Study design, setting and population

During October 2015, we conducted a cross-sectional study among high-school students in Hawalli – one of the six Governorates of Kuwait to: 1) assess the prevalence of household SHS exposure; 2) assess the prevalence of various allergic conditions including asthma, atopic dermatitis; and 3) evaluate the association between household SHS exposure and allergic morbidity including asthma and atopic dermatitis. The foremost contemplation in selecting this study population was their high expected frequency of household SHS exposure, its duration, and expected high prevalence of allergic morbidities. We obtained a list of schools located in Hawalli Governorate from the Ministry of Educations’ website, which included ten boys’ schools, nine girls’ schools in the public sector, and forty-five private schools with a co-education system. From this list, we selected nine schools as a sample of convenience, which included public-sector schools for boys (3), girls (3) and private-sector schools with co-education system (3). The students of either sex, of any nativity (Kuwaiti, non-Kuwaiti), currently enrolled in grade 11 or 12 at public or private-sector high schools, were eligible for inclusion in the study. The details of study design, setting, population and data collection have been previously described elsewhere^4,5.

Questionnaire

A structured and self-administered questionnaire was developed in English for the collection of data on sociodemographic characteristics, self-reported smoking status, and household SHS exposure, as described elsewhere^4,5. The standardized international study of asthma and allergies in childhood (ISAAC) core questionnaire was used for diagnosis of asthma and atopic dermatitis⁶. In this study, a respondent was regarded as an asthmatic, if during the past 12 months, he or she reported to have had four or more episodes of wheezing, or one or more episodes of wheezing with the use of an inhaler, or ever diagnosed as asthmatic by a physician⁶. Moreover, a respondent was considered to have had atopic dermatitis, if he or she had recurrent itchy rash for at least 6 months, and localization at the folds of the elbows, behind the knees, in front of the ankles, or around the neck, or diagnosed to have had atopic dermatitis by a physician⁶. For the diagnosis of asthma and atopic dermatitis among children and adolescents, the estimates of diagnostic parameters of the ISAAC questionnaire including sensitivity (≥87.5%), specificity (≥80.4%), internal consistency (Cronbach’s alpha=0.81)⁷, and reliability (Kappa statistic range=0.81–0.85) have been reported earlier^8,9. The questionnaire was initially developed in English and the final version was also translated into Arabic for actual use. The questionnaire was pre-tested on 20 students similar to our potential study participants and modifications were made as needed. The final questionnaire comprised 21 questions, and on average took five minutes for its completion^4,5.

Data collection

We intended to enroll about 100 senior students from grades 11 and 12 of the selected schools. We approached and sought the help of the teacher in-charge of the class for completion of the questionnaire by the students at the end of the class. The study aims were shared with the students by the respective class teachers, and were informed that their study participation was voluntary. Moreover, the students were assured of the confidentiality of the collected information. Afterwards, consenting students were invited to complete the written consent form attached to the study questionnaire. We adopted a similar procedure for data collection, both in public-sector and private schools. For this cross-sectional study, we estimated a sample size of 827 students to assess the prevalence of self-reported allergic morbidity, including asthma and atopic dermatitis, at a 95% confidence level (CI), with 2.5% bound on the error of estimation assuming a prevalence of allergic morbidity of 16% in our target population^10,11. The computed sample size was increased to 850 students to account for any potential refusals.

Data analysis

To characterize the study sample, descriptive statistics including mean with standard deviation (SD) for quantitative variables, and frequencies and proportions (%) for categorical variables, were computed. The prevalence (%) of household SHS exposure, allergic morbidities, including asthma and atopic dermatitis, were computed. Previously, we have presented elsewhere the description of statistical methods used for univariable and multivariable analyses of associations between sociodemographics, risk factors and atopic dermatitis, as well as for asthma⁵. All the statistical tests were two-tailed at a prespecified level of significance (α=0.05). For causal mediation analysis, the directed acyclic graph (Figure 1) depicts the assumed causal relationships between household SHS exposure (exposure), self-reported smoking status, asthma status (mediators), sex, nativity (pre-exposure confounders) and the atopic dermatitis status (outcome). The total effect (TE) of household SHS exposure among adolescents was decomposed to natural direct effect (NDE) and natural indirect effect (NIE), with two mediators (asthma status, self-reported smoking status) conditioned on two pre-exposure covariates (sex, nativity). In this study, the NDE contrasts the odds of atopic dermatitis among adolescents with household SHS exposure and adolescents without household SHS exposure; if for all adolescents, the level of asthma status and self-reported smoking status were set to levels that would have been observed had the adolescent been without household SHS exposure. The NIE contrasts the odds of atopic dermatitis among adolescents with household SHS exposure with their observed level of asthma status and self-reported smoking status to the counterfactual odds, where levels of asthma status and self-reported smoking status were set to levels that would have been observed had the adolescents with household SHS exposure been without household SHS exposure. NDE and NIE were averaged over all adolescents, including asthmatic and non-asthmatic, smokers and non-smokers) and were also evaluated at mean population levels of the covariates, including sex and nativity.

Figure 1

Directed acyclic graph for hypothesized causal mediation analysis of household secondhand tobacco smoke exposure effect (E) on the atopic dermatitis (Y) risk mediated through asthma, and self-reported smoking status (mediators) conditional on sex, and nationality (pre-exposure covariates) among adolescents in Kuwait, October 2015 (N=747)

To estimate the NDE and NIE, we conducted a nonparametric causal mediation analysis using the IOW method within a counterfactual framework, which is more robust against model misspecification such as interaction and nonlinearity¹². Furthermore, when multiple mediators are considered concurrently, this is an appropriate approach, which takes advantage of the invariance property of the odds ratio. IOW synthesized information on the odds ratio (OR) between household SHS exposure, self-reported smoking status and asthma into a weight, which was then used to estimate NDE via a weighted regression analysis of atopic dermatitis. Subsequently, NIE is estimated by subtracting NDE from the TE obtained from the atopic dermatitis model without weights. The 95% CI for the parameters’ estimates were obtained by nonparametric bootstrapping with 1000 replications. The proportion of the TE jointly mediated by two mediators was computed for the model employed. We did not explicitly carry out power calculation for the mediation analysis, mainly because current literature on this topic is fairly limited, and, so far, no specific method is available¹³. However, it has been argued that for a single-level mediation analysis using the percentile bootstrap method, a sample size of 558 gives 80% power to estimate efficient NIE and NDE¹⁴. In this current evaluation with a given sample size, we hope to have a somewhat larger power than 80% for the estimation of NIE and NDE.

The validity of the nonparametrically computed NDE and NIE from the observational data depends on the identification assumptions, i.e. there are no unobserved confounding effects for the association of: 1) exposure-mediators, 2) mediators-outcome, or 3) exposure-outcome, conditional on the pre-exposure confounders. An additional requisite assumption was that there were no confounding variables of the mediator-outcome relationship affected by the exposure. To evaluate the robustness of the results to the presence of unobserved confounding, sensitivity analysis was carried out¹⁵. In the sensitivity analysis, estimates of NDE and NIE under the given level of unobserved confounding were obtained. Sensitivity parameters’ estimates were used to quantify the influence of potential unobserved confounders. The purpose was to estimate the effects that would have been obtained had the unobserved confounders’ effects been accounted for. Missing values were handled by list-wise deletion, and p<0.05 was considered statistically significant for all analyses. The study protocol was approved by the Kuwait University Health Sciences Center Ethical Committee(# 1958/10/11/2015). Formal permission was sought from the Ministry of Education to conduct this study at public-sector schools and from the principals of selected private-sector schools. The study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines¹⁶.

Strengths and limitations

This is a novel study reporting causal mediation analysis of NIE of household SHS exposure through two modifiable mediators, i.e. asthma and self-reported smoking status, along with the NDE and TE on atopic dermatitis risk among adolescents. However, the following limitations of this study should be considered in the interpretation of the results. First, the study participants were enrolled as a sample of convenience; therefore, generalizability of the results needs to be exercised with care. Second, we could not account for the exact number of smokers at home, including the relationship of smokers with the adolescents in the study. These variables could have differential effects on the household SHS exposure dose and the two mediators, i.e. asthma status, and self-reported smoking status. Future studies may evaluate the dose-response assessment of this exposure. Third, the data collection was undertaken using a cross-sectional study design; therefore, temporal ordering of exposure, mediators and outcome was difficult to determine. Therefore, causal inference needs to be considered with care. Fourth, the data for this study were collected back in October 2015. Therefore, the magnitudes of the disease conditions (i.e. asthma, atopic dermatitis) and/ or risk factors (i.e. active tobacco smoking or household SHS exposure) might have undergone temporal variation. However, a recent study among university students reported the prevalence of active smoking as 22.4%, which is substantially and understandably more than the 12.4% recorded in our study. The age difference between the participants in the two studies seems to have contributed to the variation in the prevalence of active smoking recorded in the two studies. Nevertheless, the prevalences of household SHS exposure (54.1% vs 48.9%), self-reported asthma (20.6% vs 20.0%) and atopic dermatitis (14.9% in both studies) were about the same as reported in our study and the recent study¹⁹. Finally, most of the measured variables were based on recall; however, since participants were adolescents, it is expected that they have good recall ability, with minimal recall bias in the measurement of study variables.