1. Main results
PREVED is the first intervention research dedicated to perinatal environmental health education in France. The development, implementation, and evaluation of the PREVED project showed interesting results in terms of reach, adoption, and effectiveness.
1.1. Reach and Adoption
We noted a lack of social diversity of participants. Despite the specific recruitment strategies through PMI department involved in DisProSe Consortium, the underprivileged population reached was limited.
The choice of collective workshops has many benefits. Peer education is an important health education instrument that facilitates development of new skills, acquisition of experiential knowledge and promotion of health behavior change through experience sharing and social support [26, 27].
1.2. Effectiveness = Efficacy*Implementation
In the efficacy study, we found an effect on fast-food consumption, on risk perception score and on MePB presence in urines. In the implementation evaluation, we found that the key features seemed to be carried out.
We did not find any significant effect on canned food consumption. However, the Q1 questions did not specify which kind of food containers. In fact, participants in workshops were encouraged to avoid metal food containers and to replace them by glass jars as much as possible, a point that was not clarified in the leaflet.
A more pronounced fast-food consumption effect was observed in the intervention groups with a major effect in contextualized location (Group 3).
The contextualized intervention was implemented in a pedagogical apartment, which was close to real-life. Home-based educational intervention on health could improve emotional care for chronic patients and their caregivers, especially for groups in precarious situations [28, 29]. Likewise, home-like environment could create added value in health promotion interventions [30, 31].
However, we did not find major differences between Group 2 and Group 3. In fact, we noted that the apartment used for contextualized intervention was insufficiently exploited during workshops. The difference between the two intervention groups was consequently small, and our objective of promoting experimental knowledge was not fully attained.
The increased of risk perception scores and a significant difference between control group and intervention groups were highlighted. This increase was maintained even more than a year from the intervention. In practice, risk perception represents a major lever and determinant of health behavior and motivation to change [30] ; elevated risk perception could promote healthier and safer behaviors [33, 34]. This may explain the reduction of the consumption of canned tuna observed in long-term. Furthermore, both the interaction and sharing of experiences between participants are likely to contribute to favorable evolution. Involvement in health education processes depends largely on social factors such as peer support, especially within existing groups, and choice of a relatively familiar environment [33].
In PP and ITT analysis, a difference in BuPB presence in colostrum between control group and intervention groups was only highlighted (13% versus 1%, p=0,03), as it was not significant for other EDs. In urine, we found no significant decrease between 2nd and 3rd trimesters. There were higher percentage of women having a decrease in MePB presence in the intervention groups, between control group and intervention groups (19% control group versus 32% intervention groups, p=0.09). For BPA and ClxBPA, intervention failed to demonstrate a reduced detection level. We did not find that contextualization of workshops influenced biomarker levels in urine or colostrum. Urinary BPA detection percentage (43% 2nd trimester) was lower than studies looking at pregnant women in France in 2011 (74%) [34] and in 2016 (100%) [35]. That said, among Canadian pregnant women, detected BPA in first trimester was 43% [36]. Concerning colostrum, in our study, BPA was detected at 77%, close to an American study on women who had just given birth [37]. However, comparison to the literature is complicated: detection percentage varies greatly (from 17–100%) due mainly to various analytical methods. ClxBPA, which is less studied than BPA, showed detection percentages close to those found in pregnant women of the EDDS cohort [3]. We detected more pronouncedly for 2nd trimester in PREVED study than in EDDS study: MCBPA (57% PREVED versus 34% EDDS), DCBPA (42% versus 34%), TCBPA (21% versus 35%) and TTBPA (31% versus 18%). In colostrum, because of the lipophilic nature and bioaccumulation of ClxBPA, we expected to find higher percentages of detection than those found. They were higher than those of EDDS (MCBPA 46% PREVED versus 18% EDDS, DCBPA 47% versus 23%, TCBPA 66% versus 17% and TTBPA 51% versus 2%) [3].
Regarding PBs, proportions were lower than in other studies: a Japanese study [38], found stronger detection of MePB in pregnant women than our study (71% PREVED versus 94% Japanese study), EtPB (59% versus 81%), PrPB (22% versus 89%) and BuPB (3% versus 54%). Similarly, in EDDS study on detection of biomarkers in urine at 2nd trimester of pregnancy in France, detection percentage was higher than in the PREVED study: MePB (71% PREVED versus 97% EDDS), EtPB (59% versus 77%), PrPB (22% versus 84%) and BuPB (3% versus 64%). First two PBs (MePB and EtPB) are most widely used in cosmetics, since many cosmetic products combine these ingredients so as to increase their antimicrobial potential [39]. In colostrum, we found the same detection gradient for PBs. Are comparably detected for 2nd trimester in PREVED and in EDDS: MePB (88% PREVED versus 90% EDDS), EtPB (60% versus 50%), PrPB (31% versus 30%) and BuPB (5% versus 27%). As mentioned above, a drop in PBs occurred at different times for MePB in urines, which was found in ITT and PP analysis.
Other RCTs aimed at reducing ED exposure through an intervention. A first study carried out in 2011 found a decreased in BPA detection percentages with a canned-food-excluded-diet in young adult population [11]. Another study, carried out in 2013, did not find any decreased in children and their parents' urinary phthalates and BPA concentrations [13]. Two other studies carried out were aimed at reducing BPA and phthalate exposure in family members or young adult population, by avoiding plastic packaging and canned food. Both showed decreased mean concentrations for BPA in urine, by 66% for family members [12] and by 79% for young adults respectively [40]. To our knowledge, only one RCT has been carried out in pregnant women in view of reducing phthalate exposure by eating only fresh and organic diet, without founding any decrease [41].
2. Strengths and limits
2.1. Evaluation model
We adopted a parallel RCT with a randomization at individual level, with both quantitative and qualitative analyses. Our choices in this PHIR are questionable. Some authors have recommended in complex interventions to choose stepped-wedge cluster randomized trial where clusters are randomly allocated to different sequences, with each sequence defining the timing of cluster switch from the control to the experimental condition [42]. Even if individual randomization is rarely well-adapted to PHIR because interventions of interest are generally delivered at a group level (e.g. schools, health centers, geographical areas) [42], our targeted population (pregnant women) was not an identifiable group, and group contamination was consequently not possible.
Experimental methods for the evaluation of complex interventions, such as RCT, are the reference methods because they seek to determine the effects of the intervention with high internal validity, avoiding confusion bias. The RCT is clearly an option for PHIR [42] even if it has some drawbacks: i) interventions are often carried out in the same controlled environment, excluding the interaction between intervention and environment from analysis; ii) the recruited individuals are highly motivated, decreasing external validity [43]; iii) the focus of the trial is on effectiveness (efficacy), which de facto excludes adaptation factors, targets achieved and institutionalization; iv) RCTs are also limited for the study of behavioral factors [44] because the outcome is highly dependent on health determinants (individual characteristics, cultural and social environment and health systems); v) the standardization of the intervention, as expected in trials, is not favorable from a continuous learning perspective, which requires variation of the intervention [43].
Instead of RCT, some authors propose alternative assessment models such as the “out of control” test method, where only key functions are standardized while the form is adaptative [43] or RCT adaptations such as cluster randomized trials, pragmatic trials, cluster and pragmatic or non-RCT designs (quasi-experimental, cohort, realistic evaluation, case-cohort studies) [17]. These new evaluation models raise questions of feasibility, acceptability, fairness and sustainability of interventions and their adaptation to the results of qualitative studies, which provide better understanding of how and why we obtain these assessment results [44]. The process studies are largely designed to assess contextual factors and causative mechanisms [18].
We chose the RE-AIM model because it provided a solid framework to assess the implementation of the PREVED project. It enabled simultaneous examination of both participant-level outcome data and detailed organizational and site-level data [45].
2.2. Outcome choice of the randomized control trial
The main outcome was canned food consumption and not biomarker presence or concentration. A pre/post comparison of biomarker presence could have been a relevant primary endpoint, as suggested by a California study in which a diet devoid of deleterious packages for three days led to a significant decrease in urinary concentrations of BPA metabolites from an average of 3.7 ng/mL to 1.2 ng/mL [12]. However, other studies suggest that urinary BPA concentrations may vary during pregnancy [46–48], while urinary PB concentrations are minimally impacted [49]. These studies have not explored changed consumption patterns that may have occurred between two samples. As a result, it is difficult to conclude that changes in urinary concentrations are due solely to the physiology of pregnancy. Colostrum, which begins to form in the middle of pregnancy [50], may be a good biomarker of cumulative pregnancy exposure to lipophilic molecules such as BPA, Clx-BPA and PBs. However, it is a rare and difficult matrix to collect under the conditions required to avoid BPA contamination (manual sampling, without gloves or breast pumps): only, a single sample can be taken. That is why before/after comparison of biomarker presence in urine, was not the primary outcome in the PREVED study. We preferred to choose various outcomes: consumption, psychosocial and biomarkers with an interdisciplinary contribution of analytical chemistry, social psychology, epidemiology, sociology, and health promotion disciplines.
2.3. Construction
1.3.1. Consortium
The consortium creation involving the actors in the evaluation system from the construction of the intervention as part of a continuous improvement approach (researchers, field actors, decision-makers) has enabled scientific projects to mature, as recommended [51]. The solution science of the PHIR is strongly embodied in the practices and especially in the ability to support the meeting and sharing between different expertise to promote their hybridization and thereby the production of new expertise and a non-unidirectional transfer of knowledge [52]. However, in the PREVED project, this did not suffice insofar as there was a lack of exchange between partners; it could be called an epistemic misunderstanding [53]. As it is recommended that evaluation should be considered and built at the same time as the intervention itself, the consortium focused too early on the evaluation phase, which appeared too rigid to build an action of environmental health promotion.
1.3.2. Behavior theory model
The conception of PREVED workshops was based on the HBM including 12 behavior modification techniques from the taxonomy of Michie et al. [54]. According to the Medical Research Council, the main components of an intervention include the development of a theoretical model to better understand the change process [18]. Also, it has been interesting in PREVED project to integrate sociocultural and economic factors [55], particularly social determinants of health and local context [56] using a solid theoretical base and a consistent and rigorous methodology [57].
1.3.3. Diagnosis with Pregnant women
According to PREVED sociologic study, the project omitted a thorough diagnosis of our target population of pregnant women's needs. The behavior of pregnant women has been the subject of a normative and stereotypical approach. Despite the qualitative study on 12 women conducted before the cross-sectional study, a deeper exploration of lifestyle should have been performed. In citizen science, co-creation approach is recommended [58].
1.3.4. Public Heath deployment
As part of evidence-based-health-promotion [59], this study could contribute to health policy deployment of the environmental health intervention in the care pathway during pregnancy. However, while this type of intervention has been disseminated in maternities, there is evidence of its impact. With PREVED project, we have proposed effective intervention on risk perception and behavior change. To integrate these workshops into the pregnant women’s care pathway, it is necessary to involve medical doctors, nurses and/or midwives. However, preventive medical recommendations of ED exposure avoidance during pregnancy are not yet being followed [60]. Medical doctors are not educating pregnant women on environmental risk prevention [7] because they are essentially focused on infectious biological risks. They also usually have non-specialized experimental knowledge of emergent risks. However, they have recently become increasingly cognizant of those risks [61].
We have previously recommended to educate pregnant women using simple words, taking the time to understand their representations, and questioning them about their knowledge, risk perception and behavior (pre/post-test). This exchange could use an educational tool, which could be part of the PREVED© questionnaire to help health professionals [15]. This new tool, which has been informatized in a smartphone via QR code available in general practitioners' (GP) waiting rooms, has been tested and perceived by GPs as useful to initiate a discussion about environmental health with patients [62]
That said, we have described and reported on our public health interventions with the TIDieR tool which, like other tools (Astaire, Trend ..), does not make the distinction between key function (potentially transferable dimensions) and form (dimensions associated with translation with a specific context) proposed by Villeval, who introduced a "key function/implementation/context” model, which makes a distinction between transferable element assessment and context-specific element assessment [63]. This model is the best alternative for transferability insofar as it takes into account the context [64].