HMS Theses and Dissertations

The rapid integration of artificial intelligence (AI) in healthcare has outpaced formal training in nursing and medical education. This study explores the perceptions of Master of Nursing students toward AI in clinical learning, and evaluates the use of an AI-enabled Virtual Integrated Patient (VIP) platform in developing clinical reasoning and knowledge.

Part 1 of this mixed-methods study involved surveys and interviews to examine students' views. Quantitative findings showed strong support for AI in education: most students believed AI could enhance learning and should be included in the curriculum. However, concerns were also prominent, with students citing risks such as over-reliance, data security, and the potential loss of humanistic care. Qualitative data echoed these findings—students valued AI for improving efficiency, supporting knowledge acquisition, and enhancing simulation training. Still, they expressed greater trust in human clinicians and stressed the need to build institutional trust in both AI systems and healthcare professionals.

Part 2 focuses on the VIP platform. Students appreciated its adaptability, ability to organise information, and support for structured, case-based learning. The platform was seen as a helpful educational tool, with some students using it to guide clinical reasoning. However, critiques included the repetitive nature of responses, lack of intuitive interaction, and insufficient narrative nuance, leading some to prefer more traditional learning approaches. Analysis of VIP usage showed a weak correlation with final exam scores, but this relationship strengthened among students who engaged with the platform for more than 10 minutes. Multi-regression analysis highlighted variability in learning outcomes, indicating that while the VIP platform may support clinical learning, its effectiveness differs across learners.

This study reveals both enthusiasm and caution among postgraduate nursing students about AI in education. While AI-powered tools like VIP offer innovative ways to promote self-directed, clinically relevant learning, they should be integrated thoughtfully alongside traditional pedagogies. As AI becomes increasingly embedded in healthcare, cultivating both technological fluency and critical thinking is vital to prepare learners for responsible, autonomous clinical practice in a digitally augmented future.

Continuing professional development (CPD) is an intricate form of medical education intended to foster ongoing growth in medical practice. It emphasizes lifelong learning and a holistic approach to professional development, aligning with the evolving needs of patients, healthcare systems, and society. In Chile’s public health sector, the Chilean General Practitioner Training Program (EDF Program), established in 1955, provides the national health workforce with 2,500 general practitioners (GPs) working in rural and vulnerable areas at the primary healthcare level. All are mandated by law to receive CPD, yet planning, implementation, and evaluation deficits have been identified. This research aimed to assess the CPD needs of the EDF Program in the Valparaiso Health Care District, incorporating perspectives from intersectoral stakeholders, EDF GPs, and healthcare network users. A qualitative descriptive approach was selected as the study design. An indicative and purposefully stratified sample was constructed, based on three participant groups from Valparaiso’s local healthcare district. Semi-structured, in-depth interviews were conducted. A data-driven codebook was piloted, refined, and then applied to all transcriptions. Axial coding and constant comparative analysis were systematically conducted using MAXQDA 2024 software. Interviews continued until saturation was achieved, requiring a total of 22 interviews. The study identified CPD benefits, enablers, barriers, and areas for improvement amid ongoing challenges and gaps. Key findings revealed a structural and functional mismatch between the CPD program for EDF physicians and the actual needs of the primary healthcare system they serve. This mismatch, both a technical shortcoming and policy opportunity, has led to calls for a shift in the program’s framework. As a result, several pathways to realign the EDF CPD training with the realities of Chilean primary healthcare have emerged from this research.
This research highlights the inherent complexity of a formal CPD system, as it comprises vast and diverse interrelated components influencing training provision and health workforce development outcomes. Intersectoral collaboration appears vital for advancing the development of practitioners' capacities for primary health care. Future directions include developing a tailored needs assessment tool, which could improve readiness for health workforce CPD innovations.

While the Liaison Committee on Medical Education emphasizes the teaching of cultural competence in medical education, the concept of cultural humility, focusing on self-reflection and lifelong learning, has been proposed as a more effective approach. Although there have been numerous discussions on both cultural competence and cultural humility, understanding how faculty in clinical settings help students develop cultural humility skills remains limited. This multimethod study aims to identify current strategies in order to inform future curriculum development.

We administered a 10-question survey to assess faculty demographics, teaching characteristics, and familiarity with cultural humility. Participants who were at least somewhat familiar with cultural humility were invited for semi-structured interviews. Survey data were analyzed using descriptive statistics and logistic regression, and interview data were thematically analyzed to identify key teaching strategies.

In our study of 49 medical faculty members, the majority of participants were female (61%) and predominantly White (67%), covering a wide range of specialties and years of teaching. Of the participants, 74% expressed being at least somewhat familiar with cultural humility, and 10 consented to the interview. Strategies for cultural humility education included one-on-one instruction, feedback, and reflections. Multivariate logistic regression showed no significant predictive relationship between demographic or professional variables and familiarity with cultural humility, except for the number of years teaching medical students (OR: 1.21, 95% CI: 0.06-2.36, p=0.03). Thematic analysis underscored fostering learner curiosity about cultures, early patient exposure, and incorporating diverse learning perspectives as essential in developing students’ cultural humility skills.

Cultural humility is an important attribute for healthcare professionals that can enhance patient-centered care. Through focused interviews with faculty in our study and subsequent thematic analysis, our results suggest the need for longitudinal and multimodal educational strategies to cultivate cultural humility among medical students. By understanding current teaching methods, educators can design and assess more effective curricula to prepare future doctors for a culturally diverse patient population.

The acquisition of procedural skills in medical training remains an ongoing challenge. Within radiology, where procedural mastery is a key milestone, virtual reality (VR) presents a potential solution. This study evaluated the effectiveness of a novel, VR CT-guided liver biopsy module in improving residents’ knowledge and confidence. A quasi-experimental pre-post study was conducted at Massachusetts General Brigham, Brigham and Women’s Hospital, and Duke University Hospital. Residents were divided into a control group (PGY-1) and an intervention group (PGY-2 and PGY-3). Both groups completed pre- and post-surveys and a final-survey one month later. The intervention group completed the VR module between the pre- and post-surveys and then before the final-survey. Surveys assessed: (1) knowledge and cognitive skills, (2) confidence, and (3) learning experience. Exam scores from the knowledge section were analyzed with a Wilcoxon signed-rank test as was confidence measured using Likert-scale questions. VR headset data recorded performance metrics. The intervention group demonstrated a significant increase in knowledge from pre- to post-survey (p = 0.03) with no significant change at follow-up (p = 0.09). Confidence in ordering procedural steps and performing the procedure increased significantly from pre- to post-survey (p = 0.03 and p = 0.02 respectfully) and from pre- to final-survey (p = 0.01 and p = 0.01). Confidence for identifying instruments did not significantly improve from pre- to post- (p = 0.09) or from pre- to final- (p = 0.01). VR headset data showed a significant increase in time to complete the module one month later (p = 0.03), suggesting a decline in VR-specific technical skills, though the number of scans to place the coaxial needle and needle placement accuracy remained stable, indicating retention of procedural knowledge. The VR module effectively improved residents’ knowledge and confidence with CT-guided liver biopsies, demonstrating its value as a teaching tool. Despite usability challenges with the headset and controllers, participants found the VR module valuable, and the majority reported they learned something new. This work highlights the clear value of integrating VR into the training of interventional radiology residents. Future studies can correlate VR module usage with clinical performance and patient outcomes.

Background: Resuscitation skills are fundamental to the practice of Emergency Medicine, yet there is a lack of structured, formalized hands-on learning sessions for emergency medicine trainees. To address this need, we developed and implemented a novel, multimodal advanced resuscitation curriculum for PGY-2 Emergency Medicine residents. This study aims to assess the impact of this dedicated advanced resuscitation curriculum on residents’ medical knowledge, clinical skills, and attitudes. Methods: This mixed-methods study used explanatory sequential design. Quantitative data were first collected through resident pre- and post- curriculum self-assessment surveys, knowledge assessments, and faculty evaluations of residents’ clinical performance. Following quantitative data analysis, qualitative semi-structured interviews were conducted with participants to explore their experiences with the curriculum and its role in shaping their learning, clinical skills, and attitudes. Interview transcripts were analyzed thematically using a deductive-inductive hybrid coding approach. Results: Resident surveys indicated highly positive reactions to the curriculum (Kirkpatrick 1) and statistically significant improvements in confidence across key domains, including resuscitation leadership, airway management, and procedural skills (Kirkpatrick 2). Additionally, residents’ post-course assessment scores improved significantly compared to pre-course assessment scores, and follow-up assessment scores outperformed the national average (Kirkpatrick 2). Faculty surveys reflected observable on-shift behavior change, with residents demonstrating improved resuscitation leadership, procedural skills, and application of evidence-based medicine (Kirkpatrick 3). Qualitative analysis provided deeper insights as to how the curriculum was impactful. Residents emphasized the effectiveness of the curriculum’s individualized structure and alignment with adult learning principles. They also noted that the curriculum improved their confidence in resuscitation leadership, clinical decision-making, and ability to apply evidence-based medicine to practice. Participants also expressed a stronger appreciation for reflective practice and reported more receptiveness to feedback after taking this course. Conclusion: Implementation of a structured resuscitation curriculum significantly improved EM residents’ confidence, knowledge, and clinical decision-making skills. The mixed-methods approach provided a comprehensive evaluation of curriculum impact, suggesting that structured resuscitation training grounded in adult learning principles is effective strategy to prepare EM residents for high-stakes clinical care.

Background: Few evidence-based tools exist to holistically assess applicants during the admissions process for pre-doctoral dental school. Prior studies focus more on the interview process itself rather than the application that the candidate submits. Without validated assessment tools, institutions risk selecting applicants that may not fit their mission or values. This study aimed to assess the validity of a Holistic Admissions Assessment Rubric (HAAR) created for Kansas City University College of Dental Medicine for the 2024-2025 application cycle.

Methods: Messick’s universal validation framework was used to assess the validity of the HAAR in the five domains of content, response process, internal structure, relationship to other variables, and consequence. A nominal group process was first held with admissions content experts to solidify the final version of the HAAR for application cycle use. The HAAR was used to assess 100 applicants during the 2024-2025 pre-doctoral dental admissions cycle at Kansas City University College of Dental Medicine, and data were collected for each item score as well as overall score between two separate raters. Admissions committee members were blind to the HAAR results when making decisions of admission.

Results: Validity evidence was collected in all five validity domains. Inter-rater reliability with two-raters was excellent with ICC= 0.85 (ICC 95% confidence interval (CI) 0.70-0.90) for overall HAAR score. Internal consistency was high with item discrimination scores ranging from 0.14-0.67, and inter-item correlation scores ranging from 0.05-0.08. Logistic regression showed that a higher HAAR score was significantly associated with a greater likelihood of admission (OR=1.28, p.001).

Conclusions: These data results indicate that the HAAR has a high predictive value for admissions acceptance and there is high inter-rater reliability throughout the 16-criteria in the rubric. This initial validity evidence in Messick’s five validity domains strongly supports the use of the HAAR in the pre-doctoral dental admissions process.

The increasing demands of competency-based medical education have amplified the need for scalable, timely, and pedagogically sound formative assessment tools. This thesis explores the design, implementation, and evaluation of the Automated Learning Assessment System (ALAS), a generative AI-based platform developed to automate grading of short-answer responses in Harvard Medical School’s Case-Based Collaborative Learning (CBCL) curriculum. ALAS aims to reduce grading burden for faculty while improving feedback quality and enabling data-informed instruction.

This study evaluated ALAS’s grading reliability, usability, and impact on instructional decision-making. Two datasets informed the quantitative analysis. The Multi-Grade Dataset (n=822) included student responses independently graded by ten teaching assistants (TAs), allowing comparison of ALAS scores to a human consensus standard. The Single-Grade Dataset (n=8,964) included responses each graded by one TA, allowing direct comparison of ALAS scores to individual human-assigned grades. Inter-rater agreement was measured using Spearman correlation and Fleiss’ Kappa. A faculty survey (N=6) explored ALAS’s usability, feedback utility, and integration into teaching workflows.

ALAS demonstrated strong alignment with collective human judgment, achieving higher correlation with average human scores (Spearman ρ=0.678) than human raters did with each other (ρ=0.422). Faculty found the system moderately to highly usable and appreciated its ability to surface both known and previously unrecognized student learning gaps. Visual dashboards and performance summaries were highlighted as useful tools for refining instruction. Despite a small survey sample and technical limitations, ALAS was noted for reducing grading workload, mitigating cognitive load, and enhancing formative assessment responsiveness.

ALAS represents a novel and effective application of generative AI in medical education. It supports formative assessment theory by providing timely, individualized feedback and aligns with cognitive load theory by offloading repetitive grading tasks. Future goals include Canvas integration, real-time student feedback, interactive faculty dashboards, and broader adoption supported by AI literacy training. ALAS offers a promising model for ethical and scalable AI integration in health professions education.

Background: Despite being one of the most frequently ordered laboratory tests in clinical practice, the complete blood count (CBC) is underutilized by most physicians, particularly regarding graphical outputs, flag indicators, and many overlooked indices. Most doctors focus solely on basic parameters such as hemoglobin, RBC, WBC, and platelet counts, overlooking valuable diagnostic data embedded in other indices and graphical representations. Furthermore, many laboratories omit CBC graphs in routine reports, contributing to a limited understanding and usage of these features. Objective: This study developed and tested Visual CBC, an online educational module aimed at improving physicians’ proficiency in interpreting the complete spectrum of CBC data including graphical representations and underutilized indices—to enhance diagnostic accuracy and clinical decision-making Methods: A multi-centric, experimental study was conducted in teaching hospitals across Algeria. Initially, a needs assessment survey was distributed to 488 primary care physicians and interns to assess current practices in CBC interpretation. Subsequently, 63 participants completed a pre-test, underwent the Visual CBC module, and took a post-test to evaluate knowledge gains. Results: The needs assessment revealed that approximately 30% of respondents do not receive CBC graphs in their laboratory reports. Among those who do, nearly 60% reported never interpreting them, while only 3.2% consistently use these graphical elements in daily clinical practice. After completing the Visual CBC module, participants demonstrated significant improvement (P .0001 Overall z score = −6.90) in interpreting RBC and platelet graphs, as well as often-overlooked indices such as Mean Platelet Volume (MPV) and Red Cell Distribution Width – Standard Deviation (RDW-SD). Conclusion: There exists a global educational gap in the comprehensive interpretation of CBC results among physicians leading to misdiagnosing some cases and ordering unnecessary tests. The Visual CBC module effectively addresses this gap, highlighting the need for integrating graphical data into routine CBC reports and strengthening CBC interpretation training in medical education so it can cover the commonly neglected parts.

This master’s thesis on improving genetic diagnosis revolves around two projects. Project 1: Participant Perspectives on Rare Disease Genomic Research Abstract Introduction: Although advances in genomic sequencing technology have improved diagnostic capabilities, access to these technologies is variable. Furthermore, novel approaches to genomic analysis may be outside of the scope of usual clinical testing. Rare disease genomic research thus represents a potential pathway to genetic diagnosis. However, factors influencing the decision to participate in a genomic research study remain incompletely understood. We therefore investigated motivators for participation in a rare disease genomic research study and the impact of diagnostic findings. Methods: Prospective quantitative analysis of survey data from participants in the Rare Genomes Project was conducted. Surveys were sent upon enrollment to the study and at 3 and 12 months after result disclosure and included both novel items and the Perceived Stress Scale. Results: 509 participants responded out of 884 targeted surveys, either affected persons (226) or their caregivers (283). For all respondents, diagnostic importance, improved clinical understanding, future preparedness, and treatment optimization were the strongest motivators for seeking a genetic diagnosis. Perceived stress upon entry to the study was significantly higher than population norms (17.61 vs 13, P 0.00001). Variables associated with perceived diagnostic importance are mainly related to the financial burden of health conditions (OR 3.859 (95% CI 1.306–11.404), P = 0.015). At follow-up timepoints, diagnosed participants endorsed statements related to a positive impact of the diagnosis on group connections. Perceived benefit from social connectedness became more prominent over time. Conclusion: We described the psychosocial context of research enrollment for a diverse cohort of rare disease genomic research participants and identified financial problems as the prominent determinant for diagnostic importance perception. We also highlighted a potential positive impact of the genetic diagnosis in reducing the perceived stress and shifting feelings towards genetic diagnosis from vigilance to acceptance.

Project 2: Analysis of Splicing Coding and Noncoding Genes and Pseudogenes Reveals Novel Gene-Disease Relationships Abstract Splicing is a complex molecular mechanism occurring in each living human cell where the precursor messenger RNA (pre-mRNA) is processed into the mature messenger RNA (mRNA). It involves more than 300 protein-coding genes (PCGs) and around 43 small nuclear RNA (snRNA) genes. However, fewer than 30 diseases have been described as spliceosomopathies to date. This discrepancy points towards splicing machinery as an underexplored area for genetic discoveries. Furthermore, there are almost 1700 spliceosome-related snRNAs pseudogenes that are overlooked in diagnostic analyses, but recently a novel gene-disease relationship has led to reclassification of a snRNA pseudogene to a gene associated with a developmental and epileptic encephalopathy, highlighting the importance of further evaluation of snRNA pseudogenes. We evaluated both coding and noncoding snRNA genes and their pseudogenes using adapted approaches to each gene category. Analysis strategies involved evaluating heterozygous variants in PCGs with loss of function constraint (pLI > 0.9) and in snRNAs in variant depleted regions in gnomAD, with biallelic variants reviewed throughout both gene sets. For snRNA pseudogenes, we prioritized candidates having similar epigenomic, genomic, and hypermutability features to functional snRNA genes. These signals, located upstream, downstream, and within a gene, indicate the likelihood of it being a functional gene. For around 23,000 families with rare disease sequenced through the GREGoR consortium, we identified 30 variants of interest across 9 PCGs with established gene-disease relationships (GDRs) and 13 genes not yet associated with disease, including one pseudogene. Of note, one gene, CWC25, had candidate variants with dominant and recessive inheritance patterns. For snRNAs genes, we identified 55 variants of interest located in seven established GDR and 11 genes not yet associated with disease, including two pseudogenes, prioritized by one or two of the prioritization strategies. Our results expanded the phenotypic expression of one variant located in the three-way junction in RNU4-2 gene from retinal dystrophy (RD) to RD and Neurodevelopmental disorder (NDD). We also suggested pleiotropy expansion for RNU6 variants with NDD and oculo-muscular fibrosis. Finally, we identified biallelic variants in RNU5 family, suggesting inheritance pattern extension. This study highlighted the importance of splicing-related PCG and snRNA in rare disease diagnosis. The genes prioritized through this research represent selected candidates for future confirmatory studies.

Chimeric antigen receptor (CAR) T-cell therapy has transformed the treatment of hematologic malignancies, but its application remains limited by manufacturing constraints, toxicity, and antigen escape. CAR-engineered natural killer (CAR-NK) cells represent a promising alternative due to their favorable safety profile, reduced risk of graft-versus-host disease, and capability for off-the-shelf use. However, a main limitation of CAR-NK therapy is poor in vivo persistence, largely due to dependence on exogenous cytokine support. This study addresses this challenge by developing a targeted cytokine delivery strategy, the CAR-Enhancer (CAR-E) platform on NK cells. CAR-Es are bifunctional molecules composed of a CAR-targeting antigen domain fused to a low-affinity mutant cytokine, enabling selective delivery of cytokine signaling exclusively to CAR-expressing NK cells. This design concentrates survival and activation signals at the CAR interface while minimizing systemic toxicity and off-target immune activation. First, we demonstrate that IL-2–based CAR-E constructs selectively bind CAR-NK cells and induce potent STAT5 signaling in a CAR-dependent manner, while sparing non-transduced immune cells. This targeted signaling results in preferential expansion and functional activation of CAR-NK cells. Functionally, CAR-E enhances CAR-NK cytotoxicity against tumor targets without impairing CAR-antigen engagement. Notably, CAR-E also boosts NK-intrinsic cytotoxic pathways, enabling enhanced tumor killing even in the absence of canonical CAR signaling. This dual mechanism demonstrates a broader immunomodulatory role for CAR-E, beyond traditional CAR-mediated targeting. In vivo, exogenous CAR-E administration promotes robust CAR-NK cell expansion, multi-organ persistence, and complete tumor control in a xenograft model of multiple myeloma. Persisting CAR-NK cells retain an activated, non-exhausted phenotype and preserved cytotoxic function, demonstrating durable functional competence. To further improve translational feasibility, we engineered armored CAR-NK cells that autonomously secrete CAR-E via a bicistronic T2A construct. This self-secretion system increases cytotoxicity, metabolic fitness, and resistance to exhaustion under chronic tumor exposure. In vivo, armored CAR-NK cells demonstrate prolonged persistence and improved tumor control even when tumor challenge is delayed, showing their potential for long-term immunosurveillance. This work confirms CAR-E as a novel and versatile strategy to enhance CAR-NK cell persistence and functionality. By enabling targeted cytokine delivery without systemic toxicity, the CAR-E platform addresses a central limitation of CAR-NK therapy and provides a broadly applicable framework for improving adoptive cell therapies in cancer immunotherapy.