We found a significantly lower incidence in parental-reported doctor diagnosed asthma and use of asthma medication at 2 years of age in the intervention cohort compared to the control cohort. The reduced incidence for both asthma and the use of asthma medication was observed among girls. There was no significant difference in the incidence of AD and wheeze between the intervention and control cohorts, and similarly no difference was observed when the groups were stratified by sex.
The behavioural changes are described in details elsewhere [4, 23]. Briefly, the participants in the intervention cohort smoked significantly less, had an increased intake of oily fish and n-3 PUFAs, but the same levels of indoor dampness compared to participants in the control cohort.
As we did not find any difference between the cohorts regarding indoor dampness, the reduced incidence of asthma could not be ascribed to this interventional measure. The study was not designed to identify the magnitude of effect of individual interventions, and it remains therefore uncertain if the observed reduction in asthma is solely due to the reduction in tobacco smoke exposure or dietary alterations or a combination of both. Prenatal and postnatal smoke exposure is generally considered to be a strong risk factor for the development of respiratory symptoms and asthma, particularly in early life [24, 25]. The first national comprehensive mass media campaign on tobacco and health for many years in Norway was accomplished during the study period in 2003, and a total ban on smoking in restaurants and cafés first took effect on June 1st 2004. According to WHO´ s European health for all database, the prevalence of regular daily smokers above 15 years of age in Norway has gradually declined during the study period, from 31.2% in year 2000 to 21% in 2008 . In the sub-analysis of this study, the impact of the intervention among children with smoking and non-smoking parents was comparable. This may indicate that the reduced incidence of asthma in the intervention cohort was not a consequence of reduced parental smoking per se. Rather, reduced smoking in Norway in general and smoking bans in public places could have contributed to less ETS exposure of the children in the intervention cohort independent of the smoking status of their parents. In randomised controlled trials dietary supplementation with n3-PUFAs does not seem effective in reducing the incidence of asthma as showed in a meta-analysis . However, fish intake during pregnancy and during the first year of life has been found to be associated with a reduced risk of childhood asthma [28–30]. Summarised, the body of evidence of a preventive effect of reduced ETS exposure on childhood asthma is stronger than the evidence of a preventive effect of fish. As such, the reduced incidence of asthma and use of asthma medication in the intervention cohort may be attributed first and foremost to the reduced ETS exposure of the children and secondly to the increased intake of fish. The trend analysis showed no annual trends in either cohorts regarding asthma. This indicates that the reduced risk of asthma in the intervention cohort was not a consequence of an annual time trend.
We found no differences between the cohorts regarding AD. We have previously reported a negative association between the intake of fish in childhood and AD at 2 years of age . The increased fish consumption in the intervention cohort compared with the control cohort was likely too small to have an impact on AD on population level .
Our results demonstrated a strong differential impact on asthma and use of asthma medicines when stratified by sex. For wheeze, a similar differential impact was found, but was not statistically significant. An observational study has showed prenatal and postnatal exposure to tobacco smoke to be associated with asthma in 2 years old girls only , and cigarette smoking is associated with more respiratory symptoms among adolescent and adult women than among men [32, 33]. A previous randomised intervention to reduce house dust mite, pet and food allergens, and passive smoking in a high-risk population of infants found an effect of the intervention on asthma-like symptoms in girls, but not on boys . In line with this, our results indicate that the asthma phenotype among girls at 2 years of age is more susceptible to ETS and intake of oily fish than that of boys. The differential impact on boys and girls may indicate that the pathophysiology of asthma depends on the sex of the child.
The strengths of this study are the interventional, non-selected, population-based design with a large number of participants. The PACT study recruited participants from a primary health care population, regardless of parental history of atopy. The control cohort was established over a 2 year period immediately before the intervention was started, and from the same socio-economic areas. Participants in both cohorts followed the same scheduled program for pre- and postnatal follow-up in primary health care. Choosing a controlled design including whole birth cohorts made it possible to test the impact of the intervention program with a real life approach. A randomised control trial, with a co-existing cohort, would not be feasible in this type of broad population intervention, as both participants and health care workers move or communicate with each other introducing contamination of the control group. Secondly, comparing total birth cohorts also ensured high conformity between the cohorts regarding population size, race/ethnicity, maternal educational level, income, environment, urbanisation and social characteristics. The assessment of allergy related diseases were consistent through the observation period in both cohorts, and questions reporting factual information such as doctor diagnosed asthma were found reliable in a separate study .
Some 35% of the eligible pregnant women in Trondheim completed the questionnaires at 2 years of age. We had almost no active withdrawals in either cohort. The participation rate dropped during the study period , and could be a consequence of the long duration of the study resulting in a decreasing awareness, causing low recruitment activity among many GPs and midwifes, and not a consequence of self-selection among women. Decreased participation rate has been an increasing problem in epidemiological studies the recent years, but the participation rate experienced in this study might not introduce biased estimates if the participants are representative . Results from the additional non-participants study (n = 391), showed no significant differences between non-participants and participants in the PACT study . This is reassuring and indicates that the PACT population is comparable to the general maternal population in Trondheim.
The drop-outs in in the intervention cohort were different from the participants at 2 years of age regarding maternal age, maternal smoking at start pregnancy and homeowner status, but the clinical differences were small. Due to the design of the study, only a portion of the control cohort was followed prospectively from pregnancy, however similar differences were seen between the drop-outs and participants in the intervention and control cohorts.
The participation rate varied between each maternal health care centre and they had a somewhat different proportion of participants from each cohort. Multilevel logistic regression models compensated for possible clustering on maternal health care centre. The proportion of participants from each cohort in each public health centre was added as a separate variable to the model to compensate for the different participant rates from each cohort in each maternal health centre. We were able to adjust for a large number of confounders, including RS virus infection and other potential clinical confounders. Summarised in the final adjusted model, the aOR of allergy related diseases was comparable with the crude analysis (Table 3). This indicated that the effect estimates were not biased. We were unable to adjust for education due to lack of parental educational data on all participants. Parental homeowner status was tested in the adjusted model as a proxy for socioeconomic status, but did not alter the effect estimates, and was excluded in the final adjusted model.