You are here

  • Home
  • Archive
  • Volume 27, Issue e2
  • Smoke-free legislation and paediatric hospitalisations for acute respiratory tract infections: national quasi-experimental study with unexpected findings and important methodological implications

Article Text

Article menu

Download PDFPDF

Research paper
Smoke-free legislation and paediatric hospitalisations for acute respiratory tract infections: national quasi-experimental study with unexpected findings and important methodological implications
  1. Jasper V Been1,2,3,4,
  2. Daniel F Mackay5,
  3. Christopher Millett6,
  4. Ireneous Soyiri3,
  5. Constant P van Schayck3,4,
  6. Jill P Pell5,
  7. Aziz Sheikh3,4,7
  1. 1 Division of Neonatology, Department of Paediatrics, Erasmus MC – Sophia Children’s Hospital, Rotterdam, The Netherlands
  2. 2 Department of Obstetrics and Gynaecology, Erasmus MC – Sophia Children’s Hospital, Rotterdam, Netherlands
  3. 3 Centre of Medical Informatics, Usher Institute of Population Health Sciences and Informatics, The University of Edinburgh, Edinburgh, UK
  4. 4 Care and Public Health Research Institute, Maastricht University, Maastricht, The Netherlands
  5. 5 Institute of Health and Wellbeing, University of Glasgow, Glasgow, UK
  6. 6 Public Health Policy Evaluation Unit, School of Public Health, Imperial College, London, UK
  7. 7 Division of General Internal Medicine and Primary Care, Brigham and Women’s Hospital/Harvard Medical School, Boston, Massachusetts, USA
  1. Correspondence to Dr Jasper V Been, Division of Neonatology, Department of Paediatrics, Erasmus MC – SophiaChildren’s Hospital, Rotterdam 3000 CA , Netherlands; j.been{at}erasmusmc.nl

Abstract

Objectives We investigated whether Scottish implementation of smoke-free legislation was associated with a reduction in unplanned hospitalisations or deaths (‘events’) due to respiratory tract infections (RTIs) among children.

Design Interrupted time series (ITS).

Setting/participants Children aged 0–12 years living in Scotland during 1996–2012.

Intervention National comprehensive smoke-free legislation (March 2006).

Main outcome measure Acute RTI events in the Scottish Morbidity Record-01 and/or National Records of Scotland Death Records.

Results 135 134 RTI events were observed over 155 million patient-months. In our prespecified negative binomial regression model accounting for underlying temporal trends, seasonality, sex, age group, region, urbanisation level, socioeconomic status and seven-valent pneumococcal vaccination status, smoke-free legislation was associated with an immediate rise in RTI events (incidence rate ratio (IRR)=1.24, 95% CI 1.20 to 1.28) and an additional gradual increase (IRR=1.05/year, 95% CI 1.05 to 1.06). Given this unanticipated finding, we conducted a number of post hoc exploratory analyses. Among these, automatic break point detection indicated that the rise in RTI events actually preceded the smoke-free law by 16 months. When accounting for this break point, smoke-free legislation was associated with a gradual decrease in acute RTI events: IRR=0.91/year, 95% CI 0.87 to 0.96.

Conclusions Our prespecified ITS approach suggested that implementation of smoke-free legislation in Scotland was associated with an increase in paediatric RTI events. We were concerned that this result, which contradicted published evidence, was spurious. The association was indeed reversed when accounting for an unanticipated antecedent break point in the temporal trend, suggesting that the legislation may in fact be protective. ITS analyses should be subjected to comprehensive robustness checks to assess consistency.

  • prevention
  • priority/special populations
  • public policy
  • secondhand smoke

http://dx.doi.org/10.1136/tobaccocontrol-2017-053801

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Background

    Tobacco smoking continues to cause a considerable burden of death and disease worldwide.1 2 It is estimated that 40%–50% of children globally are regularly exposed to secondhand smoke (SHS).3 4 Children are particularly vulnerable to the adverse effects of SHS exposure as their bodies are still undergoing development and, especially when very young, cannot influence their own degree of exposure. Among children under 5 years of age, exposure to SHS has been estimated to cause 165 000 deaths and almost six million disability-adjusted life years annually through lower respiratory tract infections (RTIs) alone.3 Additional adverse paediatric health outcomes associated with SHS exposure include otitis media with effusion,5 invasive meningococcal disease6 and wheezing disorders including asthma.7 8

    There is a pressing need to identify effective approaches to reducing child SHS exposure and through so doing improve child health. The effectiveness of individual or family-level interventions to reduce SHS exposure has generally been disappointing.9 At a population level, governmental policies aimed at reducing tobacco smoking and SHS exposure have the potential to also reduce child SHS exposure. Comprehensive smoke-free laws and tobacco tax increases have been shown to be associated with improved respiratory health among children.10–13 Evaluation of the effectiveness of such policies is however complicated by the fact that they are generally not amendable to implementation in a randomised fashion.14

    Quasi-experimental studies, such as interrupted time series (ITS) studies, are advocated as a next best alternative to randomised designs when evaluating the impact of population-level policy changes.15 Attribution of causality is however challenging because of the inherent risks of bias and confounding. Findings are in addition sensitive to choice and specification of the statistical modelling technique used. Prespecification of a detailed statistical analysis plan has been advocated to reduce the associated risk of data dredging, to encourage publication irrespective of a study’s findings and to promote the reproducibility of science.16 17

    We investigated whether the implementation of comprehensive smoke-free legislation in Scotland was associated with changes in the number of unplanned paediatric hospitalisations or deaths due to RTIs. Introduction of the Scottish smoke-free law was followed by significant reductions in reported SHS exposure in public places and the home environment among school-age children, along with a –40% (95% CI –47 to –32, p<0.001) reduction in mean salivary cotinine concentrations.18 Previous studies in England, Hong Kong and the USA have identified significant reductions in hospital admissions for acute RTIs among children following introduction of smoke-free laws.10 12 19 20 Based on these previous studies and the wider evidence of the health impact of smoke-free legislation,11 21–24 we hypothesised that the Scottish smoke-free law would be associated with a reduction in acute RTI events.

    Methods

    This study was conducted according to a protocol developed a priori (National Services Scotland reference: PAC 04/12 IR – XRB13092; online supplementary file 1). We analysed the association between introduction of smoke-free legislation in Scotland and the incidence rate of unplanned hospital admissions or deaths due to acute RTIs among children aged ≤12 years. STROBE and RECORD statements were followed to guide reporting. Given use of fully anonymised routinely collected healthcare data, we received an exemption from formal ethical assessment for this study.

    Supplementary file 1

    Smoke-free legislation

    On 26 March 2006, a national law came into force overnight in Scotland prohibiting smoking in enclosed public places (ie, bars, restaurants, hotels, shops, shopping centres, libraries, archives, museums, galleries, entertainment premises (eg, cinemas, concert halls, theatres, gaming and amusement arcades and discotheques), film studios, assembly halls, conference centres, exhibition halls, public toilets, clubs premises, educational institutions (eg, schools), care homes and shelters, healthcare premises (eg, hospitals), child care premises (eg, nurseries), religious premises (eg, churches), sports centres, public transportation facilities (eg, airports) and vehicles and public telephone kiosks) and workplaces (including offshore facilities and work vehicles), with very few exemptions (ie, residential accommodation, designated rooms in adult care homes and adult hospices, designated rooms in psychiatric hospitals and psychiatric units, designated hotel bedrooms, designated detention or interview rooms, designated laboratory rooms, Her Majesty’s submarines and ships of the Royal Fleet Auxiliary and private vehicles).25 Of inspected premises, 96%–99% were compliant with the law in the first year following the law’s introduction and 94%–97% and 95%–97% in the second and third year, respectively.26

    Study population and period

    We included data from all children aged ≤12 years who were resident in Scotland at any time during the study period and had not yet experienced an RTI event before the study period. Children aged 13 years and above were excluded in an attempt to limit potential confounding by active smoking. We included data on all first unplanned hospital admissions or deaths due to acute RTIs (composite outcome: ‘events’) occurring between 1 January 1996 and 31 December 2012 (ie, the most recent data available at the time of data extraction).

    Outcome definitions

    Our primary outcome was the incidence rate of acute RTI events. Deaths were included to account for children that died due to an acute RTI before having reached the hospital. Secondary outcomes were the separate incidence rates of acute upper and lower RTI events. In these analyses, events that contained both a code for an upper and a lower RTI were counted as lower RTI events. For the purpose of this study, an event was considered to be associated with an acute RTI if this had been registered as either a primary or secondary diagnosis. The following International Classification of Diseases, 10th revision codes were included: upper RTIs: A37, H66-67, J00-06 and J09-11 (not J10.0/J11.0) and lower RTIs: J10.0/J11.0, J12-18, J20-22 and J40-42 (online supplementary file 2). To avoid contamination of our outcome of interest with asthma exacerbations that may or may not have been due to RTIs, events where asthma was recorded as the primary diagnosis were excluded. In order to prevent dependency of data due to individual children experiencing multiple RTI events during the study period, only first events (ie, admission or death, whichever came first) were included. For eligible children born before start of the study period (ie, 1 January 1996), a look-back period of 12 years was applied to determine whether the child previously experienced an acute RTI hospitalisation.

    Supplementary file 2

    Data sources

    Data on acute RTI hospitalisations were retrieved from the Scottish Morbidity Record-01, a national database collecting data on all hospital admissions among Scottish residents. Deaths due to acute RTIs were identified from National Records of Scotland Death Records. Data on individual pneumococcal conjugate vaccine (PCV) vaccination status were retrieved from the Scottish Immunisation and Recall System database, which collects national vaccination data. Individual-level data were linked across the different databases by electronic Data Research and Innovation Service (eDRIS) staff at Information Services Division Scotland using the unique Community Health Index (CHI) identifier before being made available to the researchers. Air quality data were obtained from the UK Governmental Department for Environment Food and Rural Affairs and linked to the main data document by the researchers.27

    Data handling and covariates

    The numbers of children at risk and those experiencing a first acute RTI event were aggregated by eDRIS staff into strata based on all possible combinations of the following covariates: month, year, sex (male; female); age group (<5 years; ≥5 years), region (according to health board of residence: South-West; South-East; North), urbanisation level (according to residential post code: urban; rural), socioeconomic status (quintiles of Scottish Index of Multiple Deprivation (SIMD28; 2006 version) based on residential postcode) and PCV vaccination (yes; no). On 4 September 2006, PCV was introduced into the childhood immunisations schedule at 2, 4 and 13 months of age, with a catch-up programme for children born from 5 September 2004.29 Given the close temporal proximity of PCV introduction to that of the smoke-free law in Scotland, we linked data on RTI events to PCV vaccination status at an individual level to address potential confounding.10

    In a post hoc analysis, we extracted air quality data for all Scottish monitoring sites that collected data during the study period: carbon monoxide (mg/m3), nitric oxide (µg/m3), nitrogen dioxide (NO2; µg/m3), nitrogen oxides as NO2 (µg/m3), sulfur dioxide (µg/m3), ozone (µg/m3) and particulate matter of diameter <10 µm (µg/m3). For each monitoring site, mean monthly values were calculated from mean daily values, and missing values were imputed using linear interpolation. Availability of air quality data from fixed stations was patchy both across time and the different monitoring sites, hampering combination of data across sites. Fairly consistent data throughout the study period was only available for Glasgow city centre.

    Statistical analyses

    Negative binomial regression analysis was prespecified as our primary analysis, in which the number of acute RTI events was the dependent variable. Predictors included: time (a continuous variable ranging from ‘1’ in January 1996 to ‘204’ in December 2012, designed to account for the underlying temporal trend in acute RTI events), timing of the smoke-free law (a dummy variable coded ‘0’ prior to March 2006 and ‘1’ otherwise), an interaction variable ‘time × smoke free law’ (to account for a change in temporal trend in acute RTI events following the law), month (a categorical variable to account for seasonality), sex, age group, region, urbanisation level, SIMD quintile and PCV vaccination. Akaike’s information criterion (AIC) was used to select the optimal model among three options according to the temporal change in RTI events following the smoke-free law: immediate (‘step’) change, gradual (‘slope’) change and step+slope change. The size of the population at risk was used as an offset in the models. We modelled acute RTI events using three variants of negative binomial regression: NB1 (constant dispersion), NB2 (mean dispersion) and NBP (‘NB rho’), which uses a second dispersion parameter that is allowed to vary freely across the data observations.30 The most appropriate negative binomial variant was selected using AIC. We tested for non-linear time effects using a restricted cubic spline with 4 df compared with 1 df using the ‘mvrs’ module in Stata V.14.31 In November 2003, there was an unusually high incidence of RTI events, which was modelled using a dummy variable.

    Post hoc exploratory analyses

    We performed a number of exploratory post hoc analyses to assess the robustness of the findings from our primary analyses, which were felt to be implausible (discussed below).

    First, we performed stratified analyses according to age group and sex to explore whether the association between smoke-free legislation and acute RTI events differed across categories of these variables. Young age and male sex are important risk factors for RTIs in childhood.

    Second, we explored whether the association between smoke-free legislation and RTI events was robust to accounting for temporal trends in air quality. Given the patchiness of air quality availability, we performed an exploratory analysis adding Glasgow city centre air quality data as parameters to the model, using data on acute RTI events occurring in South-West Scotland only. Backward selection of air quality indicators was based on AIC.

    Third, we ran time series regressions on the acute RTI events with seasonal autoregressive integrated moving average ((S)ARIMA) errors to account for regular and seasonal autoregression in the data. The models contained an underlying trend, a dummy variable for the postban period and a postban temporal trend, allowing a number of intervention effects to be tested. Again, we tested for non-linear time effects using Stata’s ‘mvrs’ module.31 Candidate error models were identified from autocorrelation and partial autocorrelation plots. The most appropriate model was selected using the AIC statistic and was subjected to standard diagnostic tests for white noise residuals using the Ljung-Box Portmanteau statistic as well as graphically using autocorrelation plots and correlograms.32

    Fourth, we tested for structural breaks in the time series data using Stata’s ‘estat sbsingle’ command. The procedure searches for a possible trend break over a stipulated range of dates by calculating the value of the test statistic (Wald or likelihood ratio) at each date and then using the maximum value of the test statistic to identify the potential break.33 For the test to work, the series needed to be trimmed prior to the search so as to avoid using dates too close to the end or beginning of the series that would result in breakdown of the procedure. After 20% trimming, we tested for a break in the intercept (step change) as well as a break in the trend (slope change). As a test of robustness, we used the econometrics software EViews to also check for a structural break using its comprehensive suite of break point detection options with 20% trimming of the data. The selected break point and its form (step and/or slope) was then included as an additional regressor in the negative binomial model, and the models were re-estimated.

    All analyses were performed within the National Services Scotland’s Safe Haven environment using Stata MP V.14 except for the (S)ARIMA models. These were analysed separately on aggregated data that were supplied to the authors by the National Services Scotland (NSS) Safe Haven after statistical disclosure control. The (S)ARIMA models and the structural break analysis were developed using Stata SE V.14 with the structural break point further corroborated using EViews 9.5.

    Results

    During 155 million patient-months of observation, 135 134 acute RTI events were recorded: 79 153 upper RTI events and 56 011 lower RTI events. There was substantial variation in the incidence rate of acute RTI events over time (figure 1) and across demographic subgroups (table 1).

    View this table:
    Table 1

    Demographic characteristics of children experiencing an acute respiratory tract infection event

    Figure 1
    Figure 1

    Monthly time series of acute respiratory tract infection (RTI) event rates. Dashed grey line indicates introduction of smoke-free legislation.

    In our primary analysis, introduction of smoke-free legislation was associated with an immediate rise in acute RTI events (incidence rate ratio (IRR) 1.24, 95% CI 1.20 to 1.28) and an additional gradual increase over time (IRR 1.06 per year, 95% CI 1.05 to 1.06; table 2). This finding was consistent when upper and lower RTI events were considered separately (table 1 and online supplementary file 3).

    Supplementary file 3

    View this table:
    Table 2

    Multivariable negative binomial regression model for acute respiratory tract infection events

    Post hoc subgroup analyses to assess whether the unanticipated findings were confined to certain subgroups demonstrated consistency across sex and age subgroups (online supplementary files 4 and 5). In an analysis of data from South-West Scotland only, addition of mean monthly air quality indicators measured in Glasgow improved model performance (online supplementary file 6). This did not have a major bearing on the association between smoke-free legislation and acute RTI events: immediate change IRR 1.25, 95% CI 1.19 to 1.32; gradual change IRR 1.08 per year, 95% CI 1.06 to 1.10.

    Supplementary file 6

    In further post hoc analyses, the strength of association between timing of smoke-free legislation and acute RTI events was very similar when evaluated using a reg(S)ARIMA model of order autoregressive term(1,7) multiplicative seasonal autoregressive term(3,12): IRR 1.15, 95% CI 1.02 to 1.28. However, automatic break point detection suggested that the increase in acute RTI events started well before introduction of smoke-free legislation, that is, in November 2004 (online supplementary file 7). Using this break point rather than timing of smoke-free legislation in the primary negative binomial regression analysis indeed improved model performance as compared with the primary model (online supplementary file 8). When timing of smoke-free legislation was then added to the model that included the November 2004 break point, smoke free legislation was associated with a gradual decrease in acute RTI events (IRR 0.91 per year, 95% CI 0.87 to 0.96), with no evidence of a ‘step’ change at that time (table 2).

    Supplementary file 7

    Supplementary file 8

    Discussion

    In this large national ITS study conducted according to a prespecified analysis plan, introduction of comprehensive smoke-free legislation in Scotland was associated with a significant increase in paediatric RTI events. Findings from our exploratory post hoc analyses, which were undertaken to further investigate this unexpected finding, however, indicated that the increase in RTI events most likely preceded the legislation by 16 months, making a direct causal link between the legislation and increased risk of RTIs implausible.

    Our study has a number of strengths. It was conducted according to a predefined protocol, including a detailed statistical analysis plan, which was developed a priori in an attempt to promote scientific transparency and reproducibility.17 We used over 10 million patient-years of high-quality data routinely collected over a 17-year period. Virtual universal availability of the CHI number minimises risks of incorrect data linkage across the datasets. We accounted for underlying temporal trends in RTI events as well as changes in population size and demographic structure. We applied a look-back period to reduce bias from RTI events occurring prior to the study period. Our modelling approach is widely applied in the evaluation of national public health interventions, including national smoke-free laws.11 20 22 23

    Given these strengths, the implausible findings are of considerable concern. It is important to consider the limitations of the study to see if these may have contributed to the observed findings. Allocation of a nationwide intervention cannot be randomised, and a quasi-experimental approach is considered a potentially valid method to evaluate impact such interventions.15 Reanalyses of cluster randomised controlled trials using an ITS approach have demonstrated that their findings can in fact be highly similar.34–36 While residual bias could possibly have influenced our findings,14 we consider it unlikely that this would explain the implausible results of our primary analysis.

    Results of our prespecified primary analysis were unanticipated and contradicted prior evidence on the topic.10 12 19 Studies in other countries, including in the UK, previously identified consistent associations between comprehensive smoke-free legislation and subsequent reductions in paediatric RTI hospitalisations.10 12 19 In line with these studies, there is consistent evidence for reductions in severe asthma exacerbations among children and in respiratory admissions among adults following implementation of smoke-free laws.11 20 22 Post hoc analyses showed that our findings were consistent across demographic subgroups and unlikely to have resulted from residual confounding by air pollution. As findings from ITS studies have previously been reported to be sensitive to choice of the modelling approach,37 we performed additional aggregate-level time series analyses, again confirming the findings of our primary analysis. Automatic break point detection is a method to explore whether a change in the incidence of the outcome under study indeed co-occurred with the known timing of the intervention, and its use has been promoted as a routine validity test in single-group ITS analyses.38 Using such an approach in a previous study, Kabir and colleagues were indeed able to pinpoint timing of the observed reduction in small-for-gestational-age births to introduction of comprehensive smoke-free legislation in Ireland.39 Using two different approaches and software packages to perform automatic break point detection in our time series, the unanticipated increase in RTI events was shown to have preceded the smoke-free law by 16 months. This earlier break point corresponds quite closely with temporal trends in paediatric RTI hospitalisations in England, which rose consistently from 2003 onwards.40 In a previous study, implementation of English comprehensive smoke-free legislation in 2007 was associated with a reduction in paediatric RTI admissions when accounting for this rising underlying trend.10 This closely corresponds to the findings of our post hoc analysis in Scotland, where RTI events were shown to increase consistently from 2004 onwards, this rise being attenuated after implementation of smoke-free legislation. Whereas an exploration of the underlying causes of the increasing trend in RTI events in Scotland was outside the scope of our study, several potential explanations for the corresponding rise in England have been postulated at the levels of the carer (eg, decreasing threshold to take children to primary care or straight into hospital for evaluation), the health professional (eg, decrease in threshold for referral by primary care doctors or for hospital admission) and the healthcare system (eg, introduction of 4-hour waiting target at emergency departments and unintended financial incentives for admission).40

    Perhaps the main value of this study therefore is that it uncovers a number of important methodological challenges, which have not previously received adequate attention in the applied ITS literature. We used advanced methods and followed a prespecified analytical approach—including a detailed statistical analysis plan—in an attempt to promote transparency.16 17 41 Despite this, our study yielded findings that were implausible and highly likely to be spurious. We therefore conducted a number of exploratory post hoc analyses, which added weight to our assessment that the findings of our primary analysis were indeed spurious. It is important to acknowledge that these post hoc analyses were unlikely to have been conducted should the findings from our primary analysis have confirmed our initial hypothesis. In such a scenario, our study would still have biased the literature on the topic.

    At present, most public policy interventions remain unevaluated. ITS studies are arguably the most robust approaches we have at our disposal to evaluate the public health impact of these major experiments, which are seldom amendable to being implemented in a randomised fashion.14 Despite their limitations, there is thus a need for many more ITS studies to be undertaken to continue to inform policy making at national, regional and global levels. To address the issues highlighted by our study, we propose that future ITS studies evaluating public health interventions should analyse the association under study using a number of different modelling approaches; ideally, these should be prespecified and include approaches based on both individual and aggregate level data (ie, formal time series approaches). Also, we recommend that automatic break point detection approaches are included to validate temporal co-occurrence of the intervention and the change in the outcome under study. This and other machine learning approaches are likely to become increasingly applicable in discerning unusual patterns in time series data over and above variations due to natural changes and those explained by temporal, environmental and individual-level confounding.38 Confidence in the results from individual ITS studies can be further increased by reproducing findings in other settings and interpreting findings from individual studies in the light of the totality of evidence on the topic.

    Going back to the primary hypothesis under investigation, we are reluctant to draw firm conclusions on the impact of Scotland’s smoke-free legislation on paediatric RTIs given the inconsistent findings of the various analyses. Building on the existing evidence base on the topic,20–24 we feel it is highly unlikely that smoke-free legislation was indeed responsible for a rise in paediatric RTI events, as our primary analyses seemed to suggest. However, given these findings, it is also difficult to be confident that the result of our additional exploratory analyses—which were post hoc—provides a more valid representation of the actual impact of the legislation.

    Given the continuing need for formal quasi-experimental evaluations of public health interventions to inform policy making, we propose additional steps to improve the robustness of such studies, including: exploring the association between the intervention and outcome using a number of different (prespecified) modelling approaches, analysing time series using both individual-level and aggregate-level approaches, assessing for potential confounding by unmeasured factors and using automatic break point detection or alternative machine learning approaches to scrutinise findings from prespecified primary analyses, irrespective of whether these support the underlying hypothesis. We hope that the lessons drawn from this experience will increase the validity of future ITS studies in the medical and public health literature.

    What this paper adds

    • Implementation of comprehensive smoke-free legislation is associated with significant early-life health benefits, including reductions in severe paediatric respiratory events.

    • Evidence on the impact of national policies typically is derived from quasi-experimental studies; given their inherent risks of bias and confounding, replication of such studies across various settings is essential.

    • We investigated if the March 2006 national implementation of comprehensive smoke-free legislation in Scotland was associated with a reduction in hospital admissions/deaths due to acute respiratory tract infections among children.

    • Our prespecified interrupted time series approach suggested that implementation of smoke-free legislation in Scotland was associated with an increase in paediatric respiratory tract infection admissions/deaths.

    • We were concerned that this result, which contradicted published evidence, was spurious. The association was indeed reversed when accounting for an unanticipated antecedent break point in the temporal trend, suggesting that the legislation may in fact be protective.

    • We discuss the findings from this national study and propose approaches to enhancing the methodological quality of interrupted time series studies.

    Supplementary file 4

    Supplementary file 5

    Acknowledgments

    We would like to thank Bradley Kirby, Katrina Smith and Sian Nowell at eDRIS for their help in data extraction and linkage. We furthermore thank John Ioannidis, Rachael Wood and Stefan Unger for providing expert advice.

    References

    1. ↵
      World Health Organization. WHO report onthe global tobacco epidemic, 2017. Monitoring tobacco use andprevention policies. Switzerland: World Health Organization, 2017. http://www.who.int/tobacco/global_report/2017/en/(accessed 6 Oct 2017).
    2. ↵
      US Department of Health and Human Services. The Health consequences of smoking—50 years of progress: a report of the surgeon general. Washington DC: US Department of Health and Human Services, 2014. https://www.surgeongeneral.gov/library/reports/50-years-of-progress(accessed 6 Oct 2017).
    3. ↵
      2. Oberg M ,
      3. Jaakkola MS ,
      4. Woodward A , et al
      . Worldwide burden of disease from exposure to second-hand smoke: a retrospective analysis of data from 192 countries. Lancet 2011;377:139–46.doi:10.1016/S0140-6736(10)61388-8
      OpenUrlCrossRefPubMedWeb of Science
    4. ↵
      2. Mbulo L ,
      3. Palipudi KM ,
      4. Andes L , et al
      . Secondhand smoke exposure at home among one billion children in 21 countries: findings from the Global Adult Tobacco Survey (GATS). Tob Control 2016;25:e95–100.doi:10.1136/tobaccocontrol-2015-052693
      OpenUrlAbstract/FREE Full Text
    5. ↵
      2. Jones LL ,
      3. Hassanien A ,
      4. Cook DG , et al
      . Parental smoking and the risk of middle ear disease in children: a systematic review and meta-analysis. Arch Pediatr Adolesc Med 2012;166:18–27.doi:10.1001/archpediatrics.2011.158
      OpenUrlCrossRefPubMedWeb of Science
    6. ↵
      2. Murray RL ,
      3. Britton J ,
      4. Leonardi-Bee J
      . Second hand smoke exposure and the risk of invasive meningococcal disease in children: systematic review and meta-analysis. BMC Public Health 2012;12:1062.doi:10.1186/1471-2458-12-1062
      OpenUrlCrossRefPubMed
    7. ↵
      2. Tinuoye O ,
      3. Pell JP ,
      4. Mackay DF
      . Meta-analysis of the association between secondhand smoke exposure and physician-diagnosed childhood asthma. Nicotine Tob Res 2013;15:1475–83.doi:10.1093/ntr/ntt033
      OpenUrlCrossRefPubMed
    8. ↵
      2. Vardavas CI ,
      3. Hohmann C ,
      4. Patelarou E , et al
      . The independent role of prenatal and postnatal exposure to active and passive smoking on the development of early wheeze in children. Eur Respir J 2016;48:115–24.doi:10.1183/13993003.01016-2015
      OpenUrlAbstract/FREE Full Text
    9. ↵
      2. Baxi R ,
      3. Sharma M ,
      4. Roseby R , et al
      . Family and carer smoking control programmes for reducing children’s exposure to environmental tobacco smoke. Cochrane Database Syst Rev 2014:CD001746.doi:10.1002/14651858.CD001746.pub3
    10. ↵
      2. Been JV ,
      3. Millett C ,
      4. Lee JT , et al
      . Smoke-free legislation and childhood hospitalisations for respiratory tract infections. Eur Respir J 2015;46:697–706.doi:10.1183/09031936.00014615
      OpenUrlAbstract/FREE Full Text
    11. ↵
      2. Been JV ,
      3. Nurmatov UB ,
      4. Cox B , et al
      . Effect of smoke-free legislation on perinatal and child health: a systematic review and meta-analysis. Lancet 2014;383:1549–60.doi:10.1016/S0140-6736(14)60082-9
      OpenUrlCrossRefPubMedWeb of Science
    12. ↵
      2. Lee SL ,
      3. Wong WH ,
      4. Lau YL
      . Smoke-free legislation reduces hospital admissions for childhood lower respiratory tract infection. Tob Control 2016;25:e90–4.doi:10.1136/tobaccocontrol-2015-052541
      OpenUrlAbstract/FREE Full Text
    13. ↵
      2. Ma ZQ ,
      3. Kuller LH ,
      4. Fisher MA , et al
      . Use of interrupted time-series method to evaluate the impact of cigarette excise tax increases in Pennsylvania, 2000-2009. Prev Chronic Dis 2013;10:E169.doi:10.5888/pcd10.120268
      OpenUrl
    14. ↵
      2. Been JV ,
      3. Sheikh A
      . Investigating the link between smoke-free legislation and stillbirths. Expert Rev Respir Med 2016;10:109–12.doi:10.1586/17476348.2016.1125784
      OpenUrl
    15. ↵
      2. Kontopantelis E ,
      3. Doran T ,
      4. Springate DA , et al
      . Regression based quasi-experimental approach when randomisation is not an option: interrupted time series analysis. BMJ 2015;350:h2750.doi:10.1136/bmj.h2750
      OpenUrlFREE Full Text
    16. ↵
      2. Thomas L ,
      3. Peterson ED
      . The value of statistical analysis plans in observational research: defining high-quality research from the start. JAMA 2012;308:773–4.doi:10.1001/jama.2012.9502
      OpenUrlCrossRefPubMedWeb of Science
    17. ↵
      2. Ioannidis JP
      . How to make more published research true. PLoS Med 2014;11:e1001747.doi:10.1371/journal.pmed.1001747
      OpenUrlCrossRefPubMed
    18. ↵
      2. Akhtar PC ,
      3. Currie DB ,
      4. Currie CE , et al
      . Changes in child exposure to environmental tobacco smoke (CHETS) study after implementation of smoke-free legislation in Scotland: national cross sectional survey. BMJ 2007;335:545.doi:10.1136/bmj.39311.550197.AE
      OpenUrlAbstract/FREE Full Text
    19. ↵
      2. Hawkins SS ,
      3. Hristakeva S ,
      4. Gottlieb M , et al
      . Reduction in emergency department visits for children’s asthma, ear infections, and respiratory infections after the introduction of state smoke-free legislation. Prev Med 2016;89:278–85.doi:10.1016/j.ypmed.2016.06.005
      OpenUrl
    20. ↵
      2. Faber T ,
      3. Kumar A ,
      4. Mackenbach JP , et al
      . Effect of tobacco control policies on perinatal and child health: a systematic review and meta-analysis. Lancet Public Health 2017;2:e420–37.doi:10.1016/S2468-2667(17)30144-5
      OpenUrl
    21. ↵
      2. Faber T ,
      3. Been JV ,
      4. Reiss IK , et al
      . Smoke-free legislation and child health. NPJ Prim Care Respir Med 2016;26:16067.doi:10.1038/npjpcrm.2016.67
      OpenUrl
    22. ↵
      2. Tan CE ,
      3. Glantz SA
      . Association between smoke-free legislation and hospitalizations for cardiac, cerebrovascular, and respiratory diseases: a meta-analysis. Circulation 2012;126:2177–83.doi:10.1161/CIRCULATIONAHA.112.121301
      OpenUrlAbstract/FREE Full Text
    23. ↵
      2. Frazer K ,
      3. Callinan JE ,
      4. McHugh J , et al
      . Legislative smoking bans for reducing harms from secondhand smoke exposure, smoking prevalence and tobacco consumption. Cochrane Database Syst Rev 2016;2:CD005992.doi:10.1002/14651858.CD005992.pub3
      OpenUrlPubMed
    24. ↵
      2. Hoffman SJ ,
      3. Tan C
      . Overview of systematic reviews on the health-related effects of government tobacco control policies. BMC Public Health 2015;15:744.doi:10.1186/s12889-015-2041-6
      OpenUrl
    25. ↵
      The prohibition of smoking in certain premises (Scotland) regulations. 2006. Scotland: Scottish Government. http://www.gov.scot/Publications/2005/08/02143247/32478(accessed 24 Jul 2017).
    26. ↵
      Web Archive UK.Latest situation. Smoke-free legislation - national compliance data: summary. Scotland: Scottish Government, 2015. https://www.webarchive.org.uk/wayback/archive/20150309182815/http://www.clearingtheairscotland.com/latest/index.html(accessed 25 Mar 2017).
    27. ↵
      Department for Environment Food and Rural Affairs. Air information resource. UK: Department for Environment Food and Rural Affairs. https://uk-air.defra.gov.uk(accessed 25 Mar 2017).
    28. ↵
      Scottish Government. The Scottish index of multiple deprivation. Scotland: Scottish Government. http://www.gov.scot/Topics/Statistics/SIMD(accessed 27 Jul 2017).
    29. ↵
      ISD Scotland. Pneumococcal conjugate vaccine catch-up campaign, 2015. http://www.isdscotland.org/Health-Topics/Child-Health/Immunisation/Pneumococcal-Conjugate-Vaccine.asp. (accessed 25 Mar 2017).
    30. ↵
      2. Hardin JW ,
      3. Hilbe JM
      . Regression models for count data based on the negative binomial (p) distribution. Stata J 2014;14:280–91.
      OpenUrl
    31. ↵
      2. Royston P ,
      3. Sauerbrei W
      . Multivariable modeling with cubic regression splines: a principled approach. Stata J 2007;7:45–70.
      OpenUrlWeb of Science
    32. ↵
      wntestq-Portmanteau(Q) test for white noise. http://www.stata.com/manuals13/tswntestq.pdf(accessed 25 Mar 2017).
    33. ↵
      estat sbsingle - test for a structural break with an unknown break date. http://www.stata.com/manuals14/tsestatsbsingle.pdf(accessed 25 Mar 2017).
    34. ↵
      2. Fretheim A ,
      3. Soumerai SB ,
      4. Zhang F , et al
      . Interrupted time-series analysis yielded an effect estimate concordant with the cluster-randomized controlled trial result. J Clin Epidemiol 2013;66:883–7.doi:10.1016/j.jclinepi.2013.03.016
      OpenUrlCrossRefPubMed
    35. ↵
      2. Fretheim A ,
      3. Zhang F ,
      4. Ross-Degnan D , et al
      . A reanalysis of cluster randomized trials showed interrupted time-series studies were valuable in health system evaluation. J Clin Epidemiol 2015;68:324–33.doi:10.1016/j.jclinepi.2014.10.003
      OpenUrlCrossRefPubMed
    36. ↵
      2. StClair T ,
      3. Hallberg K ,
      4. Cook TD
      . The validity and precision of the comparative interrupted time-series design: three within-study comparisons. J Educ Behav Stat 2016;41:269–99.doi:10.3102/1076998616636854
      OpenUrl
    37. ↵
      2. Huesch MD ,
      3. Østbye T ,
      4. Ong MK
      . Measuring the effect of policy interventions at the population level: some methodological concerns. Health Econ 2012;21:1234–49.doi:10.1002/hec.1783
      OpenUrlPubMed
    38. ↵
      2. Linden A ,
      3. Yarnold PR
      . Using machine learning to identify structural breaks in single-group interrupted time series designs. J Eval Clin Pract 2016;22:855–9.doi:10.1111/jep.12544
      OpenUrl
    39. ↵
      2. Kabir Z ,
      3. Daly S ,
      4. Clarke V , et al
      . Smoking ban and small-for-gestational age births in Ireland. PLoS One 2013;8:e57441.doi:10.1371/journal.pone.0057441
    40. ↵
      2. Gill PJ ,
      3. Goldacre MJ ,
      4. Mant D , et al
      . Increase in emergency admissions to hospital for children aged under 15 in England, 1999-2010: national database analysis. Arch Dis Child 2013;98:328–34.doi:10.1136/archdischild-2012-302383
      OpenUrlAbstract/FREE Full Text
    41. ↵
      International Epidemiological Association. Good Epidemiological Practice (GEP).IEA guidelines for proper conduct in epidemiologic research. 2010 http://ieaweb.org/good-epidemiological-practice-gep (accessed 25 Mar 2017).

    Footnotes

    • Contributors JVB conceived the study, obtained funding, developed the methods, analysed the data, interpreted the findings and drafted the manuscript. DFM developed the methods, analysed the data, interpreted the findings and contributed to drafting the manuscript. CM, CPvS and JPP developed the methods, interpreted the findings and provided feedback on the manuscript. IS extracted data, interpreted the findings and provided feedback on the manuscript. AS conceived the study, obtained funding, developed the methods, interpreted the findings and supervised drafting of the manuscript.

    • Funding This work was funded by a Thrasher Research Fund Early Career Award (NR-0166) and the International Pediatric Research Foundation Young Investigators Exchange Programme. JVB is furthermore supported by fellowship grants from the Erasmus University Medical Centre and the Netherlands Lung Foundation (4.2.14.063JO). AS is supported by the Farr Institute. The funders had no role in study design, data collection and analysis, interpretation of the findings, decision to publish, or preparation of the manuscript.

    • Competing interests None declared.

    • Ethics approval National Health Service South East Scotland Research Ethics Service; The University of Edinburgh’s Centre for Population Health Sciences Ethics Review Group.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Data sharing statement Those interested in accessing the data are advised to contact eDRIS.