Method
Study Population and Design
This nationwide population based register study included Swedish children born to women who were pregnant between July 2005 and December 2010, identified from the Medical Birth Register. The cohort was linked to the Multi-Generation Register with the Swedish personal identity number to identify parents and siblings. The personal identity number enables unambiguous linkage between the registers held by the Swedish National Board of Health and Welfare and Statistics Sweden.
Variables
We collected information on exposure and outcome from the Swedish Prescribed Drug Register, which contains complete data on dispensed drugs since July 2005, and the National Patient Register with complete nationwide information on all inpatient diagnoses since 1987 and specialist outpatient diagnoses since 2001.
Exposure. Information on generic names and dates of filled prescriptions of systemic antibiotics (Anatomical Therapeutic Chemical (ATC) classifications system code J01A-J01X) was retrieved from the Swedish Prescribed Drug Register from July 2005 to the end of follow-up. All antibiotics were divided into three groups, modified from Marra and colleagues and as described elsewhere: "any antibiotics" included any type of antibiotics; "airway antibiotics" included antibiotics used to treat respiratory infections (amoxicillin, penicillin, cephalosporin, and macrolides), and "urinary tract or skin and soft tissue antibiotics" (pivmecillinam, trimethoprim, sulphonamide, ciprofloxacin, norfloxacin, nitrofurantoin, cloxacillin, flucloxacillin, and dicloxacillin).
Outcome. Children were classed as having asthma if they had a diagnosis of asthma (ICD-10 (international classification of diseases, 10th revision) J45) registered in the National Patient Register and fulfilled one or both of two criteria for asthma drugs (ATC: R03) from the Swedish Prescribed Drug Register: two or more dispensed inhaled corticosteroids, leukotriene receptor antagonists, or fixed combinations of β2 agonists and corticosteroids, with gaps of two or more weeks between distributions or three or more dispensed asthma drugs as previously mentioned or short acting β2 agonists within a year. Date of onset was set as the date of prescription of any of the drugs or date of diagnosis, whichever came first.
Other Variables. We collected information on child's birth year, sex, birth weight (grams), gestational age (days), mode of delivery (vaginal, instrumental, emergency, or elective caesarean section), maternal smoking at first visit to the antenatal care clinic (yes/no), respiratory diagnosis as newborn (ICD-10: P22-P28), maternal age at delivery, parity (child's birth order), family situation (mother cohabits with child's father, or other), mother's country of birth (Sweden, other Nordic countries, or other) from the Medical Birth Register, which includes data on more than 98% of all births in Sweden. The highest level of education of either parent (0-9, 10-12, >12 years) was retrieved from the Longitudinal integration database for health insurance and labour market studies. Maternal asthma was defined as the mother having either fulfilled the criteria of dispensed asthma drugs or a diagnosis of asthma during the study period.
Start date of pregnancy was estimated as the date of birth minus gestational age in days. Duration of pregnancy was further divided into three trimesters (first: 1-91 days; second: 92-189 days; third: ≥190 days). To ensure full coverage of exposure to antibiotics during fetal life and childhood from the Swedish Prescribed Drug Register, we restricted the cohort to children from pregnancies with an estimated start date of pregnancy as on or after 1 July 2005. Information on date of migration and death was collected from the Total Population Register. We excluded 9631 children born to women who migrated to Sweden during the pregnancy.
Statistical Analysis
The analyses were performed in two steps. Firstly, we analysed the association between antibiotic exposure and asthma in the whole cohort of children using Cox proportional hazard regression, with attained age as analysis time scale and sandwich estimator of standard errors to account for clustering within sibling groups. Follow-up was censored at study outcome, emigration, death, or end of study period (31 December 2011), whichever occurred first. Exposure to antibiotics in childhood was included in the model as a time (age) varying exposure. The proportional hazards assumption was tested with Schoenfeld's residuals. Non-proportional hazards were found for childhood exposure to antibiotics. Consequently, we allowed for time varying effects.
Secondly, we used a sibling control design to adjust for familial factors. For the controls we used all full siblings (excluding 9896 children for whom the father's identity was unknown) who did not have the outcome of interest yet and who were still in the study at the at which the index child developed asthma. While all siblings with discordant outcomes contribute to the hazard ratio estimates of the covariates, only siblings who are also discordant on exposure are informative for the estimation of the hazard ratios for the exposure. Thus, for tabulated frequencies in the sibling control analyses we report the number of families with siblings discordant on both outcome and exposure. There were one to four controls per case and cases without eligible sibling controls were excluded. We used a stratified Cox proportional hazard model conditional on sibling group to analyse the sibling control design data.
Results are reported both as crude and adjusted for maternal smoking during pregnancy, parity, family situation (cohort and sibling analyses), mother's country of birth, parental education, and maternal asthma (only cohort) for the analyses of antibiotic exposure in fetal life. Analyses of exposure to antibiotics in childhood were adjusted for child's sex, birth weight, gestational age, mode of delivery, respiratory diagnosis as newborn, maternal age at delivery, parity, family situation, maternal antibiotic exposure in fetal life (cohort and sibling analyses), and mother's country of birth, parental education (only cohort) and for interaction between age and parity (only sibling analyses).
We performed subanalyses to explore the association of exposure during first, second, and third trimester of the pregnancy and asthma, and between exposure to antibiotics in the first year of life and incident asthma from the age of 2.
A Poisson regression model, with restricted cubic splines (six knots) for time since first antibiotic exposure, was fitted to further assess potential reverse causation.
Sensitivity Analyses. The Swedish Prescribed Drug Register reports only on antibiotics dispensed outside hospitals and most of antibiotics in Sweden (90%) are captured in the register. In a sensitivity analysis, however, we included in the group "any antibiotics" those children who had been admitted to hospital and given a diagnosis previously shown to have a high correlation with antibiotic treatment. The effect of antibiotic exposure on asthma in the full cohort was further investigated by inclusion of interaction terms between birth order (first born compared with not first born) and antibiotic exposure. The association between antibiotics in childhood and asthma was further tested for a dose-response relation between number of prescriptions (0, 1, 2-3, or ≥4 as a continuous variable) and asthma. All analyses of exposure to antibiotics in fetal life and childhood and subsequent asthma were also analysed with asthma drug and diagnosis separately and further restricted to mothers born in Sweden.
All analyses were performed with Stata statistical software, version 13.