Study design

We conducted interrupted time series (ITS) analyses to determine the associations between the implementation of the July 2008 Dutch national smoke-free restaurants and bars policy34, and changes in the incidence rates of wheezing/asthma, RTIs, and OME among children.

Setting

We used data from the Integrated Primary Care Information (IPCI) database, a longitudinal database of electronic medical records (EMRs) from primary care practices in the Netherlands. The IPCI database was founded in 1989 and is acclaimed for its comprehensive medical information35. The data in IPCI are an open cohort: GP practices and patients may enter and leave the database at any time. As the EMRs were anonymised, the geographical locations of the patients and GP practices were unknown to the researchers using these data. Details of the IPCI database have been described previously35. To perform the ITS analysis around the key year 2008 (when policy changed), we decided a priori to include data from January 2000 through December 2016.

Participants

For our study, children were included if they were aged ≤12 years and were registered with a GP practice in the IPCI database for at least six consecutive months during the study period. Children who entered the GP practice as a newborn (i.e., within 6 months after birth) and who contributed at least 6 months of data were eligible. Participants were followed-up until they turned 13 years of age, changed to a GP practice not registered in IPCI, left the country, or died. The age cut-off of 12 years of age was selected to minimise the potential confounding effect of active smoking among participants.

Outcomes

Our primary outcomes of interest were the monthly incidence of new diagnoses of wheezing/asthma, RTIs, and OME. Our secondary outcomes of interest were the monthly incidence of new diagnoses of upper RTIs (URTI) and lower RTIs (LRTI).

We specified a new diagnosis of wheezing/asthma when a relevant diagnostic International Classification of Primary Care (ICPC) code was recorded in a child’s medical records and/or when a prescription for asthma-related medication was recorded in the medical records of a child who had no previous recording of wheezing/asthma diagnostic codes or asthma-related prescriptions (the relevant ICPC codes are listed in the Online Supplement). Children did not have a new diagnosis if wheezing/asthma was recorded before or on the first day of registration with an IPCI practice, as they were considered to be prevalent cases of wheezing/asthma. Asthma-related medications included selective beta-2 adrenoreceptor agonists, anticholinergics, inhalation corticosteroids, and leukotriene receptor antagonists (for details, see the Online Supplement).

Incident diagnoses of RTIs were specified when an ICPC code for either an URTI or LRTI was recorded in a child’s medical records. As most RTIs among children resolve within 15 days, we considered a new RTI diagnosis only when registered at least 21 days after any prior RTI consultation, to minimise repeated registration of GP visits for the same RTI episode36.

We defined children to have a new OME diagnosis if the relevant ICPC code was recorded in a child’s medical records and there was no prior registration of an OME code in the preceding six months. We applied this six-month window to exclude repeated GP visits for the same episode of OME as around 72% of OME cases (95% CI 68–76) in children are known to resolve spontaneously within 6 months37.

We considered separate diagnoses of URTIs and LRTIs as secondary outcomes. Similar to RTIs in general, we defined multiple visits within 14 days as belonging to the same URTI/LRTI diagnosis. However, if an URTI was followed by a LRTI within 14 days, both the URTI and LRTI were recorded as separate diagnoses.

The GPs used codes from the first edition of the ICPC to register disease diagnoses in EMRs38. Medication codes were registered using the World Health Organization’s (WHO) Anatomical Therapeutic Chemical (ATC) codes39.

Person-time at risk

A child was considered to be at risk for an outcome in a particular month if:

The child had no prior registration of an asthma code, including in registration of prior disease history in patients newly entering GP practice (Online Supplement) [for wheezing/asthma]

The child had no prior issuing of wheezing/asthma-related medication prescription, including in registration of prior medication history in patients newly entering GP practice (Online Supplement) [for wheezing/asthma]

No OME code was registered within the six preceding months (Online Supplement) [for OME].

Main predictor of interest

The intervention was the introduction of a law on 1 July 2008 prohibiting smoking in hospitality venues, i.e., hotels, bars and restaurants, sports, arts and culture venues, amusement arcades, tobacco shops, and international passenger transport34. This law allowed for designated indoor smoking areas34. The law was accompanied by a mass-media campaign and an 8% excise tax increase on tobacco products40.

Confounding factors

In our analyses, we adjusted for the following potential confounders: the underlying time-trend in the outcome, seasonality (i.e. month of diagnosis; categorical), age group (0–4; 5-12 years of age), sex (female; male), EMR software system (HetHIS; Medicom; MicroHIS, MicroHIS Old; Mira; Promedico ASP; Promedico VDF Old; WebHIS Zorgdossier), urbanisation level (urban: ≥1500 inhabitants per km2; rural: <1500 inhabitants per km2), and social deprivation (yes: living in an area in the bottom 5% of that year’s national list of postal codes ranked according to social deprivation; no: in the top 95% of this list).

Due to a high proportion of missing values for urbanisation and social deprivation (i.e. 35% and 13%, respectively), we re-coded missing values into a third category, i.e. urban, rural, missing for urbanisation level, and yes, no, missing for social deprivation. Urbanisation level and social deprivation were completely missing for the years 2000–2003, as registration of postal code only started in 2004.

Statistical methods

Our primary analysis was an ITS negative binomial regression analysis in which we investigated the association between the implementation of the tobacco control policies in 2008 and the change in incidence of each of our outcomes of interest. A negative binomial model (rather than a Poisson model) was adopted to account for over-dispersion (variance > mean) of case incidences over time.

Our models allowed for both an immediate change in level of the outcomes (step change) using a dichotomous time-variant dummy variable, as well as a gradual change in temporal trend of our outcomes (slope change) using an interaction term between the dummy variable and year (continuous). We accounted for seasonality (categorical variable for month) and the underlying temporal incidence trend (year as a continuous variable, centred at 1 July 2008). We modelled the underlying temporal trend via linear, quadratic, and cubic B-splines in separate models to account for possible non-linearity. We selected the optimal model using Aikaike’s and Bayesian Information Criteria (AIC and BIC, respectively). Using comparison of predicted values from our models and counterfactual models with step and slope changes set to zero, we estimated the absolute number of events averted across the post-legislation study period for each outcome. We performed all analyses using Stata SE 15.1 (Statacorp, TX).

Sensitivity analyses

In post hoc sensitivity analyses, we assessed the potential impact of the substantial proportion of children with missing postal code data on our findings via: 1. including only cases with complete data on all covariates, 2. not adjusting for urbanisation level and social deprivation, and 3. imputing values for urbanisation level and social deprivation. For the latter purpose we conducted and analysed 20 imputations of these values based on all the available variables in the dataset, using the Stata commands mi impute monotone (logit) and mi estimate.

Ethical considerations and reporting

The IPCI Governance Board approved this study (no. 03/2015). The IPCI data are not subject to the Medical Research Involving Human Subjects Act (WMO) and therefore do not require approval from a medical research ethics committee. We conducted this study using a pre-specified study protocol. We performed all methods in accordance with the relevant guidelines and regulations. Meta-data and data are property of IPCI. Researchers who are interested can contact the IPCI project team at www.ipci.nl. Our study protocol and statistical codes are available on request from the corresponding author. We used the STROBE and RECORD guidelines to report our findings.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.