We conducted a single-blind randomized controlled trial with 2 parallel arms. The experimental arm received a blended course comprising an asynchronous e-learning module (2.5 hours; 45% of the course) followed by a face-to-face session (3 hours; 55% of the course). The control arm received the usual face-to-face traditional course (5.5 hours; 100% of the course). The trial was conducted between May 2024 and November 2024 at Université Claude Bernard Lyon 1 (France). It followed the CONSORT (Consolidated Standards of Reporting Trials) guidelines [24] and was prospectively registered on ClinicalTrials.gov (NCT06409273).

All first-year family medicine residents at Université Claude Bernard Lyon 1 were administratively enrolled, with no exclusion criteria. The course was delivered over a single day. Residents were assigned to 1 of 2 available dates. Participants were told that the study compared different teaching methods, with no further detail. They were block randomized using random sequences of block sizes of 4 and 8 in a 1:1 allocation ratio. Randomization was performed by the first author using random permuted blocks generated online [25]. On each teaching date, 4 groups were formed (2 per arm), with a maximum of 23 participants per group. Classrooms were configured similarly, with tables arranged in a U shape and spaces reserved for breakout work.

The 5 teachers were randomized via coin toss to 1 of the 2 study arms. Teachers who taught on both dates crossed over to the other arm on the second day. For each arm, we developed detailed, time-stamped scripts to ensure instructional fidelity and equivalence of content and duration.

The course prepared residents to complete the thesis required for full qualification in France. This is a professionally oriented doctoral thesis publicly defended before an examination board. It is required for the award of the national Doctor of Medicine degree, which permits registration with the French Medical Council for independent practice. Distinct from a PhD, it is a concise, supervised scholarly work demonstrating the candidate’s ability to design, conduct, and communicate research relevant to medical practice. We specified 5 process-oriented learning objectives: 3 for knowledge acquisition and 2 for skill development (Table 1).

The blended course sequenced active knowledge construction online with supervised skill development in person. We made the alignment between the 2 sessions explicit to residents to underscore the coherence between the learning objectives and the learning methods. The blended course arm began with an e-learning session grounded in self-regulated learning (SRL). SRL has been defined as self-generated thoughts, feelings, and actions that are planned and cyclically adapted to the attainment of personal goals [26]. Process-oriented learning objectives (Table 1) for knowledge acquisition were designed to activate the SRL forethought phase (goal setting and strategy selection) [27]. We aimed to help residents identify immediate next steps (securing a supervisor, conducting a literature review, and framing a research question), whereas strategy selection focused on learning how to choose an appropriate research method. Formative checkpoints and immediate feedback were embedded through interactive HTML5 Package activities—videos with decision points (“crossroads”), multiple-choice questions, drag-the-words items, and navigation hot spots. This design supported ongoing self-monitoring and adaptive control (comparison to standards and adjustment), thereby reducing discrepancies between current understanding and target performance. The accompanying explanatory text and curated external links to websites or documents enabled resource management and self-paced elaboration, further supporting SRL. The e-learning module was implemented online in the course management system Moodle (Moodle HQ; Multimedia Appendix 1).

The traditional course arm began with the collaborative construction of a knowledge base anchored in the knowledge-oriented learning objectives. Teachers acted as facilitators, providing key information and externalizing group thinking in mind maps that made success criteria and progress visible for monitoring.

Skill-oriented learning objectives were addressed identically in both arms during a face-to-face session. For the socioconstructivist activities, peer evaluation of thesis projects was organized in breakout groups, which discussed the research question and feasibility. These activities provided external feedback according to standards explicitly set by teachers, closing the control loop and supporting the SRL reflection phase (self-evaluation and calibration). To stimulate motivational processes and goal setting, we highlighted the timeline leading up to the thesis [28].

Data were collected at the end of the teaching day via individualized links to an online form (Multimedia Appendix 2). Participants provided demographics (gender, age, research experience, and progress stage of the thesis project before teaching) and completed two 7-point Likert questionnaires: self-assessed learning (3‐8 items per objective) and satisfaction (1 item per objective). To ensure content validity and constructive alignment, all items were copied verbatim from prespecified intended learning outcomes (Multimedia Appendix 3), yielding a criterion-referenced measure. Item order was randomized.

The primary outcome was a composite score of 20 points reflecting residents’ self-assessment of knowledge and skills. It was the sum of 2 secondary outcomes: a knowledge self-assessment score (10 points) and a skill self-assessment score (10 points).

The secondary outcomes were as follows:

Analysis was performed using R (version 4.4.3; R Foundation for Statistical Computing) in RStudio (version 2024.12.1; Posit PBC) [29]. Because the primary outcome was assessed only at the end of the intervention, participants with missing outcome data were excluded from the analysis, resulting in a modified intention-to-treat approach with a complete-case analysis. Missing outcomes were not imputed, and reasons for missing data were documented. Multiplicity across secondary outcomes was controlled using a hierarchical fixed-sequence procedure that followed the prespecified order listed above (1 to 6). Each hypothesis was tested 2 sided at α=.05, and testing stopped at the first nonsignificant result (subsequent analyses were exploratory).

We modeled the primary outcome using multivariable linear regression, estimating the arm effect (blended vs traditional) while adjusting for a prespecified set of the following predictor variables. Precourse framing of a research question and prior research experience were included as proxies for baseline research preparedness. Full completion of another Moodle e-learning course was included as a proxy for familiarity with digital learning environments. Unblinding was included to account for potential expectancy or performance effects linked to perceived allocation. Teacher and course day were included to account for contextual instructional and organizational variability. Age and gender were included as standard demographic covariates. Model assumptions were checked (Multimedia Appendix 4). We report β coefficients with 95% CIs and adjusted R². For power, a sample size of 144 (72 per arm) was planned to provide 85% power to detect a 2-point difference on a 20-point scale (SD 4; 2-sided α=.05). The SD was conservatively inflated given limited prior data. Internal consistency per objective was estimated using the Cronbach α.

This trial received ethics approval from the Comité d’Ethique de la Recherche du Collège Universitaire de Médecine Générale (CUMG-IRB 2024-04-30-02). Participants received oral and written information about the study and provided written informed consent. The data collected complied with the European Union’s General Data Protection Regulation and with the data management policy of Université Claude Bernard Lyon 1 as confirmed by the university’s data protection officer. No compensation was provided to participants.

We included 163 family medicine residents randomized to the blended (n=84, 51.5%) and the traditional (n=79, 48.5%) course. A total of 155 participants completed the online form at the end of the day: 78 (50.3%) in the blended course, and 77 (49.7%) in the traditional course (Figure 1). Table 2 presents the participants’ characteristics. The mean age of the participants was 25.7 (SD 3.54) years, and most (112/155, 72.3%) were women. Most participants (114/155, 73.5%) had an idea for their thesis topic. A minority had framed their research question (24/155, 15.5%) or already had a thesis supervisor (22/155, 14.2%). In total, 29.7% (46/155) of the participants reported having research experience, and 3.9% (6/155) had completed a full master’s degree or a PhD in addition to their medical studies. Less than a quarter of the participants (32/155, 20.6%) had already completed a full e-learning course on the university’s digital learning platform (Moodle). Nearly a third of the participants (48/155, 31.0%) were unblinded at the end of the course. Five teachers offered the course, and 3 of them taught in both arms.

The proportion of postrandomization exclusions was low (8/163, 4.9%) and similar in both group allocations (Multimedia Appendix 5). It was due to missing outcome data, mostly because of work-related early departure.

Knowledge and skill self-assessment did not differ significantly between the blended and traditional courses (mean difference 0.40 [maximum possible 20] points, 95% CI −0.21 to 1.02; P=.19; Cohen d=0.21; Table 3). This result remained not statistically significant after adjustment (Table 4). The most significant predictor of the primary outcome was having formulated a research question before the course, with a β coefficient of 1.87 (95% CI 1.06-2.68; P<.001). Unblinding during the day and teacher identity were not significantly associated with the primary outcome (P=.13).

As an exploratory result, the knowledge self-assessment was significantly higher in the blended course (mean difference 0.40 [maximum possible 10], 95% CI 0.07-0.71; P=.01; Cohen d=0.38).

The skill self-assessment by residents was not significantly different between the traditional and blended courses (P=.76). Their overall satisfaction or specific satisfaction regarding knowledge- or skill-oriented learning objectives did not differ between the traditional and blended courses. The submission of a thesis proposal form was also not found to be different between the 2 groups (11/78, 14.1% in the BL arm and 10/77, 13.0% in the traditional course arm).

Items assessing each educational objective showed acceptable Cronbach α values (>0.70) for all but one educational objective, which exhibited lower internal consistency (Cronbach α=0.47; Multimedia Appendix 6).

Among French family medicine residents, self-assessed knowledge and skills did not differ between the blended and traditional courses. The between-arm mean difference was small, and the 95% CI was narrow, indicating low uncertainty around the estimated difference. At 6 months, academic achievement (thesis proposal submission) did not differ between arms. The only statistically significant finding—an exploratory, slightly higher knowledge score in the blended arm—was not educationally meaningful.

To our knowledge, no randomized controlled trial has evaluated whether partially replacing an already active, socioconstructivist face-to-face course with a blended format affects residents’ self-reported knowledge and skills. A mixed methods randomized controlled trial compared a blended evidence-based medicine course with a lecture-based (behavioral) traditional course and found no significant difference in objectively measured knowledge and skills. However, the blended group reported higher self-efficacy, attitudes, and self-reported behaviors. Focus groups suggested a clear learner preference for the blended, integrated approach [31]. In health profession education, a meta-analysis comparing flipped classrooms with traditional, lecture-based (behavioral) courses found modest improvements in learning [32]. Flipped classrooms are a common BL configuration comparable to our study intervention combining preclass online preparation with active in-class work. Other observational studies have demonstrated the relevance of BL in medical education, most notably since the COVID-19 pandemic [33,34]. A recent meta-analysis of systematic reviews recommended further research to determine the relative benefits of BL in each individual context. It also confirmed our exploratory finding of improved knowledge acquisition with BL [35]. Moreover, a randomized controlled trial with undergraduate medical students demonstrated the noninferiority of BL vs traditional education in technical skill performance (life-saving trauma skills) and retention [36]. A randomized controlled trial of postgraduate medical students was conducted to compare BL to traditional education, and it showed significantly greater efficacy of BL for developing defibrillator technical skills [37]. Our trial extends the evidence to academic skills related to developing a research project (thesis); it suggests that shortening face-to-face time through a structured e-learning component can yield comparable educational outcomes when the face-to-face component remains active and socioconstructivist.

More broadly, syntheses of digital education in the health professions consistently report large benefits vs no intervention [38] but little or no differences vs nondigital instruction [39].

This study may reassure teaching teams that BL can produce short-term educational outcomes comparable to those of traditional active and socioconstructivist education among residents, particularly in settings with constrained teaching capacity. Still, the implementation of BL does not address all the issues associated with the growing disparity between the number of teachers and residents. First, BL may not always be appropriate, either because of a misalignment between educational objectives aimed at developing complex competencies and the learning modality or because of the substantial resources required to develop online courses that effectively address such competencies [40]. For example, a course focused on professional communication competencies might be better suited to face-to-face relational simulation as the sole learning modality. Second, monitoring residents’ acquisition of generic knowledge, skills, and competencies based on a national framework still necessitates frequent, individualized human contact. Third, strategies facilitating students’ SRL may imply their supervision at the asynchronous e-learning phase, which would not allow for a significant reduction in teaching time.

The transition from traditional to blended education represents a substantial organizational shift. Teachers must allocate time to review asynchronous assessments and prepare collective feedback for the face-to-face session. It is important to estimate the time required for residents to complete all e-learning modules to avoid unintentionally increased workload. Strategies to promote SRL are also needed to optimize educational outcomes [41,42]. For example, providing dedicated time and space at the university to complete the e-learning module may help residents allocate time for it. Furthermore, working independently at home may increase feelings of stress and anxiety, as was observed during the COVID-19 pandemic [43].

One limitation of this study was the reliance on self-assessment rather than objective measures. Incorporating objective performance measures would enhance validity and mitigate self-report bias, for example, through standardized pretest-posttest knowledge tests and rubric-based scoring of residents’ performance on key skills (framing a research question and assessing feasibility) rated by trained assessors. Still, formative self-assessment is integral to self-regulation, enabling feedback on progress relative to intended learning outcomes [44,45]. Moreover, the e-learning module was an initial version. A well-established, iteratively refined module might have produced higher scores among residents randomized to the BL arm. The e-learning module was completed immediately before the face-to-face session, which constrained residents’ autonomy to plan their study and is partly misaligned with SRL. This design ensured full completion of the e-learning module, reduced the risk of early unblinding or contamination between groups before the face-to-face session, and standardized learning conditions (same setting and time on task). However, it did not capture real-life environmental variability, and the immediate sequencing in BL may also have facilitated knowledge activation for the subsequent face-to-face session. Although a high proportion of participants were unblinded by the end of the teaching day—a potential limitation—this variable was not associated with the primary outcome in multivariate analysis. The complete-case analysis may have introduced attrition bias; however, the proportion of missing outcome data remained low (8/163, 4.9%) and similar in both group allocations.

Regarding strengths, the narrow 95% CI suggests that an educationally meaningful difference of ≥10% (≥2/20 points) between groups would have been detected even though the study was not designed to test equivalence or noninferiority. A formal noninferiority trial is still needed. Furthermore, the lack of significant teacher effects on the primary outcome indicates that the standardized, timed script effectively minimized instructor-specific variability.

Self-assessment of knowledge and skills by family practice residents did not differ between blended and traditional courses. This study can reassure educators that BL may produce short-term educational outcomes (knowledge and academic skills) comparable to those of traditional active and socioconstructivist education among residents, particularly in settings with constrained teaching capacity.