Smoke-free legislation and the incidence of paediatric respiratory infections and wheezing/asthma: interrupted time series analyses in the four UK nations

We investigated the association between introduction of smoke-free legislation in the UK (March 2006 for Scotland, April 2007 for Wales and Northern Ireland, and July 2007 for England) and the incidence of respiratory diseases among children. We extracted monthly counts of new diagnoses of wheezing/asthma and RTIs among children aged 0–12 years from all general practices in the Clinical Practice Research Datalink during 1997–2012. Interrupted time series analyses were performed using generalised additive mixed models, adjusting for underlying incidence trends, population size changes, seasonal factors, and pandemic influenza, as appropriate. 366,642 new wheezing/asthma diagnoses and 4,324,789 RTIs were observed over 9,536,003 patient-years. There was no statistically significant change in the incidence of wheezing/asthma after introduction of smoke-free legislation in England (incidence rate ratio (IRR) 0.94, 95% CI 0.81–1.09) or any other UK country (Scotland: IRR 0.99, 95% CI 0.83–1.19; Wales: IRR 1.09, 95% CI 0.89–1.35; Northern Ireland: IRR 0.96, 95% CI 0.76–1.22). Similarly no statistically significant changes in RTI incidence were demonstrated (England: IRR 0.95, 95% CI 0.86–1.06; Scotland: IRR 0.96, 95% CI 0.83–1.11; Wales: IRR 0.97, 95% CI 0.86–1.09; Northern Ireland: IRR 0.90, 95% CI 0.79–1.03). There were no demonstrable reductions in the incidence of paediatric wheezing/asthma or RTIs following introduction of smoke-free legislation in the UK.


Supplementary Table S1: Covariates included in models via backward selection process
All models furthermore included a thin plate spline to account for the underlying time trend, a cubic spline to account for seasonality, terms to account for variations in the number of days in a month and the number of days GP practices were open (there was no strong correlation between these), and an offset variable to indicate the number of patients at risk. In the QOF sensitivity analysis the underlying time trend was non-linear. RTI

LAY SUMMARY
Exposure to other people's tobacco smoke (second hand smoke; SHS) conveys an important health risk. Children are particularly vulnerable to adverse effects of SHS and are unable to influence their own degree of exposure. To protect people from SHS the World Health Organization (WHO) recommends implementation of smoke-free laws. Evidence is now emerging that smoke-free laws improve paediatric health, for example by decreasing hospital admissions for asthma. The effect of smoke-free legislation on the occurrence of asthma and respiratory infections among children that present to the GP has never been studied. We will study this effect of smoke-free across the UK, where smoke-free legislation was introduced at different time points in each of the four countries. Trends in the occurrence of asthma and respiratory infections will be investigated from 1997 to 2012 using data from the Clinical Practice Research Datalink (CPRD), which currently covers about 5-10% of GP practices in the UK. Effect of other factors that might influence trends in these diseases will be taken into account. To the best of our knowledge this is the first explorative study to assess the effect of smoke-free legislation on infant and child health in primary care and as such may help inform the development and implementation of additional measures to protect infants and children around the world from adverse effects of SHS.

OBJECTIVE, SPECIFIC AIMS AND RATIONALE
We aim to describe the association between introduction of smoke-free legislation in different countries in the UK and incidence changes in asthma and respiratory tract infections (RTIs) in primary care among children aged 12 and younger.

BACKGROUND
Tobacco use kills more than five million people annually making it the leading global cause of preventable death 1 . It is estimated that exposure to other people's tobacco smoke ("second-hand smoke" (SHS)) kills an additional 600.000 people worldwide each year, including 166.000 children under 15 years of age 1, 2 . Among non-smoking adults, SHS exposure also increases the risk of developing asthma, lung cancer and ischaemic heart disease 2 . In an attempt to reduce this substantial burden on passive smokers, the World Health Organization (WHO) has recommended that smoke-free indoor public environments are enforced through national legislation and that educational strategies are pursued in parallel to reduce SHS exposure in the home 3 . Studies have since shown that smoking bans effectively reduce SHS exposure, and through doing so reduce respiratory and sensory symptoms, sudden cardiac arrest, and admissions for acute myocardial infarction as well as associated mortality 4-6 .
As developing individuals, children are particularly vulnerable to the negative effects of SHS and they are unable to influence their own degree of exposure. Currently over 40% of children worldwide are regularly exposed to SHS 1 . SHS exposure both before and after birth increases the risk of adverse health outcomes in early life, and has been shown to increase the risk of stillbirth, preterm birth, low birth weight, and RTIs as well as asthma in childhood 7-12 . Children are thus particularly likely to benefit from smoke-free laws. Indeed, recent evaluations of the 2006 Scottish smoking ban showed reductions in preterm birth and low birth weight, as well as hospital admissions for asthma among children after its introduction 13, 14 . The latter finding has recently been confirmed in a similar study in England 15 .

Knowledge gap
It is important that evaluations of smoke-free laws are reproduced in several countries where smokefree laws were introduced at different time points given concerns that the evidence inevitably comes from quasi-experimental studies, which are inherently at high risk of bias 16 . Furthermore, an evaluation of child health effects of smoke-free legislation has hitherto been mainly based on the study of hospitalisations. We propose to build on this work by using data from the Clinical Practice Research Datalink (CPRD) database, which currently covers 5-10% of GP practices across the UK. This will provide an important community-based perspective to the current evidence base. From CPRD we will obtain disease trends for paediatric asthma and RTIs in primary care. Smoke-free laws were introduced at different time points in Scotland (March 26, 2006), Wales (April 2, 2007), Northern Ireland (April 30, 2007), and England (July 1, 2007), facilitating comparison of associated disease incidence trend changes among countries. As such the proposed study will serve to further inform the development and implementation of global policy and strategies to further reduce SHS exposure in a particularly vulnerable population.
We hypothesise that introduction of smoke-free laws across the UK is associated with decreases in the incidence of asthma and RTIs in primary care among children aged 12 and younger.

STUDY TYPE
Hypothesis generating / descriptive study

STUDY DESIGN
Interrupted time series using generalised additive mixed models (GAMM)

SAMPLE SIZE / POWER CALCULATION
Sample size calculation for time-oriented analyses is complicated given the complexity of the models. Formal power calculations are non-existent for GAMM models. A previous epidemiological evaluation of the 2007 English smoking ban among children <15 years of age showed a significant sudden drop in asthma hospitalisations of 9% (95% confidence interval (CI) 7-11%) and a subsequent annual drop of 3% (95% CI 2-4%) among children <15 years of age 15 . A similar evaluation of the 2006 Scottish smoking ban demonstrated an 18% (95% CI 15-22%) annual reduction in asthma hospitalisations among children <15 years of age 13 . To the best of our knowledge no prior studies have evaluated the effect of smoke-free legislation on paediatric RTIs. Meta-analyses of observational studies show that among children the association between second-hand smoke exposure and RTIs is of greater magnitude than that of smoke exposure and asthma 2, 10-12 . Any effect of smoke-free legislation on paediatric RTIs is therefore likely to be larger than that estimated for asthma. Considering the much higher incidences of asthma and RTIs in primary care as opposed to secondary care and the larger time span of our proposed study, we are confident that our study is adequately powered to detect statistically significant and clinically relevant drops in incidences of the proposed outcomes, even though CPRD holds primary health care data for 'only' 5-10% of the UK population.

STUDY POPULATION
The study population will consist of all children aged 12 or younger registered with a CPRD practice for at least part of the study period (January 1 st 1997 to December 31 st 2012). The age cut-off is selected in an attempt to minimise the potential confounding effect of self-smoking.

SELECTION OF COMPARISON GROUP(S) OR CONTROLS
An interrupted time series analysis does not employ a comparison or control group in the usual sense of this term. GAMM models will be used to model monthly counts of the number of children diagnosed with asthma or a RTI from January 1997 to December 2012 taking into account the number of children registered in CPRD during the month and who are thus 'at risk' of developing the outcome. An incidence rate ratio (IRR) with 95% CI will be calculated to quantify any change in the incidence of asthma and RTI in the months after the introduction of smoke-free legislation compared to the months beforehand. The model will account for any underlying long-term trend in asthma/RTI incidence (e.g. a long-term increase or decrease), as well as seasonal variations in the outcome.

EXPOSURES, OUTCOMES AND COVARIATES
Exposure Every patient will be followed from the start of valid data collection or 1 January 1997, whichever comes latest, until the end of valid data collection, the end of the study period, or until 31 December of the year before the child becomes 13 years old. If the month of birth is available for a child, he/she will be followed until the latest day of the month before he/she becomes 13 years old. We have found however that month of birth is not available for about 58% of children in the database. For these children the date of first registration with a CPRD practice will be taken as a proxy for date of birth if they occur in the same year. If birth year precedes the year of registration, July 1 will be assigned as the birth month.
The exposure in this study is the intervention under investigation, namely exposure to a national smoking ban.

Asthma
Incident asthma diagnoses in primary care will be evaluated, using the following definition: The earliest recording wheezing / asthma related medication defines the index date.
A new asthma diagnosis is thus made when any diagnostic code for asthma and/or any prescription of medication used for asthma are recorded for the first time in a particular patient. A full list of Read codes used for asthma diagnosis is provided in Appendix 1. A list of medications considered to be indicative of asthma for the purpose of this study can be provided upon request.
An inclusive definition of asthma is used for this study given the clinical heterogeneity of wheezing disorders in childhood and the inability to reliably distinguish between different wheezing phenotypes using routine health care data. Furthermore, most if not all wheezing phenotypes are expected to be affected by SHS exposure and are therefore potentially sensitive to an effect of smoke-free legislation.

Respiratory infection
The primary outcome will be all acute RTIs in primary care, using the following definition: -Primary outcome: o recording of a medical code indicative of a RTI (Appendix III). o and no recording of an RTI during the 14 days preceding the diagnosis The primary outcome will involve all acute upper and lower RTIs. A full list of Read codes used for respiratory tract infection diagnoses is provided as Appendix 2. A respiratory infection is considered new when being registered at least 14 days after any prior respiratory infection in order to exclude repeated GP visits for the same RTI.
Secondary outcomes will be all acute upper RTIs and all acute lower RTIs using the following definitions: -Secondary outcomes: o Upper RTIs: recording of a medical code indicative of an upper RTI (Appendix III) and no recording of any RTI during the 14 days preceding the diagnosis o Lower RTIs: recording of a medical code indicative of a lower RTI (Appendix III). and no recording of a lower RTI during the 14 days preceding the diagnosis

Covariates
In GAMM modelling data are evaluated at aggregate level producing a rate ratio based on case counts. GAMM models therefore do not allow incorporation of individual-level covariates. Time series modelling is based on the assumption that the population structure does not change over time. Although this is expected to hold for most covariates, a change in age structure of the population over time may be expected given the increasing birth rate in the UK. Given the fact that both asthma and RTIs are more common among younger children, this may produce a spurious increasing trend in both outcomes over time. We will therefore perform a sensitivity analysis to assess the effects of the smoking ban among age subgroups, as discussed below.
National-level mean monthly temperature will furthermore be included as a predictor variable in each model, using data freely available from the UK Meteorological Office.

Production of incidence rates
For every month, a denominator population will be defined. The denominator population consists of all patients aged 12 years and under who are enrolled on the 15 th day of that month in CPRD practices that contribute up-to-standard data. For each month we will identify the number of children with each of our outcomes. We will calculate incidence rates and plot these over time in order to visualise obvious changes in temporal patterns.

Model description
We will use GAMM to model monthly counts of the number of children with each outcome from January 1997 to December 2012 taking into account the number of children registered in CPRD during the month and who are thus 'at risk' of developing the outcome. An incidence rate ratio with 95% confidence interval will be calculated to quantify any change in the incidence of asthma and RTI in the months after the introduction of smoke-free legislation compared to the months beforehand. The model will account for any underlying long-term trend (linear or non-linear, as appropriate) in asthma/RTI incidence (e.g. a long-term increase or decrease), as well as seasonal variations in the outcome (using a cyclic cubic spline). Model residuals will be checked for any remaining autocorrelation, which will be accounted for using autoregressive and/or moving average terms if necessary. National-level mean monthly temperature will also be included as a predictor variable in each model. The primary analysis will be performed in England and will involve both primary and secondary outcomes. Secondary analyses will separately be performed in a similar manner -but for primary outcomes only -in Northern Ireland, Scotland, and Wales. All analyses will be performed using the R statistical software package.

Validity issues
A number of potential threats to validity exist within this study. Given the multitude of primary and secondary analyses as well as potential validity threats, we will not perform separate sensitivity analyses to address these. Instead, these issues will be outlined and discussed in the report, with a scope of potentially addressing them in future work. The following potential validity threats are considered: 1. Validity of recording An important assumption of time series analysis is that the quality and degree of recording are constant over time. Quality criteria will be applied as discussed previously. A theoretical threat to the consistency of the degree of recording is the 1 April 2004 introduction of the Quality and Outcomes Framework (QOF) which provides GPs with an incentive to record asthma diagnoses (not RTIs). Recording changes are expected to produce sudden changes in incidence rates that cannot be explained by external factors. Such changes are typically not present in databases measuring the same outcome using a different method. We will perform a basic assessment of this issue by visually comparing incidence plots obtained in this study to incidence rates of hospitalisations for asthma (as derived from HES and published by Millett et al 15 ), as well as RTIs (as derived from HES, unpublished data; ClinicalTrials.gov NCT01920165).

Data dependency
In contrast to the asthma outcome, RTI diagnoses are allowed to occur recurrently within the same individual. Such intra-individual RTIs will have a greater level of interdependency than inter-individual RTIs. As GAMM models work with stratum-specific counts, they cannot account for this data dependency. Previous similar studies using hospital department visits and hospitalisations as outcomes experienced the same validity threat, which was not specifically addressed in any of the studies as the consequential risk of bias was deemed to be low 15, 17, 18 .

Age
As mentioned previously, a change in age composition of the study cohort over time may result in spurious incidence changes of the outcomes of interest. As such changes occur gradually, this issue will in part be accounted for by accounting for underlying temporal trends in the models. As the primary aim of PCV is reducing severe pneumococcal disease, which is likely to represent a very small proportion of all RTIs presenting to the GP, consequential bias of PCV introduction is likely to be negligible in the current study.

Asthma definition
The primary outcome definition may result in misclassification of prevalent asthma cases entering the database as 'new' (incident) cases upon their first recorded visit for asthma or first prescription after database entry. This has been shown to be a particular problem for subjects entering a practice that is already in the database (e.g. after the up-to-standard date for that practice 19 . As most children will enter a new practice by being born rather than moving house and the cohort is dynamic, with new patients entering the database continuously, potential consequential bias is expected to be small.

RTI definition
Using the primary definition, RTIs are considered incident when not preceded by another RTI in the 14 days preceding the event. This pragmatic cut-off may potentially introduce bias in two directions: 1. Mild upper RTIs may follow each other up at <14 day intervals, resulting in underestimation of the true incidence of 'new' RTIs; and 2. Chronic RTI disease episodes may extend beyond the 14-day period thus resulting in overestimation of the true incidence when the GP is visited for the same infection at a >14 day interval. This will however only introduce bias if the likelihood of any of these two situations occurring, changes over time. This is considered unlikely and the consequential bias is therefore expected to be low in this study. 7. Denominator definition A very small proportion of patients will contribute to only part of the month, as they enter or leave the database during that month. In prior studies evaluating temporal smoking rate patterns, sensitivity analyses excluding such patients have shown negligible changes in outcome estimations.

PATIENT / USER GROUP INVOLVEMENT
We will liaise with members of The Netherlands Asthma Foundation.

LIMITATIONS OF THE STUDY DESIGN, DATA SOURCES AND ANALYTIC METHODS
The main limitation of this explorative study will be that a causal interpretation of the findings is not possible. For example, other policy changes may have taken place at the same time as the introduction of smoke-free legislation, which themselves may have led to changes in the outcomes of interest. However, interrupted time series analysis is among the strongest observational designs to infer possible causation where a randomised controlled trial is not ethical or feasible, as is the case with national smoke-free legislation. We will fully discuss the limitations of our analysis when preparing our work for publication. Other limitations have been discussed in the section "Validity issues" above.