The results showed that the dietary intake of lean and oily fish and cod liver oil was statistically significant higher in the intervention cohort, both for mothers during pregnancy and for children during the first 2 years of life. Parental smoking prevalence was generally low postnatal, particularly among the mothers, with a statistically significant difference between the cohorts. There was, however, a statistically significant annual trend in the control cohort. There was no difference between the cohorts regarding an indoor dampness index ≥3. Pregnancy and the first years of life are a period of frequent contact with health professionals in many countries and a favorable period for implementing relevant life-style interventions. Which health professionals are responsible and what recommended schedules to follow may differ between nations, but primary care health professionals are generally encouraged to inform, educate and promote healthy behavior among pregnant women and parents during infancy. To assess the efficacy of such interventions should be of general interest although the premises may be somewhat different. The PACT study was conducted over an 8 years period with a historical control cohort established over a 2 years period immediately before the intervention started. The intervention was implemented as a guideline by all primary care health professionals to all women in ordinary pre- and postnatal health care, regardless of participation in the study or not. This implied for participants in the intervention cohort that they could return self-reported questionnaires at any scheduled follow-up even if they had failed to do so on a previous occasion. This is indicated with figures and arrows in Figure 1.
The comparisons at different age levels permit presentation of behavioral trends. A behavioral trend in the control cohort or in both cohorts simultaneously implies that a possible difference between the cohorts must be interpreted with caution and other explanations than the intervention program should be considered.
Parents in the intervention group seemed to be more persistent in continuing cod liver oil supplement for their infants. There was an annual trend towards increased oily fish intake among children in both cohorts during first 2 years of life probably reflecting a gradual introduction of fish in the diet for all children. There was, however, a persistent and significant difference in oily fish intake between the cohorts at both 1 and 2 years and a shift towards an increased share of lean fish in the diet in the control cohort during the period. A probable interpretation of this may be that oily fish was substituted for lean fish in the children's diet in the intervention group, which was in accordance with the intervention program.
The low smoking prevalence and annual trend towards less smoking in both cohorts during pregnancy are in accordance with earlier findings showing significantly increasing difference in smoking cessation between pregnant women in Trondheim and the comparable city of Bergen and all of Norway in the actual time period . The PACT study period coincided with new legislation on smoking in public places and ongoing national campaigns against smoking. The increased smoking cessation rate observed among pregnant women in Trondheim compared to Bergen and Norway could possibly be a consequence of the ongoing PACT-project as such, with the increased focus on life-style factors during pregnancy and infancy in general and smoking cessation in particular. Interestingly, the continuous smoking intervention did not seem to have any additional effect on the few remaining smokers in the intervention group.
We observed no difference in the housing dampness index between the cohorts. This may reflect a low adherence to the housing dampness intervention. Indoor climate was an unknown and unaccustomed subject for intervention among both health professionals and recipients. Even more expensive and extensive actions as improved roofing and drainage of buildings could have been recommended, but the program had no resources to follow up on this level. The stable fraction of approximately 4% reporting indoor dampness index ≥3 within and between both cohorts at all ages indicates that the question on this topic was highly reliable.
The effectiveness of the three interventions will be assessed by changes in incidence of allergic disease in an upcoming separate paper. The impacts of the dietary intervention however, can be based on inference from an earlier PACT study on the associations between cod liver oil and oily fish and atopic disease. We found that weekly intake of oily fish at 2 years of age gave an OR for atopic eczema/dermatitis (AD) of 0.57 (95%CI: 0.35-0.94) compared to intake less than weekly. This gives an absolute risk reduction of 7.5% (from 18.2% to 10.7%) at 2 years. With a prevalence of eczema of 15.1%, some 184 of 1213 children would have AD of which 13 cases would be prevented with a 14.5 percent points increase in oily fish consumption, corresponding to a 1.1% reduced prevalence of AD at 2 years of age. It is however, a possibility that a simultaneous intervention on smoking and indoor dampness can give a different outcome.
The strengths of the study are the controlled cohort design with a large number of pregnant women followed prospectively in the intervention cohort, and the assessment of risk-factor behavior that was consistent through the observation period and across cohorts. The non-randomized design was adapted to comply with the assignment to investigate the effectiveness of interventions implemented in the way new guidelines usually are in ordinary primary health care. We decided on a design with a control cohort 1 year in advance of the intervention primarily because a public and community based randomized intervention including the entire primary health care in the municipality would have been impossible to implement without contaminating a co-existing control cohort. Secondly, this design also ensured high conformity between the cohorts regarding population size, race/ethnicity, maternal educational level, income, environment, urbanization and social characteristics . This was supported by the results from the additional non-participants-study that included 391 parents, indicating no major selection bias. Only self reported questionnaires were used, as this is a common and feasible way of assessing information in large epidemiologic studies[32, 40]. For smoking behavior self reported questionnaires are known to have equal or better reliability, compared to interviews using a structured questionnaire [41, 42].
A potential weakness may be that the cohort design with a 1-year difference between the control cohort and intervention cohort might have biased the results toward a better effectiveness of the intervention because of possible annual trends. Although the intervention program was adopted as the official prophylactic program in the community, and nearly 100% of pregnant women visit their GP regularly, only some 34% of the eligible pregnant women participated in the PACT study. The dropouts at 2 years in the intervention cohort was statistically different from the participants regarding maternal age, maternal smoking at start pregnancy and homeowner. This indicates some selection bias at two years in the intervention cohort to be taken into consideration when evaluating the findings.
The inclusion rate decreased substantially during the inclusion period and it was decided to stop at 2860 included. With time health professional reduced their follow-up intensity resulting in a falling follow-up rate, much greater than anticipated. We have however, reason to believe missing was mainly at random in this study. Almost no participants withdraw from the study in the study period, and the reports we have from health professionals indicate that investigators experienced weariness or forgot to present new questionnaires to the participants at ordinary visits as the main reason for missing. Decreasing awareness in long lasting prospective studies causing low participation has been described earlier. The 2100 needed for analysis was based on power estimates for 40% reduction in prevalence of asthma at 6 years, and do not necessarily apply to the reported differences in behaviour in this paper. We think the differences we have found are reliable. If the differences are small, it is certainly a possibility for type II error, meaning that due to the lack of power we erroneously claim a difference non-significant. On the other hand, it is unlikely that the effect of an intervention is overestimated.
It was a requirement specification in this study to have as close to a real-life approach as possible, meaning that the intervention was implemented in the form of a guideline to the participating health professionals. It was a deliberate choice not to follow up on this intervention more than what is usual for guidelines in general when introduced in primary health care. The implementation strategies were considered modest in accordance with the real-life demand, and the effectiveness of the interventional program was exclusively based on parental self-reported risk-factor behavior questionnaires [44, 45].