A competency-based digital medical education model in conflict-affected settings Authors: Karim Qayumi, Farid Omar, Masood Noora, Seyedeh Toktam Masoumian Hosseini, Mohsen Masoumian Hosseini, Asmat Naebkhilh, Sohaila Aziz, Parwin Mansurih, Somaya Waqef, Nasrin Oryakhil, Anoosha Wahidi, Hasiba Karimi, Anmol Kapoor, Sayed Fekrat, Manizha Noorandish, Farida Homa Tanin, Zahra Rostami, Hamida Ebadi, Hamasa Ebadi, Roohullah Shabon, Yaar Mohammed, Shafiqa Allahyar, Aurelia Soltan, Mariam Hazhir, Stefan Wisbauer, Mohibullah Waziri, Alberto Ferrereso Published: 2026 ABSTRACT SummaryBackgroundMedical education in conflict-affected settings faces severe barriers, including collapsed infrastructure, faculty shortages, and sex-based exclusion. This study evaluated first-term learning outcomes among Afghan female medical trainees whose university education was interrupted following restrictions on women’s access to higher education. Most participants resided in Afghanistan, with a minority in neighbouring countries. The Canadian Virtual Medical University Initiative (CVMUI) delivers a fully online, competency-based curriculum aligned with Entrustable Professional Activities (EPAs).MethodsWe implemented a multi-system, pre-internship digital curriculum spanning five EPAs (cardiovascular, respiratory, gastrointestinal, musculoskeletal, and central nervous systems). Lecturio provided theoretical instruction, and CyberPatient enabled simulation-based clinical training. The study was conducted from November 2024 to July 2025 at a single virtual site (CVMUI). Ninety-two Level I–II female students who had previously been enrolled in medical school in Afghanistan were recruited; inclusion criteria were English proficiency (TOEFL iBT ≥70), reliable internet access, and commitment ≥15 h/week. The main outcomes were changes in knowledge (multiple-choice questions) and clinical skills (virtual Objective Structured Clinical Examinations) measured pre- and post-training. Secondary outcomes included student satisfaction and curriculum quality. Analyses used Shapiro–Wilk tests for normality, paired t-tests or Wilcoxon tests, and effect sizes.FindingsLevel I students demonstrated significant knowledge gains across all systems (mean increases +52 to +58 points; p < 0.0001). Level II students showed substantial improvements in knowledge, clinical performance, and total scores (mean differences 57–62 points; large effect sizes). More than 90% of participants rated learning objectives, feedback, and educational climate as very good or excellent. Convergent findings across knowledge, skills, and satisfaction supported strong construct validity and alignment with prior pilot data.InterpretationA fully digital, simulation-enabled, competency-based model can deliver high-quality pre-internship medical education in conflict-affected settings, offering a scalable and equitable pathway for women excluded from traditional training. Future work should examine integration of this model into national and international credentialing frameworks.FundingCanHealth International and philanthropic contributions. Research in contextEvidence before this studyWe systematically searched PubMed, Scopus, and Google Scholar from database inception to December 31, 2024, using the terms “digital medical education”, “conflict-affected regions”, “Afghanistan”, and “women in medicine”. The search yielded 127 articles after deduplication. Limited empirical data evaluated fully online competency-based medical curricula for learners in conflict zones. Previous publications, including our 2024 eClinicalMedicine pilot, validated a single-system (cardiovascular) digital training model.Added value of this studyThis multi-system evaluation of a digitally enhanced, Entrustable Professional Activity-based curriculum for Afghan female medical students demonstrates significant improvements in knowledge and clinical skills outcomes, high learner satisfaction, and validation against international benchmarks. It confirms that virtual, simulation-based competency-based medical education can serve as a viable alternative in contexts where traditional training is not accessible.Implications of all the available evidenceDigital competency-based medical education models, such as the Canadian Virtual Medical University Initiative (CVMUI), can sustain medical training in crisis and low-resource settings, particularly for women who are denied access to physical campuses. With rigorous design, psychometric reliability, and supportive mentorship, these programmes can produce outcomes comparable to conventional training and offer a scalable, equitable pathway toward rebuilding global health-workforce capacity. Next steps include longitudinal follow-up of learners and assessment of clinical practice outcomes. Research in context Evidence before this study We systematically searched PubMed, Scopus, and Google Scholar from database inception to December 31, 2024, using the terms “digital medical education”, “conflict-affected regions”, “Afghanistan”, and “women in medicine”. The search yielded 127 articles after deduplication. Limited empirical data evaluated fully online competency-based medical curricula for learners in conflict zones. Previous publications, including our 2024 eClinicalMedicine pilot, validated a single-system (cardiovascular) digital training model. Added value of this study This multi-system evaluation of a digitally enhanced, Entrustable Professional Activity-based curriculum for Afghan female medical students demonstrates significant improvements in knowledge and clinical skills outcomes, high learner satisfaction, and validation against international benchmarks. It confirms that virtual, simulation-based competency-based medical education can serve as a viable alternative in contexts where traditional training is not accessible. Implications of all the available evidence Digital competency-based medical education models, such as the Canadian Virtual Medical University Initiative (CVMUI), can sustain medical training in crisis and low-resource settings, particularly for women who are denied access to physical campuses. With rigorous design, psychometric reliability, and supportive mentorship, these programmes can produce outcomes comparable to conventional training and offer a scalable, equitable pathway toward rebuilding global health-workforce capacity. Next steps include longitudinal follow-up of learners and assessment of clinical practice outcomes. Introduction The World Health Organization projects a global shortage of 18 million health workers by 2030, with significant deficits in developing nations and conflict-affected regions.1 The UN’s Sustainable Development Goals (SDG) 3 and 4 aim to ensure healthy lives and promote well-being while advancing inclusive, equitable, and high-quality education with lifelong learning opportunities for all.2 To eliminate this gap in medical education and bring equity and access to all, medical education must undergo a profound transformation.3 Nowhere is this transformation more urgent than in conflict-affected regions, where war, sex-based restrictions, and systemic instability have dismantled traditional higher education systems.4 In countries such as Afghanistan, Sudan, Ukraine, and Gaza, political upheaval has severely curtailed opportunities, particularly for women to pursue professional careers in medicine, exacerbating an already critical shortage of healthcare workers.5,6 Conventional medical training depends heavily on synchronous in-person lectures, pre-internship hospital-based clinical rotations, and access to trained faculty and resources, which are often unavailable or inaccessible in fragile settings.7 Students from rural areas, marginalised groups, and particularly women are disproportionately excluded.8 In such contexts, pre-internship digital medical education may offer an alternative that can deliver quality instruction and credentialed training despite systemic, geographic, political, or cultural barriers.9 Recent advances in digital medical education, simulation-based training, and competency-based curricula support clinical training in resource-constrained environments. However, evidence remains limited regarding the effectiveness of fully remote clinical training models for learners who are completely excluded from traditional educational institutions10, 11, 12 The Canadian Virtual Medical University Initiative (CVMUI) was established to address these challenges by offering a fully online, competency-based, student-centred programme aligned with the international medical curriculum concept.13 Grounded in pedagogical best practices and Entrustable Professional Activities (EPAs), and using digital platforms, CVMUI targets students in low-resource and conflict-affected regions, particularly women who are barred from attending physical institutions. This model incorporates video lectures, clinical simulations, virtual Objective Structured Clinical Examinations (OSCEs), case-based discussion forums, feedback, and psychometrically validated assessments. A pilot project with eight universities in Afghanistan proved the efficacy of the CVMUI curriculum for the cardiovascular system.13 This paper evaluates the implementation of the CVMUI model among Afghan female medical students, examining learning outcomes and student satisfaction. By employing a mixed-methods approach that incorporates pre- and post-testing, skill-based assessments, and survey-based evaluations, we aim to assess the effectiveness, reliability, and validity of the first term of the CVMUI programme. The findings provide insights for policymakers, academic leaders, and global health advocates on alternative educational pathways in regions where traditional medical schools are no longer viable. Methods Programme design and setting This explanatory sequential mixed-methods evaluation was conducted from November 2024 to July 2025 across five EPAs at the CVMUI. The project employed a QUAN → QUAL design: Phase 1 collected and analysed quantitative outcomes (knowledge, skills); Phase 2 used qualitative data to explain how and why the digital curriculum produced these outcomes. Level I comprised foundational sciences; Level II included systems-based pathology and clinical skills for trainees. The study cohort represents a census of all eligible Afghan women who applied and enrolled in the programme. Participation required English proficiency and reliable internet access, which may limit generalisability to the broader population of displaced or excluded learners. These requirements reflect the real-world conditions necessary for participation in digital medical education and therefore define, rather than limit, the population to which this model applies. Participants and sampling All eligible learners were enrolled via complete enumeration. Participants were Afghan female medical students who were barred from continuing their traditional education and who met the inclusion criteria: (1) previous medical school enrolment; (2) English proficiency (TOEFL iBT ≥70); (3) reliable internet access; and (4) informed consent. Exclusion criteria: active enrolment in another medical programme or inability to commit ≥15 h/week. Because the programme enrolled a defined cohort of eligible learners, the study sample represents complete enumeration rather than recruitment based on a priori statistical power calculations. The total CVMUI student body included 92 young Afghan women with an average age of 23·3 years (range 18–35). Of these, 84 completed Phase 1, resulting in a 92·1% retention rate. Eight participants withdrew due to relocation, marriage, or socioeconomic factors (Table 1). Participant sex was female for all individuals; sex was self-reported. Because the study population was uniform, comparisons by sex were not applicable.Table 1Baseline characteristics of study participants.CharacteristicLevel 1 (n = 58)Level 2 (n = 26)Total/SummaryParticipants582684Sex (female)58 (100%)26 (100%)84 (100%)Age, mean (range)23.1 (18–35)23.8 (20–26)Early 20s overallTraining levelEarly preclinicalAdvanced preclinicalMixed cohortLocation in Afghanistan50 (86%)MajorityMajorityResiding abroad8 (14%)≈5 (19%)MinorityKabul residence36 (62%)13 (50%)Majority urbanDisplacement (left Afghanistan)7 (11%)≈5 (19%)≈14% overallAcademic performanceMean GPA 92.3%Mean score 82.4%VariableEducational continuity53 (92%) active26 (100%) activeHigh retentionEnrollment period2023–20252024–2025Recent cohortsNote: Q4 (fourth quarter) refers to the period of October–December. In the CVMUI enrollment analysis, “Q4 2024—Q1 2025” represents the peak intake period from October 2024 through March 2025. Table 1Baseline characteristics of study participants.CharacteristicLevel 1 (n = 58)Level 2 (n = 26)Total/SummaryParticipants582684Sex (female)58 (100%)26 (100%)84 (100%)Age, mean (range)23.1 (18–35)23.8 (20–26)Early 20s overallTraining levelEarly preclinicalAdvanced preclinicalMixed cohortLocation in Afghanistan50 (86%)MajorityMajorityResiding abroad8 (14%)≈5 (19%)MinorityKabul residence36 (62%)13 (50%)Majority urbanDisplacement (left Afghanistan)7 (11%)≈5 (19%)≈14% overallAcademic performanceMean GPA 92.3%Mean score 82.4%VariableEducational continuity53 (92%) active26 (100%) activeHigh retentionEnrollment period2023–20252024–2025Recent cohortsNote: Q4 (fourth quarter) refers to the period of October–December. In the CVMUI enrollment analysis, “Q4 2024—Q1 2025” represents the peak intake period from October 2024 through March 2025. Baseline characteristics of study participants. Note: Q4 (fourth quarter) refers to the period of October–December. In the CVMUI enrollment analysis, “Q4 2024—Q1 2025” represents the peak intake period from October 2024 through March 2025. Curriculum and EPAs The unit of education at CVMUI is the EPA. EPAs are units of professional clinical practice that integrate multiple competencies and can be entrusted to trainees once adequate proficiency has been demonstrated.12 System-based modules (cardiovascular (CV), respiratory (Resp), gastrointestinal (GI), central nervous system (CNS), and musculoskeletal (MSK) were designed to develop the knowledge, clinical reasoning, and patient-management skills required for relevant EPAs. The theoretical content was mapped to Lecturio (minimum 120–150 min/day), and the clinical competency milestones for Level II, including history taking and physical examination were delivered through the CyberPatient online simulation platform (two cases/week and eight cases/EPA). Each EPA spanned at least four weeks. Psychometric properties for the knowledge tests (45-item MCQ/EMQ bank) were validated by a six-member expert panel (κ = 0·82). A modified Angoff method14 used ten borderline-student items per EPA, with cut-scores set at 65% (Cronbach’s α = 0.76–0.84 per EPA15 (Appendix, pp. 1)). Testing effects were reduced by rotating 30% of alternate items between pre- and post-tests, with item analysis confirming comparable difficulty (p = 0.42).16 For virtual OSCEs (four stations, 20 min each), examiner training included a 2-h calibration session; inter-rater reliability was assessed using intraclass correlation coefficients (target >0·75). Final scores were averaged across two independently trained examiners; ICCs ranged from 0.78 to 0.9116 (Appendix, pp. 1). Ethics Ethical review by the CVMUI Ethics Committee was waived because the analysis used anonymised data collected for institutional quality improvement (non-research educational programme evaluation). All participants provided written informed consent for the anonymised use of their educational data. Statistical analysis Quantitative data analysis (phase 1) Continuous variables were summarised as means with standard deviations or medians with interquartile ranges, as appropriate. Within-participant changes in assessment scores were evaluated using paired statistical tests. Normality of difference scores (post-test minus pre-test) was assessed per module and per outcome using the Shapiro–Wilk test (Table 2), supplemented by visual inspection of histograms and Q–Q plots.Table 2Pre- and post-training scores, mean differences, and 95% confidence intervals.MeasureCVRespGIMSKCNSLevel 1 knowledge Number of students3938405058 p value of test for normality (Shapiro–Wilk test)0.140.090.320.260.15 TestPair t-testPair t-testPair t-testPair t-testPair t-test Pre-training mean (SD)27.10 (4.55)30.55 (4.57)28.57 (5.68)27.78 (8.39)28.13 (7.96) Post-training mean (SD)82.29 (4.24)87.92 (6.29)87.00 (14.76)80.10 (21.77)82.45 (20.10) Mean difference55.1857.3758.4352.3254.31 95% CI53.15–57.2254.63–60.1052.98–63.8744.68–59.9647.76–60.86 p value0.0010.0010.0010.0010.001Level 2 knowledge Number of students1312132625 p value of test for normality (Shapiro–Wilk test)0.0220.0370.0410.160.24 TestWilcoxonWilcoxonWilcoxonPair t-testPair t-test Pre-training mean (SD)18.15 (7.10)18.50 (7.05)18.92 (5.79)20.53 (8.68)22.28 (7.10) Post-training mean (SD)58.03 (15.10)63.95 (11.89)67.69 (14.76)75.84 (14.36)76.21 (14.39) Mean difference39.8745.4548.7755.3053.93 95% CI28.70–51.0436.32–54.5839.92–57.6248.56–62.0447.65–60.21 p value<0.0001<0.0001<0.0001<0.0001<0.0001Level 2 clinical performance Number of students1312132625 p value of test for normality (Shapiro–Wilk test)0.040.0130.0150.240.34 TestWilcoxonWilcoxonWilcoxonPair t-testPair t-test Pre-training mean (SD)25.30 (4.57)28.41 (4.23)25.23 (6.55)20.53 (8.68)23.80 (5.60) Post-training mean (SD)82.06 (9.50)95.66 (4.45)86.15 (17.45)89.38 (10.96)85.60 (13.28) Mean difference56.7567.2560.9268.8461.72 95% CI48.30–65.2062.01–72.4951.47–70.3762.98–74.7055.43–68.01 p value<0.0001<0.0001<0.0001<0.0001<0.0001CV, cardiovascular; Resp, respiratory; GI, gastrointestinal; MSK, musculoskeletal; CNS, central nervous system; SD, standard deviation; CI, confidence interval. All p-values two-sided; analyses by paired t-test or Wilcoxon signed-rank test as appropriate. Table 2Pre- and post-training scores, mean differences, and 95% confidence intervals.MeasureCVRespGIMSKCNSLevel 1 knowledge Number of students3938405058 p value of test for normality (Shapiro–Wilk test)0.140.090.320.260.15 TestPair t-testPair t-testPair t-testPair t-testPair t-test Pre-training mean (SD)27.10 (4.55)30.55 (4.57)28.57 (5.68)27.78 (8.39)28.13 (7.96) Post-training mean (SD)82.29 (4.24)87.92 (6.29)87.00 (14.76)80.10 (21.77)82.45 (20.10) Mean difference55.1857.3758.4352.3254.31 95% CI53.15–57.2254.63–60.1052.98–63.8744.68–59.9647.76–60.86 p value0.0010.0010.0010.0010.001Level 2 knowledge Number of students1312132625 p value of test for normality (Shapiro–Wilk test)0.0220.0370.0410.160.24 TestWilcoxonWilcoxonWilcoxonPair t-testPair t-test Pre-training mean (SD)18.15 (7.10)18.50 (7.05)18.92 (5.79)20.53 (8.68)22.28 (7.10) Post-training mean (SD)58.03 (15.10)63.95 (11.89)67.69 (14.76)75.84 (14.36)76.21 (14.39) Mean difference39.8745.4548.7755.3053.93 95% CI28.70–51.0436.32–54.5839.92–57.6248.56–62.0447.65–60.21 p value<0.0001<0.0001<0.0001<0.0001<0.0001Level 2 clinical performance Number of students1312132625 p value of test for normality (Shapiro–Wilk test)0.040.0130.0150.240.34 TestWilcoxonWilcoxonWilcoxonPair t-testPair t-test Pre-training mean (SD)25.30 (4.57)28.41 (4.23)25.23 (6.55)20.53 (8.68)23.80 (5.60) Post-training mean (SD)82.06 (9.50)95.66 (4.45)86.15 (17.45)89.38 (10.96)85.60 (13.28) Mean difference56.7567.2560.9268.8461.72 95% CI48.30–65.2062.01–72.4951.47–70.3762.98–74.7055.43–68.01 p value<0.0001<0.0001<0.0001<0.0001<0.0001CV, cardiovascular; Resp, respiratory; GI, gastrointestinal; MSK, musculoskeletal; CNS, central nervous system; SD, standard deviation; CI, confidence interval. All p-values two-sided; analyses by paired t-test or Wilcoxon signed-rank test as appropriate. Pre- and post-training scores, mean differences, and 95% confidence intervals. CV, cardiovascular; Resp, respiratory; GI, gastrointestinal; MSK, musculoskeletal; CNS, central nervous system; SD, standard deviation; CI, confidence interval. All p-values two-sided; analyses by paired t-test or Wilcoxon signed-rank test as appropriate. Level I knowledge All five modules satisfied the normality assumption (CV: p = 0.14; Resp: p = 0.09; GI: p = 0.32; MSK: p = 0.26; CNS: p = 0.15). Paired t-tests were therefore used for every Level I comparison. Level II knowledge The MSK (p = 0.16) and CNS (p = 0.24) modules were normally distributed; the CV (p = 0.022), Resp (p = 0.037), and GI (p = 0.041) modules were non-normal. Accordingly, paired t-tests were used for MSK and CNS, and Wilcoxon signed-rank tests for CV, Resp, and GI. Level II clinical performance Normality was met for MSK (p = 0.24) and CNS (p = 0.34), but not for CV (p = 0.04), Resp (p = 0.013), or GI (p = 0.015). Paired t-tests were applied to MSK and CNS, and Wilcoxon tests to CV, Resp, and GI. For comparisons between independent groups (e.g., the benchmark cohort), independent-samples t-tests were used when both groups’ data met the normality assumption; otherwise, Mann–Whitney U tests were applied. Given the within-participant design, fixed individual-level confounders are inherently controlled. As a comprehensive sensitivity analysis, for every EPA module (CV, Resp, GI, MSK, and CNS) at both Level 1 and Level 2 (knowledge and clinical performance), we explored multivariable linear regression models with post-test score (or post-test performance score) as the outcome and pre-test score, age, and prior university semester (Level 1: semesters completed categorically; Level 2: number of semesters completed continuously) as covariates. These analyses confirmed that the significant pre-to-post improvements persisted after adjustment, with negligible changes in effect estimates across all modules (Appendix, pp. 10–11). Future studies with larger samples may explore more comprehensive multivariable modelling approaches. Mean differences with 95% confidence intervals (CIs) were calculated. Missing data were minimal (<5% of observations); therefore, primary analyses used complete-case methods (listwise deletion). Correlation analyses were conducted on subsets with complete data (n = 61 for grade point average [GPA]-related analyses; n = 86 for perception-based analyses). Effect sizes were estimated using Cohen’s d. All statistical tests were two-sided, and a p-value < 0.05 (the significance threshold) was considered statistically significant. Analyses were conducted using SPSS version 29.0 (IBM Corp., Armonk, NY, USA). Qualitative evaluation (phase 2) Phase 2 was designed to explain the quantitative patterns through universal qualitative data collection across the entire census (n = 84). All students participated in weekly reflection surveys (5-point Likert + open-ended questions), monthly one-on-one semi-structured interviews (45–60 min via Zoom), and end-of-term evaluation surveys with qualitative components. A six-member team analysed all qualitative data using thematic analysis (Braun & Clarke, 2006) in NVivo v14.17 Trustworthiness was established through member checking, triangulation, audit trail, and inter-coder reliability (Cohen’s κ = 0.88). Quantitative perception data and qualitative themes were integrated using joint display matrices (Fig. 6). Role of the funding source CanHealth International provided free access to the virtual learning platform for all students and faculty described in this study. KQ is the founder and current CEO of CanHealth International, and STMH, MMH, NO, FO, MN (Masood Noora), and AN are affiliated with CanHealth International. CanHealth and other funders (philanthropic donors) had no role in the collection, statistical analysis, or interpretation of study data. The funders had no role in study design, data interpretation, or writing of the report. Results Participant characteristics A total of 92 participants were enrolled, of whom 84 completed the training programme. Participants were distributed across training levels and clinical modules, with the majority residing within Afghanistan. Baseline demographic characteristics are summarised in Table 1. Differences in total counts across tables reflect variations in data completeness and EPA-specific inclusion criteria. Eight participants withdrew due to relocation, marriage, or socioeconomic circumstances (Fig. 1). Level I (knowledge) Mean knowledge scores increased significantly across all modules. Improvements ranged from 52.3 to 58.4 points, with narrow confidence intervals. The largest improvements were observed in the GI (mean difference 58.43, 95% CI 52.98–63.87) and Resp modules (57.37, 54.63–60.10), while improvements remained substantial in CV (55.18, 53.15–57.22), CNS (54.31, 47.76–60.86), and MSK modules (52.32, 44.68–59.96). All comparisons were statistically significant (p < 0.0001) (Fig. 2, Table 2) (Appendix, pp. 11).Fig. 1Schematic diagram of the present CVMUI observational project. CVMUI, Canadian Virtual Medical University Initiative; EPA, Entrustable Professional Activity; MCQ, multiple-choice question; CB, case-based; OSCE, Objective Structured Clinical Examination.Fig. 2This figure shows Level I assessment results across EPAs 1–5, with post-test scores significantly higher than pre-test scores (p < 0.0001 for all). The improvements demonstrate substantial learning gains, with error bars (SEM) confirming consistency of the outcomes. Schematic diagram of the present CVMUI observational project. CVMUI, Canadian Virtual Medical University Initiative; EPA, Entrustable Professional Activity; MCQ, multiple-choice question; CB, case-based; OSCE, Objective Structured Clinical Examination. This figure shows Level I assessment results across EPAs 1–5, with post-test scores significantly higher than pre-test scores (p < 0.0001 for all). The improvements demonstrate substantial learning gains, with error bars (SEM) confirming consistency of the outcomes. Level II (knowledge) Significant improvements in knowledge were observed across all modules, with mean differences ranging from 39.9 to 55.3 points. The greatest increases were seen in the MSK (55.30, 95% CI 48.56–62.04) and CNS modules (53.93, 47.65–60.21), followed by GI (48.77, 39.92–57.62), Resp (45.45, 36.32–54.58), and CV modules (39.87, 28.70–51.04). All changes were statistically significant (p < 0.0001) (Fig. 3, Table 2) (Appendix, pp. 12–15).Fig. 3This figure shows significant improvements in Level II assessment scores across all five systems—CV, Resp, digestive, MSK, and CNS—with post-test scores markedly higher than pre-test scores (p < 0.0001). Error bars indicate SEM, confirming consistent learning gains with the CVMUI curriculum. This figure shows significant improvements in Level II assessment scores across all five systems—CV, Resp, digestive, MSK, and CNS—with post-test scores markedly higher than pre-test scores (p < 0.0001). Error bars indicate SEM, confirming consistent learning gains with the CVMUI curriculum. Level II (clinical performance) Clinical performance scores improved markedly across all modules. Mean improvements ranged from 56.8 to 68.8 points. The largest improvements were observed in MSK (68.84, 95% CI 62.98–74.70) and Resp modules (67.25, 62.01–72.49), followed by CNS (61.72, 55.43–68.01), GI (60.92, 51.47–70.37), and CV modules (56.75, 48.30–65.20). All comparisons were statistically significant (p < 0.0001) (Fig. 4, Table 2) (Appendix, pp. 13).Fig. 4This figure shows significant improvements in Level II performance assessed by OSCE across all five systems—CV, Resp, GI, MSK, and CNS—with post-test scores markedly higher than pre-test scores (p < 0.0001). Error bars indicate SEM, confirming consistent learning gains with the CVMUI curriculum. This figure shows significant improvements in Level II performance assessed by OSCE across all five systems—CV, Resp, GI, MSK, and CNS—with post-test scores markedly higher than pre-test scores (p < 0.0001). Error bars indicate SEM, confirming consistent learning gains with the CVMUI curriculum. Level II total (knowledge + skills) Level II students demonstrated significant gains across knowledge, skills (virtual OSCE), and total scores (all p < 0.0001). Total mean score improvements ranged from approximately 57–62 points, with very large effect sizes (Cohen’s d 1.85–2.89). High-score attainment (≥70%) exceeded 85% across systems, reaching 100% in the respiratory module (Appendix, pp. 14–15). Comparison with benchmark cohort At Level I, programme scores were substantially higher than benchmark values across all modules, with mean differences ranging from 33.6 to 41.4 points (all p < 0.0001). At Level II (knowledge), improvements relative to the benchmark cohort were more variable, with statistically significant differences observed in GI, MSK, and CNS modules (Table 3).Table 3Contextual comparison of knowledge and clinical performance scores between the benchmark cohort and the current programme.MeasureCVRespGIMSKCNSLevel 1 knowledge Benchmark46.4846.4846.4846.4846.48 Program82.2987.9287.0080.1082.45 Mean difference35.8141.4440.5233.6235.92 p value0.00100.00100.0010<0.0001<0.0001Level 2 knowledge Benchmark56.0356.0356.0356.0356.03 Program58.0363.9567.6975.8476.21 Mean difference2.007.9211.6619.8120.18 p value0.610.110.030<0.0001<0.0001Level 2 clinical performance Benchmark73.5073.5073.5073.5073.50 Program82.0695.6686.1589.3885.60 Mean difference8.5622.1612.6515.8812.10 p value0.16<0.00010.0200.0100.020CV, cardiovascular; Resp, respiratory; GI, gastrointestinal; MSK, musculoskeletal; CNS, central nervous system; SD, standard deviation; CI, confidence interval. All p-values two-sided; analyses by paired t-test or Wilcoxon signed-rank test as appropriate. Table 3Contextual comparison of knowledge and clinical performance scores between the benchmark cohort and the current programme.MeasureCVRespGIMSKCNSLevel 1 knowledge Benchmark46.4846.4846.4846.4846.48 Program82.2987.9287.0080.1082.45 Mean difference35.8141.4440.5233.6235.92 p value0.00100.00100.0010<0.0001<0.0001Level 2 knowledge Benchmark56.0356.0356.0356.0356.03 Program58.0363.9567.6975.8476.21 Mean difference2.007.9211.6619.8120.18 p value0.610.110.030<0.0001<0.0001Level 2 clinical performance Benchmark73.5073.5073.5073.5073.50 Program82.0695.6686.1589.3885.60 Mean difference8.5622.1612.6515.8812.10 p value0.16<0.00010.0200.0100.020CV, cardiovascular; Resp, respiratory; GI, gastrointestinal; MSK, musculoskeletal; CNS, central nervous system; SD, standard deviation; CI, confidence interval. All p-values two-sided; analyses by paired t-test or Wilcoxon signed-rank test as appropriate. Contextual comparison of knowledge and clinical performance scores between the benchmark cohort and the current programme. CV, cardiovascular; Resp, respiratory; GI, gastrointestinal; MSK, musculoskeletal; CNS, central nervous system; SD, standard deviation; CI, confidence interval. All p-values two-sided; analyses by paired t-test or Wilcoxon signed-rank test as appropriate. Baseline knowledge comparison Objective pre-test scores indicated low baseline competence (mean 30%), whereas self-reported knowledge was slightly lower (24%) and showed reasonable alignment (mean difference 6 percentage points; Cohen’s d = 0.40). In contrast, prior academic performance, as reflected by converted university GPA, markedly overestimated baseline knowledge (mean 81%), with large discrepancies compared with objective pre-test scores (51 percentage points; Cohen’s d = 3.40) (Table 4).Table 4Comparison of baseline knowledge indicators and their agreement with objective performance.Indicator of baseline knowledgeMean/value (%)ComparisonEffect size (Cohen’s d)InterpretationObjective pre-test score30%––Direct measure of actual medical knowledge; indicates low baseline competenceSelf-reported background knowledge24%Pre-test vs. self-report: 6-point difference0.40Good alignment between perceived and measured knowledgePrevious university GPA (converted)81%GPA vs. pre-test: 51-point difference3.40GPA substantially overestimates actual knowledgeGPA vs. self-report3.8057-point DifferenceGPA poorly reflects perceived and actual competenceOverall statistical comparison–GPA vs. pre-test: p < 0.000001–Strong statistical evidence of GPA inflationGPA, Grade Point Average; p, p-value. Table 4Comparison of baseline knowledge indicators and their agreement with objective performance.Indicator of baseline knowledgeMean/value (%)ComparisonEffect size (Cohen’s d)InterpretationObjective pre-test score30%––Direct measure of actual medical knowledge; indicates low baseline competenceSelf-reported background knowledge24%Pre-test vs. self-report: 6-point difference0.40Good alignment between perceived and measured knowledgePrevious university GPA (converted)81%GPA vs. pre-test: 51-point difference3.40GPA substantially overestimates actual knowledgeGPA vs. self-report3.8057-point DifferenceGPA poorly reflects perceived and actual competenceOverall statistical comparison–GPA vs. pre-test: p < 0.000001–Strong statistical evidence of GPA inflationGPA, Grade Point Average; p, p-value. Comparison of baseline knowledge indicators and their agreement with objective performance. GPA, Grade Point Average; p, p-value. Student satisfaction Students rated the learning climate as highly supportive (mean 4.78/5), feedback as constructive (4.71/5), and learning objectives as clear (4.74/5). More than 90% rated these domains as very good or excellent (Fig. 5) (Appendix, pp. 16–27).Fig. 5Student satisfaction with the CVMUI learning environment and instruction. Student satisfaction with the CVMUI learning environment and instruction. Triangulation of outcomes Convergence across knowledge gains (MCQs), skills (virtual OSCEs), and affective measures (satisfaction) demonstrated strong construct validity and a coherent programme effect (Table 5, Fig. 6).Table 5Data sources and indicators for triangulation of outcomes.Data typeIndicatorStatistical resultsKnowledge gainsPre- vs. post-test scores in levels 1 & 2p < 0.0001Skill developmentPerformance on procedural/clinical tasksp < 0.0001Learner satisfaction>90% “Excellent/Very Good” in surveys>90% across all categoriesFig. 6Data triangulation demonstrates the conversion of cognitive, psychomotor, and affective domains within the CVMUI curriculum. Integration occurred through joint display analysis linking quantitative outcomes with qualitative explanatory themes. Table 5Data sources and indicators for triangulation of outcomes.Data typeIndicatorStatistical resultsKnowledge gainsPre- vs. post-test scores in levels 1 & 2p < 0.0001Skill developmentPerformance on procedural/clinical tasksp < 0.0001Learner satisfaction>90% “Excellent/Very Good” in surveys>90% across all categories Data sources and indicators for triangulation of outcomes. Data triangulation demonstrates the conversion of cognitive, psychomotor, and affective domains within the CVMUI curriculum. Integration occurred through joint display analysis linking quantitative outcomes with qualitative explanatory themes. Barriers and supports The challenges fell into three main categories:•Technological: intermittent electricity and connectivity (approximately 30%).•Academic: gaps in foundational knowledge and the need for improved study skills.•Personal: caregiving responsibilities, stress, and limited space for studying. •Technological: intermittent electricity and connectivity (approximately 30%). Technological: intermittent electricity and connectivity (approximately 30%). •Academic: gaps in foundational knowledge and the need for improved study skills. Academic: gaps in foundational knowledge and the need for improved study skills. •Personal: caregiving responsibilities, stress, and limited space for studying. Personal: caregiving responsibilities, stress, and limited space for studying. Modular pacing, academic staff mentorship, and the virtual student-support clinic helped to mitigate these barriers (Appendix, pp. 27–29). Discussion A fully online, competency-based, pre-internship model can substantially improve theoretical knowledge and clinical skills among medical students in conflict-affected settings. Gains across CV, Resp, GI, MSK, and CNS EPAs support the curriculum’s internal, external, and construct validity and confirm that high-fidelity competency-based medical education can be delivered effectively in a virtual environment. Post-test improvements were consistently large, with large effect sizes across all systems, though cardiovascular outcomes were modestly lower than those of respiratory and digestive modules. Faculty debriefings and learner interviews indicated that this initial module coincided with students’ acclimation to unfamiliar digital platforms and simulation tools, consistent with Hrastinski’s framework on digital readiness18 and Sweller’s cognitive load theory.19 As students progressed, their fluency with the platforms increased, and performance across subsequent systems improved accordingly. This aligns with global shifts toward competency-based medical education, which prioritises measurable outcomes and progressive entrustment over time-based progression.11,20, 21, 22 Anchoring instruction in EPAs ensured that learning objectives were explicit, observable, and contextually relevant. Integration of Lecturio for theory delivery and CyberPatient for simulation created a feedback-rich environment that supported the development of applied clinical skills. The significant improvements in virtual OSCEs, particularly in history-taking and physical examination, reinforce evidence from Issenberg and colleagues22 that simulation enhances clinical competence and support findings from health professional education that online environments can match or exceed traditional instruction. Psychometric indicators (Cronbach’s α = 0.87; ICC = 0.82) and triangulation of quantitative and qualitative data further strengthen confidence in the curriculum’s validity and reproducibility. A central methodological finding concerns the reliability of baseline knowledge measures. Students reported a mean GPA of approximately 80% from their previous universities; however, Afghan medical schools are not accredited by the World Federation for Medical Education, and their examination systems operate without standardised psychometric oversight. GPA may reflect grade inflation, inconsistent assessment practices, variable faculty expertise, or sociocultural and structural pressures, rather than true mastery of biomedical knowledge.23 These concerns were borne out in our data: GPAs (≈80%) were markedly discordant with objective CVMUI pre-test scores (30%) and student self-assessed background knowledge (24%). The >50-point gap between GPA and pre-test results (p < 0.0001) represents a substantial divergence and mirrors the literature showing that GPA in non-standardised systems often lacks construct validity.23,24 By contrast, the close alignment between the pre-test and self-assessment (a six-point difference) supports the diagnostic pre-test as a more accurate and equitable measure of baseline competence. Standardising pre- and post-assessments further eliminated institutional covariates and ensured that knowledge gains could be attributed to the CVMUI curriculum rather than prior educational variability. The CVMUI model addresses profound equity gaps in Afghanistan, where sex-based restrictions have halted medical education for thousands of women. Prior work by Glass et al.25 and Turkmani26 has underscored the urgency of alternative training pathways to sustain maternal and child health services amid political crises. CVMUI provides a secure, accessible, and academically rigorous solution for displaced female learners and marginalised populations. High satisfaction ratings – over 90% of learners rated learning climate, clarity of objectives, and quality of feedback as very good or excellent – affirm the programme’s affective and pedagogical strengths. Challenges reported by students, including foundational knowledge gaps, intermittent connectivity, and psychosocial stressors, mirror barriers described in other digital-learning studies from marginalised populations.27,28 CVMUI’s modular design, mentorship structures, technical support, and virtual student clinic mitigated many of these issues, promoting continuity and inclusion. Although the present study evaluates educational outcomes rather than professional licensure pathways, competency-based training models such as this are intended to preserve learners’ clinical competence and facilitate reintegration into formal training systems when educational access is restored. Future work will examine recognition of such training within national and international credentialing frameworks. Beyond educational efficacy, CVMUI contributes directly to global health workforce development. Traditional medical education models, reliant on physical campuses and clinical sites, exclude many learners in fragile settings. By reducing geographic, financial, and sociopolitical barriers, digital competency-based medical education offers a scalable and cost-effective mechanism for expanding medical training capacity, aligning with global recommendations for workforce expansion and universal health coverage.13 As Afghanistan and similar contexts face critical shortages of female healthcare providers, CVMUI’s model provides an urgently needed, internationally aligned pathway. Several limitations should be acknowledged The study cohort represents a selected group of learners with sufficient digital access and English proficiency, which may limit generalisability. The evaluation lacked a concurrent control group and relied on internal programme assessments. A concurrent control group was not feasible due to the absence of alternative educational pathways for this population following institutional exclusion; therefore, a within-participant pre-post design with large effect sizes and triangulated outcomes was used to strengthen causal inference. Attrition, although limited, may reflect socioeconomic pressures that could influence participation. Comparisons with prior benchmark cohorts should be interpreted cautiously, given potential contextual differences across training environments. The study population consisted exclusively of women; comparisons by sex were not applicable. Although the within-participant pre-post design inherently controls for fixed individual-level confounders (e.g., baseline ability and prior training background), residual confounding from time-varying factors cannot be excluded. Given the modest total sample size and limited independent observations per EPA, extensive multivariable modelling was not the primary analytical strategy, but the consistency of findings across all modules strengthens confidence in the intervention effect. Future studies with larger samples should incorporate multivariable approaches to address time-varying confounding. A digitally delivered, competency-based pre-internship curriculum can achieve substantial improvements in knowledge and clinical skills, even in crisis-affected and resource-constrained settings. The CVMUI model demonstrates that high-quality CBME can be delivered virtually, with strong psychometric performance, high learner satisfaction, and clear gains across multiple clinical domains. By providing an equitable, scalable, and WFME-aligned educational pathway for learners excluded from traditional institutions, particularly women in Afghanistan, this model offers a viable and urgently needed strategy for strengthening the global health workforce. Contributors KQ (Karim Qayumi) conceptualised the study, designed the curriculum, and acquired funding. FO (Farid Omar) and MN (Masood Noora) supervised programme implementation and data collection. STMH (Seyedeh Toktam Masoumian Hosseini) and MMH (Mohsen Masoumian Hosseini) performed the statistical analysis and qualitative data analysis. AN (Asmat Naebkhilh), SA (Sohaila Aziz), PM (Parwin Mansurih), SWa (Somaya Waqef), NO (Nasrin Oryakhil), AW (Anoosha Wahidi), HK (Hasiba Karimi), AK (Anmol Kapoor), SF (Sayed Fekrat), MN (Manizha Noorandish), FHT (Farida Homa Tanin), ZR (Zahra Rostami), HE (Hamida Ebadi), HE (Hamasa Ebadi), RS (Roohullah Shabon), YM (Yaar Mohammed), SA (Shafiqa Allahyar), AS (Aurelia Soltan), MH (Mariam Hazhir), SWi (Stefan Wisbauer), MW (Mohibullah Waziri), and AF (Alberto Ferrereso) contributed to teaching, assessment, and student mentorship. Authors with identical initials: SA (Sohaila Aziz and Shafiqa Allahyar), MN (Masood Noora and Manizha Noorandish), HE (Hamida Ebadi and Hamasa Ebadi). KQ and FO accessed and verified the underlying data. All authors read and approved the final version of the manuscript. The corresponding author (KQ) had final responsibility for the decision to submit for publication. Data sharing statement Aggregate data are presented in the Appendix. De-identified individual participant data are available from the corresponding author upon reasonable request. Declaration of interests KQ (Karim Qayumi) is Founder and President of CanHealth International and Founder and Chief Executive Officer of Interactive Health International (IHI), the organization that developed and operates the CyberPatient platform used in this study. CVMUI (Canadian Virtual Medical University Initiative) is an educational initiative of CanHealth International. KQ and contributors affiliated with CanHealth International participated in this project in a voluntary capacity and did not receive salaries or financial compensation related to this study. STMH (Seyedeh Toktam Masoumian Hosseini), MMH (Mohsen Masoumian Hosseini), NO (Nasrin Oryakhil), FO (Farid Omar), MN (Masood Noora), and AN (Asmat Naebkhilh) are affiliated with CanHealth International and served as members of the independent Research Committee for this project. SWi (Stefan Wisbauer) is Co-Chief Executive Officer of Lecturio. Lecturio provided access to its digital learning platform free of charge for educational purposes within the study. CyberPatient and Lecturio were provided free of charge to participating students and instructors for the purpose of this educational initiative. Neither IHI nor Lecturio had any role in the collection, statistical analysis, or interpretation of study data. To support research independence, KQ and SWi were not members of the independent Research Committee and were not involved in the statistical analysis or independent interpretation of outcome data, which were conducted by the study Research Committee. All other authors declare no competing interests.