INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Asthma is one of the most common childhood health problems. In order to identify potential etiologic factors, many epidemiologic studies have charted prevalence rates across geographic region, age, sex, and cohort. Findings reveal considerable regional differences within countries (1), as well as substantial variation across countries, with prevalence rates ranging from 3% (1) to 15% (3). Asthma rates also vary across age and sex, with higher prevalence in childhood than in adolescence, and the greatest incidence of new cases among children (5). Rates tend to be higher among males than among females early in life, are similar for both sexes later on in childhood, and then there is a predominance of female cases during the teen-age years (7, 8). Although variability associated with demographic factors is sizable, the causes of asthma remain unknown. More recently, the question of secular increases in prevalence rates for asthma has received particular attention. Here it is argued that changes in types and levels of environmental exposures, and not genetic risk, could account for increases in prevalence that occur over a short time period.

One would expect that many of the environmental exposures experienced during childhood and of potential importance for the occurrence of asthma are shared by siblings who grow up in the same family environment. Such exposures include, for example, air pollution, number of siblings, parental smoking, mites, domestic pets, exposures in indoor air, and factors associated with parental socioeconomic status. If these shared environmental exposures are important for the development of asthma then the risk to co-twins of affected twins should be significantly elevated for both MZ and DZ pairs.

This study investigates whether there is evidence for shared environmental effects in the development of asthma. We present population-based age-specific incidence and prevalence rates of asthma, estimate the absolute and relative risks for asthma according to the presence of asthma in the co-twin, and analyze genetic and environmental sources of variance of the liability for asthma.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Sample

The data are from a population-based sample of Norwegian twins who are participants in a study on health and development at the National Institute of Public Health in Oslo. All twins born between 1967 and 1974 were identified through the national Medical Birth Registry (9). In total, 5,078 pairs of twins were born during this period, and both twins survived past the age of 3 for 4,231 of these pairs. The unique person numbers for these twins were then matched to the National Census Registry to obtain current address and vital status. In November, 1992, questionnaires were sent to 7,992 twins from the 3,996 pairs who were at least 18 yr of age, where both members in the pair were alive, and for whom a current address was available. A detailed description of the sample and zygosity classification procedure is provided elsewhere (10).

Questionnaire responses were received from 5,864 twins (response rate, 73%), comprising 2,570 pairs and 724 twins whose co-twin did not respond. Among the intact pairs there are 416 identical (MZ) male pairs and 387 fraternal (DZ) male pairs, 528 female MZ pairs, 443 female DZ pairs, and 796 pairs of unlike sex. Among the 724 single responders, 363 were men and 361 were women.

Measures

The questionnaire included items for zygosity classification and a physical health history. The illness and symptom checklist was prefaced with "have you now, or have you ever had, any of the following illnesses or health problems," where asthma was one item in the list. If the respondent answered "yes" two further questions inquired about their ages at the first and last episodes (if they no longer had asthma).

Analyses

To describe the occurrence of asthma across age and sex, incidence rates and point prevalences were computed using actuarial methods and summarized by 2-yr age bands, from birth through 25 yr of age, for male and female participants separately. Survival analyses using the LIFETEST procedure in SAS (11) were conducted to investigate the absolute risk of developing asthma through 25 yr of age. This procedure is applicable for data such as these that are right censored (asthma could occur by age 25 among twins who were younger than 25 when they responded to the questionnaire). Because of dependency in the twin data this analysis was conducted on a split file constructed by randomly assigning each twin within a pair as "twin a" or "twin b." The analyses were repeated for each half of the data. Tests of sex differences in the absolute risks for asthma were conducted using the log-rank test.

Next, the absolute risk of asthma contingent upon the presence of asthma in the co-twin was estimated using the LIFETEST procedure. For these analyses the sample was selected to include only those pairs where at least one in the pair had a history of asthma, but age of onset in the co-twin was not analyzed. Thus, the risk estimates the probability for developing asthma by 25 yr of age given a positive history of asthma in the co-twin by 25 yr of age. Zygosity differences were tested using the log-rank test and 95% confidence intervals. The conditional risk of asthma given the presence of asthma in the co-twin can also be estimated using probandwise concordance rates (12). Probandwise concordance, defined as the ratio of twice the number of concordant pairs divided by twice the number of concordant pairs plus the number of discordant pairs, was calculated here to facilitate comparison with earlier publications (13, 14).

Data from the full sample were then analyzed to estimate the relative risk of asthma. This relative risk reflects the incidence of asthma in the co-twin (regardless of age of onset) divided by the incidence of asthma given that the co-twin does not have asthma. These analyses are based on Cox proportional hazards models using the SAS-PHREG procedure with the single covariate indexing the co-twins history for asthma (Yes /No).

The usual requirement for doing Cox (or proportional hazards) regression is that the data represent independent observations. This assumption is clearly violated here because there is dependency between the occurrence of the event among members of a twin pair. However, as demonstrated elsewhere (15, 16), the risk estimates obtained using dependent data are valid, but variances may be underestimated. Consequently, the standard errors, log-rank tests, and confidence intervals may be inaccurate. Following the guidelines proposed by Bonney (17), we have adjusted these statistics by simply doubling the variances estimated by SAS. Bootstrapping techniques, in which twin pairs were sampled with replacement, were then conducted to verify this adjustment.

Twin resemblance for the liability to develop asthma was measured using intraclass correlations for groups defined by zygosity and sex. In these calculations each pair of twins is treated as a class, and the correlations represent the portion of total variance (between plus within) accounted for by between-pair variance. Tetrachoric correlation, based on the assumption of an underlying, normally distributed, liability for asthma, were used here because the coding for asthma is dichotomous. The intraclass correlations are interpreted based on biometric models regarding how genes and environment contribute to twin similarity and differences in MZ and DZ pairs. Specifically, twin resemblance for asthma can arise from shared genes, shared environments, or both. Identical twins share in common all of their genes, but fraternal twins share, on the average, 50% of their segregating genes. Consequently, genetic effects contribute only to similarity among MZ pairs, but they contribute to similarity and differences in DZ pairs. Typically, genetic effects are classified into those caused by additive (A) genes and nonadditive (D) effects, where A refers to the added effects of the genes at a locus and D refers to intralocus gene interactions. Greater MZ than DZ intraclass correlations are indicative of genetic influences for variation in the liability to develop asthma. Environmental effects are also classified into two categories. Shared environmental effects (C) are common to members of a twin pair and thereby contribute to twin resemblance regardless of zygosity. Examples of shared environmental effects for asthma during childhood would include factors shared by children in their rearing environment such as indoor air exposures, parental smoking, number of siblings, and familial socioeconomic status. The importance of shared environmental influences is reflected by the degree to which twice the DZ correlation exceeds the MZ correlation. Finally, nonshared environmental effects (E) are exposures that are not shared by members of a pair and thereby do not contribute to twin resemblance but only to within-pair differences for a trait. Influences associated with nonshared environment are indicated by less than perfect MZ correlations.

Structural modeling procedures using Mx (18) were used to estimate the genetic and environmental components of variance described above. These models are widely used with twin data; applications are detailed elsewhere (19) and have been described in this journal previously (20). Briefly, the model comprises a set of simultaneous linear equations specifying how genetic and environmental influences contribute to twin covariance for asthma in groups defined by zygosity. The known genetic parameters of this model are based on the degree of biologic relatedness (MZ or DZ). Accordingly, the equations specify that all sources of genetic influences (additivity and dominance) are perfectly correlated between members of MZ pairs but are correlated 0.5 (additive) and 0.25 (dominance) between members of DZ pairs. In the classic twin design shared environmental effects contribute only to co-twin resemblance and are, thereby, correlated 1.0 in MZ and in DZ pairs. In contrast, nonshared environmental effects contribute only to within-pair differences and are uncorrelated within pairs. Because of problems with parameter identification all of the genetic and environmental parameters are not estimated in a single model. Rather, a series of models is analyzed to find the one model that best fits the data. By comparing two nested models the importance of any particular genetic or environmental parameter can be assessed using the chi-square difference, which itself is distributed as a chi-square with the degrees of freedom equal to the degrees of freedom between the two models. If excluding a parameter from the model results in a significant increase in the chi-square, then the parameter is significant to that model. For dichotomous data Mx can analyze contingency table input summarizing the number of pairs concordant and discordant for asthma by zygosity. Each table (Twin 1 asthma* Twin 2 asthma) has three degrees of freedom, yielding a total of six degrees of freedom for the MZ and DZ data.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Of the 5,864 twins who answered the questionnaire, 16 were missing data on asthma. Altogether, 332 (5.7%) endorsed a positive history of asthma. For the 40 who did not indicate the age of onset, the mean age for their birth cohort was substituted. Sex differences in the prevalence of asthma (6.0% for males and 5.4% for females) were not significant [²₍₁₎ = 1.00, p = 0.32]. However, the average age of onset was significantly lower [t ₍₃₃₀₎ = 3.53, p < 0.01] among the male twins (mean, 7.8) with asthma than among the female twins (mean, 10.2) with asthma. There was no evidence for differences in the prevalence of asthma across the eight birth cohorts from 1967 to 1974 [²₍₇₎ = 6.24, p = 0.51].

Incidence rates per 1,000 person-years and point prevalences for asthma were computed for each 2-yr interval from birth through 25 yr of age, and they are listed in Tables 1 and 2 for males and females, respectively. The incidence density varies across age for both sexes. The rates for males peak within the first 2 yr of life, and there is a second, smaller elevation between about 8 and 11 yr of age. In contrast, asthma incidence among females is not characterized by a large peak early in life. Rather, the rates are greatest between 8 and 11 yr of age and then again during teen-age years at 14 to 17 yr of age. In general, the magnitude of the peaks are more pronounced among the males, whereas the elevation during the teen-age years is more salient among the females.

The rates in Tables 1 and 2 indicate steady increases in asthma prevalence for both sexes from birth through 25 yr of age. Although the general age pattern is quite similar in male and female twins, the prevalence of asthma is consistently greater among the male twins from birth through most of the teen-age years, after which the rates are similar in both groups.

Actuarial estimates for the cumulative incidence of asthma by 25 yr of age age plotted for the full sample by sex in Figure 1. The risk of asthma is greater among the male twins across the entire age range; however, according to the long-rank tests this difference was not significant in either half of the split file [log-rank_{(twin a)} = 1.30₍₁₎, p = 0.25; log-rank_{(twin b)} = 0.09₍₁₎, p = 0.76]. In the full sample, the cumulative incidences were 0.062 for the men and 0.057 for the women.

View larger version (11K): [in this window] [in a new window]   Figure 1.   Actuarial estimates for the risk of developing asthma from birth through 25 yr of age by sex.

In the general sample, there were no zygosity differences in the lifetime prevalence for asthma among the male twins [MZ prevalence, 6.3%; DZ prevalence, 5.9%; ²₍₁₎ = 0.17, p = 0.68] or female twins [MZ prevalence, 4.9%; DZ prevalence, 5.6%; ²₍₁₎ = 0.68, p = 0.41].

Pairwise descriptive information for twin concordance for asthma, the probandwise concordance rates, actuarial estimates of the cumulative incidences of asthma if the co-twin has a history of asthma and tetrachoric, intraclass correlations are shown in Table 3. As illustrated by the values in Table 3, the actuarial estimates yield the same results as the probandwise concordance rates. There is a significantly greater probability of developing asthma among MZ twins who co-twin has asthma than among DZ twins whose co-twin has asthma. After correction for dependency in the data, the log-rank tests [male: log-rank = 7.52₍₁₎, p = 0.01; female: log-rank = 7.58₍₁₎, p = 0.01] reveal significant zygosity differences in the probability of developing asthma for male and female twins whose co-twin is affected. Additionally, the 95% confidence intervals reveal that the rate values for the MZ twins are greater than the upper confidence limit for the DZ twins in both male and female twins. Data from the unlike-sex twins were analyzed to investigate whether the risk to co-twins varied contingent upon the sex of the affected twin. Although, there were 78 pairs of unlike-sex twins where at least one member of the pair had a history of asthma. Among these there were 39 male twins with asthma from pairs whose female co-twin didn't have asthma, 33 female twins with asthma from pairs whose male co-twin didn't have asthma, and six pairs where both twins had asthma. The cumulative incidence rates were similar for the co-twins of affected male (0.14) and female (0.15) twins.

Results based on 800 bootstrap samples revealed that the variance of the bootstrap estimates were very close to those values obtained by the simple doubling approach. For comparison to the values in Table 4, the confidence intervals for the relative risks based on the bootstrap samples were 10.3 to 31.7 for the MZ pairs and 1.1 to 4.1 for the DZ pairs.

A test of the assumption for Cox regression revealed proportional hazards between the groups whose co-twin had a positive versus a negative history of asthma for the MZ [Wald ²₍₁₎ = 0.51, p > 0.47] and DZ [Wald ²₍₁₎ = 0.88, p > 0.35] pairs. The parameter estimates, standard errors, relative risks, and corrected confidence intervals are listed in Table 4 for each group defined by sex of the pair and zygosity. Among MZ twins whose co-twin had a positive versus a negative history of asthma, the increased risk of developing asthma was 22.5 times greater among the male twins, and 13.8 times greater among the female twins. As listed in Table 4, the corresponding values for the same-sex DZ twins were small and did not reach significance after correcting for the dependency in the data.

Results of the modeling outcomes are presented in Table 5, and indicate that shared environmental influences do not account for any of the variation in liability for asthma; C was estimated at zero, and the chi-square difference test between the ACE and AE models was nonsignificant [4.83₍₄₎  4.83₍₃₎ = 0.00₍₁₎, p = 1.00]. Submodels of the ADE model revealed that either A or D could be dropped without a significant increase in chi-square, but, biologically, it is rare to have genetic dominance (D) in the absence of genetic additivity (A). Rather, it is most likely that there is little statistical power to distinguish between these two effects. The sum of the modeling outcomes indicate genetic effects (additive and nonadditive) explain approximately 75% of the variation, and the remaining 25% is attributable to nonshared environmental influences.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Risk Associated with Genes and Environment

Although many studies of familial risk demonstrate the importance of a familial component for asthma, results are ambiguous, with some emphasizing the importance of genes (21, 22) and others the importance of genes and environments (23), and little power to resolve the sources of the familial effect. Twin studies, however, are particularly suited to decomposing the familial sources of variance. The main finding from this study is the lack of evidence for effects of shared family environment on the risk of asthma. These results are congruent with, but also extend, those based on other large twin studies reporting no (20) or little (13, 14) influence of shared environment. Specifically, the previous large-scale twin samples are characterized by a very wide age range, which makes it difficult to relate shared environment to the exposures assumed to play an etiologic role in the development of asthma early in life. Both the Australian (20) and Swedish (14) twin studies inquired about a lifetime history of asthma, but, the twins were 18 to 88 yr of age (Australia) and 42 to 81 yr of age (Sweden) at the time of the questionnaire. Consequently, environmental influences (both shared and nonshared) for asthma for these twins encompass effects that may come into play later in life. Furthermore, if there are age trends for the occurrence of asthma, then variation caused by age will appear as common environment because co-twins are perfectly correlated for age. The Finnish (13) data were based on record linkage assessing cases after 15 yr of age, with the oldest twins born prior to 1925, thus phenotypic variance, associated with either environments or genes that influence asthma in infancy and early childhood, could be missed. In contrast, the oldest age in this Norwegian sample was 25 yr; less than 8% of the cases reported ages of onset older than 18 yrs, and more than 80% of the cases occurred by 15 yr of age when most of the respondents (93%) still lived at home. Therefore, many exposures typically cited as important for the development of asthma from infancy through adolescence would fall into the realm of shared environment for this sample. Such exposures include, for example, air pollution, number of siblings, parental smoking, mites, domestic pets, and indoor air climate (24). Although we do not have direct measures of these exposures, there is no reason to assume that MZ twins are significantly more concordant for these exposures than are DZ twins for the age range studied here. Consequently, if these exposures are important for the development of asthma then the risk to twins of affected twins would be significantly elevated in both MZ and DZ pairs. The present findings did not support this. Rather, the risk estimates revealed an almost 18-fold increase in risk for asthma to MZ twins whose co-twin had asthma compared with MZ twins whose co-twin didn't have asthma. The corresponding risk was 2.3 for the DZ twins. Modeling results indicated that genetic effects account for 75% of the variation in liability for asthma, and there was no evidence for shared environment. These findings are in accordance with earlier twin studies (see reference 24 for a review) reporting that 50 to 75% of the variation in the liability for asthma is due to genetic effects and the remaining variation is due to nonshared environmental effects. The consistency of the collective twin results, representing four different countries with varying prevalence rates, suggests strongly that the familial risk for asthma is primarily, explained by genetic factors.

The assumption that shared environments are equal for MZ and DZ twins is critical to the interpretation of these results. Specifically, genetic effects will be overestimated if environments that predispose to asthma are shared more among co-twins of MZ than DZ pairs. In the present study we were not able to test directly for possible violations of the equal environments assumption and, therefore, cannot rule out that shared MZ environment could explain some of the excess resemblance for asthma among MZ pairs. Because of the nature of the most frequently cited putative exposures for asthma (air pollution, number of siblings, parental smoking, mites, domestic pets, and indoor air climate), it is unlikely that within-pair differences in these exposures differed substantially between MZ and DZ twins when they were infants and of preschool age. However, similarity for environmental exposure may become greater among MZ than among DZ twins as they grow older, especially if MZ twins spend more time together.

Other potential environmental risk factors could also contribute to zygosity differences in resemblance for asthma. First, the risk for perinatal insults and prematurity is greater among MZ than among DZ twins. If prematurity is a predisposing factor then asthma should be more common among MZ than among DZ twins, but we found no evidence of zygosity differences in the prevalence of asthma [²₍₁₎ = 0.10, p = 0.76]. Alternatively, if prenatal factors are important causes of asthma then greater in utero competition among identical twins could lead to reduced twin similarity for asthma in the MZ compared with the DZ twins, but this was not supported by our results. Second, a positive history of asthma in the co-twin may lead to increased awareness in the respondent or diagnostic biases whereby MZ co-twin of asthmatics are more likely to be diagnosed then are DZ co-twins of asthmatics. Again, this does not appear to affect the results appreciably because there is no zygosity difference in the prevalence rates for asthma. Third, co-twin cooperation in filling in questionnaires could pose a bias, particularly if MZ twins have closer co-twin contact than do DZ twins. A previous study of ear infections in this same twin material analyzed the effects of current co-twin contact on twin resemblance (25) and found slightly greater resemblance in both MZ and DZ pairs who had higher versus lower contact, implying a cooperation effect in answering the questionnaire items. However, extensive analyses revealed that the genetic and environmental parameter estimates were only slightly affected by the degree of contact. Finally, identical twins may be more similar than fraternal twins for other potential risk factors, such as smoking and physical activity. It is important to remember that many of these influences may not represent pure environmental effects. Increased similarity in MZ twins, for example for smoking, may be a consequence of their genetic similarity rather than a cause of phenotypic resemblance. The effect of differential concordance for smoking between MZ and DZ pairs is probably minimal here because the reported ages of onset for asthma generally precede the age of onset for smoking.

Incidence and Prevalence

In total, just under 6% of our sample, 18 to 25 yr of age, endorsed a positive lifetime history of asthma. Analysis of the data as a retrospective cohort revealed a fairly linear increase in prevalence from birth through the age range studied. Males tend to have higher prevalence but shorter duration of asthma than females. The trend for excess prevalence among the male twins persists until about 16 yr of age. Survival analyses indicated a slightly elevated risk for asthma among the male twins across all ages. Although there were no sex differences in the lifetime risk for developing asthma, age-specific analyses revealed important sex differences. In particular, the highest incidence rates occurred during the first 2 yr of life for male twins, whereas female twins tended to have a later age of onset with the highest incidence rates in later childhood. These findings follow the uniform pattern of age and sex differences reported in the literature that are based on both clinical diagnoses (7, 26) and questionnaire data (5). The reasons for sex differences in the developmental trajectories for asthma are unknown, though several have speculated about possible explanations (8). For instance, the predominance of male cases early in life may reflect male-female differences in patterns of lung growth and maturity, greater susceptibility to infections among boys, and sex differences in exposure to environmental risk factors (26). However, there may also be a diagnostic bias because it is difficult to establish differential diagnosis of asthma at such an early age when boys are more prone to infections than girls are.

Comparisons to prevalence rates from nontwin, population-based Norwegian data suggest that the twin data are representative of the general population. Although limitations arise because of interstudy differences in age ranges, diagnostic procedures, and period of study, there was close agreement between the prevalence rates from the twin data and other results. Based on physician diagnoses, the cumulative prevalence rate was 3.1% for school children in Oslo 7 to 15 yr of age (27). The analogous rate in our sample, derived by limiting age of onset to less than 15 yr, was 2.6%.

Self-report Methodology

Clearly, concerns arise regarding validity and reliability with the use of retrospective self-reports to measure a positive history of asthma as well as the age of onset. In general, the epidemiologic literature relies on large-scale studies that are, for the most part, based on self-report measures of asthma, symptoms, or medications. Despite the widespread use of this methodology, there is a marked paucity of studies that actually test the reliability of such recall data. Disease classification in this study is not based on an operational definition from symptoms but rather on a simple question inquiring about ever having asthma. The consequence of substantial recall bias could lead to either an increase or a decrease in the incidence rates. Although we do not have direct validation of the asthma cases in this study, assessment of bias using several indirect methods suggests that inaccurate recall does not pose a serious threat to the results.

First, several studies report good correspondence between questionnaire and clinical assessments. There was 95% agreement between questionnaire-based self-reports of asthma and diagnosis by an allergist for the small subsample of Swedish twins (14). Another Swedish study of teen-agers revealed close agreement between self-report and physician diagnosis (28) where the yearly incidences were 1.2 based on the self- reports (proportion of population reporting asthma) and 1.1 based on physician diagnoses. Second, consecutive questionnaire responses revealed 99% consistency for the asthma item in the Swedish twin study (14); and test-retest reliability for questionnaire items for asthma and wheezing, assessed at a 4-mo interval, was 0.98 in the Australian twin study (20). Third, studies investigating the validity and reliability of self-report measures for other illnesses find that self-reports reflect medical conditions (29). Finally, our rates of asthma were quite similar to those from another Norwegian study (27), and the pattern of prevalence associated with age and sex in these data is congruent with the general literature, which lends further support to the reliability of these data.

Twin data can provide additional methods to evaluate reliability and validity of retrospective self-reports because the pattern of twin-correlations for MZ and DZ pairs should reflect theoretical expectations for phenotypes under genetic influence (33). The tetrachoric correlations (Table 3) are consistent with expectations of substantial genetic variation for asthma, and they are strikingly similar to those reported in other twins studies (20, 34). For example, the corresponding values from the Dutch twin register are 0.76, 0.18, 0.77, 0.16, and 0.26 (34). The clear presence of zygosity differences in twin resemblance for asthma, in conjunction with the high correspondence of results between this and other twin studies, suggests that recall bias has relatively little influence on the genetic analyses based on a lifetime risk. Rather, it is more likely that bias would affect the absolute prevalence and incidence rates across age.

Finally, bias may arise if individuals with asthma were more or less likely to answer our questionnaire. This would particularly pose a problem if mortality was associated with asthma. However, the mortality rates caused by asthma among children in Norway are quite low, and were below 0.32 per 100,000 during the period 1960 to 1989 for children younger than 15 yr of age (35).

In summary, the risk for asthma contingent upon the co-twin's history of asthma is greatly increased among MZ twins. There is no evidence that many of the commonly cited environmental risk factors for asthma, which both members of a twin pair would have experienced during infancy and childhood, increase the risk for the development of asthma.