Pierre-Antoine Gourraud

Background: Academic medicine (AM) careers combine research, teaching, and patient care, yet their attractiveness has declined in recent years, particularly among women, who remain underrepresented in this field. Previous research highlights the central role of interest in research, but the cognitive and contextual processes underlying academic career intentions during residency remain insufficiently understood. Social Cognitive Career Theory (SCCT) provides a relevant framework for examining how individual cognitions and learning experiences jointly shape career intentions. Methods: This cross-sectional study included 1,570 French medical residents who completed an online questionnaire. Guided by the SCCT framework, we examined the relationships between learning experiences (academic satisfaction and perceived stress), self-efficacy, outcome expectations (operationalized through work centrality), professional interest in research, and intention to pursue a career in academic medicine. Hierarchical ordinal regression models were used to predict AM career intention, and structural equation modeling was conducted to investigate the formation of professional interest in research. Gender and prior research experiences (master’s or PhD) were examined as moderating factors. Results: Overall, residents reported low intention to pursue a career in academic medicine, with women expressing significantly lower intentions than men. Professional interest in research emerged as the strongest predictor of AM career intention, followed by research experiences and mentorship. Outcome expectations, measured through work centrality, also contributed significantly to career intentions. Structural equation modeling supported key assumptions of SCCT: perceived stress was negatively associated with self-efficacy, academic satisfaction was positively associated with outcome expectations, and work centrality played a central role in shaping professional interest in research. Gender differences in career intentions were largely explained by disparities in learning experiences and research opportunities rather than by differences in outcome expectations. Moderation analyses suggested gender-specific patterns at advanced levels of research involvement. Conclusions: These findings support the relevance of SCCT for understanding academic medicine career intentions and highlight the central role of professional interest in research and contextual learning experiences. Promoting positive learning environments, providing early and structured research exposure, and ensuring access to mentorship may represent key levers for fostering academic career engagement and addressing gender disparities in academic medicine.

Introduction: A decline in intentions to pursue careers in academic medicine (AM) has been observed, despite their importance for healthcare systems and scientific advancement. Persistent gender disparities further highlight the need to understand the factors shaping these career choices. Methods: This cross-sectional online survey study examined factors associated with AM career intentions among 1,641 french medical residents (63% women) who completed an online survey assessing individual (e.g. self-efficacy, professional interest), contextual (e.g. mentorship, perceived stress), and background variables (e.g. residency year, research experience). Regression analyses identified factors, while mediation and moderation analyses explored gender-related mechanisms. Results: Consistent with prior research, interest in research, prior research training (MSc/PhD), and mentorship were associated with academic medicine career intentions. Additional factors included work-related expectations (higher work centrality, stimulation, and lower conformity/autonomy) and intrapreneurial self-capital. Gender differences were partly explained by research-related factors (57.06%) and work centrality (15.61%), with no moderating effects observed. Discussion: Promoting AM careers requires enhancing early research exposure, providing structured mentorship, and fostering organizational flexibility to accommodate diverse professional goals. Strengthening supportive environments (especially for women) by improving mentorship access and workplace safety could sustain engagement and promote equity in academic medicine.

Objectives: Academic medical careers remain marked by persistent gender inequalities, despite the growing feminisation of the medical workforce. The objective of this study was to examine whether gender differences exist in key individual and institutional determinants of academic medical career aspirations among medical residents, including research motivation, mentoring, work centrality, self-efficacy, perceived discrimination and stress. Design: This was a national cross-sectional online survey. Setting: Multicentre study conducted across 36 medical schools in France. Participants: A total of 1570 medical residents (997 women and 573 men) voluntarily participated between November 2022 and February 2023. All participants completed validated self-report questionnaires. There were no exclusion criteria beyond being enrolled in a French residency programme. Primary and secondary outcome measures: Attitudes towards research (interest, motivation, significance) were measured using the Scale of Attitudes towards Research. Mentoring was assessed with the Mentor–Mentee Perception Questionnaire, work centrality with Hirschfeld and Feild’s scale, self-efficacy with the General Self-Efficacy Scale, perceived discrimination with the Sexual Harassment and Discrimination Experiences of Academic Medical Faculty instrument and stress with the Perceived Stress Scale. Gender differences were analysed using t-tests or χ² tests with Holm-Bonferroni corrections for multiple comparisons. Results: Compared with men, women reported lower research motivation (mean (SD) 21.0 (5.5) vs 22.8 (5.5); p _adj =0.011), lower research interest (27.1 (5.5) vs 28.1 (5.7); p _adj =0.011), lower work centrality (26.7 (7.0) vs 28.2 (8.1); p _adj =0.011) and lower self-efficacy (28.3 (5.2) vs 29.9 (5.0); p _adj =0.011). Women were less likely to report having a mentor (38.5% vs 44.5%; p _adj =0.02). They also reported substantially higher levels of experienced gender discrimination (22.8% vs 3.8%; p _adj =0.005), sexual harassment (57.7% vs 18.2%; p _adj =0.005) and perceived stress (8.48 (3.29) vs 7.27 (3.43); p _adj =0.011) Conclusions: Gender differences were observed across several individual and institutional factors associated with academic medical career aspirations. Reduced access to mentoring and greater exposure to discrimination and stress among women may contribute to lower research motivation and self-efficacy. These findings highlight the need for institutional strategies addressing mentoring, workplace culture and equity to support gender parity in academic medicine. Registration: OSF preregistration for this study is available at https://osf.io/9yseq/?view_only=1c72743f542b402ba67beed6908e597d

Objectives: The “Projections In Multiple Sclerosis” (PRIMUS) project aims to develop a precision medicine platform enabling neurologists to support therapeutic decisions in multiple sclerosis by visualizing similar patient data in a reference database. We present a data integration method to combine randomized clinical trials (RCTs) and observational studies data and optimize the informativeness of the resulting database. Material and Methods: We developed an extract-transform-load data integration pipeline to combine 13 source databases: the “mother” and “high-definition” cohorts from the French MS registry and 11 industrial RCTs (31 786 patients). We aimed to inform each treatment class initiation with at least 500 patients with a 2-year clinical and MRI follow-up. Our data integration strategy used every patient visit as a potential baseline time point to inform a specific neurologist’ query to the platform, thus tailoring the actual analysis cohort to each visiting patient. Results: The PRIMUS database had 12 953 patients with at least one informative visit, using the per-visit integration. It could inform treatment initiation scenarios with at least 485 patients for glatiramer acetate and at most 1754 for natalizumab. For instance, our per-visit integration method identified 1306 patients in the high-definition cohort against 610 with the classical epidemiological per-patient integration. Although the mother cohort’s longitudinal data were deemed sparse, we identified 6128 informative patients. Discussion: Neurologists would query the PRIMUS database through a web application to support discussions with their patients and the selection of disease-modifying treatments. Conclusion: Our data integration pipeline enabled the creation of a highly informative reference database.

The production of realistic anonymous synthetic data is of high interest for the design and the training of machine learning algorithms. The exploitation of non-anonymous data is subject to several rules and limitations related to privacy constraints. This work aims to propose an anonymization method through a signal generation framework. The method builds a statistical mixing model where each signal is expressed as a linear mixture of its neighbors in the database, used to construct a set of synthetic anonymous signals. The quality of the resulting avatar database is assessed in terms of privacy and utility. The second contribution is to propose a complete pipeline allowing the application of the proposed approach to a database of real ECG signals corresponding to several cardiac arrhythmias. The distributions of the extracted ECG features are first analyzed. The predictive performance of classifiers respectively trained on the real and the avatar database are then compared. The main result shows that the model trained on avatar data achieves a 6 percent point F1-score difference compared to the model trained on original data. Consequently, while privacy metrics confirm that the avatar dataset is anonymous, the utility of the data is preserved.

Background: Chronic lung allograft dysfunction (CLAD) leads to declining respiratory function and high mortality, representing the main barrier to long-term survival in lung transplantation (LT). We performed the first genome-wide association study (GWAS) investigating donor’s and recipient’s genetic factors associated with CLAD. Method: We genotyped 392 donor-recipient pairs from the multicentric Cohort in Lung Transplantation. We tested 4.5 million SNPs for association with CLAD using multivariable logistic regression models corrected for age, sex, initial disease and genetic ancestry. Three levels of explanatory variables were separately considered to conduct GWAS: donors-only, recipients-only, and donor-recipient mismatches. We also ran HLA-centric analyses using the same models. Results: Our analysis confirmed the deleterious impact of HLA allelic and epitopic mismatches on CLAD risk, mostly driven by class I HLA (p=0.004). No significant associations with CLAD were found for donors’ genotypes or donor-recipient non-HLA mismatches. We highlighted two independent recipient’s loci associated with CLAD, including one protective signal (0.39 in CLAD vs 0.66 in non-CLAD recipients, p-value=5.05×10 ^-7 , q-value=0.017, OR=0.35) encompassing the PLXDC2 gene, and one risk signal (0.66 in CLAD vs 0.38 in non-CLAD recipients, p-value=9.86×10 ^-7 , q-value=0.017, OR=2.83) encompassing the ZNF518A/BLNK genes. These non-coding SNPs are putative regulatory variants of gene expression. Importantly, our single-cell RNA-sequencing showed a down-regulation of PLXDC2 in monocytes and lung epithelium in CLAD vs healthy controls (p≤2.0×10 ^-16 ). Conclusion: This first LT GWAS revealed two candidate loci from the recipient’s genome, both biologically relevant for CLAD pathogenesis. Our study calls for larger LT genomic initiatives to increase power for signal discovery.

Background: Sharing information derived from individual patient data is restricted by regulatory frameworks due to privacy concerns. Generative artificial intelligence can generate shareable virtual patient populations as proxies for sensitive reference datasets. Explicit demonstration of privacy is demanded. Objective: This study evaluated whether a privacy-by-design technique called “avatars” can generate synthetic datasets replicating all reported information from randomized clinical trials (RCTs). Methods: We generated 2160 synthetic datasets from two phase 3 RCTs for patients with multiple sclerosis (NCT00213135 and NCT00906399; n=865 and 1516 patients) with different configurations to select one synthetic dataset with optimal privacy and utility for each. Several privacy metrics were computed, including protection against distance-based membership inference attacks. We assessed fidelity by comparing variable distributions and assessed utility by checking that all end points reported in the publications had the same effect directions, were within the reported 95% CIs, and had the same statistical significance. Results: Protection against membership inference attacks was the hardest privacy metric to optimize, but the technique yielded robust privacy and replication of the primary end points (in 72.5% and 80.8% of the 1080 generated datasets). Utility was uneven across the variables and end points, such that information about some end points could not be captured. With optimized generation configurations, we selected one dataset from each RCT replicating all efficacy end points of the placebo and approved treatment arms while maintaining satisfactory privacy (hidden rate: 85.0% and 93.2%). Conclusions: Generating synthetic RCT datasets replicating primary and secondary efficacy end points is possible while achieving a satisfactory and explicit level of privacy. To show the potential of this method to unlock health data sharing, we released both placebo arms as open datasets.

Background Reversible cerebral vasoconstriction syndrome (RCVS) is characterized by thunderclap headaches and transient cerebral vasoconstriction. The influence of environmental factors, particularly air pollution and weather conditions, on RCVS remains poorly understood. Methods From 2010 to 2022, clinical data from two hospitals were analyzed alongside air pollution and meteorological data provided by Air Pays de la Loire and Météo-France. A time-stratified case-crossover design was used to evaluate the association between RCVS onset and daily mean concentrations of O₃, PM₁₀, PM_2.5, and NO₂, using spatial modeling of residential exposure. Pollutant effects were examined over lag periods of up to three days, with adjustments for temperature and atmospheric pressure. Results A total of 216 patients were included (mean age: 39.9 years; SD: 12.3; sex ratio: 1:1). In the three days preceding RCVS onset, only PM₁₀ concentrations remained significantly higher compared to the day of onset (19.03 vs. 17.99 µg/m³, P = 0.0029). A significant downward trend in PM₁₀ levels was observed over this period (slope −0.735, P = 0.0286), whereas no significant changes were found for PM_2.5, O₃, or NO₂. PM₁₀ was the only pollutant significantly associated with RCVS onset across lag days (Lag 0: OR 0.747, 95% CI: 0.579–0.962, P = 0.024; Lag 1: OR 0.770, 95% CI: 0.599–0.989, P = 0.041; Lag 2: OR 0.776, 95% CI: 0.604–0.997, P = 0.047; Lag 3: OR 0.768, 95% CI: 0.599–0.984, P = 0.037). After adjusting for atmospheric pressure and temperature, these associations were no longer statistically significant. Conclusion This study suggests a short-term decrease in PM₁₀ levels prior to RCVS onset and a negative association between PM₁₀ concentration and occurrence. These findings contribute to understanding the potential environmental triggers of RCVS in the context of changing air quality and climate.