This quantitative, descriptive, monocentric, matched before-and-after cohort study evaluated the knowledge and behaviors of first-year family medicine residents at the University of Montpellier–Nîmes regarding EDCs in common household settings. The study followed the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for cohort studies (Multimedia Appendix 1).

We applied the Kirkpatrick model for training evaluation, which assesses learning outcomes across four levels: (1) reaction—participants’ satisfaction with the training; (2) learning—acquisition of knowledge, skills, and competencies; (3) behavior—application of learned content in a professional setting; and (4) results—achievement of targeted objectives, measured using predefined indicators.

Our evaluation specifically assessed participants’ ability to identify vulnerable populations, list potential health effects of EDCs, and recognize common sources of exposure. We also examined participants’ interest in environmental health, perceptions of the e-learning format, and potential thesis topics.

The study population included all first-year family medicine residents at the University of Montpellier–Nîmes. The training was mandatory for this group. Pre- and posttraining questionnaires designed using Microsoft Forms and integrated directly into the Rise 360 platform (Articulate Global, LLC) were completed at the beginning and end of the module, respectively.

Only residents who completed both questionnaires were included in the matched analysis. After matching, data were anonymized. Although the module was also made available to midwifery students, their responses were excluded from this study.

A working group was established in January 2023 comprising 1 resident, 2 university lecturers in general practice, 1 university midwife, and the audiovisual team of the University of Montpellier–Nîmes Faculty of Medicine. The module’s design was inspired by the “Endocrine disruptors – My home” website [21] (which targets the public) and was structured into 7 thematic subsections.

Pedagogical objectives were to (1) identify vulnerable populations, (2) identify primary known or potential EDCs in the home, and (3) understand the health effects of EDCs and propose preventive alternatives.

Each subsection followed a consistent structure (Table 1). The module was developed using the Rise 360 platform [22], which enables the creation of interactive and adaptive e-learning content. Educational materials included (1) virtual rooms displaying household objects containing EDCs created using the Microsoft Copilot artificial intelligence tool [23] (Figure 1); (2) educational videos (≤3 minutes each) on the health effects of EDCs, produced in collaboration with the audiovisual service of the Faculty of Medicine and filmed on the set of the Department of Pedagogical Engineering and Audiovisual Production at the Faculty of Medicine (Figure 2); (3) interactive association quizzes linking EDCs to their health effects, used as tests during training (Figure 3); and (4) flash cards presenting evidence-based alternatives to reduce exposure.

A global summary and complete bibliography were provided at the end of the module. The training is publicly accessible [24].

A literature review was conducted to identify validated pretest-posttest questionnaires for assessing knowledge and behaviors related to EDCs. No tool was fully relevant to our objectives or target population [25]. A questionnaire was then developed based on the learning objectives of the module and items from previous PCEH surveys [7,8]. It was reviewed by a panel of 3 experts, all GPs, including 2 with expertise in medical education and environmental health and 1 with specific expertise in environmental health. The experts independently assessed the items for content relevance, clarity, and alignment with the educational objectives. On the basis of their feedback, several items were revised to improve wording and reduce ambiguity. The internal consistency of the behavioral scale was assessed using the Cronbach α.

For the pretraining questionnaire, the structure was as follows: (1) sociodemographic characteristics (questions 2-7), (2) ability to identify household EDCs (yes or no; questions 8-11), (3) self-reported preventive behaviors (5-point Likert scale; questions 12, 15, 17, and 18), and (4) ancillary knowledge (multiple choice; questions 13, 14, and 16).

For the posttraining questionnaire, the knowledge and behavior items were the same (excluding sociodemographic characteristics), and the questionnaire included additional items on satisfaction with the module (questions 2-5), interest in environmental health, perception of the e-learning format, and potential thesis topics (questions 17-19).

Pseudonymization was implemented by having participants enter the first letter of their last and first names and the last 4 digits of their phone number at the start of each questionnaire, allowing for matching while preserving anonymity.

The full questionnaires are provided in Multimedia Appendices 2 and 3.

Data were analyzed by the study author using Microsoft Excel and the BiostatTGV statistical website [26]. Qualitative data (“yes” or “no” responses) were described as frequencies and percentages and analyzed using the McNemar chi-square test for matched samples. Regarding quantitative data (Likert scales), before-and-after differences (Δ) were calculated for each item, expressed as means, SDs, and ranges. They were analyzed using the paired Student t test (2-tailed). Statistical significance was set at a P value of less than .05. The sample size calculation was based on an assumed mean score of 2.5 out of 5 (SD 1) in the pretraining questionnaire and 3 out of 5 (SD 1) in the posttraining questionnaire, corresponding to a moderate effect size (approximate Cohen d=0.5). Using an α of .05 and 90% power, the required total sample size was calculated as 138 participants (69 per group) for pre- and posttest comparisons. This assumption is consistent with effect sizes commonly reported in studies evaluating educational interventions in medical education, particularly e-learning formats [9], and was considered a conservative estimate in the absence of prior data specific to endocrine disruptor training.

All participants provided informed consent in accordance with ethical guidelines. An impact analysis compliant with the MR-004 standard was submitted to the University of Montpellier–Nîmes’s data protection officer. Ethics committee approval was not required under the General Data Protection Regulation framework [27]. Data confidentiality was maintained per National Commission on Informatics and Liberty regulations. Anonymized data were stored on a secure, password-protected server. While participation in the training was mandatory as part of the residency program, study participation was voluntary, and residents could withdraw at any time by contacting the investigator via email. No financial compensation was provided.

This study was conducted over a 17-day period (September 10-26, 2024). Of 148 eligible family medicine residents, 98 (66.2%) completed the pretraining questionnaire, and 78 (52.7%; matched participation rate) completed both the pre- and posttraining assessments (Figure 4). Sociodemographic characteristics of the study population are summarized in Table 2. The mean age was 25.82 (SD 1.5) years, and 73.1% (57/78) of the participants were women. Only 11.5% (9/78) reported prior training in environmental health.

Overall satisfaction with the 4 modules (kitchen, bathroom, bedroom and living room, and outdoors) was rated at a mean of 4.0/5 (SD 0.9). The suitability of the e-learning format was rated at a mean of 4.1/5 (SD 1.0), and the extent to which the training achieved its objectives was rated at a mean of 4.3/5 (SD 0.8). Regarding module length, 77% (60/78) of the participants were satisfied, including 47% (37/78) who were completely satisfied and 30% (23/78) who were moderately satisfied; 15% (12/78) were indifferent; and 7% (6/78) were little or not satisfied. Detailed satisfaction data are provided in Multimedia Appendix 6.

Identification rates for all 16 EDCs improved significantly (P<.001; Figure 5). Before the training, correct identification ranged from 12% (9/78) to 36% (28/78) for 14 of the 16 substances. After the training, rates increased to 69% (54/78) to 97% (76/78). The largest absolute increases were observed for pesticides (64/78, 82.1% before vs 78/78, 100% after) and parabens (30/78, 38.5% before vs 73/78, 93.6% after). The mean overall improvement in correct identification across all EDCs was 55.56 (SD 13.54) percentage points. Detailed data are available in Multimedia Appendix 5.

The mean pretraining behavior score on the 5-point Likert scale was 1.63 (SD 0.89). After the training, the mean score increased to 3.77 (SD 0.88), representing a significant mean improvement of 2.13 points (95% CI 1.71-2.56; P<.001; Figure 6). This pattern was consistent across all household settings (kitchen, bathroom, bedroom and living room, and outdoors) and for both self-assessment items (“in my home” and “in patients’ homes”).

The internal consistency of the behavioral scale was assessed using the Cronbach α, which was 0.91, indicating excellent internal consistency.

Detailed data are available in Multimedia Appendix 4.

Participants reported high overall satisfaction (mean 4.0/5, SD 0.9) across all module sections, with positive ratings for the e-learning format (mean 4.1/5, SD 1.0) and module duration (mean 4.2/5, SD 1.0). Compared with earlier PCEH modules, this version integrated short videos, interactive pictures, and flash cards—pedagogical tools known to enhance learner engagement and satisfaction [28]. Previous literature suggests that high satisfaction correlates with greater learning gains [29].

Knowledge improved markedly, with a mean 56% increase in correct identification of EDCs across all substances (P<.001), exceeding gains reported in prior PCEH evaluations [7,8] and aligning with results from other environmental health training programs [30]. By the end of the module, participants could identify numerous EDCs in common household environments, enabling them to detect these exposures during consultations or home visits.

Self-reported behaviors also improved significantly, with a mean increase of 2.13 points (95% CI 1.71-2.56) on a 5-point Likert scale—greater than improvements observed in previous PCEH studies (0.74 and 1.1 points) [7,8]. These gains fulfill the third pedagogical objective, equipping participants to propose evidence-based alternatives to reduce exposure. Such counseling can contribute to exposure-based prevention strategies and may help mitigate cumulative risks from multiple substances (“cocktail effect”) [31] while balancing reduction efforts with avoidance of undue patient anxiety [32].

Baseline results confirmed limited access to reliable information on EDCs, with the media as the primary source [7,16]. The asynchronous e-learning format allowed for broad participation across the Occitania region (41/78, 52% of the cohort), with sociodemographic characteristics comparable to those in previous PCEH studies [7,8].

Gains in identification were smaller for pesticides and parabens, likely reflecting high baseline recognition (64/78, 82% and 48/78, 62%, respectively), suggesting a ceiling effect. However, some heterogeneity in learning outcomes was observed. Some items showed lower recognition rates even after the intervention, particularly behavioral disorders, alkylphenols, and phenoxyethanol. Several explanations may account for this pattern. First, these topics may be less familiar to general practice residents, either because they are less frequently encountered in routine clinical practice or because they are less emphasized during initial medical training. Second, certain substances, such as alkylphenols and phenoxyethanol, involve more technical knowledge and may be more difficult to retain following a single e-learning session. Third, health effects such as behavioral disorders may be perceived as multifactorial and less directly attributable to endocrine disruptors, which could limit participants’ confidence in identifying these associations. These findings highlight potential areas for improvement in future versions of the module, suggesting the need to reinforce these specific topics through more explicit teaching, clinical case illustrations, or repeated exposure. This pattern also suggests that a single exposure to the content may not be sufficient for more complex or less salient topics.

Secondary outcomes showed strong identification of critical exposure periods and locations, confirming the first pedagogical objective and reinforcing prior training. The weaker association between EDCs and the “behavioral disorder” item highlights a target for future module updates.

A major strength of this study is the use of a structured pedagogical framework (Kirkpatrick model) combined with an innovative instructional design integrating interactive media, artificial intelligence–generated visuals, and professionally produced short videos. The module was developed collaboratively by educators, clinicians, and audiovisual specialists, ensuring both scientific accuracy and pedagogical quality.

The evaluation benefited from a matched before-and-after design and covered all participants in a defined academic cohort. The sociodemographic profile was consistent with those of previous cohorts, supporting generalizability to similar populations.

However, the study has limitations. The 17-day data collection window limited recruitment, and results are based on 52.7% (78/148) of eligible residents. The monocentric setting limits external validity, and midwifery students were excluded due to insufficient responses. Parenthood, a potential confounder [33], was rare (2/78, 2.6%) and not analyzed.

Outcomes were assessed immediately after the training, introducing possible recall bias and precluding assessment of long-term retention; self-reporting also raises the risk of social desirability bias. Objective measures such as clinical case assessments or objective structured clinical examinations could strengthen future evaluations.

Although the internal consistency was excellent, further psychometric validation (eg, factor analysis) would be needed to fully validate the scale.

A key limitation relates to the higher levels of the Kirkpatrick model, particularly the assessment of behavior changes and impact on clinical practice. As participants were early-stage residents, their opportunities to apply newly acquired knowledge in real clinical settings may have been limited. While early exposure to environmental health concepts may increase awareness, it may not immediately translate into observable changes in practice.

However, this also highlights opportunities for reinforcing learning throughout residency. Future educational strategies could include integrating environmental health considerations into clinical reasoning, for example, by systematically exploring occupational and environmental exposures during patient history taking or by documenting potential environmental determinants of health in medical records. More broadly, longitudinal approaches combining repeated exposure, clinical case discussions, and practical application may help facilitate the translation of knowledge into practice. This aligns with educational models emphasizing spiral curricula and progressive integration of competencies over time [34].

The strong results support the continued integration of this e-learning module into the PCEH curriculum and its potential extension to other health professionals involved in environmental health prevention. Scaling could involve multidisciplinary health centers, territorial health professional communities, and continuing education programs, with evaluation of reproducibility across professional groups.

Incorporating a standardized EDC assessment tool into the module could help professionals prioritize intervention areas during consultations. Additional content could address substances with persistently low recognition rates (eg, alkylphenols and phenoxyethanol) through simplified explanations or interactive case scenarios.

Given participants’ preference for some in-person learning, hybrid formats could be explored—for example, pairing e-learning with expert-led workshops during annual PCEH conferences. Additional modules, such as those on EDCs in pharmaceuticals, are planned for 2025.

Although interest in conducting theses on EDCs or environmental health remained low, earlier exposure in the medical curriculum, thesis mentorship, and targeted workshops could increase engagement. As first-year residents must now select thesis topics by the end of year 1, introducing this training earlier—during the second cycle—could enhance its influence on research topic selection. Tracking the number of theses on environmental health since the PCEH’s inception could help assess its research impact.

Finally, integration with prevention initiatives (eg, Desbrest Institute of Epidemiology and Public Health consultations, ASALEE (Action de Santé Libérale En Équipe) nurse protocols, or standardized prevention assessments [35]) and with the Platform for Data in Primary Care [36] could strengthen monitoring and targeted interventions, particularly for individuals of childbearing age.