Sampling
In total 600 participants are selected from children attending primary schools located in the three prioritised and one less-exposed areas. A list of all schools in the study area is obtained from the Department of Education in the Western Cape Province. One or two schools from each area located nearest to the City of Cape Town metro air quality monitoring stations are selected. One hundred and fifty Grade-4 students in each study area are targeted. Grade-4 students are selected for the study since they will not leave primary school to enter senior school during the study period. The average age of Grade-4 students is 10 years; as such they are old enough to adhere to instructions during the various tests and reasonably confident to answer simple questions regarding their symptoms. This age group also provides a key phase in the onset of childhood asthma which often is preceded by atopic dermatitis and allergic rhinitis (with reference to the ‘atopic march’). Although, there are gender differences in asthma reported in the literature, all eligible Grade-4 students are included in the study and gender will be adjusted-for, stratified and assessed for interaction in the analyses.
After meeting the school principal, obtaining permission from the schools board, and obtaining the Grade-4 class lists and addresses, the houses of the school children are visited by trained field staff to obtain the caregivers (parent or guardian) consent. Random sampling is used to select 75 students from the list of consented children when the number of consented children exceeds 75 in each school.
All Grade-4 students attending selected schools qualify for inclusion in the study. Baseline data is collected between February 2015 and August 2015, while the follow-up data is obtained from May 2016 to September 2016. However, spirometry is rescheduled for participants with a positive response obtained from a pre-test questionnaire to any of the following items: If the child had a recent operation (in last 12 months); If the child feels like vomiting or has any pain; If the child is being treated for Tuberculosis; If the child had flu, sinusitis or lung infection in the last 3 weeks. However, children with positive response to ‘having epilepsy’ are excluded from doing spirometry.
Analysis protocol for objectives 3
i) Protocol for objectives 3a.
Exposures: In the panel study, the hourly and daily averages of each pollutants (24-h averages for PM, SO2 and NO2, and 8-h average for O3) and airborne pollens will be determined for the same day as well as for a lag period of 7-days.
Outcomes: Daily nadir (minimum best) and intraday variability of peak expiratory flow (PEF) and FEV1 will be used as markers for the worsening of asthma.
Covariates: The potential confounders determined apriori include: age, gender, race/ethnicity, asthma medication use, respiratory allergy history, atopic status, BMI, hour and day (of the week) of PEF, season, and temperature. Other covariates with P-value < 0.20 from the bivariate analysis will also be included in the multivariate model.
Analytical method: A general additive multiple linear regression model (GAM) will be fitted for predicting the percent change in within day PEF and FEV1, and nadir PEF and FEV1 for an interquartile range (IQR) increase in particular pollutant and airborne pollen while taking into consideration possible confounders and effect modifiers including penalized splines for seasonal and meteorological variations. The IQR scaling will enable the percentages to be directly relevant to the exposure experienced by the participants and will also make the percentages for different pollutants directly comparable to each other. To account for within-subject correlation, Generalized Estimating Equation (GEE) regression model will also be used as it focuses on estimating the average response through the fixed-effect or random-effect, as opposed to a regression parameters that predicts the effect of changing one or more covariates on a given participant.
ii) Protocol for objectives 3b.
Exposures: The hourly and daily averages of each pollutants (24-h averages for PM, SO2 and NO2, and 8-h average for O3) and airborne pollens will be used to construct a lag of up to 30 days. The annual averages of air pollutants from the LUR will also be considered.
Outcomes: Asthma symptom score (ASS) will be on an ordinal scale ranging from 0 to 8 as defined above. Airway obstruction will be defined as a dichotomous variable with reduced FEV1 (less than 80% of the predicted value) and a reduced FEV1/FVC ratio (less than 0.8). Airway inflammation will be assessed using the log-transformed FeNO levels on a continuous scale and also as a polychotomous (ordered) variable as categorized above.
Covariates: The potential confounders include; age, gender, race/ethnicity, asthma medication use, respiratory allergy history, atopic status, BMI, and day (of the week) of test date, season, and temperature. Other covariates with P-value < 0.20 from the bivariate analysis will also be included in the multivariate model.
Analytical method: Multiple linear regression will be used to model the associations between IQR range increase (or a 10 μg/m3 increase where applicable) in pollutants (air pollutants and airborne pollen) and continuous outcome of interest while taking into consideration possible confounders and effect modifiers. Multiple logistic regression will be used in the case of a binary outcome while multiple ordinal regression will be used for ordinal outcomes such as ASS. In addition, to examine the co-dependency of the various pollutants, the correlation structure will be used to guide the selection of a two-pollutant model to avoid potential multi-collinearity. Sensitivity analysis on the lag structure will also be done using different exposure windows up to 60 days prior to examinations, whenever appropriate, to evaluate the temporality of possible effects. For the purpose of this analysis, long-term effects of pollution levels will be investigated as potential confounders or effect modifiers of the short-term effects in models that will include random intercepts.
iii) Protocol for objectives 3c.
Exposure: The change in annual averages of 24-h period for NO2, SO2 and PM, and 8-h daytime average for O3, including airborne pollen levels across the cohort study period.
Outcomes: Changes in ASS at follow-up, changes in levels of FeNO (∆FeNO) or percentage predicted FEV1 (∆%FEV1) and FEV1/FVC ratio (∆FEV1/FVC) between the two study periods.
Covariates: The potential confounders include; age, gender, race/ethnicity, asthma medication use, respiratory allergy history, atopic status, BMI, hour and day (of the week) of test, season, and temperature. Adjustment for short-term effects of air pollution using appropriate lag structure at each study period together with potential confounders and effect modifiers taken into consideration, will help assess the independent effects of long-term air pollution. Time independent covariates (∆Z) such as race/ethnicity effects will be assessed together with time-elapsed between the two child-specific yearly test dates (∆Age), while time-varying covariates (∆W) effects will be assessed considering all possible transitions or changes over time for categorical and continuous covariates respectively. The most appropriate linear distributed lag models will be chosen from all possible types of lag-based models for short-term effects of air pollution, and the model selection will be based on the Akaike Information Criterion (AIC). The confounding effect of ambient temperature will be tested using the lag structure selected for the short-term air pollutant effects. Potential effect modifiers such as gender, asthma status, respiratory allergy, baseline outcome status (obstruction status, FeNO levels) and season will be tested in the final model.
Analytical method: Two strategies will be employed in these analyses. The first is a cohort analysis starting with a disease-free cohort to explore new-onset of asthma, while the second will be a change-analysis to investigate the changes in the respiratory outcomes over time. To explore the association between the asthma score and pollutant concentrations both defined at follow-up, in a subpopulation reporting neither symptoms nor asthma at baseline, a negative binomial regression model (with a log link) will be used to account for extra-Poisson variation due to the distribution of the asthma score, with a majority of zeros. The result will be expressed as ratios of the mean asthma scores (RMS). The pollutants effect will be scaled for an increase of 10 μg/m3 higher concentration. The subpopulation of interest will be considered a sample being in all likelihood free of asthma at baseline. Thus, new onset of symptoms (a reflection of asthma incidence) might be interpreted following the occurrence of symptoms at follow-up. To account for the participant reporting only one of the symptoms at follow-up, further analysis will be performed by comparing those with none or only one symptoms (coded as participants free of symptoms) at follow-up with those with at least two symptoms. This will also be followed-up by considering those reporting none, one or two symptoms as participants free of symptoms, comparing them with participants reporting at least three symptoms.
Multiple linear regression will be used to assess the relationship between changes in levels of FeNO (∆FeNO) or percentage predicted FEV
1 (∆%FEV
1) and FEV
1/FVC ratio (∆FEV
1/FVC) between the two study periods and the corresponding changes in ‘long term’ and ‘short term’ air pollution and airborne pollen levels. A 12-month period prior to the day of tests will be used to assess the effects of changes in long term pollution levels, while adjusting for short term levels based on lags of up to 60 days prior to the day of tests. The model will assume a general form provided below;
$$ \Delta \mathrm{FeNO}\ \left(\Delta \%{\mathrm{FEV}}_1\ \mathrm{or}\ \Delta {\mathrm{FEV}}_1/\mathrm{FVC}\right)={\upbeta}_0\Delta \mathrm{Age}+{\upbeta}_1{\Delta \mathrm{AP}}^{\mathrm{LT}}+{\upbeta}_2{\Delta \mathrm{AP}}^{\mathrm{ST}}+{\upbeta}_3{\Delta \mathrm{P}}^{\mathrm{LT}}+{\upbeta}_4{\Delta \mathrm{P}}^{\mathrm{ST}}+{\upbeta}_5{\Delta \mathrm{Temp}}^{\mathrm{ST}}+{\upgamma}^{\mathrm{T}}\mathrm{Z}\ \mathrm{x}\ \Delta \mathrm{Age}+{\upgamma}^{\mathrm{T}}\Delta \mathrm{W}+\upvarepsilon $$
where ∆Age, ∆AP
LT, ∆AP
ST, ∆P
LT, ∆P
ST, and ∆Temp
ST denotes the time elapsed between the two tests, changes in long-term air pollution levels, short-term pollution, changes in long-term pollen levels, short-term pollen levels, and temperature levels, respectively. The change-on-change model will allow for exploring the determinants of change in FeNO or %FEV
1 or FEV
1/FVC rather than determinants of level of the actual FeNO or %FEV
1 or FEV
1/FVC. In an attempt to explore the change in obstructive pattern defined by reduced FEV
1 less than 80% of the predicted value and a reduced FEV
1/FVC ratio less than 0.8, a binary variable will be created with participant changing from non-obstructive airways to obstructive airways coded as 1; those without change in obstructive airways pattern also coded as 1; those with changes from obstructive airways to non-obstructive airways coded as 0; and those without change in non-obstructive airways pattern also coded as 0. Thus, a multiple logistic regression will be used to model the association between change in obstructive patterns and change in exposure taking into consideration confounding and effect modifiers.
Analysis protocol for objectives 4
Predictor: Baseline FeNO levels.
Outcome: Crude incidence rates for new-onset asthma (defined by increasing ASS or presence of new asthma symptom) will be calculated by dividing the number of cases by the total person-years at risk. Follow-up is considered complete at the time of reported diagnosis for children who developed new-onset asthma.
Covariates: The potential confounders include; age, gender, race/ethnicity, asthma medication use, respiratory allergy history, atopic status, BMI, season, temperature including average ambient air pollution and airborne pollen levels on the day of the FeNO measurement.
Analytical approach: To investigate the relationship between baseline FeNO with new-onset asthma, incident rates will be calculated while exploring series of multivariate modelling approaches to adjust for potential confounders and heterogeneity of effects within subgroups of children. A cohort analyses starting with a disease-free cohort will be fitted to investigate the association between FeNO and new-onset asthma adjusting for possible confounders including lifetime history of wheezing. Fitted models with appropriate interaction terms with statistical significance tested by partial likelihood ratio test will be used to assess heterogeneity of associations among subgroups. Thereafter, stratified analyses will be performed in the presence of such significant interaction. Splines and piecewise cubic polynomials joined smoothly at a number of breakpoints (knots) will be used to explore the nonlinearity nature of FeNO effects.