Institutional approaches to artificial intelligence policy and guidance in health informatics and information management education: emerging trends and inconsistencies

Abstract

Objective

To examine the characteristics, scope, and thematic content of institutional and program-level academic artificial intelligence (AI) policies and guidance documents used by CAHIIM-accredited Health Informatics (HI) and Health Information Management (HIM) master’s programs in the United States.

Materials and Methods

This convergent mixed‑methods study examined publicly available AI‑related documents (eg, student handbooks, integrity policies, program manuals) from all 48 CAHIIM‑accredited HI and HIM master’s programs listed in November 2025. Forty programs (83%) had at least 1 qualifying document, 8 programs had no qualifying documents. Quantitative analyses summarized policy characteristics and examined associations with delivery modality using Fisher’s exact tests. Qualitative analysis applied Latent Dirichlet Allocation topic modeling to identify recurring semantic themes across policy texts.

Results

Most documents provided guidance rather than formal policies; neither policy type nor audience differed by delivery mode. Content emphasized academic integrity, responsible AI use, and expectations for student conduct. References to privacy, intellectual property, and regulatory concepts were present but less common than academic integrity-related themes. Topic modeling identified 4 dominant themes: academic integrity and appropriate AI use; generative AI in teaching; AI and data‑tool use in research; and student engagement with conversational AI systems.

Conclusion

AI guidance and policy are available across most accredited HI and HIM master’s programs; however, they vary in structure and scope. The majority of documents emphasize academic integrity and responsible AI use. Future research could examine how academic program AI policies relate to distinct but related domains such as curriculum development and organizational AI governance.

Introduction

Background and significance

Artificial intelligence (AI) technologies are increasingly integrated into healthcare practice and research, where they are utilized for classification, prediction, pattern recognition, and clinical and administrative decision support.^1,2 With AI tools becoming integral to routine practice, Health Informatics (HI) and Health Information Management (HIM) professionals increasingly engage with AI systems that need exacting evaluation and governance. Consequently, graduate training must equip students with the competencies to navigate the legal, ethical, and data‑governance requirements in AI use. The Commission on Accreditation for Health Informatics and Information Management (CAHIIM), through its Future Education Model, designed specifically for accredited HIM programs, underscores the need to develop student competencies in AI governance, technical foundations, and regulatory compliance.³

HI and HIM occupy a unique role in health education, because faculty and students routinely handle protected health information, clinical data, and use health information systems governed by HIPAA, FERPA, and human‑subject protections. Using AI tools in coursework, research, or practice-based learning raises considerations related to privacy, security, fairness, transparency, and accountability. Despite the growing number of institutional guidelines on generative AI and academic integrity, most are too broad to address the unique requirements of health‑data governance or the risks associated with applying AI in clinical or research contexts. Although many institutions have developed AI-related academic guidance, these documents primarily address issues such as academic integrity, authorship, citation, and acceptable use of AI in educational settings. Less is known about how HI and HIM programs adapt these academic policies for disciplines in which students frequently engage with health-related data, research activities, and professional practice expectations.

Academic AI policies, AI curriculum design, and healthcare AI governance represent distinct but related domains. Academic AI policies govern acceptable AI use within educational settings. Curriculum design determines competencies and educational content. Healthcare AI governance addresses organizational, regulatory, and operational oversight of AI within healthcare environments. This study focuses only on academic policies related to AI use in the educational setting. Healthcare applications are discussed to provide background and explain their broader impact on HI and HIM education.

Prior research

Research on AI policies within higher education exhibits substantial variation in approaches. Global analyses show institutional approaches range from strict prohibitions to instructor‑driven flexibility, with many institutes adopting context‑specific approaches.⁴ Studies focused on U.S. Research Level 1 institutions indicate growing support for generative AI but report the need for significant pedagogical redesign.⁵ Case studies from leading universities underscore the importance of maintaining transparency, privacy safeguards, and risk‑management competencies.⁶ Although entities like the European Union and World Health Organization work to advance ethical AI use in health‑related research, these standards have not been fully integrated into HI and HIM graduate education.^7–9

Rationale and gap in the literature

HI and HIM master’s programs are accredited by the CAHIIM, a Council for Higher Education Accreditation (CHEA)-recognized accrediting body. In addition, institutions must also meet regional accreditation requirements. CAHIIM accreditation standards require programs to meet established competencies and compliance expectations, while CAHIIM frameworks provide broader guidance for AI governance, curriculum development, and educational practices.³ However, accrediting bodies do not specify how programs should operationalize AI curriculum governance, design, or student learning activities. Some academic institutions and clinical informatics fellowships rely solely on institutional ethical policies, instead of field-specific policies. As a result, there is limited understanding of academic AI policy development across HI and HIM master’s programs. To date, no systematic review has examined the characteristics, scope, and content of AI‑related policies within CAHIIM‑accredited HI/HIM graduate programs.

Study objective

This study addressed this gap by examining the extent of how CAHIIM-accredited HI and HIM master’s programs in the United States operate under institutional academic-related AI Policies or program-specific educational policies related to AI in HI and HIM education.

The analysis examines the scope of these policies, their alignment with regulatory frameworks, and thematic patterns that emerge across programs. By comparing policy characteristics across institutions, the study provides insights that may help inform future academic AI policy frameworks in HI and HIM education.

Research questions

Quantitative

1. How do AI policy characteristics differ by program delivery mode (online, campus, hybrid) across CHEA regions?

2. To what extent do academic AI policies address legal, intellectual property, privacy, transparency, and responsible-use considerations related to AI in educational settings?

3. How do AI policies articulate the expectations related to academic integrity, ethical data use, and equitable access to AI tools for learning and research?

Qualitative

• 4) What recurring themes (eg, ethics, privacy, IP) emerge across institutional AI policies relevant to HI and HIM education?

Methods

Study design

This cross-sectional observational study analyzed publicly available AI policies from CAHIIM-accredited HI and HIM master’s programs in the United States first by quantitative analysis of summarized extracted policy characteristics and examining associations between program attributes. Second, a separate text analysis of keyword frequencies and Latent Dirichlet Allocation (LDA) topic modeling were applied to identify thematic patterns within the policy documents.

Sample

The sample included all master’s-level HI and HIM programs accredited by CAHIM and listed in the program directory as of November 2025. Certificate-only programs were excluded because they do not follow the same curricular, accreditation, or governance standards.¹⁰ Of 48 eligible programs, 5 had no publicly available AI-related documents, leaving 43 with at least 1 policy. In 3 cases, a single institutional policy applied to both HI and HIM programs; these shared policies were counted once. The final analytic sample was 40 distinct AI policy documents, which served as the unit of analysis for quantitative and qualitative analysis.

Data collection and extraction

Between December 2025 and February 2026, we manually collected publicly accessible AI-related policy documents from institutional HI and HIM program websites identified via the CAHIIM program directory. The approach constituted an exploratory environmental scan rather than a systematic review. Documents were eligible for inclusion if they addressed institutional or program-level academic policies or guidance related to AI use, specifically those governing student or faculty conduct at the institutional or program level such as academic integrity, acceptable use, or instructional guidance, rather than clinical practice outside of the educational setting, and were available without an institutional login. Policy attributes, including policy type, policy audience, instructional delivery format (online, campus-based, or hybrid), and specified content components, were extracted to a structured Excel workbook to maintain reviewer consistency.

Quantitative data analysis

Quantitative analyses were conducted in R version 4.5.2,¹¹ with statistical significance set at α = .05. Descriptive statistics (frequencies and percentages) summarized policy characteristics and program delivery modalities (online, on‑campus, hybrid). Associations between delivery mode, policy type, and audience were evaluated using Fisher’s exact tests, due to small contingency table cell counts. Due to small sample sizes, policy type was recoded to enhance analysis; “Guideline,” “Informational,” “Policy” were retained, while “Mixture,” “Statement,” or “Task Force” were recoded into “Other.” Delivery mode was recoded as a binary variable: “Any Online” (Online or Both) versus “Campus Only.”

Keywords were developed from prior literature on academic AI policy, academic integrity, intellectual property, privacy, responsible AI use, and higher-education policy. Additional terms related to healthcare regulatory frameworks were included descriptively because of the health-focused context of the programs examined.^12,13 The final keywords were determined through the study team’s consensus following discussion. Keywords were organized into thematic categories aligned with RQ2 and RQ3 for frequency analysis. Automated keyword searches identified common themes related to responsible AI use, academic integrity, intellectual property, and data privacy/security. Geographic patterns in policy availability were summarized across CHEA regions with cross‑tabulations.

Qualitative text analysis

Policy PDF documents were converted to Word (DOCX) before text preprocessing. Preprocessing included lowercasing, punctuation, stop word removal, and tokenizing text. LDA was conducted with the topicmodels package in R to identify recurring themes in the AI policy document corpus. A 4-topic solution was selected based on topic coherence and interpretability. The LDA model identified co‑occurring word groups and generated probability distributions for each topic, revealing thematic patterns not captured in keyword frequencies alone.^14,15 For each document, the model produced topic membership probabilities, which were utilized to summarize theme prevalence. LDA has been used in healthcare research, including analyses of clinical narratives and nursing documentation.^16,17

Ethical consideration

All materials in this study were publicly available institutional documents and contained no human subjects’ data or identifiable personal information. As such, the study met criteria for exemption from IRB oversight. All source URLs and retrieved files were securely archived in a protected research repository to ensure documentation and reproducibility.

Results

Sample descriptives

Among the 48 CAHIIM‑accredited master’s programs in the study, 40 (83%) had at least one publicly available AI‑related document. Among these, 8 (20%) were private, and 32 (80%) were public institutions. Most documents targeted both faculty and students (20, 51%), followed by faculty‑only policies (13, 33%) and student‑only policies (6, 15%). One document did not specify an intended audience.

Quantitative research questions

RQ 1: Content delivery, policy type, and policy audience

Table 1 presents the demographic characteristics of the programs and policy documents. Most programs used online delivery (36, 90%), with fewer campus-only (4, 10%) formats. Policy documents were primarily guidelines (21, 53%), followed by informational documents (9, 23%), formal policies (7, 18%), and other types (3, 8%). Intended audiences were predominantly faculty and students (20, 51%), followed by faculty‑only (13, 33%), and student‑only (6, 15%); 1 document (3%) did not specify an intended audience.

Table 1.

Program delivery modes, policy types, and policy audiences (N = 40).

Category	n	%
Program delivery mode
Any online	36	90
Campus only	4	10
Policy type
Guideline	21	53
Informational	9	23
Policy	7	18
Other (mixture/statement/task force)	3	8
Policy audience
Faculty and students	20	50
Faculty only	13	33
Students only	6	15
Not specified	1	3

“Other” includes mixed, statement, and task force categories combined. Percentages may not sum to 100% due to rounding.

Fisher’s exact tests for association showed no significant association between delivery mode (“Any Online,” “Campus Only”) and policy type (P = .85) or policy audience (P = .71). The finding of a lack of correlation suggests that cell counts were too low to detect meaningful differences.

The Higher Learning Commission (HLC) (n = 14) and the Southern Association of Colleges and Schools Commission on Colleges (SACSCOC) (n = 13) accounted for the majority of policies, representing 60% of the dataset, followed by the Middle States Commission on Higher Education (MSCHE) (n = 8), the Northwest Commission on Colleges and Universities (NWCCU) (n = 4), and the New England Commission of Higher Education (NECHE) (n = 3). No policies were identified for Western Association of Schools and Colleges (WASC) Senior College and University Commission or WSCUC-accredited institutions. See Figure 1 CHEA Accreditation Regions.

Figure 1.

CHEA accreditation regions.

RQ 2: AI policies for legal, IP, and privacy, and responsible-use considerations in academic AI policies

The keyword frequencies for this question are provided in Table S1. For legal keywords, general academic integrity (n = 139) and integrity‑oriented terms (ethics n = 31, legal n = 31) occurred far more frequently than regulatory references, such as HIPAA (n = 5), FERPA (n = 11), or IRB guidance (n = 8), which appeared infrequently. Policies referenced broad IP concepts, such as intellectual property (n = 57), property n = 50), copyright (n = 8), but rarely addressed detailed issues such as licensing (n = 8) or patents (n = 6).

Training terms related to general instructional activities were common (n = 168), whereas references to workshops (n = 17) were limited.

Health‑related terms, such as personal health records (n = 19) and confidentiality (n = 5), appeared infrequently, and electronic health records were not mentioned which is consistent with the academic scope of documents reviewed.

General privacy and security terminology appeared in many documents, such as security (n = 55) and privacy (n = 54), whereas references to healthcare-specific regulations such as HIPAA, FERPA, and IRB oversight were uncommon. Given that these regulations are often addressed through broader institutional compliance policies rather than academic AI policies, their limited appearance in AI-specific documents is not unexpected.

RQ 3: AI policies and academic integrity, data ethics, and equitable AI access

The keyword frequencies for this question are provided in Table S1. Academic integrity emerged as the most dominant across all keyword categories (n = 139), followed by citation (n = 59), assessment (n = 50), and plagiarism (n = 38). Less frequently mentioned were authorship (n = 13), detectors (n = 12), and misuse (n = 10). Mentions of proctoring (n = 2), disclosure of AI (n = 1), and contract cheating (n = 1) were rare.

Within the data ethics category, the most frequently mentioned terms were bias (n = 65), security (n = 55), and privacy (n = 55). Other principles appeared less frequently, such as transparency (n = 17), fairness (n = 10), accountability (n = 10), and risk assessment (n = 5). Several coded terms (civil rights, explainability, robustness) were not present in any policy.

Themes related to equitable access were limited. There were fewer frequent mentions of inclusion (n = 9), accessibility (n = 9), devices (n = 5), accommodations (n = 4), and equitable access (n = 2). Minimal mentions were present for broadband (n = 1), undeserved (n = 1), and devices (n = 5), indicating a limited focus on digital inclusion.

Qualitative results

RQ 4: Themes in AI policies

Table 2 summarizes the results of LDA modeling. LDA discovered 4 coherent clusters of terms across the AI policy documents. Topic 1 (Academic Integrity and Policy on Student Use of AI/Tools) reflected terms related to academic integrity and expectations for appropriate student use of AI tools. Topic 2 (Student Use of Generative AI Tools) captured terms associated with student engagement with generative AI tools in courses. Topic 3 (AI and Data Tool Use in University Research) represents language describing the use of AI and data analysis tools within university research contexts. Topic 4 (Student Engagement with ChatGPT in Academic Tasks) focused on students’ use of ChatGPT and related tools in academic work.

Table 2.

Topics identified with LDA topic modeling.

Topic	Theme	Key terms
1	Academic integrity and policy on student use of AI/tools	student, academic, policy, faculty, use
2	Student use of generative AI tools	tools, use, student, generative, course
3	AI and data tool use in university research	data, tools, use, research, generative
4	Student engagement with ChatGPT in academic tasks	generative, use, tools, ChatGPT

Abbreviations: AI, artificial intelligence; LDA, Latent Dirichlet Allocation.

Figure 2 displays the intertopic distance map from the LDA model, where principal component analysis (PCA) was used to position the topic in 2-dimensions (PC1 and PC2). Topics farther apart have distinctly different vocabularies; points closer together share more terms. The plot shows clear topic separation, indicating a coherent model with limited vocabulary overlap across themes. Topic 1 (Academic Integrity and Policy) appears separate on the far left of the plot, distinct from the other topics. Topics 2 (Generative AI Use) and 4 (ChatGPT Use) cluster on the right, relatively close together, reflecting the similarities between generative AI use and Chat-based AI use. Topic 3 (AI and Data Tools in Research) is in the lower-right quadrant, indicating a unique focus on the use of AI and data tools.

Figure 2.

Intertopic distance map from the Latent Dirichlet Allocation (LDA) model.

Discussion

This study is one of the first to systematically examine AI academic policies in CAHIIM-accredited HI and HIM master’s programs in the United States. The documents reviewed were primarily designed to guide acceptable AI use in academic settings rather than to define healthcare practice standards or workforce competencies. Therefore, their emphasis on academic integrity and responsible use is consistent with the intended role of academic policy. While most programs have developed some form of AI guidance, the focus remains on academic integrity and responsible AI use in student assessments. Issues related to privacy and IP were less emphasized.

Quantitative and qualitative analyses showed that the main emphasis of policy attention is on prevention of student misuse of generative AI in academic coursework. While comparatively little guidance addresses student AI use in program applied learning environments such as research settings and simulated environments where AI curricular and data governance concerns may intersect.^18,19

Interpretation of findings

Quantitative analysis showed that most AI policies were advisory, online delivered, and emphasized academic integrity. This suggests institutional efforts have appropriately focused on academic integrity and student behavior. However, policies may need to evolve as AI-use becomes more prevalent in health science and healthcare education. In agreement, keyword frequencies were dominated by academic integrity concepts, honor codes, and prohibitions on AI use (expressly prohibited). This distinction highlights how these AI academic policies are not intended to address healthcare practice, curriculum design, or organizational data governance.

Graduate policies emphasized originality in research milestones, while equity, bias mitigation, and intellectual property were broad and largely non-actionable. Ethical concerns were noted, but there were no specific procedures for risk management, indicating the challenges for higher education to operationalize AI ethics in academic settings.

LDA modeling identified 4 thematic clusters: academic integrity, student engagement with generative AI, ChatGPT-specific academic use, and AI and data tools in research. These topics indicate institutions prioritize student conduct and responsible AI use, yet gaps remain in institution-specific guidance regarding operational implementation of AI use within educational contexts.

Comparison with prior research

Prior work documented wide variation in institutional responses to generative AI, with a cross‑national review of 343 universities showing approaches ranging from bans to instructor‑driven discretion, and nearly half adopting flexible policies shaped by regional norms.⁴ Research on U.S. R1 institutions similarly found generative AI use is generally supported, but the curricular integration requires significant pedagogical revision.⁵ Surveys of high school and higher‑education administrators further indicate that many institutions still lack dedicated policies addressing ethical use, privacy, and algorithmic transparency.⁶ Case studies of AI governance at major U.S. universities highlight emerging governance models that distribute AI responsibility across institutional roles.⁷

In agreement, our results show that AI-specific guidance in CAHIIM-accredited HI and HIM programs primarily emphasize academic integrity and responsible AI use in educational platforms and settings consistent with broader trends in higher education AI policy development.

Thus, although AI policy development in higher education is advancing, HI and HIM programs appear to be creating a distinction between academic policy frameworks and the broader competencies delivered through curriculum and professional training.

Implications for practical applications

In the context of HI and HIM education, these findings suggest that institutions are increasingly adopting AI guidance and policies focused on academic integrity, acceptable use, transparency, and authorship. Practical efforts may therefore focus on improving clarity, consistency, and implementation of academic AI policies across institutions. AI curriculum development and workforce preparation should be viewed as complementary, distinct, yet interrelated activities from academic policy development.

Regarding policy and accreditation, accrediting bodies and professional organizations can help reduce variability by providing AI policy frameworks that integrate best practices in academic integrity, data ethics, and equitable access. Standardizing, contextualizing, and refining institutional AI policies would enable programs to more effectively operationalize policy frameworks that address the unique requirements of HI and HIM programs. In addition, programs should monitor regional policy trends and invest in faculty development to support AI-enabled graduates.

On February 13, 2026, the U.S. The Department of Labor released the AI Literacy Framework, which contained foundational content areas that include understanding AI principles, exploring AI use, directing AI effectively, evaluating AI outputs, and using AI responsibly.²⁰ Additionally, they listed effective delivery principles of AI consisting of enabling experimental learning, embedding learning content, building complementary human skills, addressing prerequisites for AI literacy, creating pathways for continued learning, preparing enabling roles, and designing for agility. The framework and principles guidelines were created to be flexible, allowing not only formal education, but also other AI-focused literacy efforts to effectively take part in developing AI skills that will prepare students and the workforce for the current and future technology challenges. The AI literacy framework and foundational content are echoed in AMIA’s 2026-2028 Strategic Plan section,²¹ Empowering the Member, outlining AI leadership, developing guidelines and frameworks governing AI in informatics, and career advancement, developing new educational pathways supporting informatics workforce development.

Future research

Future research should expand this study to include other non-US institutions to enhance the generalizability of the findings and examine variation in academic AI policy development across educational settings. Including longitudinal analyses may provide insight into how AI-related academic policies evolve in response to emerging technologies, institutional priorities, and changing educational practices. Furthermore, exploring the factors influencing policy development and implementation through qualitative research methods with stakeholders, such as faculty and administrators, may aid in creating AI policy frameworks. Future work may also examine how academic AI policies interact with other institutional activities, including curriculum development and organizational AI governance, which represent distinct but related domains.

Limitations

This study examined only publicly available AI‑related policies from accredited HI and HIM programs; thus, internal or unpublished documents may be excluded, creating potential visibility bias. As a result, this study may underrepresent programs with strong but private compliance structures and those with AI policies within syllabi or clinical handbooks. Findings are limited to CAHIIM‑accredited HI and HIM master’s programs within the United States and may not extend to other disciplines, non‑accredited programs, or international contexts. Differences in policy length, structure, and detail could have affected keyword search and topic‑modeling results. In addition, broader institutional, professional, and student conduct policies addressing healthcare regulatory requirements (such as HIPAA, FERPA, and research compliance) were not included but may have been mentioned in academic policies due to the program context.

Additionally, because LDA is an unsupervised algorithm, it may overlook subtle or context‑specific aspects of policy language, and topic interpretation involves some researcher judgment. Additionally, topic‑modeling results reflect only textual patterns and cannot capture institutional intent. Moreover, small cell sizes may have reduced statistical power. Finally, AI‑related policies continue to evolve quickly, so these results represent a single point in time rather than ongoing trends.

Conclusion

AI policies in CAHIIM-accredited HI and HIM master’s programs primarily address academic integrity, responsible AI use, and acceptable educational practices. Most documents function as guidance rather than formal policy and focus on student and faculty conduct within academic settings. The findings demonstrate increasing institutional attention to AI while also highlighting substantial variation in policy structure and content. Future work should examine how academic AI policies interact with curriculum design, professional preparation, and healthcare AI governance, which represent distinct but related domains.

Author contributions

Christina Eldredge (Conceptualization, Formal analysis, Investigation, Methodology, Project administration, Supervision, Writing—original draft, Writing—review & editing), Diane Dolezel (Conceptualization, Formal analysis, Investigation, Methodology, Visualization, Writing—original draft), Josette Jones (Conceptualization, Investigation, Supervision, Writing—review & editing), Neha Vooppala (Conceptualization, Investigation), and Stephen E. Bronsburg (Conceptualization, Investigation, Supervision, Writing—review & editing)

Supplementary material

Supplementary material is available at Journal of the American Medical Informatics Association online.

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Conflicts of interest

All authors have no conflicts of interest to declare.

Data availability

The data underlying this article will be shared on reasonable request to the corresponding author.