Study sample
Participation in the Nurse Health Assessment (NHA) [
25] of Understanding Society: the UK Household Longitudinal Study (UKHLS) [
26] was invited from adults living in Great Britain during wave two of the UKHLS for the General Population Sample (GPS) and during wave three for the British Household Panel Survey (BHPS) sample. The GPS is a stratified, clustered, equal probability sample of residential addresses throughout the UK in 2009 [
27]. The BHPS began as a stratified random sample initiated in 1991 with country specific boost samples [
28]. NHA interviews consisted of a nurse undertaking physical functioning measures, anthropometrics and blood samples approximately five months after the main interview, beginning in May 2010. Of the 43,747 adult, British resident members of the GPS and BHPS who gave a full interview in waves two and three respectively [
29], participation in the NHA was limited to those who gave a full English language interview and were not pregnant. In the second year of wave two, selection was restricted to 81 % of primary sampling units in England to allow interviewing of the BHPS sample. Of 35,937 eligible to participate in the NHA, 20,700 (57.6 %) took part [
30].
Measurement
Outcome variable - forced expiratory volume in the first second of exhalation (FEV
1) is used to measure lung function here. FEV
1 was captured with the electronic NDD Easy On-PC spirometer in UKHLS [
29]. Normal levels are dependent on age, height, gender and ethnicity. FEV
1 was transformed into a percentage of the expected FEV
1 (FEV
1%) for a healthy non-smoking person of equivalent age, height, gender and ethnicity using guidelines from the Global Lung Function Initiative [
32]. A FEV
1% between 80 % and 120 % is considered normal, FEV1% below 80 % is considered obstructed. FEV
1 is the most widely used parameter to measure the mechanical properties of the lungs [
33] and is more reproducible than forced vital capacity (FVC) [
34]. FEV
1/FVC was not considered a useful measure for the purposes of this analysis as if both FVC and FEV
1 are reduced as in restrictive lung diseases and lung defects then a normal FEV
1/FVC result is produced. Measurement requires participants to make an effort and it needs to be done correctly to produce a high quality measurement. Obtaining the highest quality measurement, Grade A, required participants to produce two highest FVC and FEV
1 measurements within 100ml of each other and was only achieved by 51.3 % of those who provided measurement. Poor quality graded measurements were included in this analysis. FEV
1% is used here as a continuous variable.
Exposures of interest – there are two key exposures of interest here smoking and occupational exposure to hazards for lung function. Respondents were asked if they had ever smoked a cigarette, pipe or cigar and those who responded positively were asked if they ever did so nowadays, based on their responses to this they were classified as current, former and never smokers. A job exposure matrix assessing risk for chronic obstructive pulmonary disorder (COPD) measured occupational exposures to dusts, gases or fumes [
35]. It was linked to the SOC2000 classification of occupations and derived into whether participants were exposed to COPD risks in their current or last (if not employed) occupation. Those who were students, long term ill or disabled and carers were classified as unexposed.
Moderator – the key moderator is childhood SEP, this is in part because direct measures of socially patterned exposures, such as maternal smoking and ETS, were not available nor was birth weight which might more closely indicate the environment in utero. Maternal education was used, therefore, as a marker for such exposures by measuring childhood SEP. This was prioritised over paternal or household SEP as it was hypothesised that maternal SEP would capture the experiences in utero and in early childhood that are consequential for lung function, better than other childhood SEP measures. Paternal or household SEP may not reflect the resources available to women, important for this hypothesis, due to unequal sharing of resources within the household, which is more likely to impact negatively on women. There is a strong correlation between maternal and paternal SEP, but more missingness in the measure for fathers. Maternal occupation was asked in reference to when the respondent was aged 14 and thus may not reflect the period around birth posited here to influence lung function. Responses were derived into a dichotomous variable with categories of ‘no schooling or qualifications’ indicating disadvantaged childhood SEP, ‘some qualifications or post school qualifications indicating advantaged childhood SEP’ and ‘do not know’ (n=1,426) and ‘other’ (n=41) were classified as missing. Maternal education was asked of some GPS respondents in wave one and others in wave two, while some members of the BHPS sample responded to maternal education in wave 13 of BHPS. Responses obtained to maternal education in the three different waves were combined into one variable indicating advantaged or disadvantaged childhood SEP.
Confounders – a number of key confounders – age, sex, ethnicity, height and weight are not included in the models as these factors have already been included in the standardisation of lung function (see above). There are three additional confounders at the individual level – adult SEP, physical activity and obesity – as well as measures of household and area deprivation and pollution. As the exposures of interest, smoking and occupational hazards, are socially patterned and associated with other health behaviours that influence lung function; these were included in the analysis to prevent over estimation of their association. All individual level covariates were captured in wave two except waist circumference, which was captured in wave three for the BHPS sample. Educational attainment indicated adult SEP, the derived measure of highest qualification was grouped into ‘A level and higher’ indicating advantaged SEP and ‘GCSE and lower’ indicating disadvantaged. For physical activity, respondents were divided into those who participated in mild or moderate physical activity at least once a week or less than this. Waist circumference was used here to indicate obesity, very high waist circumference is defined as 102 centimetres or greater for a man and 88 centimetres or greater for a woman [
36]. Measurement and exclusions from the waist measurement are described elsewhere [
29]. Whether respondents were GPS or BHPS members was included as a covariate due to the different time lags in the collection of confounders.
ETS was indicated by whether the household contained a smoker. As ETS is socially patterned, household tenure was used as a proxy for household SEP. When tenure was rented from a Local Authority or Housing Association, it was classified as disadvantaged household SEP, all other tenure indicated advantaged SEP. Nitrogen dioxide, particulate matter, sulphur dioxide and benzene were captured by the ‘living environment’ domain of the 2010 Index of Multiple Deprivation (IMD). The IMD is derived for Lower Area Super Output Areas (LSOA). LSOA have varying sizes with populations between 1,000 and 3,000 individuals or 400 and 1,200 households. We use air pollution captured by the IMD for LSOA, which we linked to households LSOA in UKHLS [
26]. Modelled estimates of each pollutant obtained on a 1 kilometre grid were related to a standard value defined as a risk to health or ecosystems and then summed to create an overall measure [
37]. This was derived into a binary measure of whether an area was above or below the mean level of air pollution. The income domain of the 2010 IMD was included to control for area deprivation. A binary measurement of whether an area was above or below the mean level of income deprivation was created to indicate area deprivation here. More information on the measurement of air pollution and income deprivation is provided elsewhere [
37].
Statistical method
A mixed linear regression model, with three levels (individual, household and area) with main effects for each exposure of interest was fitted and then extended to include interaction effects between childhood SEP and smoking, physical activity, obesity and occupational exposures. As ETS and air pollution were measured within households and areas respectively, cross level interactions with childhood SEP were estimated. Random intercepts captured variation in lung function at the area and household levels but the coefficients for each parameter were assumed to have the same association across households and areas [
38]. Significant interactions indicate that the association between the exposure or health behaviour and lung function was different for those with disadvantaged childhood SEP compared to those from an advantaged background.
Analysis was undertaken using Stata 14 [
39]. Complete case analysis was employed without any imputation of missing values. This approach was taken because missingness was mainly the result of two key variables. Non-response was high on the FEV
1 and was unlikely to be random as poor lung function can affect the ability to provide lung function measurement, and on maternal education. Multiple imputation is an effective approach to addressing missingness when it is spread across a range of variables. Complete case analysis produces unbiased estimates if the outcome is not associated with being a complete case once confounders are controlled for [
40]. We estimated whether the outcome was associated with being a complete case using logistic regression with control for confounders, as it was not, complete case analysis is used here (full results available on request). The profile of the study population was compared to the analytical sample (those without non-response on any measure) in the presentation of descriptive results. Given high levels of missingess on the outcome variable, we carried out a sensitivity test
1. We reran the models incorporating those without lung function measurement as having low FEV
1 to assess if estimates were consistent with the main analysis. As an additional sensitivity test, the analysis was rerun separately for those with a ‘Grade A’ quality lung function measurement and all other levels of quality. The results to both reflect those presented here and hence are not included but are available on request.
The cross sectional weight for the combined NHA sample was used throughout to adjust for unequal selection probabilities and differential nonresponse to the NHA.