Study design and phases

This exploratory qualitative study employed reflexive thematic analysis of semi-structured interviews and a co-design workshop, followed by the collaborative development of an action plan. The current study was conducted in three phases. Phase 1 involved individual semi-structured interviews with institutional leaders and faculty. Phase 2 consisted of a co-design workshop with multidisciplinary participants, utilizing the SWITCH framework as a sensitizing lens. Phase 2 focused on the collaborative development of a SWITCH-aligned action plan, synthesized from the findings of Phases 1 and 2.

Ethical considerations

Before commencement of the study permission from Aga Khan University’s Ethical Review Committee (ERC number 2023-8372-24124) was sought. Written informed consent was taken from all participants. Data collection was completed within the ERC-approved period (before March 2024).

Participants and data collection

The study was conducted from May 2023 to May 2024. Subsequent data cleaning and statistical analysis were conducted from March 2024 onward. Data were collected from various sources and participants via interviews, an interactive co-design workshop, and discussions, ensuring informed consent was obtained from all participants. The composition of participants varied across these Purposive sampling was done to gather data from various educational leaders [associate deans, departmental chairs, section heads, education leads of Postgraduate Medical Education (PGME) and Undergraduate Medical Education (UGME) and multidisciplinary faculty, clinicians] and UGME students, PGME students at AKU and AKUH [10]. The study’s objectives and the absence of conflicts of interest were communicated to all the participants. Considering the study’s objective, the sample of interest, the theoretical perspectives of the study, and the purposive sample selection, a sample of 10–12 participants was estimated for the interviews. However, data collection through interviews continued until thematic saturation was achieved. Workshop included 22 participants with almost negligible overlap with interview participants. Subsequent meetings for action plan development primarily involved the research team, and authors to refine and validate emerging insights.

Prior to the interviews, a pilot study was conducted with two participants to evaluate the interview schedule’s effectiveness, leading to refinements in language and style. Interviewees were contacted through invitation emails and follow-up phone calls to schedule interviews. Semi-structured interviews, lasting 20–30 minutes, were conducted by trained research colleagues with qualitative research experience who also participated in pilot interviews prior to data collection. The research team (n = 4) included both men and women who used interview schedules, based on literature and with clear prompts, to carry out the process. The researchers framed the interviews as discussion sessions rather than performance evaluations to encourage open dialogue and reduce participant anxiety. Face-to-face interviews were audio recorded and thematically analyzed by the research team and the principal investigator. Clarifications were sought during and post-interviews to ensure accuracy, with key findings reviewed in team meetings.

Using interview findings, a workshop aimed to discuss the actionable plan for integrating AI into healthcare. The workshop gathered collaborative data focused on how the SWITCH model of change can be utilized to create a comprehensive framework for AI integration in medical education and in-service practices at an academic medical center. A qualitative validation approach was implemented during the workshop, where participants discussed preliminary themes and findings, offering feedback. To mitigate social desirability bias, confidentiality of all participants was ensured, discussions were held neutrally, and open-ended questions were utilized to foster honesty. Researchers also considered their own biases and experiences to limit their impact on participants, while data triangulation was employed to enhance credibility. This process facilitated the triangulation of perspectives, ensuring that the findings accurately reflected the participants’ viewpoints.

An action plan was developed and subjected to rigorous review through closed meetings, allowing for collaborator feedback and refinement to enhance effectiveness and feasibility. The plan was reviewed among authors via meetings and email. All interviews, workshop sessions, and planning meetings were audio recorded, transcribed verbatim, and anonymized. Data management was conducted manually using Microsoft Word and Excel.

Data analysis

Reflexive thematic analysis was conducted using Braun and Clarke’s six-step framework with slight modifications to suit the objective of the study [11], While steps 1–4 followed a systematic, interview based thematic analysis, steps 5 and 6 transitioned into participatory validation through a workshop and the subsequent development of an action plan for change. Microsoft Word and Excel were used to organize and analyze the notes and transcripts from interviews and workshop. As described in Fig 1, the analysis began with an immersion phase, during which research team reviewed the transcripts several times to achieve thorough familiarity with the data. During this step, preliminary observations and reflexive notes were documented to identify initial patterns relevant to the study objectives. As step two of the process, the transcripts were divided into smaller units and initially coded by two researchers. Segments of data were labelled to capture meaningful patterns. The team held regular discussions to address coding discrepancies and presented data with illustrative quotes from participants, with their informed written consent. To ensure rigor in this process, the research team collaborated and resolved discrepancies through consensus, thereby enhancing inter-coder agreement. As a third step the codes were grouped into higher level categories to form overarching themes, with the research team reviewing the process for recurring patterns and key concepts. An inductive approach allowed themes to emerge from the data while acknowledging the researchers’ interpretative role. The research team mitigated potential bias by grounding the synthesis in participant accounts and reflective coding notes. All records of coding decisions, modifications and rationales were documented. As fourth step, the themes were iteratively reviewed, raw data was revisited and refined to ensure internal homogeneity (all data within a specific theme supported a unified interpretation) and external heterogeneity (ensuring themes were distinct). In the fifth step of the process, specific themes were established and further refined during a multidisciplinary workshop. This refinement aimed to ensure that the themes were both coherent and distinct from one another. Additionally, the workshop participants compared the insights and findings generated during their discussions with the themes identified in the interviews, ensuring a comprehensive evaluation of the data collected from multiple sources. Triangulation was achieved by comparing interview insights with data collected during workshops, capturing multiple perspectives and ensuring consistency across different settings. The use of collaborative problem solving and anonymity further validated the findings by reducing bias and confirming results directly with participants. Finally, as last step, thematic mapping and reporting were conducted, where the validated themes were aligned with the SWITCH model of change. The analytic integration utilized data source triangulation to synthesize interviews, workshop, and meetings into a cohesive action plan. To mitigate potential power imbalances, participants were guaranteed anonymity, ensuring that individual identities remained disconnected from the data. Potential bias in the analysis phase was managed through data source triangulation and the use of reflective notes to bracket pre-existing assumptions. By reconciling individual interview themes with broader workshop discourse and institutional planning data, the team ensured the final SWITCH-based framework was grounded in collective evidence rather than researchers led narratives. Findings were compared for thematic complementarity and subsequently reconciled through a rigorous process based on viability, stakeholder relevance, and institutional alignment. This iterative validation process ensured that the final SWITCH based action plan was not merely a thematic summary, but a strategically calibrated system designed to shape the path of AI change while addressing the documented elephant resistance of the workforce. By categorizing the emergent themes under the domains of directing the rider, motivating the elephant, and shaping the path, the analysis moved from descriptive findings to a structured framework for guiding AI related institutional change.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Conceptual framework for ai integration in academic medical centers: Study phases and thematic analysis with integrated quality check measures.

https://doi.org/10.1371/journal.pone.0346582.g001

Interview findings

A total of eighteen interviews were conducted, representing professionals from diverse departments including Pathology and Laboratory Medicine, Women and Child Health, Dean’s Office, Hospital Operations, Department for Educational Development, Emergency Medicine, Pediatrics, Community Health Sciences, Surgery, School of Nursing, Quality Network of Quality, Teaching, and Learning (QTL_Net) and the Medical College of AKU. The male-to-female ratio among the participants was approximately 1.25 to 1. The range of years of experience among the professionals varied from 9–32 years, with a mean experience of 14.5 ± 5.1 years. Content analysis revealed following overarching themes: the potential of AI opportunities, resistance toward AI-induced changes, maximizing learning development and resources in AI to drive transformation through leadership styles, and importance of trainings/ capacity building to sustain change (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Leadership insights from structured interviews: themes, categories, and codes.

https://doi.org/10.1371/journal.pone.0346582.t001

AI opportunities for improvement

The interviewees framed AI not only as a technological upgrade of healthcare systems but as a performance enabling method for healthcare professionals. Predictive analytics and integrated AI-powered diagnostics were frequently mentioned by participants as becoming standard in clinical workflows to address labor shortages and improving efficiency. Several participants expressed enthusiasm about integrating AI into their administrative responsibilities to streamline hospital operations including providing care to patients at the bedside. Furthermore, responses from participants illustrated a wider awareness among healthcare professionals that AI solutions could be used for patient interaction, data flow, and patient management in addition to acting as channels for service delivery. This could indicate a change in hospital operations and professional duties within the healthcare system by shifting responsibility from hospital workers to digital interfaces. The AI tools were viewed as facilitators of timely and informed clinical decision making through rapid access to patient data and decision support systems. As one respondent (P16) noted, “AI finds utilization in triage in emergency medicine for better patient outcomes” highlighting the role of AI in strengthening patient flow and time management in emergency settings. Participants envisioned an integrated healthcare system enabling patients to connect handheld devices to hospital networks for instant transmission and analysis of medical data, promoting individual care and overcoming cognitive limits. One of the participants (P12) remarked, “integrated healthcare system may not only enhance convenience for patients but also ensure quicker and more accurate decision-making by healthcare professionals, ultimately leading to improved patient outcomes.” Participants accounts suggested that AI use shifts the clinical decision making from reliance on individual judgement alone to a more data supported and system assisted process. A significant tension emerged between the initial expectations of AI as a productivity facilitator and the emergent risks of clinical dependence. Some also expressed apprehensions that increased dependence on AI may lead to over reliance to automated systems raising concerns about bias in clinical decision making. While participants anticipated that AI would alleviate cognitive load, they simultaneously expressed a drift toward automation bias, where over reliance on algorithmic outputs threatened to undermine individual clinical accountability and lead to the deskilling of the medical workforce.

Participants further indicated that benefits of AI use could extend across many clinical departments. They associated AI related transformation with improvements in diagnostics across many disciplines. As one respondent (P2) explained “AI is poised to advance in diagnostics, leveraging image analysis in pathology, radiology, enabled by rapid digitization of glass slides, mass spectrometry chromatograms and digital imaging technology” suggesting that integration of AI and diagnostic technologies are perceived as important drivers for accurate and timely patient management. One of the respondents (P5) associated AI related advancements is reshaping expectations in diagnostic laboratories, “in a laboratory setting, natural intelligence is utilized for analyzing organism colony morphology, biochemical reactions, microscopy, and susceptibility patterns based on previous experiences. my expectation with AI is going to be much higher than what I deliver right now and what my seasoned technologists deliver right now.” The participants responses indicated that they expect AI to handle tedious tasks, complex clinical data exceeding current performance standards thus raising their expectations significantly compared to current human capabilities.

Across the interviews, participants highlighted the potential of AI tools to personalize learning pathways for medical students automate grading tasks with real-time feedback, paper setting and exams scoring. Additional uses include allowing students to set their own schedules and advance at different rates. This implied a more comprehensive rethinking of physicians as learning facilitators, interpreters of AI-generated insights, and not just content providers. As AI takes over the role of content provider by allowing students to advance at their own rates, the physician’s role may shift toward learning facilitation and the interpretation of AI generated insights. This represents a sociotechnical tension where the opportunity for student autonomy creates an existential challenge for clinicians, who need to redefine their professional value beyond their traditional expertise.

Reluctance to embrace change

The healthcare providers’ accounts of challenges linked to AI integration in their hospital highlighted both structural and psychological dimensions of organizational change. The participants consistently identified that any attempt at implementing AI would be met with resistance due to the fear of losing their job. Some participants expressed that AI being a new subspecialty would create jobs, but a significant hurdle would be acceptability. Concerns regarding job security highlight the tension between AI as an innovation and its potential threat to professional identity, revealing occupational anxiety and uncertainty about workforce stability and responsibilities in hospitals. Some participants framed AI as an enhancer of professional skills while maintaining job security, viewing its inclusion in healthcare systems as a complement to their roles rather than a substitute. As one participant (P14) explained, “AI was never meant to and is never meant to replace people” highlighting the continued importance of judgement in clinical care.

Participants also acknowledged the challenge of introducing a major organizational change, emphasizing the need for meticulous planning to effectively adopt AI technology. This necessitates ongoing communication to articulate needs, expectations, and seek expert guidance for AI implementation. While the institute’s checks and balances ensure organizational safety and integrity, they can also impede or halt beneficial changes. Concerns regarding data privacy, patient consent, and cybersecurity were highlighted as both regulatory and moral obligations in the context of AI integration in hospitals. They were hesitant to fully endorse AI as a tool in medical education due to unresolved ethical issues and inadequate regulatory oversight. Participants’ acceptance coexisted with mild reservations, indicating that organizational, psychological, and ethical concerns had just as big of an impact on AI adoption as functionality and apprehension about employing AI technologies in healthcare. Participants identified the import of AI technology as a significant barrier to change, particular in a country like Pakistan where reliance on imported technology increases procurement complexity and cost. As a result, AI integration is a resource-intensive shift in which the availability of AI tools, cost, and infrastructure are critical to driving AI integration into healthcare. The participants highlighted concerns regarding reliance on imported AI solutions noting limited local development capacity as a potential barrier to sustainable implementation. As one respondent (P3) stated, “all our lab consumables even Eppendorf are imported, implying that AI tools are costly to import, and overall, AI integration costs make it difficult to sustain. If Pakistan had its own local AI systems for hospitals, we wouldn’t worry about import costs and integration would be sustainable.” Locally developed AI systems could help overcome barriers associated with high cost and import dependency. Interviewees expressed concerns regarding the importance of ensuring that imported AI solutions are matched to local needs, are validated on clinical data and conform to all relevant regulations. As one respondent (P9) noted, “I have seen lots of AI algorithms lack the evidence needed to support implementation. Clinicians will have to demand substantial resources either from the institute or the vendor to validate these tools in our local population” highlighting the need for local validations, extra work and time and stronger support from leadership. Another participant (P2) stated “AI makes diagnostics faster but makes the doctor’s day slower because of the validation burden”. The findings reveal a contradictory drift where the opportunity for efficiency was not viewed as a simple benefit, but as a catalyst for anxiety regarding job security and the potential de-skilling of the clinical workforce. In the domain of diagnostics, a sharp contradiction emerged between the technical speed of AI tools and the temporal reality of clinical practice. While participants identified the opportunity for quick diagnostic interpretations, this was overshadowed by a validation burden. This created a drift from efficiency to increased workload, as clinicians reported that the cognitive effort of auditing an uninterpretable AI would exceed the time required for a traditional manual diagnosis. This matters because it suggests that AI does not simply save time but rather reconfigures clinical workload, shifting the burden from task execution to high-stakes oversight, which can lead to increased mental fatigue and burnout.

Leadership styles and practices

Participants emphasized that the extent of AI integration hinges upon the vision of its leaders. Leaders possess the ability to empower and facilitate AI-related change within the institution. To achieve this, they themselves must first acquire a comprehensive understanding of AI-based healthcare solutions and subsequently endeavor to maximize the benefits while mitigating associated challenges. Effective leadership entails advocating for change in a manner that alleviates fears and encourages acceptance of AI technology among others. Strong leadership to handle this change with robust IT security, are essential for the successful integration of AI in healthcare stated another respondent (P13), “Strong AI leaders in our hospital can reframe its use from just speed and efficiency to accountability and reduction in clinical errors”. The role of AI integration in hospitals as a real-time auditing mechanism was discussed. One of the respondents (P9) stated, “leadership should clearly communicate that AI transition is non-negotiable and as a tool for patient safety and transparency” thereby emphasizing that the transition towards AI systems enhances leadership oversight and transparency through the availability of traceable audit trails and real time monitoring of clinical activities.

Leadership for AI integration in hospitals is strategic and relational rather than just administrative. The capacity of the leader to incorporate AI into the current workflow, cultivate trust, and reaffirm the crucial role of human health care workers in addition to AI tools will determine if the transition is effective. Participants varied opinions regarding leader’s approach to change management. Some preferred a more authoritative stance, while others advocated for a collaborative approach, involving users and stakeholders in planning and execution. Some also rationalized that AI-leadership can also emerge from individuals at lower levels who are passionate about the change and committed to supporting and guiding others. Like one participant (P11) stated, “a tech-savvy AI leader must be found in the institution, who does not have to be a departmental or institutional head and may be anyone as long as they are knowledgeable and enthusiastic in driving this change,” suggesting that expertise and AI competence may play a more important role than hierarchical position in leading such a massive project. A significant hierarchical drift emerged, where participants identified a shift from top-down management to distributed leadership. Technical proficiency, often found among junior staff or frontline clinicians, superseded traditional seniority. This creates a sociotechnical tension where the expert is no longer defined by years of clinical experience, but by their ability to navigate and implement AI tools, effectively decentralizing power within the hospital structure. Regardless of the approach, participants stressed that team building and strategic planning were crucial components for effective AI integration. Putting together committed cross-functional teams with people with a range of leadership philosophies, operational expertise, and clinical and AI understanding is essential. Forming a team is an effective organizational strategy to integrate AI practices, promote group ownership, and moderate possible resistance to change.

Capacity building of healthcare providers and students

Based on the interview data, limited infrastructure and shortage of AI-specialized professionals were the areas identified for attention. Participants consistently underlined that several training exercises prior to implementation were necessary to acquaint stakeholders and everyone else with AI to reduce errors, foster trust and confidence and clarify shifting roles brought by AI tools. According to the responses, the success of integrating AI into healthcare would also depend on trained personnel. It is crucial to have instructors or hospital officials who are knowledgeable about AI technology as role models and to instruct peers and pupils. Besides trainings, concerns about ethical issues prompt calls for regulatory policies, procedures and rigorous validation of AI tools to ensure reliability, patient safety and efficacy prior to use. Once formulated all involved (medical teachers and students) must be aware and practice these ethics in AI related regulations and policies. Validation of AI tools was framed not only as technical prerequisite but a confidence building mechanism to allowing healthcare providers to use AI tools with reduced fear of errors. Participants identified how AI integration can be a threat to patient care if healthcare providers are replaced with non-validated AI tools. One participant (P14) stated that “validating AI algorithms is crucial. Many overlook its importance and ethical implications for patients” highlighting concerns about accountability and responsible implementation. Similarly, another participant (P5) emphasized that “AI integration can jeopardize patient care if we gatekeepers are replaced with non-validated automated solutions” underscoring the importance of clinical validation and regulatory oversight prior to clinical adoption. Hence, the staff must be aware of AI tools validation processes and policies before implementing it. A significant thematic drift emerged within capacity building, where the focus expanded from individual technical proficiency to institutional policy literacy. Participants identified building capacity as a process of understanding the regulatory and ethical boundaries of AI use, shifting the clinician’s role from a simple end-user to a policy aware practitioner. This creates a tension between the desire for rapid innovation and the necessity for procedural compliance, where the speed of AI adoption is naturally constrained by the workload of policy validation. Participants assessed financial resources as necessary not just for acquiring AI technologies and an effective integrated clinical database, but also for sustaining training validation and system maintenance, in addition to cyber protection. Participants noted that AI algorithms require clean, curated clinical data. The Electronic Health Record (EHR) offers hope by centralizing medical records and data, improving accessibility and efficiency. Participants identified gaps in technical expertise and infrastructure as barriers to AI integration in healthcare. As one of the participants (P4) stated; “there is a deficiency of expertise, resources, and finances. Achieving progress requires investment in human resources and systems like EHR,” reflecting concerns of readiness of institutions to support this change.

Workshop findings

A workshop to strategize AI implementation in healthcare was held at AKU in collaboration with the Quality Network of Quality, Teaching, and Learning (QTL_Net) AKU. A diverse group of 22 individuals, including research analysts, faculty, PhD students, and undergraduate and postgraduate medical students from various departments (information technology, internal medicine, oncology, pathology, radiology, and psychiatry) participated in the workshop. Table 2 describes the collected data from this workshop concerning the identification of SWITCH components.

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. SWITCH components for change management to integrate AI in education, service and research in a hospital identified by workshop participants.

https://doi.org/10.1371/journal.pone.0346582.t002

The action plan for institutional change

An analytical framework that captures how healthcare providers think about, negotiate, and react to AI in their current situations was developed from participant narratives. The proposed action plan conceptualized AI integration as multidimensional change process comprising of three domains: strategic actions for clinicians and healthcare providers (Direct the Rider), change pathway enablement and motivation and attitude (Motivate the Elephant) and environmental readiness (Shape the Path). The resulting action plan represents an analytic synthesis of the qualitative themes identified in Phase I. The action plan represents an analytic synthesis of the primary study themes, including opportunities for improvement, reluctance to change, leadership styles, and capacity building. By mapping these emergent findings onto the SWITCH framework, the plan moves beyond a generic policy guideline to provide a contextually grounded roadmap for institutional change. The proposed action plan for incorporating AI into research, education, and services at an academic healthcare facility is as follows:

Direct the rider

1. Engagement with AI Experts: To direct the rider, the action plan includes engagement with AI experts to provide the technical clarity and strategic guidance identified as necessary during the interviews. This action is an analytic synthesis of the capacity building theme, addressing the knowledge gaps reported by participants regarding AI integration and ethical governance. The need for multidisciplinary collaborations and the engagement of AI experts is highlighted by the fact that clinical expertise alone is insufficient to address the technical, ethical, and infrastructural aspects of AI-healthcare systems. AI experts can offer important guidance and expertise in implementing AI solutions, assisting in the successful navigation of complex technologies and methodologies. Collaborating with interdisciplinary teams, local and external AI experts, and individuals with experience ensures the accuracy, reliability, and safety of AI-driven healthcare solutions, leveraging diverse perspectives and resources for successful implementation.
2. Curriculum Integration: The analysis highlights how a thorough medical curriculum (undergraduate and graduate) that covers both theoretical knowledge and real-world applications in healthcare settings must incorporate AI concepts and skills. To direct the rider, the plan proposes curricular reforms as a primary strategy for capacity building. This action is an analytic synthesis of participant concerns regarding the current lack of technical readiness. Specialized AI-related courses and continuous professional development (CPD) for physicians, supervisors, and teachers were identified as crucial enablers in enhancing both foundational knowledge and practical skills in the application of AI in healthcare. Capacity building and training in essential AI concepts such as machine learning, deep learning, and neural networks are emphasized as fundamental to the integration of AI in healthcare. These concepts support various applications including diagnostics, personalized medicine, predictive analytics, data management, medical image analysis, and clinical decision support systems. By integrating AI competencies into the formal curriculum, the framework ensures that future healthcare professionals possess the foundational knowledge required for ethical and effective AI utilization.
3. Ethical, Legal, and Regulatory Considerations: By providing concrete legal guidelines, the plan gives the ‘Rider’ the technical clarity needed to proceed with AI adoption without the barrier of ambiguity. This is an analytic synthesis of the institutional challenges identified by participants, specifically the uncertainty regarding data privacy and liability. The action plan emphasizes the importance of ethics as a critical domain in the planning process for bringing AI in healthcare. It highlights the necessity to discuss and address ethical issues, including patient privacy, data security, algorithmic bias, and accountability. The study reveals a significant gap between the concerns surrounding AI in healthcare and the actual needs of healthcare providers. It highlights the necessity for an inclusive policy framework aimed at standardizing operations and addressing the potential risks noted by study participants. Such a policy should focus on protecting both patients and healthcare professionals by promoting the responsible advancement and application of AI technologies, while also safeguarding privacy, autonomy, and data security. Adherence to ethical standards will also help uphold equity in healthcare delivery and comply with legal and regulatory requirements.
4. Strategic Prioritization: The findings highlight a sequential approach to prioritizing AI applications within organizations. This includes assessing their impact, scalability, feasibility, and alignment with organizational goals. Additionally, fostering a supportive AI culture and focusing on outcomes rather than processes in strategic planning are emphasized as crucial elements of this approach. To operationalize this component there is value in leveraging existing AI platforms in the institute like Computerized Physician Order Entry (CPOE), AKU’s Center of Innovation and Medical Education (CIME), Total Laboratory Automation (TLA), Virtual Learning Environment (VLE) and the ongoing initiative of AKU of Electronic Health Record systems. This approach not only minimize redundancy but capitalize on staff familiarity. One such example is AI-supported simulators in CIME which are in current times enabling professionals to safely practice procedures using real-world scenarios, utilizing AI algorithms for training purposes. The integration of CIME with VLE can be used as a medium to certify and re-certify for credentialling of faculty and staff to provide safe care to the patients. The VLE can also be strategically strengthened to provide clinicians with access to teaching and learning resources, case studies, and examples of clinical best practices. These examples are presented as illustrative options, showing possibilities or approaches. This approach directly addresses the capacity building theme by providing the clear roadmap that institutional leaders often lack. By dividing projects into quick wins (high-impact, low-effort) and long-term goals, the framework gives staff the specific, measurable targets needed to track progress and stay accountable.
5. Objective Setting and Measurement: Importance of clear evidence-based objectives ensures alignment between strategy and execution of AI integration in the healthcare institutes. Specifying measurable objectives along with key performance indicators (KPI) is crucial for the successful implementation of AI, as it allows for the development of realistic timelines and key milestones. Furthermore, it establishes clear metrics or KPIs for evaluating performance effectively and will strengthen governance. This is an analytic synthesis of the institutional challenges and capacity building themes, addressing the lack of measurable objectives found in the data. By defining specific technical and clinical targets, the framework provides the rider with the necessary clarity to evaluate success and maintain accountability.
6. Pilot Programs and Prototyping: The action plan incorporates pilot projects as an essential stage for feasibility assessment. This is an analytic synthesis of the institutional challenges and capacity building themes, as it allows for the iterative testing of protocols and data collection instruments in a controlled setting. By identifying technical flaws or workflow deficiencies early, the pilot provides the ‘Rider’ with the empirical evidence needed to refine the plan before full-scale implementation. These pilot studies are intended to evaluate viability, gather feedback, and refine implementation plans. Suggested actions include integrating AI into existing initiatives like EHR, CPOE, and AI-supported simulators in CIME and VLE to enhance healthcare outcomes through targeted projects.