Casselbrant ML, Mandel EM, Fall PA, Rockette HE, Kurs-Lasky M, Bluestone CD, Ferrell RE. The Heritability of Otitis MediaA Twin and Triplet Study. JAMA. 1999;282(22):2125-2130. doi:10.1001/jama.282.22.2125
Author Affiliations: Department of Otolaryngology, University of Pittsburgh School of Medicine (Drs Casselbrant, Mandel, and Bluestone); Department of Pediatric Otolaryngology, Children's Hospital of Pittsburgh (Drs Casselbrant, Mandel, and Bluestone, and Ms Fall); and the Departments of Biostatistics (Dr Rockette and Ms Kurs-Lasky) and Human Genetics (Dr Ferrell), University of Pittsburgh, Pittsburgh, Pa.
Context Anatomical, physiological, and epidemiological data indicate that there
may be a significant genetic component to prolonged time with and recurrent
episodes of otitis media in children.
Objective To determine the genetic component of time with and episodes of middle
ear effusion and acute otitis media (AOM) during the first 2 years of life.
Design Prospective twin and triplet cohort study with enrollment from 1982
Setting Otitis Media Research Center in the ear, nose, and throat clinic of
Children's Hospital of Pittsburgh, Pittsburgh, Pa.
Patients A total of 168 healthy same-sex twin and 7 triplet sets were recruited
within the first 2 months of life; zygosity results were available for 140
sets; 138 (99%) of these were followed up for 1 year and 126 (90%) for 2 years.
Main Outcome Measures Proportion of time with middle ear effusion, episodes of middle ear
effusion, and episodes of AOM by zygosity status.
Results At the 2-year end point, the estimate of heritability of time with middle
ear effusion was 0.73 (P<.001). The estimates
of discordance for 3 or more episodes of middle ear effusion were 0.04 for
monozygotic twins and 0.37 for dizygotic twins (P
= .01). The estimate of discordance of an episode of AOM in monozygotic twins
was 0.04 compared with 0.49 in dizygotic twins (P
Conclusions Our study suggests there is a strong genetic component to the amount
of time with middle ear effusion and episodes of middle ear effusion and AOM
The predisposition of infants and young children to recurrent episodes
of or prolonged time with middle ear effusion (MEE) may have a significant
genetic component that can be quantified. Anatomical, physiological, and epidemiological
data have suggested a heritable aspect of the disease.1- 14
Twin studies provide a powerful method of determining the contribution
of genetics to a disease, because the potentially confounding effect of environmental
factors is significantly reduced. The children in our monozygotic (MZ) or
dizygotic (DZ) twin pairs were the same age, lived in the same household,
and therefore presumably shared similar environments. A significantly higher
concordance rate of a trait in MZ twins compared with DZ twins is suggestive
of a significant genetic component. Conversely, similar concordance rates
for MZ and DZ twins suggest that environmental factors play a larger role.
This article reports the results of a prospective twin and triplet study
with a clinically characterized patient population designed to provide data
on the heritability of MEE.
The study was conducted at the Otitis Media Research Center in the ear,
nose, and throat clinic at Children's Hospital of Pittsburgh. Newborn twin
and triplet sets were recruited mainly from Magee-Women's Hospital in Pittsburgh,
Pa, the largest maternity hospital in the area, with a total of 68,745 births
(yearly mean, 8593; range, 8256-8821) from 1990 through 1997—approximately
30% of all births in the area. From 1995 through 1997, 900 sets of twins and
108 sets of triplets were born. Some sets were enrolled from recruitment information
received from Mothers of Multiples clubs in the Pittsburgh area.
In a pilot project (Twin Study I), same-sex and opposite-sex twin pairs
were enrolled, whereas in Twin Study II, only same-sex twin and triplet sets
were enrolled. A child was ineligible if he or she had any major congenital
malformations, a medical condition with a predisposition for otitis media
(OM) (eg, cleft palate, Down syndrome), received assisted ventilation, been
cared for in an intensive care unit, or was at high risk for sensorineural
hearing loss. Infants whose parent(s) gave informed consent were enrolled
in the study, which was approved by the human rights committee at Children's
Hospital of Pittsburgh and the internal review board at Magee-Women's Hospital.
Information obtained at entry included history of illness and treatment
received; method of feeding; type of daily care; family history with respect
to OM, allergy, and recurrent infection; number and age of sibs, and socioeconomic
status. An ear, nose, and throat examination, including pneumatic otoscopy
by a validated otoscopist,15 and behavioral
audiometry appropriate for the age of the child were performed.
Follow-up visits were scheduled at the Otitis Media Research Center
clinic or at home at bimonthly (Twin Study I) or monthly (Twin Study II) intervals
and whenever symptoms of ear disease intervened. Each visit included an extensive
interval history and pneumatic otoscopy. Beginning at 7 months of age, tympanograms
and acoustic reflex measurements were obtained at each visit. If a child was
diagnosed as having middle ear disease, he or she was treated according to
a standardized treatment protocol. At age 12 months, blood samples were obtained
for zygosity testing.
Prior to 1993, tympanometry at the Otitis Media Research Center clinic
was performed on a Madsen Z073 tympanometer, and the presence of effusion
was determined using an algorithm combining otoscopy and tympanometry.16 As of January 1993, a GSI-33 middle ear analyzer
(Lucas-Grason-Stadler, Inc) was used for testing at the Otitis Media Research
Center clinic and a GSI-38 middle ear analyzer was used for home visits to
meet the specifications of the American National Standards Institute.17 An algorithm consistent with American Speech-Language-Hearing
Association guidelines18 was developed by Nozza
et al19 to determine the presence of effusion,
categorizing middle ear status by tympanometric width and otoscopy. If the
infant was younger than 7 months old or if tympanometry could not be obtained,
the diagnosis was based on otoscopy alone.
The diagnosis of acute otitis media (AOM) required the presence of effusion
by otoscopy and at least 1 symptom (fever, otalgia, or irritability) and 1
sign (erythema and/or white opacification of the tympanic membrane, bulging
or fullness, or otorrhea) of inflammation. Otitis media with effusion (OME)
was defined as effusion in the middle ear without the symptoms of AOM. The
determination of the presence of effusion was based on the previously described
algorithms (see "Acoustic Immittance Measurements" section). Otoscopically
OME was diagnosed by a semi-opaque tympanic membrane with decreased mobility
or the presence of fluid levels or bubbles behind the membrane. Middle ear
effusion was used to designate middle ear fluid diagnosed as either OME or
AOM. The term OM represents varying types of middle
ear disease (ie, OME and AOM).
During the pilot study (Twin Study I), zygosity was assessed using a
battery of 6 red blood cell antigens (ABH, Rh, MNS, K, Kp, FY) and 7 plasma
protein and red blood cell enzyme loci (HP, GC, PLG, PGMI, ESD, ACP, GPT).20 In Twin Study II, zygosity was assessed using 6 microsatellite
loci with a minimum heterozygosity of 0.7, using standard polymerase chain
reaction–based methods.21 Each battery
of markers provides a probability of excluding monozygosity of greater than
0.99. At follow-up, most participants in Twin Study I were resampled for DNA
genotyping, and no discrepancies between serological and biochemical testing
and DNA testing were observed. For subjects for whom permission to obtain
blood samples was denied, an attempt was made to determine zygosity status
using cheek scrapings.
In Twin Study II, the treatment of first choice for a new episode of
AOM or an episode of OME was a 10-day course of amoxicillin. If this regimen
was not effective for AOM, the children received amoxicillin-clavulanate.
Patients allergic to amoxicillin were given alternative medication, such as
cefaclor, erythromycin-sulfisoxazole, or 1 of the newer cephalosporins. If
OME persisted for 4 consecutive months or 180 days (cumulative) during a 1-year
period, tympanostomy tubes were recommended. Antibiotic prophylactic treatment
for a 3-month period or tympanostomy tubes were recommended for children who
had 4 episodes of AOM in 6 months or 5 episodes in 12 months. For children
in Twin Study I, there was no treatment protocol, but they were treated in
a similar fashion with amoxicillin given initially for episodes of AOM and
The primary outcome was the proportion of time with MEE (proportion
of time × 100 = percentage of time). Time with MEE was estimated by
dividing the length of time that the child participated in the study into
intervals in which the end points of the intervals were the midpoints between
the dates of 2 successive visits. Middle ear status for the entire interval
was assumed to be the same as middle ear status at the visit within the interval.
In cases in which there were more than 91 days between 2 successive visits,
the interpolation was applied for a maximum of 45.5 days. Middle ear status
was considered to be unknown for the remaining days in the interval. The beginning
and resolution of an episode of MEE was defined as a change in status from
no MEE to MEE and back to no MEE. If tympanostomy tubes had been inserted,
the child was not considered at risk for MEE when the tympanostomy tube was
Estimates of heritability were obtained using the model proposed by
DeFries and Fulker.22 An observation of each
child was used once as the independent observation, and the degrees of freedom
were appropriately adjusted. Because the variable of interest was the proportion
of time with MEE, an arcsine transformation was applied to the data. Because
twin or triplet sets in which none of the children had disease may exert a
disproportionate influence on the regression model, sets with no disease were
excluded from the DeFries-Fulker model. This had little effect on the analysis
at later ages, because fewer sets were excluded. To accommodate the 5 sets
of triplets, we fit all possible regression models from a random pairing of
2 members of the triplet set, which resulted in 243 possible regression models
from the total twin or triplet populations. For the estimates of the genetic
component (h2), shared environment (c2), and the corresponding
significance levels, we used a weighted-probability estimate from the 243
samples. We also followed the strategy proposed by Cherny et al,23
who recommend eliminating the proband term from the model when the estimate
of the environmental effect is low to obtain a more unbiased estimate of heritability.
Episodes of MEE and AOM were used as secondary end points in this study.
The degree of discordance in MZ twins and DZ twins was compared using the
method of Olson et al.24 In applying this method,
presence of disease was defined to be greater than or equal to 3 episodes
for MEE and at least 1 episode for AOM.
A total of 168 same-sex twin pairs and 7 triplet sets (33 twin pairs
in Twin Study I and 142 twin or triplet sets in Twin Study II) were recruited
within the first 2 months of life from 1982 through 1995 (Table 1). Only same-sex twin or triplet sets were included due to
reported differences in incidence of OM between males and females.25- 27 Zygosity results
were available for 140 twin or triplet sets: 64 (46%) DZ sets and 76 (54%)
MZ sets. Of the 135 twin sets, 61 (45%) were DZ and 74 (55%) were MZ. Of the
5 triplet sets, 3 (60%) were DZ and 2 (40%) were MZ. Twenty-three (66%) of
the 35 twin or triplet sets with undetermined zygosity dropped out prior to
zygosity testing at 1 year of age. Zygosity status was determined in 131 twin
or triplet sets from blood samples and in 9 sets from cheek scrapings.
Of those with known zygosity, 138 (99%) twin or triplet sets were followed
up to 1 year of age: 63 (98%) DZ sets and 75 (99%) MZ sets. One hundred twenty-six
twin or triplet sets were followed up for at least 2 years: 61 (95%) DZ sets
and 65 (86%) MZ sets (P = .17).
By the 2-year end point, 1 or both sibs from 17 twin or triplet sets
had undergone tympanostomy tube insertion. There were 9 sets in which both
sibs received tubes (6 MZ, 3 DZ) and 8 sets in which only 1 sib received tubes
(3 MZ, 5 DZ). The average age at the time of tympanostomy tube insertion was
21.2 months in the DZ group and 21.4 months in the MZ group (P = .64).
For the 140 twin or triplet sets with known zygosity, 54% of the sets
were male and 46% were female; 84% were white. This population appears representative
of the general population born at Magee-Women's Hospital. In the present study,
there was a slightly greater percentage of males in MZ twin or triplet sets
than in DZ twin or triplet sets (60% and 45%, respectively, P = .08). There were no statistically significant differences at entry
between DZ and MZ twin or triplet sets in the following subject characteristics:
birth order in the family, ear disease in natural family, ear disease in natural
sibs, occupation of principal wage earner, mother's education, or number of
smokers in the household.
Table 2 summarizes the average
proportion of time with MEE by 3-month age intervals for the 131 children
in DZ sets and the 154 children in MZ sets. Although an estimate is presented
in the interval from 0 to 3 months, it should be viewed with caution, because
the average age at entry was 2 months, and the first interval is likely to
include only 1 month of observation. After 3 months of age, the proportion
of time with MEE increased; for both groups of children, the maximum value
occurred in the intervals from 6 to 15 months, when it ranged from 0.24 to
0.30. For the first year of life, the average proportion of time with MEE
in the DZ group and MZ group was 0.21 (P = .99).
The average proportion of time with MEE for the first 2 years was 0.20 in
the DZ group and 0.23 in the MZ group (P = .72).
No episodes of MEE were observed by 1 year of age in any child in 11
(17%) DZ sets and in 18 (24%) MZ sets (P = .46).
In the remaining sets, there was no difference between members of the set
in the proportion of time with MEE by the 1-year end point in 4 (8%) DZ sets
and 11 (19%) MZ sets (P = .14). An absolute difference
in proportion of time with MEE of greater than or equal to 0.25 was found
in 14 (26%) DZ sets and 9 (16%) MZ sets (P = .24).
The mean difference in the proportion of time with MEE between sibs within
a set was 0.17 for the 53 DZ sets and 0.13 for the 58 MZ sets (P = .39).
Middle ear effusion had not been observed in 5 (8%) of 64 DZ sets and
9 (12%) of 76 MZ sets (P = .61) by 2 years of age.
In the remaining sets, there was no difference in the proportion of time with
MEE in 3 (5%) DZ sets and 9 (13%) MZ sets (P = .20).
An absolute difference in proportion of time with MEE of greater than or equal
to 0.25 was found in 12 (20%) DZ sets and 4 (6%) MZ sets (P = .03). The mean difference in proportion of time with MEE between
sibs within a set was 0.15 for the 59 DZ sets and 0.10 for the 67 MZ sets
(P = .02).
The correlation of the proportion of time with MEE between 2 sibs in
a twin or triplet set for selective cumulative time intervals is shown in Figure 1. Estimates of correlation during
the first interval are more variable, because there was often only 1 visit
made during the first 3 months. The other time intervals show a correlation
between sibs in the range of 0.34 to 0.39 in DZ sets and of 0.63 to 0.83 in
MZ sets. The correlations were consistently higher in the MZ twin or triplet
sets than in the DZ sets, regardless of length of follow-up, and were significantly
different from each other at 6 (P = .04), 12 (P = .04), 18 (P<.001), and
24 (P<.001) months.
The occurrence of 3 episodes of MEE by 2 years of age was used as the
definition of recurrent disease. Measures of discordance in MZ and DZ twin
or triplet sets as defined by Olson et al24
were obtained. At 2 years, the discordance estimate for recurrent episodes
of MEE was 0.04 for MZ and 0.37 for DZ twin or triplet sets (P = .01), based on the 126 twin or triplet sets that completed follow-up
to age 2 years.
The method of Olson et al24 was also
used to measure the discordance in MZ and DZ twin or triplet sets of children
having at least 1 episode of AOM.
At 2 years, the discordance estimate for episodes of AOM was 0.04 for
MZ and 0.49 for DZ twin or triplet sets (P = .005).
By 1 year, there were 111 twin or triplet sets in which at least 1 child
had 1 or more episodes of MEE; by 2 years, this was true for 126 twin or triplet
sets. Figure 2 plots the proportion
of time with MEE during the first 2 years of 1 child (x-axis) vs the proportion
of time with MEE for a second child (y-axis) of a twin or triplet set. The
slopes of the lines fitted to the data would be expected to be similar for
MZ and DZ twin or triplet sets, but if there is a genetic component of disease,
one would expect the variability to be higher in DZ twins or triplets than
in MZ twins or triplets. In the present study, the variability of sibs in
DZ sets was significantly greater than the variability of sibs in MZ sets
(P = .007). When the DeFries-Fulker22
model was applied to the data at the 2-year end point, the estimate of heritability
(h2) was 0.76 (P = .02) and the estimate
of shared environment (c2) was −0.03. Because c2
was not significant, it was eliminated from the model, and the resultant estimate
of h2 was 0.73 (P<.001). The corresponding
estimates for males and females separately were 0.64 (P<.001) and 0.79 (P<.001), respectively.
Although males appear to have a lower heritability estimate, the test for
interaction was not statistically significant (P
= .36), indicating that the difference in the heritability estimates between
males and females may be due to chance.
In this prospective twin study with a highly clinically characterized
population, the estimated heritability of proportion of time with MEE was
0.73 (P<.001) by age 24 months, and the strong
correlation between members of MZ twin or triplet sets compared with DZ sets
in regard to experience with MEE was not dependent on length of time of follow-up
(Figure 1). This suggests that genetics
plays a large role in MEE. Previous studies have indicated that hereditary
factors play a role in the development of OM. For example, the degree of pneumatization
of the mastoid process, a trait believed to be linked causally to OM, was
found to be more similar in MZ than in DZ twins.1,2
Racial differences in eustachian tube anatomy and function have also been
reported: the shorter, straighter eustachian tube found in American Indians
is associated with a higher incidence of chronic middle ear disease.3,4 The frequency of HLA-A2 antigens was
significantly higher in children with recurrent AOM than in children with
OME, yet that of HLA-A3 antigens was lower in children with recurrent AOM
than in healthy children.5,6 Also,
the genetically determined IgG2 marker IgG2m(23) has been shown to be significantly
associated with recurrent AOM.7,8
Several epidemiologic studies have also indicated strong support for
a hereditary component of OM. A study of Apache Indian children adopted into
middle-class homes outside the reservation reported that although the incidence
of most infectious diseases decreased, that of OM remained comparable to the
incidence reported for children on the reservation.9
Familial clustering was found in studies of OME from Pittsburgh,10
of AOM and chronic OME from Minnesota,11 and
of AOM from Sweden,12 suggesting a genetic
component to the disease. Rich et al13 estimated
that the genetic component accounted for up to 60% of the liability for OM
in the presence of "permissive environment." A retrospective self-reporting
study in 2750 Norwegian twin pairs estimated the heritability for liability
to AOM at 0.74 in females and 0.45 in males.14
The authors of that study found that individual environmental factors accounted
for 26% of the susceptibility to AOM in both males and females and common
environmental factors accounted for the remaining 29% of the susceptibility
to AOM in males. We found a trend toward a lower heritability estimate in
males (0.64) than in females (0.79). However, the difference in heritability
between males and females was not statistically significant. Environmental
factors accounted for 8% of the susceptibility to time with MEE in males but
were slightly negative in females.
Investigation of secondary end points, such as number of episodes of
MEE and number of episodes of AOM, supports the hypothesis of greater discordance
in DZ twins than in MZ twins.
There are several strengths of the present study. First, the prospective
design allowed for accurate assessment of disease state, eliminating dependency
on recalled information. Second, the twin model permits better control of
the effect of environmental factors on disease. Also, the children were treated
for MEE according to a standardized treatment protocol, thereby reducing as
much as possible any potential impact of different treatment interventions
on duration and recurrence of disease. In addition, this study summarizes
the correlations of MEE between sibs over time (Figure 1), because the magnitude of the genetic component of disease
may be a function of age as well as outcome measure. The correlation between
MZ twins is consistently higher than the correlation between DZ twins, regardless
of the length of follow-up.
There are limitations to the study. The children were examined together
by the same examiner, not by 2 examiners blinded to their zygosity status.
However, to minimize this effect, we used an algorithm incorporating a more
objective measure of middle ear status (tympanometry) with otoscopy to determine
the presence of MEE. Also, that there is no single accepted definition of
ear disease during the first 2 years of life is an issue in genetic studies
of ear disease.
The DeFries-Fulker model was used in the analysis of the present study.
Other methods to estimate heritability have been proposed that have better
statistical properties, but the DeFries-Fulker model has the advantage of
being presented in the familiar framework of a linear regression model. There
is good agreement between the DeFries-Fulker model and the primary alternative
method of a variance component model.23
These results suggest that there is a strong genetic component to the
amount of time with MEE and episodes of MEE and AOM in children. This finding
may influence the primary care physician to identify the sibs and offspring
of affected patients as high-risk cases. Closer surveillance of patients at
risk for MEE could result in earlier detection and treatment of the disease,
as well as prevention of possible developmental problems.