The topics outlined in detail in Table 1 and summarized in Figure 1 would provide a strong foundation for a stand-alone course on data science in health care for medical students, but integration, reinforcement, and application of the content throughout the curriculum will be essential to achieve meaningful learning outcomes. Here we outline suggestions and considerations for how each of the major topic areas in Table 1 may fit well into a focused anchor course positioned early in the Undergraduate Medical Education (UME) curriculum, paired with deliberate longitudinal integration across a medical education program. The content proposed has been derived through discussion among the subject matter experts in this paper along with a review of the Clinical Informatics Accreditation Council for Graduate Medical Education subspecialty curricula to extract topics relevant to health data science.

Early stages of a preclerkship curriculum represent an ideal opportunity to introduce fundamental concepts in data science, such as its definition, the important role it plays in the evolving landscape of health care delivery, and an overview of the types of health data and data quality. Delivery of this content can take place through a combination of an anchor course on the subject, along with brief required web-based modules created in collaboration with subject matter experts in health data science. These brief web-based modules may integrate well into preclerkship courses covering epidemiology, public health, and health system science.

Many medical schools incorporate longitudinal clinical experiences early in the preclerkship curriculum [10]. As students prepare to interact with clinical data in the electronic health record (EHR), reinforcement and further exploration of themes around data quality and data coding can be discussed. For example, a discussion on structured data (such as vital signs documented in discrete fields) versus unstructured data (such as free-text notes) may ideally take place when students are first being introduced to the use of EHRs in primary care, particularly in relation to their role in the management of chronic diseases such as type 2 diabetes. The integration of AI into health care and the increasing ability of LLMs to transform free text into structured data can be introduced. The importance of having structured data to observe trends necessary in making clinical decisions (such as looking at the trend of a patient’s blood sugar levels) or leveraging risk calculators (such as those for cardiovascular risk in patients with diabetes) should be highlighted.

Learning about major types of databases and data schemas used in health records may fit best in an anchor course dedicated to clinical informatics or data science in health care. Such a focused course would allow students to explore key types of health records (such as EHRs and personal health records); the role they play in generating, storing, and analyzing health data; and the respective benefits and limitations of choosing local versus cloud storage of data.

Clerkship curricula can reinforce concepts in data sources deliberately through exploring case examples in small group discussions. For example, during a clinical rotation where students care for patients with rheumatologic conditions that require the tracking of a patient’s function and pain scores over time, the role of patient-reported outcomes through previsit questionnaires and data from consumer medical devices (such as step counters and sleep trackers) can be explored. As web-based care usage increases and medical device innovations evolve to provide novel types of patient-generated health data, medical trainees should be taught an approach to evaluate these novel data sources for appropriateness in integrating into clinical decision-making. Similarly, the expanding availability of “omics” data, which refers to comprehensive data sets generated from the analysis of different molecular aspects of biological systems such as genes, proteins, and metabolites, is important to discuss due both to its potential implications in advancing precision medicine and to its present practical and ethical limitations [11]. The abovementioned example of a patient with a rheumatologic condition could be a starting point for discussion around targeted biologic treatments based on genomic and proteomic screening, including AI-generated personalized predictive analytics.

Understanding the range of sources of health data is of particular importance in considering research questions and study design, with potential data sources including administrative data, billing and claims data, and population health data. The COVID-19 pandemic has resulted in the emergence of a number of such population health databases, where data are tracked on the administration of vaccines, viral testing, and contact-tracing, in many cases entirely outside of the patient health record [12]. The strengths and limitations of the various data sources are critical considerations, and teaching about data sources in the context of a research methods course represents an opportunity for integration of data science themes.

Some traditional analytical methodologies, such as regression methods, may already be covered in existing components of the curriculum such as epidemiology or research methods [13,14]. An introduction to novel analytic tools and methodologies could be integrated into such existing courses or could be included in a data science anchor course.

Didactic sessions can explore how data science methodologies have evolved over time and what new capabilities are made possible through the use of advanced tools such as machine learning and neural networks, particularly in handling very large data sets to help produce insights tailored toward the needs of an individual patient (ie, precision medicine). It will be valuable for students to learn about the major machine learning models that exist (eg, unsupervised learning models, supervised learning models, and reinforcement learning models), along with their limitations, such as of a lack of explainability of results that is prevalent among such tools [15]. Students can consider specific case examples in an interactive, small group format, exploring benefits, barriers, and complexities that arise with implementing such rapidly evolving and sophisticated methods. An example for discussion would be the use of machine learning algorithms, such as general adversarial networks, to analyze and augment large data sets for the purposes of improving data quality and ensuring representation of an adequately broad spectrum of patient populations [16]. This improves the ability to build downstream applications that can better augment the ability of human radiologists through automated triaging, segmentation, and diagnosis of imaging modalities such as computed tomography and magnetic resonance imaging.

Students typically receive practical orientation to the particular EHR in use at core clinical clerkship sites [17], but these introductions could be made more robust with the exploration of data analytic tools embedded in the EHR, or tools being planned for near-term development. Clerkship directors can engage local subject matter experts in clinical informatics and data science to develop and implement interactive modules to learn about novel tools that are relevant locally. For example, a medical center may be adopting an application of machine learning in which optical character recognition automatically reads handwritten clinical notes. Students could learn about this new tool and consider its potential benefits (eg, facilitating medical documentation and coding; enhancing the quality of structured information in health records), as well as its inherent limitations.

Clinician educators may themselves lack knowledge on data science and the functioning of the tools available [18]. Learning outcomes around real-world usage may thus best be achieved through collaboration with data and IT professionals present at the clinical sites who could participate in small group sessions and simultaneously help raise awareness for clinician educators. In addition, subject matter experts in IT and data science can contribute to faculty session guide documents for small group work, outlining key teaching points that allow for faculty development of core clinical teachers who can then teach in these new content areas more independently.

An initial introduction to the use of health data would educate students on the functional benefits and uses of health data as applied to clinical care delivery. This introduction to usage would fit well in an anchor course and should be delivered early enough in the preclerkship curriculum to allow for real-world examples to be highlighted within organ systems–based preclerkship courses. For example, when learning about renal function and discussing fluid balance, the crucial role of health data visualization within the electronic medical record (EMR) can be highlighted. Even in the preclerkship setting, educators can demonstrate how visual trending of relevant data parameters (including laboratory values such as serum creatinine and measurements such as blood pressure and weight) facilitates patient care.

Throughout the clinical rotations, as students encounter real-world examples of data visualization and CDS tools in the EMR, educators can ask students to notice and report on examples of important applications, benefits, and limitations of data use. Students should develop proficiency in using the EMR to visualize data that aids in clinical decision-making for their primary patients, starting with basics such as graphical trending of vital signs, laboratory test results, and medication dosing [19]. Students can explore existing CDS tools, recognizing important attributes of the data leveraged by these tools, including generalizability, data shift, and accuracy [20]. During an internal medicine rotation, students could discuss the implementation of an alert that uses machine learning algorithms to predict the risk of intensive care unit (ICU) transfer for an admitted inpatient on a medical ward [21] and the differential outcomes that may take place when data (such as the patient’s blood pressure or body temperature) are not collected accurately or recorded in a timely fashion. Similarly, a recent study revealed how the pandemic resulted in a data shift in the demographics of patients being admitted to the ICU, thereby reducing the accuracy of some sepsis prediction tools and leading to a surge in false positive alarm triggers [22]. Highlighting such differential impacts sets the stage for discussion around the validity of such algorithms in different patient populations and other biases that may impact accuracy of the CDS tool.

Similarly, the emerging role that novel AI tools will play in helping deliver both the administrative and clinical elements of health care is optimally discussed during clinical rotations and primarily through real-world examples, including attention to what may be on the horizon to alleviate present-day challenges; for example, growing evidence to support the ability of LLM-powered chatbots to serve as an interface for patient history taking and responding to common medical questions, providing both high-quality and empathetic responses [23]. It is appropriate to discuss this in the context of the strain on the health care system due to excessive amounts of administrative or nonclinical tasks assigned to health care providers [24].

Courses early in the preclerkship curriculum addressing medical ethics and professionalism present an opportunity for an introduction to concepts in privacy and ethics of health data usage. Prior to beginning any patient care activities, students complete the required Health Insurance Portability and Accountability Act (HIPAA) and patient privacy training, which address key components of privacy [25]. In later sessions focused on transition to clerkship, the curriculum should provide a more in-depth exploration of these topics, using specific case-based examples and interactive instructional methods. Finally, case studies during clinical rotations may be helpful in allowing students to apply this knowledge and integrate concepts into clinical practice. For example, students could consider a case study in which a pediatrician caring for a child with a mental health diagnosis discusses with parents the option of using data from the patient’s social media accounts to monitor mental health status, allowing for discussion of consent, privacy, access, bias, and authorization to use data from third-party platforms [26]. Additionally, the tools used by pediatricians to monitor mental health status themselves pose an opportunity to explore whether they are appropriate for the patient based on demographic factors, allowing for the exploration of fairness and equity [27].

As with topics in health data ethics, basic cybersecurity objectives should be included in modules required before students begin any patient care activity. More advanced cybersecurity topics may fit well in the transition to clerkship, with opportunities to reinforce and apply concepts as students transition between different clinical rotations.