Statistical analyses
The analyses were performed using the software IBM SPSS Statistics for Windows, v26, Armonk, New York, USA. Internal missing on specific questions in the questionnaire was low in all surveys, <5 %. Internal missing about asthma and respiratory symptoms was treated as a negative response while internal missing about exposures was treated as missing. The chi-square test was used to test for differences in proportions and the Students t-test for differences in means. Poisson regression analysis with physician-diagnosed asthma as outcome was utilized to estimate risk ratios (RR) with 95% CI. The models were analyzed both unadjusted and adjusted with age, sex, allergic rhino-conjunctivitis, family history of asthma, smoking habits and socioeconomic status included as covariates.
The cumulative incidence of physician-diagnosed asthma was estimated from two models of defining the population at risk. In model A, individuals with any of the following variables at baseline were excluded from the population at risk: physician-diagnosed asthma, ever having had asthma, asthma medication use, recurrent wheeze or attacks of SOB last 12 months. In model B, individuals were excluded from the population at risk if reporting having physician-diagnosed asthma at baseline. The cumulative incidence of asthma medication use and recurrent wheeze was also estimated by model A and B, and additionally by a third model C, where the population at risk excluded those reporting asthma medication use at baseline when asthma medication use was analyzed as the outcome variable, and where the population at risk excluded those reporting recurrent wheeze at baseline when recurrent wheeze was the outcome. As we did not have complete data on person-years we used the cumulative incidence during the ten years observation period in each cohort to estimate annual incidence rate by assuming the incidence to be constant. The estimated annual incidence rate was calculated as\(Incidence\ rate=\frac{a}{(10\times(b-\left(\frac{a}{2}\right))}\)where a is the number of incident cases and b the number of subjects in the population at risk at the start of the study, as also estimated in other studies [15,18,19].