Precision Education: Adjusting Learning Through Data

Adjusting Educational Experiences Based on Data

Precision education adjusts educational experiences using data on the learner’s current performance, prior experience, learning pace, and goals. In medical education, this approach defines common competencies while varying the time, methods, and pathways through which learners attain them and develop their areas of interest. Assessment data, learning logs, and AI support these adjustments. Personalized learning requires distinguishing among adjustments to pace, method, and goals, while examining the conditions under which instructors and institutions implement them.

Designing Precision Education to Connect Initial and Formative Assessments

The concept of adjusting education to the learner’s starting point and subsequent progress stems from the expansion of medical knowledge and the changing role of clinicians (Coates, 2025). The cited doubling time of medical knowledge has shortened from 50 years in 1950 to approximately 73 days in 2021. This underscores the clinician’s role in selecting and managing knowledge. The editorial maps the diagnosis, prediction, treatment, and prevention of precision medicine onto education. This approach establishes the learner’s starting point through test scores, GPA, prior education, and behavioral observation. It predicts support needs using online learning logs and performance data. The core of precision education lies in assigning educational experiences based on the learner’s current performance and needs, then revising these assignments using subsequent assessment results.

Specific plans incorporate the needs of learners, instructors, and institutions alongside the learner’s self-assessment, future goals, preferred learning styles, and personal background. For learners who have already demonstrated competence, relevant completion requirements can be waived or modified. The timing of learning, amount of repetition, and degree of support from materials and instructors can vary. The design calls for communicating expected levels of performance and measurable outcomes to learners. Strategies and materials are adjusted through immediate and periodic formative assessment. Dashboards and electronic health records (EHR) provide data for verifying progress alongside human feedback. AI supports monitoring and adjustment. Implementation involves collaboration among multidisciplinary teams and IT professionals, privacy protection, and periodic reviews for accuracy and bias.

The learner profile underlying this design aligns with the Master Adaptive Learner (MAL) framework of the American Medical Association’s Accelerating Change Initiative. MAL is a metacognitive model for self-regulating knowledge acquisition, decision-making, and the development of adaptive expertise. It positions curiosity, self-motivation, and a growth-oriented attitude as the foundation of learning. While AI reduces the time spent finding necessary materials, the learner selects information sources and regulates learning. The coach supports efforts to address difficulties. The editorial suggests applying CBME—which aims for predetermined outcomes—alongside MAL. Implementation includes training leaders and instructors in AI, technology, and coaching, employing internal IT personnel, and iteratively improving small-scale AIEd programs. The approach compares the conditions under which AI-based precision education is useful with those in which traditional education is more reliable. Its long-term aim is to extend learning strategies developed during training into lifelong learning and maintenance of certification activities.

Learning Plans that Co-design Common Competencies and Individual Career Paths

Personalization in CBME supports both the achievement of common competencies and the development of individual potential (An, 2025). Synthesizing four papers from a special issue on personalization in medical education, this editorial describes personalization as a process that deepens the core philosophy of CBME. It argues that education can be structured around the learner’s pace, interests, and career path while maintaining the essential competencies required to meet national and community healthcare needs. The editorial defines individualized learning as adjusting the pace of progress toward a common goal, and differentiated learning as adjusting the method to learners’ characteristics and preferences. Both approaches are instructor-led and based on Mastery Learning. Personalized learning accommodates diversity in the learning goals themselves.

The internal medicine residency program at VinUniversity in Vietnam addresses both competency gaps and career development through an individualized learning plan (ILP). The common goal is to achieve the core competencies of the Accreditation Council for Graduate Medical Education-International (ACGME-I). The Clinical Competency Committee (CCC) reviews performance through multisource feedback. VinUniversity conducts comprehensive assessments every 6 months and uses the results to establish each resident’s individual learning goals and strategies. If oral case presentations fall short of common milestones, intensive mentoring or additional practicums are planned. If a resident seeks a cardiology fellowship, the plan can include leading a journal club or pursuing advanced learning in related skills. At Seoul National University College of Medicine, the Curriculum Path Model (CPM) allows students to pursue study equivalent to a minor in management, mathematics and statistics, or literature and philosophy. The global health curriculum combines required and elective courses. The continuous research program connects research experiences from the pre-medical years through the fourth year of medical school.

The editorial suggests that such assessment practices and pathway designs could evolve into precision medical education (PME). This approach analyzes learning patterns, assessment results, and clinical performance data to adjust support. It brings together learning experiences from off-campus platforms in portfolios for evaluation. At VinUniversity, faculty had limited time for ILP follow-ups. The proposed response involves faculty development in collaboration with the University of Pennsylvania and training graduates as future faculty. At Seoul National University College of Medicine, participating faculty receive substantial teaching credits and funding to run courses. Workloads are distributed through shared lectures, team teaching, and the assignment of dedicated staff.

Practical Use of an AI Teaching Assistant Based on Course Materials

A mixed-methods study of an AI teaching assistant at Dartmouth’s medical school combined anonymous chat logs and end-of-course surveys in a second-year Neuroscience & Neurology course (Thesen & Park, 2025). The cohorts comprised 92 students from October 2023 to January 2024 and 98 students from October 2024 to January 2025. Students used the tool voluntarily during their respective 14-week courses. Through getcody.ai, the first cohort received access to GPT-4 and the second to GPT-4o. The knowledge base consisted of 145 English documents, including lecture slides, textbook excerpts, pre-learning materials, instructor handouts, and syllabi. The research team applied a RAG approach, dividing the documents into units of approximately 200–300 words and retrieving materials and sources relevant to each question to generate answers. Responses were restricted to the selected documents. Students had continuous access through LMS and chat links. The analysis drew on the Technology Acceptance Model (TAM) to examine perceived usefulness and ease of use, and on cognitive load theory to anticipate increased use before exams.

Across the two academic years, 360 conversations and 2,946 messages were recorded, with a median conversation length of 2 turns. From the first academic year to the second, conversations decreased from 256 to 104, and messages from 1,922 to 1,024. Across both cohorts, the daily average number of conversations during the 3 days before an exam was 5.11, compared with 1.14 during regular periods; the pooled comparison yielded t(252) = 3.95, p < 0.001, Cohen’s d = 1.74. When the cohorts were analyzed separately, the difference in the second cohort was not statistically significant (p = 0.1073). The increase in pre-exam use was larger in the first cohort. Questions focused on neuroanatomy and physiology (65.9%) and clinical syndromes and diseases (53.7%), followed by clinical case discussions (14.3%) and imaging and diagnostic techniques (7.0%). These percentages reflect the assignment of multiple topics to a single conversation. Many questions sought to verify learned concepts such as the function of the cerebellar vermis or Brown-Sequard syndrome. Thematic analysis combined GPT-4o and human review. Agreement between human coding (reviewing 15% of conversations) and LLM coding was 78.4%, with a weighted Cohen’s κ of 0.64.

Among end-of-course survey respondents, 39.3% (22/56) in the first cohort and 26.4% (23/87) in the second had used the tool at least once. Usefulness was rated at 2.8 (SD = 1.4) and 3.3 (SD = 1.0) out of 5, respectively. Students valued the ability to ask questions immediately and the reduction in time spent searching through multiple slides. The use of official materials also contributed to trust. Students reported that responses fell short of expectations when the tool was asked to create study guides from specific slides or construct comparison and summary tables. The study drew on logs of voluntary use and surveys without tracking individual use. The results describe usage patterns in relation to the course schedule and students’ perceptions of usefulness. The research team proposed introducing the tool in early courses when learning habits are formed, providing links to source locations for verification, and offering training in its use. It also discussed combining direct answers during exam preparation with Socratic tutoring during regular periods.

Distinguishing the Granularity of Records from the Precision of Educational Systems

An article examining the meaning of precision draws on existing definitions and experience in program development and educational leadership to examine the targets and levels at which precision can be pursued (Patocka et al., 2025). The authors’ institution has prioritized precision in medical research, education, and technology initiatives since 2015. A 2016 program development framework outlined 7 domains of precision education. These domains include individualized learning opportunities and pathways, teaching and assessment methods, data/analytics/evaluation, admissions/credentialing/hiring, learning environment, systems, and social accountability and workforce needs. The article extends the meaning of precision beyond accuracy as proximity to the true value and consistency across repeated measurements to include fidelity, exactitude, clarity, and meticulousness. Because the meaning of precision changes with its target, the article proposes pattern theory. This accommodates similar definitions and their variations within a common framework rather than fixing a single definition.

In assessment, observing and recording performance requires different levels of precision from interpreting records and making progression decisions. The article divides the process of translating performance into assessment into 6 stages, examining how improvements at one stage translate into subsequent judgments. In an account drawn from the authors’ experience as residency program directors, the introduction of CBME replaced a single summative evaluation form with over 28 EPAs. The account indicates that competency records became more precise, while feedback became less specific and actionable. Another example concerns concentrating resources on increasing the inter-rater reliability of interview scores. This may reduce the capacity to examine the relationship between admission criteria and long-term health workforce needs.

Deciding where and to what extent to increase precision requires considering resource constraints, sustainability, and effects on other educational responsibilities together. The article considers CBME a form of precision education because it adjusts education to each learner’s pace. The time and tools needed to track and measure progress also affect care delivery and physician workforce planning. High-quality data can identify gaps in learning opportunities and support diverse needs. Standardization methods, resource allocation, and data bias shape who receives that support and how it is provided.

Reading notes

The AI teaching assistant study did not systematically evaluate response accuracy, limiting interpretation of RAG’s role. The account of changes in records and feedback following the introduction of EPAs draws on program operations and requires interpretation within that context. In research connecting HRD and medical education, the activities learners select after assessment, subsequent instructor feedback, and the time and compensation structures that make them possible can be tracked together.

References

• An, S. (2025). Personalizing medical education in the era of competency-based medical education. Korean Medical Education Review, 27(3), 191–193. https://doi.org/10.17496/kmer.25.035
• Coates, W. C. (2025). Precision education – a call to action to transform medical education. International Journal of Emergency Medicine, 18(1), Article 21. https://doi.org/10.1186/s12245-025-00819-1
• Patocka, C., Harper, L., & Ellaway, R. H. (2025). Focusing on “Precision” in Precision Education. Academic Medicine, 100(12), 1421–1424. https://doi.org/10.1097/ACM.0000000000006293
• Thesen, T., & Park, S. H. (2025). A generative AI teaching assistant for personalized learning in medical education. npj Digital Medicine, 8(1), 627. https://doi.org/10.1038/s41746-025-02022-1