing a particular comorbid phenotype (e.g., ++, +-, -+, or - -) in a cotwin appears to be influenced by the comorbid phenotype through which the proband was selected (e.g., ++ or +-), such that similar proband--cotwin phenotypes are more commonly observed. Thus, subtypes of RD may differ etiologically and phenotypically depending on whether the disorder occurs in isolation or in conjunction with ADHD. Because there is evidence for heterogeneity, it is reasonable to test whether the ++ phenotypic subtype reflects a genetically comorbid condition. There were 32 MZ and 24 DZ pairs with at least one member possessing both RD and ADHD. Eighteen MZ pairs and three DZ pairs were concordant for this subtype, yielding probandwise concordances of 72% for MZs and 22% for DZs (z = 3.70, p < 0.001). As was explained in the Analysis section, a correction must be applied before drawing the conclusion that the significantly greater MZ concordance rate is indicative of a genetically mediated comorbid subtype. This was done by taking the cross-products ofthe RD and ADHD concordance rates and using these cross-products as an estimate of the rates to be expected if the comorbidity observed is not due to a common genetic mechanism. Before taking the crossproducts, the RD and ADHD concordance rates were first adjusted by removing the ++ concordant pairs from the analysis. Given the conservativeness of this correction, 346
should a significant MZ-DZ concordance difference remain, strong evidence for an RD-ADHD comorbid subtype of genetic etiology would be provided. Removing the 18 MZ and three DZ ++ concordant pairs and recalculating concordance rates yielded probandwise values for RD of 0.76 and 0.63 for MZs and DZs, respectively. Rates for ADHD were 0.57 for MZs and 0.17 for DZs. Thus, assuming etiological heterogeneity, the expected ++ probandwise concordance rate for MZs, if the two disorders are not coheritable, is 0.43. For DZs, this expected rate is 0.11. A test of the MZ-DZ ++ concordance difference, after correcting for these expected rates yielded nearly statistically significant results (z = 1.33; P < 0.10, one-tailed). Discussion Twin data have been presented that examine the etiology ofRD-ADHD comorbidity, using qualitative diagnoses. Using conventional levels of statistical significance, the data suggest that RD and ADHD may cooccur because of reasons other than a common genetic etiology. However, patterns in the data suggest that in some cases the RD-ADHD comorbidity observed in children may be due, in part, to heritable influences. Although the results of this study suggest that RD and ADHD may not be purely genetically isomorphic disorders, this conclusion must be tempered for the following reasons. First, these data are preliminary in the sense that they are based on a relatively small sample, albeit the largest one reported on this topic to date. Larger sample sizes will facilitate more complex analyses that may yield additional data relevant to the relationship between RD and ADHD. That the general cross-concordance and the ++ subtype analyses yielded results approaching conventional significance levels suggest that there may be a subset of individuals with both RD and ADHD because of a common genetic etiology. Second, the method of diagnosing ADHD solely on the basis of the DICA may be subject to criticism because it relies on single versus multiple informants, and it is not confirmed by objective measures of attention and other ADHD symptoms. Again, these criticisms can be empirically examined using more extensive interview and objective measures of ADHD, which are currently being obtained. Finally, the subtype analyses are preliminary. There may be additional subtypes within the comorbid subtype, and only in some may RD and ADHD co-occur because of a common genetic etiology. As the sample size increases, more detailed subtype analyses can be conducted. The twin method listed in this paper can also be applied to the study of comorbidity found for other psychiatric and medical disorders (e.g., Gilger, in press; Gilger et aI., in press). When working with these types of data, it is important to consider alternative explanations for comorbidity other than a common genetic etiology, especially when a significantly greater MZ than DZ cross-concordance is found. Because comorbidity is so important in psychiatry, a more detailed discussion of the relevance of behavioral genetic studies to this issue is given below. It is noteworthy that this discussion not only applies to twin data but to family data as well. J. Am. Acad. Child Adolesc. Psychiatry, 3 J :2, March 1992
TWIN STUDY OF ETIOLOGY OF COMORBIDITY
Other than the mechanism of common genetic etiology, there are three additional alternative explanations for comorbidity: 1) RD leads to ADHD as a secondary symptom; 2) ADHD leads to RD as a secondary symptom; or 3) the observed association is an artifact. Again, it is likely that comorbid cases are etiologically heterogeneous; where in some, ADHD is secondary to RD; in others, RD and ADHD share a common etiology; and, in still others, the observed cooccurrence of ADHD and RD is due to chance. This having been said, these three additional explanatory mechanisms can be described as follows. Explanations 1 and 2 would suggest that having one disorder is a necessary (but it need not be a sufficient) condition for the other disorder. There is in fact evidence that at least some cases of ADHD may represent secondary symptoms ofRD (e.g., Cantwell and Baker, 1991; Shaywitz and Shaywitz, 1988). For example, the prevalence of comorbid ADHD and RD depends on the method of subject ascertainment. If subjects are selected on the basis of having RD, about 33% are also ADHD, but when subjects are first selected for ADHD, approximately 11 % are RD as well (Shaywitz and Shaywitz, 1988). This pattern of association suggests that RD may somehow facilitate the expression of ADHD symptoms, rather than ADHD being instrumental in the development of RD. Additional support for Explanation 1 and the conclusion that RD-ADHD comorbidity reflects little or no genetic covariance was found in two other studies conducted in Colorado. One study was a multiple regression analysis of the CRP twins, where the heritability of ADHD was assessed in terms of a continuous dimension, rather than as a dichotomous diagnosis (Gillis et aI., in press). Partialling out the variance resulting from IQ and/or reading ability-disability in these analyses had essentially no effect on the heritability estimates for ADHD. These analyses in a sense "controlled" for shared genetic variance between RD and ADHD. Finding little subsequent change in the heritability of ADHD suggests that there is minimal genetic covariance between RD and ADHD in this twin sample. In another study where data based on a separate nontwin sample was examined, additional evidence favoring the secondary symptom hypothesis (i.e., RD leads to ADHD) was found using a neuropsychological approach (Penningtonet aI., submitted manuscript). The third explanatory mechanism suggests that the observed RD-ADHD association may be artifactual because of sample selection methods (e.g., Berkson's bias, Berkson, 1946) or perhaps definitional overlap. Selection biases are particularly prevalent in clinically ascertained samples, where there is a greater tendency for the most severely affected children to seek care. Thus, the high rates of RD-ADHD comorbidity often reported may simply reflect artificially inflated statistical estimates of this association, rather than some causal (Explanations I and 2) or etiological (genetic) interrelatedness of RD and ADHD. Definitional and diagnostic overlap or ambiguity may also contribute to the observed rates of RD-ADHD comorbidity. For example, the symptoms of both RD and ADHD were once considered part of the minimal brain damage syndrome (Strauss and Lehtinen, 1947). Moreover, such symptoms as inattenJ. Am. Acad. Child Adolesc. Psychiatry, 31:2, March 1992
tiveness in the classroom and poor school performance are still considered by some to be hallmarks of both RD and ADHD. To the degree that RD and ADHD share diagnostic criteria, comorbidity may be inflated irrespective of whether or not the two disorders share a common neurological substrate, are causally related, or have a common etiology. Whereas each of these possible mechanisms responsible for comorbidity is distinct, they can lead to very similar patterns of results in twin cross-concordance or family cosegregation studies. If, for example, ADHD were secondary to RD in some percentage of RD cases in the population, a larger MZ than DZ RD-ADHD cross-concordance rate is expected, similar to the rate expected if the comorbidity were genetic in origin. Given that RD is heritable, more MZ than DZ cotwins will have RD, and, given that RD is expected to lead to ADHD in an equal number of MZ and DZ twins, more MZ than DZ cotwins will also exhibit ADHD. Thus, at first glance, the twin data would falsely suggest a genetic etiology to comorbidity when none may exist. The causal basis for a greater MZ than DZ cross-concordance could be elucidated by a more detailed examination of the cotwin data. If the MZ-DZ difference in cross-concordance was due to secondary factors, then a large proportion of ADHD cotwins should also have the primary disorder of RD. On the other hand, if the comorbidity and RD-ADHD cross-concordance was due to genetic factors, one would expect cotwins to exhibit ADHD even when they do not have RD (e.g., ADHD is in a sense a variable expression of the RD gene). Of course, the distinction between genetic and secondary factors in comorbidity becomes cloudy as the association between the primary disorder and the secondary disorder increases. In those cases where the primary disorder has directly led to the secondary disorder, two disorders in a sense share the same genetic basis, although one disorder may appear later in development than the other. It is also noteworthy that the potential confusion between secondary and common etiological effects does not exist when examining a comorbid subtype as was done in the second set of analyses presented in this paper. The subtype test is a robust yet conservative measure of whether a genetically based comorbid subtype exists in the twin sample. In the case of Explanation 3, where comorbidity is artifactual, there is little reason to expect to observe differences in the MZ and DZ rates. Artifact effects are probably best identified after the other possible mechanisms have been dismissed. Another way to test an artifact hypothesis in twin data or family data ascertained through a clinic or special referral is to compare the comorbidity rates in probands (or the person through which the pair or family was identified) with the comorbidity rates observed in their cotwins or relatives. Artifacts of sampling would be suggested by higher comorbidity rates in the probands than in the cotwins or relatives. In the case of twins, such a comparison is best performed on the DZ pairs. Because MZs are genetically identical, probands and cotwins may be very similar even if a selection bias is operating. In spite of the intricacies involved, the results of this study begin to address why RD and ADHD cooccur in children and suggest a method by which other comorbid disorders in 347
GILGER ET AL.
psychiatry may be studied. Such studies would help clarify the reliability, validity, and usefulness of combining primary and secondary diagnoses such as RD and ADHD in research and clinical practice. For example, it is clinically important to recognize that RD-ADHD comorbidity may not simply be a complex phenotype of a single biological or neurological mechanism in all individuals. This suggests that interventions may need to be directed at the various symptoms and causes of each "disorder"; treating comorbid RD and ADHD the same as ADHD by itself (e.g., with Ritalin®) may be ineffective, because many of these children may need tutoring for their primary RD. References American Psychiatric Association (1987), Diagnostic and statistical manual of mental disorders (3rd edition-Revised). Washington, D.C.: APA Press. Biederman, J., Faraone, S. V., Keenan, K, Knee, D. & Tsuang, M. T. (1990), Family-genetic and psychosocial risk factors in DSM-I1I attention deficit disorder. J. Am. Acad. Child Adolesc. Psychiatry, 29:526-533. Berkson, J. (1946), Limitations of the application of fourfold table analysis to hospital data. Biometrics, 2:47-51. Cantwell, D. P. & Satterfield, J. (1978), The prevalence of academic underachievement in hyperactive children. J. of Pediatr. Psychol., 3:168-171. - - Baker, L. (1991), Association between attention deficit-hyperactivity disorder and learning disorders. Journal ofLearning Disabilities, 24:88-95. DeFries, J. C. (1985), Colorado Reading Project. In: Biobehavioral Measures of Dyslexia, eds. D. B. Gray & J. F. Kavanagh. Parkton, MD: York Press, pp. 107-122. - - Fulker, D. W. (Eds.) (1986), Special issue: multivariate behavioral genetics and development. In: Behav. Genet., 16: 1-235. - - Gillis, J. J. (1991). Etiology of reading deficits in learning disabilities: quantitative genetic analysis. In: Advances in the Neuropsychology of Learning Disabilities: Issues, Methods, & Practices, eds. J. E. Obrzut & G. W. Hynd. Orlando, FL: Academic Press pp. 29-47. Dunn, L. M. & Markwardt, F. c., Jr. (1970), Manualfor the Peabody Individual Achievement Test. Circle Pine, MN: American Guidance Service. Dykman, R. A. & Ackerman, P. T. (1991), Attention deficit disorder and specific reading disability: Separate but often overlapping disorders. Journal of Learning Disabilities, 24:96-103. Felton, R. H., Wood, F. B., Brown, I. B., Campbell, J. & Harter, M.
348
R. (1987), Separate verbal memory and naming deficits in attention deficit disorder and reading disability. Brain and Language, 31:171-184. Gilger, J. W. (in press), A comment on twin analyses of comorbidity in psychiatry. Am. J. Psychiatry. - - Pennington, B. F., Green, P., Smith, S. D. & Smith, S. A. (1992), Dyslexia immune disorders, and left-handedness: twin and family studies of their relations. Neuropsychologia. (in press). Gillis, J. J., Gilger, J. W., Pennington, B. F. & DeFries, J. C. (1992), Attention-deficit hyperactivity disorder in reading disabled twins: evidence for a genetic etiology. J. Abnorm. Child Psycho/. (in press). Hannah, M. c., Hopper, J. L. & Mathews, J. D. (1983), Twin concordance for a binary trait: I. Statistical methods illustrated with data on drinking status. Acta Genet. Med. Gemello/., 32:127-137. Herjanic, B., Campbell, 1. & Reich, W. (1982), Development of a structured psychiatric interview for children: agreement between child and parent on individual symptoms. J. Abnorm. Child Psycho/., 10:307-324. Holobrow, P. & Berry, P. (1986), A multinational, cross-cultural perspective on hyperactivity. Am. J. Orthopsych. Assoc., 56:320-322. Hopper, J. L., Hannah, M. C., Macaskill, G. T. & Mathews, J. D. (1990), Twin concordance for a binary trait: III. A bivariate analysis of hay fever and asthma. Genet. Epidemio/., 7:277-289. McGee, R., Williams, S. & Silva, P. A. (1987), A comparison of girls and boys with teacher-identified problems of attention. J. Am. Acad. Child Adolesc. Psychiatry, 26:711-714. Nichols, R. C. & Bilbro, W. C. (1966), The diagnosis of twin zygosity. Acta Genet. Med. Gemello/., 16:265-275. Pauls, D. L., Hurst, C. R., Kruger, S. D., Leckman, S. D., Kidd, K & Cohen, D. J. (1986), Gilles de al Tourette's Syndrome and attention deficit disorder with hyperactivity: evidence against a genetic relationship. Arch. Gen. Psychiatry, 43:1177-1179. Rutter, M., Macdonald, H., Couteur, A. L., Harrington, R., Bolton, P. & Bailey, A. (1990), Genetic factors in child psychiatric disorders. II: Empirical findings. J. Child Psycho/. Psychiatry, 31 :39-83. Shaywitz, S. E. & Shaywitz, B. E. (1988), Attention deficit disorder: current perspectives. In: Learning Disabilities: Proceedings of the National Conference, eds, J. F. Kavanaugh & T. J. Tress. Parkton, MD: York Press, pp. 369-523. Strauss, A. & Lehtinen, L. (1947), Psychopathology and Education in the Brain Injured Child. New York: Grune & Stratton. Weiner, Z., Reich, W., Herjanic, B., Jung, K G. & Amado, H. (1987), Reliability, validity, and parent-ehild agreement studies of the Diagnostic Interview for Children and Adolescents (DICA). J. Am. Acad. Child Adolesc. Psychiatry, 26:649-653. Wechsler, D. (1974), Examiner's Manual: Wechsler Intelligence Scale for Children-Revised. New York: The Psychological Corporation. - - (1981), Examiner's Manual: Wechsler Intelligence Adult ScaleRevised. New York: The Psychological Corporation.
J. Am. Acad. Child Adolesc. Psychiatry, 3 I :2, March 1992
