Copyright 1998 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.1998
To determine the association between environmental tobacco smoke and middle ear disease in preschool-age children.
A population-based case-control study with verification of disease history and exposure reporting in a subsample.
Setting and Participants
Participants were identified through a population-based probability sample of 1320 first-grade students in 36 schools in Calgary, Alberta. The parents of 625 children meeting case (n=227) or control (n=398) definitions were interviewed by telephone for their children's exposure history. The adequacy of exposure and disease measures was assessed using hair cotinine measurements, home visits, and physician medical records for 92 children.
Main Outcome Measures and Results
A history of middle ear disease was found in 23.9% of the sample. Relationships were found between middle ear disease and 2 or more household smokers (crude odds ratio [OR], 1.85; 95% confidence interval [CI], 1.15-2.97), 10 or more cigarettes smoked by the mother per day (crude OR, 1.68; 95% CI, 1.12-2.52), and 10 or more cigarettes smoked in total in the household per day (crude OR, 1.40; 95% CI, 0.98-2.00) during the first 3 years of life. In logistic regression modeling, these effects persisted after adjusting for child care (type, age started, duration, and group size), infant feeding (type and duration), socioeconomic status, maternal educational level, number of prenatal ultrasonographic examinations, and health services use. The mean current hair cotinine levels were higher for children living in homes with 1 or more smokers vs no smokers (0.51 vs 0.31 ng/mg, P=.01). There was fair agreement (75.3%) between physician medical records and parental report of disease history, but some misclassification bias toward the null hypothesis is likely.
Environmental tobacco smoke is an important risk factor for middle ear disease in urban preschool-age children, even in a relatively affluent population.
MIDDLE EAR disease (MED) or otitis media (OM) affects up to 46% of children by the age of 3 years1 and is the most frequent reason for ambulatory care visits and the administration of prescription drugs in this age group.2- 5 In addition, OM is the most common reason for surgery in children in developed countries.2,6,7 The total costs associated with OM are estimated to be more than $3.5 billion (1989 data) in the United States alone.8 The fluctuating hearing losses associated with frequent or recurrent MED may place young children at risk of adverse language, cognitive, and motor developmental outcomes.9,10
Environmental tobacco smoke (ETS) may be the most common indoor air contaminant for young children, with estimates ranging from 38% to 60% of all children exposed to passive smoke in their homes.10- 13 The US Environmental Protection Agency linked numerous adverse outcomes in young children with passive smoke exposure, including MED; however, the evidence is suggestive but not conclusive. Following the Environmental Protection Agency's report, controversy has continued in the English-language literature; the ETS-MED association seems well accepted in some professional literature but denied in other literature.14- 17 As recently as March 1997, the ETS-MED relationship was found to be positive in about as many studies as it was found to be negative in.17 This inconsistency in findings may be partly due to methodological issues, such as insufficient power, inadequate measurement of exposure and disease, lack of confounder control, and non–population-based study subjects.17- 25
This study examines the relationship between ETS and MED in a population-based case-control study with adequate sample size, independent collection of information on the adequacy of exposure and disease measures, multivariate analysis for confounder control, and blinding of the interviewer and study participants.
The subjects were identified through stratified cluster sampling of all 207 first-grade classrooms in publicly funded elementary schools in Calgary, Alberta. Schools were stratified into 5 categories based on a social risk index of the immediate surrounding community. The social risk index was computed using relative proportions of 1-parent families, families receiving social assistance, adult unemployment, and median annual income in each community, ranking these values according to distance from the "average" community risk score and taking the average rank overall. Thirty-six schools were sampled using the first-grade census.
All parents of the first-grade students received a preschool health history questionnaire. The preschool health history questionnaire was developed using questions from standard child health surveys (ie, the National Population Health Survey and the Longitudinal Survey of Children) and included items on various early childhood health problems, such as physician-diagnosed asthma and developmental concerns, as well as questions on MED history. It was pretested with 33 families to identify ambiguous or incomprehensible questions. The preschool health history questionnaire was implemented using modified Dillman total survey design techniques.26 Participants were blind to the primary hypothesis in that it was presented as a general study of environmental factors in child health, and ETS and MED questions were imbedded among various other items.
Children in the case and control groups were identified based on the responses to MED questions. Children were classified as cases if they had a history (to their sixth birthday) of any of (1) persistent middle ear fluid that lasted 3 months or longer; (2) 4 or more episodes of OM in any 12-month period; (3) antimicrobial treatment or prophylaxis that lasted 3 months or longer for OM; or (4) myringotomy, auditory tube surgery, or both or otolaryngological recommendation for surgery. Children were classified as controls if they had a history (to their sixth birthday) of 2 or fewer episodes of OM and neither of 1 through 4 previously noted. We chose not to use "pure" controls (ie, children with no episodes of OM), given that OM is prevalent in this age group such that 1 or 2 episodes likely represent the normal course. Children were deemed ineligible for the study if they were born out of the country (to avoid disease detection bias because of lack of primary health care), if they had health conditions associated with increased frequency of OM (sensorineural hearing loss, craniofacial anomalies, and some congenital syndromes), or if they did not meet either the case or control definition.
The parents of children classified as cases or controls completed a telephone interview adapted from the Environmental Protection Agency's Environmental Inventory Questionnaire.27 This questionnaire ascertained exposure to ETS and other indoor air contaminants as well as information on possible confounding variables during 3 age periods: 0 to 12, 13 to 36, and 37 to 72 months. Environmental tobacco smoke or passive smoke was defined as the particulate and gas-phase compounds released into the air from a burning tobacco product (including cigarettes, cigars, and pipe tobacco), the same compounds exhaled by the active smoker, or both. Questions on ETS exposure included household, child care setting, and other outside home exposures (such as public places, vehicles, and other homes). Exposure management behaviors in the household (eg, restricting smoking to specific rooms or outdoors) were also ascertained. The primary exposure measure was number of household smokers during the child's first 3 years of life to ensure that exposure preceded disease status; this measure was dichotomized as 0 to 1 or 2 or more household smokers, based on evidence in this study that exposure was frequently minimized by spouses in single-smoker households. The number of cigarettes smoked in the household per day by the mother, father, and all household smokers was used as a secondary exposure measure and categorized as fewer than 10 and 10 or more using standard national cut points.28
Information was also collected on the following potential confounding variables: age, sex, low birth weight, preterm birth, history of allergies, history of asthma, upper and lower respiratory tract infections, health care use (hospital admissions and emergency department and physician visits), resistance and susceptibility to illnesses, general health status, family history of respiratory morbidity (including OM), child care (type, age started, duration, and group size), infant feeding (type and duration), number of maternal prenatal ultrasonographic examinations (PUEs), maternal and paternal educational level, household income, tenure (home owned or rented by occupants), age and size of the home, crowding, type of heating, gas appliances, fireplace (type and use), relative humidity, condensation and molds, pets, and chemical-producing renovations and hobbies. Telephone interviews for the collection of exposure and confounder data were conducted by a single interviewer blinded to case or control status and trained in cognitive interviewing techniques.29,30
Sample size was determined using the national survey-based estimates of the prevalence of passive smoke exposure and population-based studies of the prevalence of MED, a 1:2 case-control ratio with a power of 80%, an α=.05, and the typical effect size found in previous studies. The required sample size was then inflated for an estimated nonresponse of 25%.
A 15% random sample of 92 children was selected from which independent information on OM history and exposure reports were collected. For OM history, agreement between parental reports and physician medical records was reviewed by mailing questionnaires to family physicians containing the same standard items about OM episodes and surgery as had been presented to the parents. In the exposure validation substudy, homes were visited for the observation of smoking behaviors and the sampling of hair for the measurement of cotinine. The objective of this part of the study was to assess the extent to which reported current household smoking was congruent with current evidence of smoke exposure in the home and in the hair of the children. We assumed that the extent to which respondents were truthful about the reporting of current household smoking would be generalizable to the extent to which they were truthful about the reporting of household smoking for the exposure period used in the analysis. Hair samples were collected by clipping about a dozen 1-cm strands from the back of the neck with sterile scissors; they were shipped in new coin envelopes wrapped in plastic bags via courier to the laboratory. Cotinine, the major metabolite of nicotine, was measured in the samples using a radioimmunoassay technique.31,32
The Conjoint Medical Research Ethics Board of the University of Calgary, Calgary, reviewed and approved the study. Written informed consent was obtained from the parents for interviews, and verbal informed consent was obtained from the children prior to the collection of hair samples.
Differences between the cases and controls on categorical variables were tested using the z test. Odds ratios (ORs) were also computed along with Mantel-Haenszel χ2 tests and Cornfield approximate 95% confidence intervals (CIs).33 Unconditional logistic regression models were computed to examine the association between number of household smokers and number of maternal cigarettes smoked per day and history of MED after adjusting for potential confounders. The Pearson χ2 goodness-of-fit test was applied to evaluate the fit of the models.34 Analyses of variance and F tests were used to compare mean cotinine values (in nanograms per milligram) by reported exposure level. The Mantel-Haenszel extension χ2 test was used to explore differences in effect by level of exposure.35 The percentage agreement and the κ were used to compare physician and parent disease classifications.33
Thirty-six (90%) of 40 schools agreed to participate, and 1780 children made up the initial sample. Of these children, 146 were ineligible because of country of birth and OM-related birth defects and chronic health problems, leaving a pool of 1634 children. Parents of 1320 (80.8%) of the 1634 eligible children completed the preschool health history questionnaire; 852 (64.5%) of the 1320 parents consented to be interviewed. Parents provided interviews for 230 (72.8%) of 316 children who fit the case criteria and 401 (64.8%) of 619 children who fit the control criteria; the other 385 children did not fit either criterion. The parents of 3 cases and 3 controls who consented to interviews were unavailable for follow-up, leaving data for 227 cases and 398 controls (n=625).
The cumulative incidence of parent-reported MED was 23.9%; and of auditory tube surgery, 9.8%. These values were comparable with those found in other population-based studies.36- 38 As seen in Table 1, cases were more likely to have been cared for in a day care or day home (government-sponsored home-based child care), to have been cared for in larger groups, and to have started child care during their first year. Cases were also less likely to have mothers with 12 or fewer years of education, less likely to have mothers who underwent fewer than 2 PUEs, and more likely to have had 4 or more visits to health care providers (hospital or clinic physicians or hospital, public health, or walk-in clinic nurses) during the past year.
Of the cases, 19.8% had lived with 2 or more smokers during their first 3 years of life compared with 11.8% of the controls. Table 2 displays the crude Ors and 95% CIs for this primary finding and other indicators of exposure.
Table 3 and Table 4 provide the results of logistic regression modeling for number of household smokers and number of maternal cigarettes smoked per day, respectively, as these were the exposures shown to be most important in previous analyses. The model reduction strategy involved an examination of the contribution of groups of related variables to the overall significance of the models until the most parsimonious model was found. In separate models, the number of household smokers, the number of cigarettes smoked by the mother per day during the child's first 3 years of life, and child care location were strong and independent predictors of MED, after adjustment for other significant predictors in the models. Thirteen or more years of maternal education, 3 or more PUEs, and past year health services use were also significant correlates of MED in the models. Attendance at day care was a significant correlate of MED when the exposure was coded as number of maternal cigarettes smoked per day, but this variable failed to reach statistical significance when the exposure was coded as number of household smokers.
Table 5 displays a pattern suggestive of a dose-response relationship for level of exposure by number of maternal cigarettes smoked per day and MED. In this analysis, exposure categories were adjusted slightly because of small numbers in the 1 to 10 cigarettes smoked per day exposure level.
In the validation study, as shown in Figure 1, cotinine levels in the hair samples were compared by number of household smokers. Because the cotinine measures represented recent exposures, parental reports of smoke exposure for the most recent exposure period (age 3-6 years) were used for comparison. Cotinine distributions were positively skewed so distributions were transformed such that assumptions for statistical testing were more adequately met. The results were similar with and without variable transformation. The mean cotinine values were statistically higher according to reported number of household smokers, grouped as 0, 1, and 2 or more (P=.03). There was concordance between reports of current household smoking and in-home observations conducted by pairs of study personnel (including C.E.A.-B. for approximately 10% of the visits) for 95.7% of the visits.
Distribution of cotinine values by number of current household smokers according to parental report. The shaded areas represent the proportion of the distribution of cotinine values falling between the 25th and 75th percentiles; the middle bars, the distribution medians, which separate the number of total values in half; and the open circles, outliers, which are data points that lie outside the distribution. The distribution means are reported adjacent to each plot. The difference between groups was significant (P=.03, F test).
The percentage agreement on case or control status as reported by parents compared with physician records was 75.3% (55 of the 73 children for whom physician medical record information was available for at least 3 years of the preschool-age period). The κ for agreement between proportions was 0.56 (95% confidence interval, 0.39-0.72; P<.001). Of the 18 children for whom there was disagreement, 15 (83.3%) were classified by physicians as nondiseased when they had been classified by parental report as diseased and 3 (16.7%) were classified as diseased by physician report but nondiseased by parental report. For 7 of these children, sufficient information was available from the interview to confirm the original classification. However, uncertainty in status for the remaining 11 suggests a potential level of disease misclassification in the study of about 15%.
This study showed an elevated risk of frequent or persistent MED in children exposed to ETS during their first 3 years of life, even after consideration of many possible confounders. The biological mechanism for this association is unknown but could include impaired mucociliary clearance and irritation of adenoid tissue, resulting in increased histamine production, both of which can lead to eustachian tube dysfunction, or reduced immune response through altered phagocytic antibacterial defenses leading to more upper respiratory tract infections.
Because the proportion of the target population consenting to interviews was less than ideal, selection bias is a possible alternate explanation for this finding if smoking parents were less likely to participate. Because a low socioeconomic status (SES) is associated with smoking prevalence, we examined this issue with a regression analysis. Schools with a higher social risk (ie, those with more students with a low SES) had a slight tendency to provide fewer responses (response to interviews correlated with the social risk score of the school community, but the correlation was small [r=−0.25, r2=0.06]). The possibility of selection bias based on exposure classification was also assessed by reviewing the prevalence of exposure among controls. The proportion of households with 2 or more smokers was similar but slightly below the anticipated proportion based on recent Canadian population-based surveys (12% vs 16%).11,39 This type of bias would likely distort the finding toward the null hypothesis.
Selection bias for disease was also reviewed. The proportions of cases and controls whose parents consented to the interview were similar to those eligible overall (23.9% vs 27.0% for cases and 46.9% vs 47.1% for controls). Finally, the exposure-disease relationship was calculated for high-response schools (n=3 schools [53 children]), defined as those with 73% or more parents consenting to interviews, and the calculated OR for the primary hypothesis was higher than that of the original analysis (OR, 2.76; 95% CI, 0.44-18.35). Thus, several lines of evidence suggest that the direction of selection bias was toward the null hypothesis and that its magnitude was not large.
Blinding of the subjects to the primary study hypothesis probably provided some protection against selection bias based on differential exposure. Other indications also suggested that the reporting of exposure was reasonably valid. Cotinine values were significantly (P=.03) higher in a direction that concurred with reported number of household smokers. In-home observations noted clear evidence of household smoking in only 1 home that had been reported to have no smokers, and there was indirect evidence in 3 others. Thus, the proportion of incongruence in smoking reporting would have ranged from 1.1% to 4.3% at a maximum, which is consistent with that found in the literature.40 Finally, most parents interviewed were mothers, and associations were stronger for exposure indicators associated in part or exclusively with the mother. The pattern of associations found does not suggest large-scale deception about self-reported smoking status.
On the other hand, well-documented difficulties in the measurement of MED retrospectively,41,42 along with evidence from physician medical records in the substudy, suggest a greater magnitude of misclassification of disease. Whether this level of inaccuracy in reporting MED differed according to exposure status is a critical question in the assessment of misclassification bias. Smoking is more prevalent among adults with lower educational attainment and SES. However, in this study, a higher maternal educational level was independently associated with MED, possibly through better detection of symptoms, increased use of health services, or better reporting. Therefore, parents in smoking households, in which the maternal educational level and SES were lower, may have reported less disease than those in nonsmoking households.
There was good consistency with the literature for the finding of elevated risk of MED associated with out-of-home child care during the first 3 years of life.43 In bivariate analyses, elevated risk was evident for all child care indicator variables, including group size, age beginning group care, and duration of out-of-home care. However, collinearity problems required removal of some of these variables from the final models. On the other hand, breast-feeding was not independently associated with MED; in fact, the sign of model coefficient sizes suggested an association of risk. This is not congruent with the results of other studies, including a recent, well-designed, large, prospective cohort study that demonstrated a protective effect of breast-feeding.44 The study population may have been too homogeneous for SES and infant nutrition for such effects to manifest themselves. Mothers whose infants had early episodes of OM may have chosen or may have been counseled to extend the period of breast-feeding as a preventive measure against more episodes, as earlier episodes are predictive of subsequent episodes. Also, retrospective reporting of breast-feeding duration may have lacked precision. Finally, more highly educated mothers might have been more likely to breast-feed and less likely to smoke45; they also may have been better reporters of disease history.
None of various SES indicators, including tenure, crowding, income adjusted for family size and cost of living, and paternal educational level, was independently associated with MED apart from maternal educational level. The study population may have been too socioeconomically homogeneous for such an effect to manifest itself. The statistically significant elevated risk of MED found for maternal postsecondary education is counterintuitive, but it fits with an explanation proposed previously, ie, these mothers display a better awareness of early life symptoms, increased use of health services, and more accurate reporting. If the study was not sensitive enough to detect previously demonstrated associations for other risk or protective factors under the conditions of imperfect reporting and a fairly homogeneous population, this would suggest that passive smoke exposure must indeed be a robust risk factor.
The statistically significant independent association between the number of PUEs and MED was an unexpected finding. Estimates of the actual number of examinations were requested, but it is unknown how accurate reporting may have been. A study conducted in this same geographic area suggested that there is an association between number of PUEs and delayed speech (for which MED could be an intermediate outcome),46 but much more research, as well as a plausible biological mechanism, would be required before causality could be inferred. Perhaps women whose pregnancies are considered to be at risk tend to undergo more PUEs, and MED might be 1 of several adverse outcomes of high-risk pregnancies.
Although health services use is clearly a consequence of MED (ie, on the causal pathway), it was retained in the models for several reasons. There was some evidence from preliminary analyses that exposed children had increased health services use. Also, Froom and Culpepper43 have shown that children attending group child care are taken for health service visits more frequently. Under either of these scenarios, health care use may have operated as a confounding variable. Increased use may also lead to the identification of more episodes of MED.
The findings reported in this article are reasonably congruent with those reported in other recent articles. Kitchens47 and Ey et al,48 despite using different exposure and disease measurement strategies, found similar ORs of 1.65 (95% CI, 1.02-2.66) and 1.75 (95% CI, 1.03-2.95), respectively, for MED and ETS exposure. Stenstrom et al24 found a significant association between MED and having 1 or more smoking caregivers (OR, 2.68; 95% CI, 1.27-5.65). The larger effect size may have been the result of a comparison with pure controls. Finally, Collet et al49 found a statistically significant OR of 1.80 for 20 or more maternal cigarettes smoked per day after adjustment for maternal educational level, family income, family history of OM, group child care attendance, and sex.
Evidence was found in this study that heavy exposure to ETS among preschool-age children is associated with a history of MED, even after consideration of the role of confounding variables. In addition, a pattern suggestive of a dose-response relationship was found for the number of cigarettes smoked per day by mothers during their children's first 3 years of life and a history of recurrent or persistent OM in their first-grade children. However, the results may not be generalizable beyond relatively affluent and homogeneous urban populations; the ETS-MED association might well be stronger in children with greater social disadvantage, increased exposure to ETS, or both. The avoidance of passive smoke exposure by preschool-age children may have a considerable effect on the burden of morbidity due to MED.
Accepted for publication October 7, 1997.
This study was supported in part by funds from the Alberta Heritage Foundation for Medical Research (Dr Adair-Bischoff), Edmonton, and the Alberta Children's Hospital Foundation, Calgary.
Presented in part at the International Society for Environmental Epidemiology Conference, Edmonton, Alberta, August 18, 1996, and at the American Pediatric Society/Society for Pediatric Research Meeting, Washington, DC, May 5, 1997.
We thank the study staff and volunteers Cheryl Heynen, Jeanne Vidacovich, and Orasa Kovindha; staff of the Calgary Board of Education schools and the Calgary Catholic School District schools, Calgary, for allowing recruitment of families in their facilities; Barry Kimberley, MD, PhD, for clinical input; and Rollin Brant, PhD, for statistical consultation.
Editor's Note: Here is another example of children "taking it in the ear" for adults' trangressions.—Catherine D. DeAngelis, MD
Corresponding author: Carol E. Adair-Bischoff, MD, Department of Community Health Sciences, Faculty of Medicine, The University of Calgary, 3330 Hospital Dr NW, Calgary, Alberta, Canada T2N 4N1 (e-mail: email@example.com).
Adair-Bischoff CE, Sauve RS. Environmental Tobacco Smoke and Middle Ear Disease in Preschool-Age Children. Arch Pediatr Adolesc Med. 1998;152(2):127-133. doi:10.1001/archpedi.152.2.127