Clinical AI and Downstream Care Capacity in Viet Nam: Managing Diagnostic Demand, Bottlenecks and Patient-Safety Risks
IMPLEMENTATION INSIGHTS
Clinical AI and Downstream Care Capacity in Viet Nam: Managing Diagnostic Demand, Bottlenecks and Patient-Safety Risks
IMPLEMENTATION INSIGHTS
Last reviewed: 8 October 2026
A hospital introduces an AI application to help identify suspicious abnormalities in diagnostic images. The technology performs well during evaluation, and clinicians begin identifying more findings that may require further investigation.
Initially, the results appear encouraging. Potentially important abnormalities receive attention earlier, and more patients are referred for additional assessment. However, the number of appointments available for confirmatory imaging, specialist consultations and biopsy procedures remains unchanged. Waiting times begin to increase, while clinical teams face growing pressure to accommodate the additional workload.
The AI may be performing its intended function. Yet the hospital may not have sufficient capacity to provide the care that follows.
This illustrates an important challenge in responsible healthcare AI implementation: Identifying more patients who may need care does not automatically mean that the healthcare system can assess, diagnose and treat them more effectively.
Clinical AI may improve the detection of abnormalities, support risk assessment and help clinicians prioritize attention. However, the value of these capabilities depends partly on whether subsequent services can accommodate the clinical demand created, accelerated or redirected by the technology.
The central governance question is therefore not simply whether AI improves diagnostic performance or workflow efficiency. It is whether the healthcare organization can respond to the resulting clinical needs without creating unacceptable delays, unnecessary procedures or additional risks for other patients.
For healthcare organizations in Viet Nam, this means looking beyond the immediate AI application to understand its effects on patient flow, diagnostic demand, specialist availability and the capacity of services that patients may need next.
Clinical AI can change demand in different ways
Healthcare organizations sometimes expect AI to improve efficiency and reduce workload. In practice, its effects depend on what the technology does and how it changes clinical decisions.
An AI application that helps identify previously overlooked abnormalities may increase the number of patients requiring further investigation. If those findings are clinically meaningful, the additional demand may be appropriate and beneficial. Another application may generate more positive findings without a corresponding increase in confirmed disease, potentially leading to additional examinations, specialist referrals or procedures that ultimately prove unnecessary.
AI may also change the timing of demand rather than its total volume. Earlier identification of potentially important findings can bring forward appointments that would otherwise have occurred later. A service may therefore experience a short-term increase in referrals even when the total number of patients requiring care changes relatively little.
In other circumstances, AI may reduce downstream demand. An appropriately evaluated application could help avoid unnecessary repeat examinations, support more selective referrals or improve processes that previously delayed care.
These effects should not be treated as interchangeable. They depend on the system's intended purpose, performance in the local patient population, clinical thresholds, professional decisions and the workflow into which the technology is introduced.
For example, AI identifying suspicious pulmonary findings may influence demand for CT scans and respiratory consultations. An AI-assisted endoscopy system may affect the number of lesions requiring professional assessment, removal or histopathological examination. A diabetic retinopathy screening application may change the number and urgency of referrals to ophthalmology services.
Each application creates a different pattern of clinical activity. Even hospitals using the same technology may experience different consequences because their patient populations, clinical practices and available services are not identical.
Healthcare organizations should therefore avoid assuming that better detection automatically leads to better outcomes or that every additional AI-generated finding represents a patient who needs a new procedure.
The first question is what clinical activity AI is expected to change and which subsequent services will need to respond.
Downstream capacity is more than the availability of equipment
When preparing to introduce AI, healthcare organizations may focus on technical requirements such as network connectivity, software integration, servers and access to clinical information. These are important, but they do not establish whether the organization can manage the clinical consequences of AI use.
Downstream care capacity refers to the practical ability of the healthcare system to provide the examinations, professional assessments, clinical decisions and treatments that may become necessary after an AI-supported finding.
For a diagnostic imaging application, this may involve additional imaging appointments, specialist interpretation, biopsy procedures, pathology services and clinical consultations. In oncology, the pathway may extend to multidisciplinary assessment, molecular testing, surgery, systemic treatment, radiotherapy and supportive care.
Capacity is not simply the number of machines or examination rooms available. It also depends on qualified personnel, operating schedules, consumables, equipment maintenance, laboratory turnaround times and coordination between services.
A hospital may own a CT scanner but have limited appointment availability because the equipment is already heavily utilized. A pathology laboratory may have sufficient physical space but limited specialist interpretation capacity. An oncology department may identify additional patients requiring treatment while infusion appointments or radiotherapy sessions remain constrained.
The capacity a service can reliably provide may therefore be substantially lower than its theoretical maximum.
Healthcare organizations also need to consider how resources are shared. An additional imaging appointment allocated to a patient with an AI-supported finding may occupy a slot that would otherwise have been available to someone else. A faster diagnostic process may improve access for one group while increasing waiting times elsewhere.
This does not mean clinically necessary referrals should be restricted simply because they create additional demand. Rather, the organization needs to understand how competing demands can be accommodated according to appropriate clinical priorities.
Responsible AI implementation requires attention to the services receiving additional clinical work, not only the technology generating the initial finding.
What international research shows about system-level bottlenecks
A study published in the Journal of Medical Internet Research in May 2026 provides a useful illustration of how AI-related changes can affect an entire healthcare service [1].
May and colleagues developed a proof-of-concept framework combining scenario planning with discrete-event simulation to examine AI automation and data interoperability in oncology.
The researchers modeled patient movement through a capacity-constrained cancer care system. The simulated pathway included administrative processing, diagnostic imaging, biopsy, pathology, molecular testing, multidisciplinary coordination and treatment initiation. Four scenarios represented different combinations of AI automation and data interoperability.
Within the model, higher levels of automation combined with effective interoperability improved patient flow and reduced the simulated time from referral to treatment. By contrast, isolated automation without sufficient data interoperability increased administrative verification work and worsened overall system performance.
The simulation also illustrated a phenomenon known as bottleneck migration. When upstream diagnostic processes became more efficient, the service limiting patient flow could shift elsewhere.
In one modeled scenario, the main capacity constraint moved from diagnostic CT scanners to chemotherapy infusion units. The model also incorporated additional demand associated with false-positive AI-supported triage results, illustrating how confirmatory investigations with limited clinical yield could consume resources needed by other patients.
These findings are relevant because they show how improving one process may create pressure on another service. However, the study's limitations are important.
It used a stylized simulation with assumed capacities, service times and operational mechanisms. The model was designed to explore possible system behavior rather than reproduce the performance of a particular hospital. It was not calibrated to demonstrate actual outcomes in Viet Nam, and its numerical results should not be used to predict how many additional patients, appointments or treatment delays a Vietnamese healthcare organization will experience.
The simulation also did not directly evaluate patient outcomes, financial costs or differences in access between population groups. These remain important considerations for real-world implementation.
The broader lesson is that improving an upstream process does not necessarily improve the entire care pathway when downstream services are already operating near their practical capacity.
An AI implementation may therefore appear successful when evaluated within one department while creating new delays elsewhere.
Better detection requires accessible diagnostic and treatment services
The relationship between early detection and downstream care capacity is not new.
WHO's Guide to Cancer Early Diagnosis, published in 2017, emphasizes that earlier recognition of cancer-related symptoms needs to be supported by timely diagnostic evaluation, referral and access to treatment [2].
Its 2020 guide on screening programmes similarly stresses that screening should be assessed in relation to demonstrated benefits, potential harms, available resources and quality assurance arrangements [3].
Early diagnosis and population screening are not the same activity. Early diagnosis focuses on identifying and evaluating disease in people who have symptoms or clinical concerns, while screening generally aims to identify possible disease or elevated risk among people without recognized symptoms. Nevertheless, both depend on the availability of appropriate confirmatory assessment and subsequent care.
These principles are relevant to clinical AI, although the WHO publications are not specific AI governance frameworks.
An AI application identifying a suspicious abnormality does not establish a confirmed diagnosis or ensure that appropriate treatment will follow. Additional clinical assessment may be necessary, and subsequent investigations may have their own risks, costs and capacity requirements.
If patients cannot access those services, the potential benefit of earlier recognition may not be realized. Conversely, increasing the number of findings without appropriate clinical indications can expose patients to unnecessary examinations, anxiety or procedures.
It is also important to distinguish false-positive findings from overdiagnosis. A false positive suggests the presence of a target condition that is not confirmed through an appropriate reference standard. Overdiagnosis concerns the identification of a genuine condition that would not have caused symptoms or harm during the patient's lifetime. Both can contribute to unnecessary care, but they are different problems and require different evidence to assess.
The objective is therefore not to maximize the number of positive AI outputs. It is to improve appropriate patient care while minimizing avoidable harm.
Why downstream capacity matters in Viet Nam
Healthcare organizations in Viet Nam differ considerably in specialist staffing, equipment, patient volumes, information systems and referral arrangements.
Some large hospitals provide extensive diagnostic and treatment services within the same organization. Other hospitals and clinics depend on external laboratories, specialist centers or referral facilities for particular examinations and treatments.
An AI application introduced at one healthcare facility may therefore create demand that another organization must accommodate.
For example, a clinic using AI-assisted diagnostic imaging may identify more patients requiring specialist assessment. If confirmatory imaging or biopsy services are available only at a referral hospital, the value of earlier recognition depends partly on whether patients can access those services.
Similar concerns may arise in AI-supported diabetic retinopathy screening, cancer-related diagnostics and other activities that can increase referrals for specialist care. The relevant constraints may differ according to the location of the facility, available workforce, service capabilities, patient financial circumstances and access to follow-up.
Published research provides some context.
A study by Nguyen and colleagues, published in JCO Global Oncology in February 2025, examined the availability and geographic distribution of radiotherapy and nuclear medicine equipment in Viet Nam [4]. Using data collected through October 2024, the researchers identified equipment shortages and uneven distribution that could affect access to cancer services.
These findings remain relevant, but they should not be treated as a complete or current inventory of radiotherapy and nuclear medicine capacity across Viet Nam in October 2026.
Healthcare capacity continues to change. On 10 March 2026, the Ministry of Health reported that Viet Nam National Cancer Hospital, commonly known as Hospital K, had brought two new-generation Versa HD linear accelerators and other medical equipment into operation at its Quán Sứ campus [5].
This development illustrates ongoing investment in clinical capacity. It does not, by itself, establish the extent to which national demand for cancer treatment is being met.
The practical implication is that healthcare organizations should rely on current, service-specific information rather than assume that published national averages or historical equipment counts accurately describe their own circumstances.
The important question is not whether every Vietnamese hospital faces the same capacity limitation. It is whether each organization understands which services its AI application may affect, who provides those services and whether patients can access them within clinically appropriate timeframes.
Estimate how AI may change the number of patients requiring care
Before deploying AI that may influence diagnostic or referral activity, healthcare organizations should consider how the technology could change demand for subsequent clinical services.
The assessment should begin with current activity. How many patients undergo the relevant examination? How often do clinicians identify suspicious findings? What proportion of those findings leads to additional imaging, specialist consultation, pathology or treatment?
These questions establish a starting point for understanding what may change after AI introduction.
If AI identifies additional abnormalities, some findings may represent patients who could benefit from earlier diagnosis. Others may be false positives, incidental findings or abnormalities that do not require intervention.
The number of additional referrals cannot be estimated reliably from AI sensitivity alone. Disease prevalence, specificity, thresholds, patient characteristics, professional interpretation and referral criteria all influence how many patients may require follow-up.
The organization should also distinguish between AI findings and individual patients. One patient may have several findings but need only one specialist appointment. Another may require multiple investigations before a final clinical decision is possible.
Demand should therefore be estimated in terms of actual clinical services rather than simply counting alerts, suspected lesions or positive AI outputs.
A useful approach is to examine the sequence from an AI-supported finding to professional assessment, a clinically justified referral and the actual service required. Each step may change the number of patients who proceed further.
The timing of demand is equally important. An AI application may not substantially increase the total number of patients requiring a service over a year, but it may bring more patients forward during a shorter period.
Where reliable local data are unavailable, healthcare organizations can use plausible scenarios. These might include little change in total demand, a moderate increase in clinically appropriate referrals or a more substantial increase requiring additional confirmatory testing.
Such scenarios are planning assumptions, not validated predictions. Their purpose is to understand the range of changes the service may need to accommodate rather than forecast one precise number that creates false confidence.
Identify where the next bottleneck could develop
Once the likely effects on demand are understood, the organization should identify the services most likely to experience additional pressure.
Consider an AI application that identifies suspicious pulmonary findings. Patients may need CT imaging, respiratory consultation, biopsy or other investigations. The availability of these services influences how quickly a possible finding can be evaluated.
If CT appointments are readily available but biopsy services are already operating near capacity, the main delay may occur after imaging. If biopsy capacity improves but pathology interpretation remains constrained, the bottleneck may shift again.
In oncology, further constraints can arise during multidisciplinary review, treatment planning or the delivery of surgery, systemic therapy or radiotherapy.
This is why evaluating only the capacity of the immediately following service may be insufficient. The organization needs to understand the practical sequence of services that patients are expected to use and where waiting times may increase.
Some services may have adequate average capacity but still struggle during predictable periods of high demand. Pathology workloads can vary according to procedural schedules, specialist clinics may operate on limited days, and some examinations may depend on particular personnel or equipment.
Capacity planning should therefore consider both routine workload and the service's ability to respond to fluctuations.
A service operating near its practical capacity may also have little flexibility to accommodate urgent patients, equipment interruptions or temporary staff shortages. Additional demand that appears manageable on paper may create substantial delays when these everyday operational realities are considered.
An AI application may improve the speed of an upstream process while leaving the overall pathway constrained by a department that was not involved in the original technology decision.
The clinically important bottleneck is the service limiting timely patient care, not necessarily the department where AI is deployed.
More findings do not always mean more appropriate care
A capacity review should not focus only on whether additional appointments can be provided. Healthcare organizations also need to determine whether the clinical activities generated by AI are appropriate.
An AI-positive finding may require professional review before further investigation is justified. The need for additional imaging, biopsy or specialist consultation should depend on the relevant clinical indication and applicable professional standards.
Without appropriate review, AI may contribute to unnecessary diagnostic activity. For individual patients, this can involve procedural complications, incidental findings, psychological distress and additional financial burden. For the healthcare organization, unnecessary investigations may occupy capacity needed by patients with stronger clinical indications.
For example, if an AI application increases referrals for confirmatory testing but a substantial proportion of those referrals is not clinically indicated, the additional workload may reduce access for other patients.
Conversely, restricting referrals simply to protect capacity may also be unsafe if patients genuinely need further assessment.
The objective is not to maximize or minimize referral volume as an isolated performance measure. It is to ensure that referrals are clinically justified and that patients who need further care can receive it appropriately and in a timely manner.
This requires distinguishing between an AI-generated finding, a professional decision that further assessment is indicated and a confirmed diagnosis.
Healthcare organizations should evaluate how AI changes these decisions rather than measuring success only by the number of patients entering the next service.
Capacity constraints can affect patients outside the AI-supported service
AI-related demand may affect patients who never directly use the technology.
A hospital may introduce AI to improve identification of a particular condition and allocate more specialist appointments to patients with AI-supported findings. If available capacity remains unchanged, other patients may experience longer waiting times. Some of those patients may have equally important or more urgent clinical needs.
This creates a patient-safety concern that cannot be assessed solely by examining patients who received an AI-positive result.
Healthcare organizations should consider how the additional demand affects everyone sharing the relevant resources. Potential consequences include delayed appointments, displacement of other clinically necessary services, longer diagnostic intervals and reduced availability for patients requiring repeated follow-up.
These effects should not be assumed to occur in every implementation. They depend on available capacity, changes in demand and how appointments are managed. Nevertheless, they are foreseeable possibilities that deserve consideration before expanding a service.
Clinical prioritization should remain grounded in appropriate patient needs and professional judgment, not simply in whether a finding originated from AI. An AI-positive result should not automatically displace another patient whose condition requires more urgent attention.
Access also matters. A patient who lives near a major hospital may have a different experience from someone who must travel a considerable distance for specialist assessment. An AI application may identify a need for care accurately without resolving the barriers that prevent the patient from receiving it.
The relevant equity concern therefore extends beyond differences in model performance. It also includes whether patients can complete the services that follow an AI-supported finding.
Capacity may depend on another healthcare organization
Not every healthcare facility can provide all subsequent clinical services internally.
A clinic may perform an initial examination but refer patients for specialist consultation or confirmatory testing. A hospital may outsource certain laboratory services or depend on another institution for treatment modalities that are unavailable locally.
In these arrangements, downstream capacity may be outside the direct control of the organization deploying AI.
A screening service may have sufficient personnel and technology to assess a large number of patients but depend on an external specialist clinic to evaluate positive findings. If referrals increase beyond the receiving service's capacity, earlier identification may not result in timely assessment.
Healthcare organizations should therefore consider realistic appointment availability, referral volume and accessibility of external services that may be affected.
Where appropriate, participating organizations may need to discuss how increased demand will be accommodated and how clinically important delays will be recognized.
The relevant arrangements must remain consistent with applicable legal and professional responsibilities. This does not mean one organization automatically becomes responsible for every subsequent service, nor does it mean contractual arrangements can eliminate responsibilities established by law.
The distinctive capacity question is whether the services required after an AI-supported finding can realistically accommodate the patients being referred, including when those services belong to another organization.
A healthcare organization should understand this dependency before promising that AI-supported detection will necessarily lead to faster diagnosis or treatment.
What Viet Nam's current legal framework means
Viet Nam's Law No. 134/2025/QH15 on Artificial Intelligence has been effective since 1 March 2026 [6].
Article 4 establishes human-centred principles, including that AI should serve people rather than replace human authority and responsibility. Article 6 requires stricter risk management appropriate to essential sectors directly affecting life, health and other protected interests. For healthcare, the Law specifically refers to patient safety, reliability under real-world conditions and protection of health data [6].
These provisions are relevant when AI influences clinical decisions or changes the number and urgency of patients requiring further care. However, they should not be interpreted as establishing a statutory formula for calculating additional appointments, staffing requirements or diagnostic capacity before every AI deployment.
Decree No. 142/2026/NĐ-CP, effective since 1 May 2026, provides further requirements concerning AI risk classification, conformity assessment and the responsibilities of relevant participants [7].
The applicable obligations depend on the system's legal classification, the roles of providers and deployers, and its actual functions and deployment arrangements. Where deployment or changes to an AI system create new or higher risks under the relevant legal conditions, provisions concerning risk-classification review may become applicable. An increase in patient referrals does not, by itself, automatically trigger statutory AI reclassification in every situation.
Decision No. 33/2026/QĐ-TTg, effective since 15 August 2026, establishes Viet Nam's current list of high-risk AI systems [8]. Its healthcare-specific entries concern defined AI-assisted surgical and robotic-surgery systems meeting particular conditions.
An AI application used in diagnostic imaging or screening should not automatically be described as legally classified high-risk merely because it influences clinical decisions or changes demand for further services. However, an application that falls outside the statutory high-risk list may still have important patient-safety implications.
The Law on Medical Examination and Treatment No. 15/2023/QH15, as amended, provides a foundation for professional responsibility and patient safety [9].
Article 71 addresses the prevention of medical incidents through identification, reporting, analysis of causes and preventive measures. Circular No. 19/2013/TT-BYT continues to provide a framework for hospital quality management [10].
Where AI-enabled software or equipment is regulated as a medical device, the applicable medical-device framework must be considered separately, including Decree No. 98/2021/NĐ-CP, as amended, and Circular No. 24/2026/TT-BYT where relevant [11, 12].
When services involve processing or exchanging personal information, applicable data-protection requirements may also need to be considered, including Law No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP, both effective since 1 January 2026 [13, 14].
These legal frameworks apply according to the activities and circumstances involved. They do not prescribe one universal approach to downstream capacity assessment for every clinical AI application.
The downstream capacity review proposed in this article is a practical patient-safety and operational governance approach. It is not a universal statutory requirement or prescribed Ministry of Health approval procedure for clinical AI.
Plan for uncertainty rather than one optimistic forecast
Healthcare organizations should be cautious about basing implementation decisions on a single expected increase in demand.
Actual referral patterns may differ from initial estimates because of local disease prevalence, AI performance, clinical thresholds, staff behavior, examination volume and patient participation. A system may produce fewer additional findings than anticipated, while another may generate more referrals during early deployment.
Demand may also change as clinicians gain experience with the application or as the characteristics of the patient population change.
Scenario-based planning can help organizations prepare for this uncertainty.
A relatively conservative scenario may assume little change in the total number of patients requiring further assessment, with AI mainly influencing the timing or priority of existing referrals. Another may assume a moderate increase in clinically appropriate follow-up. A more demanding scenario may examine the consequences of substantially higher referral volumes or more patients requiring confirmatory investigations.
The organization should also consider what happens if anticipated efficiencies do not materialize. A new AI application may initially require additional professional review and increase workload before the intended operational benefits are achieved.
These scenarios should use the best available local data and relevant evidence. They should not be presented as validated forecasts simply because they have been incorporated into a planning document.
The objective is to understand whether the organization can respond to plausible changes in demand without creating unacceptable delays.
A practical AI-Driven Clinical Demand and Capacity Readiness Review
Healthcare organizations in Viet Nam can assess these issues through an AI-Driven Clinical Demand and Capacity Readiness Review.
This is proposed as a practical organizational method. It is not a prescribed Vietnamese regulatory framework, Ministry of Health form, WHO standard or international certification requirement.
The review should begin with one clearly defined clinical AI application and the services that may be affected after it produces a relevant finding.
The organization should identify what changes in clinical activity are expected. These may include additional confirmatory examinations, specialist consultations, repeat investigations, treatment referrals or changes in the timing of existing appointments.
The next consideration is how much demand may change. Rather than counting AI-positive outputs alone, the organization should estimate how many patients may require each subsequent service after appropriate professional assessment.
The review should distinguish clinically meaningful additional findings from false positives, incidental findings and cases where further investigation is not indicated.
The organization should then examine the practical capacity of the services receiving additional demand. Relevant considerations may include appointment availability, qualified personnel, equipment, procedure rooms, laboratory capacity, operating schedules and the amount of capacity already committed to existing patients.
Where follow-up takes place at another healthcare organization, the assessment should consider the capacity and accessibility of that service rather than assuming that a referral destination has sufficient availability.
The next task is to identify where delays could develop. The organization should examine whether additional demand can be accommodated without unacceptable waiting times, excessive pressure on a particular department or displacement of other clinically necessary care.
The review should also consider how the organization will respond if demand exceeds expectations. Depending on the circumstances, this may involve increasing relevant service capacity, improving referral arrangements, introducing additional clinical sessions or adjusting the pace of AI expansion.
These decisions should remain consistent with clinical responsibilities. Capacity limitations should not be used to justify withholding urgently required assessment or treatment.
Finally, the organization should determine how it will monitor the actual effects of implementation. This includes whether patients receive the intended follow-up, how long they wait, where backlogs emerge and whether other patient groups experience reduced access.
The findings should inform decisions about maintaining, modifying or expanding the AI service.
For a relatively simple application, a focused discussion involving clinical, operational and quality leaders may be sufficient. A more complex service affecting several departments may require structured capacity analysis and coordination with external providers.
The depth of the review should reflect the potential effect of AI on patient flow and the consequences of delayed care.
The purpose is not to construct a sophisticated forecasting model for every AI application. It is to ensure that healthcare leaders have examined whether the services required after AI-supported detection can realistically be delivered.
Monitor whether patients actually receive the next service
After implementation, healthcare organizations should evaluate whether anticipated changes in clinical demand are occurring and whether subsequent services can respond appropriately.
The number of positive AI findings may be useful information, but it is not a sufficient measure of success.
A clinically meaningful assessment should consider how many patients require further investigation after professional review, how many receive the recommended service and whether that assessment occurs within an appropriate timeframe.
For example, an AI application may identify more suspicious findings, but some patients may not complete the recommended CT scan or specialist consultation. Others may undergo additional testing but experience substantial delays before a diagnosis is confirmed.
The organization should distinguish between identifying a possible need for further care, arranging the service and completing the clinically indicated assessment.
Useful measures may include additional referral volumes, the proportion of referrals judged clinically appropriate, appointment availability, waiting times for confirmatory testing and the number of patients whose recommended follow-up remains incomplete.
Where relevant, the organization may also examine the interval between an initial AI-supported finding and a confirmed diagnosis, treatment decision or treatment initiation.
These measures should be selected according to the clinical service. They are not universal reporting requirements.
Average waiting times may also conceal important problems. A service may maintain a reasonable average while a smaller number of patients experience substantial delays. The distribution of waiting times and cases exceeding clinically justified expectations may therefore provide important safety information.
Where feasible, healthcare organizations should also examine whether patients outside the AI-supported pathway experience changes in access. For example, if a new AI programme increases demand for specialist appointments, the organization may need to determine whether other patients are waiting longer for the same service.
Not every change can be attributed to AI. Staffing, equipment availability, seasonal demand and concurrent service improvements may also influence outcomes.
Nevertheless, monitoring should help leaders understand whether AI implementation is contributing to better patient flow or creating pressures that require intervention.
The clinically important result is not simply how many additional findings AI generates, but whether patients receive appropriate subsequent care safely and in a timely manner.
Capacity constraints should influence implementation decisions
A healthcare organization may find that an AI application performs well but generates more demand than subsequent services can accommodate.
This does not automatically mean the technology should be rejected. It may indicate that additional preparation is needed before wider deployment.
Possible responses include increasing clinical capacity, strengthening referral arrangements, reviewing the appropriateness of follow-up requests or coordinating AI introduction with other planned service improvements.
An application intended to improve early recognition of a particular condition may deliver greater value when introduced alongside improvements in confirmatory testing or specialist appointment availability.
However, implementation decisions must remain clinically responsible.
Restricting AI use solely to keep referral numbers low can be problematic if the system is identifying patients with genuine clinical needs. Conversely, expanding an AI service without a realistic response plan can expose patients to delays and unnecessary investigations.
Healthcare leaders should distinguish between changing the pace of technology expansion and denying clinically necessary care to patients who have already been identified as requiring assessment.
If demand exceeds safe operational capacity, the organization should respond through appropriate clinical prioritization, capacity management and escalation rather than leaving an increasing backlog unmanaged.
Decisions to expand, modify or restrict the AI service should consider the evidence of benefit, patient-safety risks and the organization's ability to maintain continuity of care.
International guidance supports a health-system perspective
WHO's Guideline: Recommendations on Digital Interventions for Health System Strengthening, published in 2019, emphasizes that digital interventions should be evaluated in relation to health-system feasibility, benefits, harms, resource use and equity [15].
The guidance also recognizes that digital technologies are not substitutes for functioning healthcare systems.
WHO's Implementation Playbook: Delivering Impact for Health provides a broader approach to implementation, emphasizing the need to connect planning, action, progress monitoring and adjustment with meaningful outcomes [16].
These resources are not clinical AI downstream capacity standards and do not create specific regulatory requirements for healthcare organizations in Viet Nam. However, they reinforce a useful principle: digital innovation should be integrated into the practical delivery of healthcare rather than evaluated only as an isolated technical improvement.
The same principle appears in WHO's cancer early-diagnosis and screening guidance, which highlights the importance of diagnostic capacity, timely referrals and access to appropriate treatment [2, 3].
For healthcare leaders, this means that AI should be evaluated as part of the care patients actually receive. An application that improves recognition of possible disease may provide genuine value, but the organization must also understand whether patients can obtain the assessment and treatment that follow.
What healthcare organizations in Viet Nam should prioritize
Healthcare organizations do not need to begin with a complex capacity-modeling programme.
A practical first step is to identify AI applications most likely to change demand for diagnostic examinations, consultations or treatment. The organization should then determine which services receive patients after an AI-supported finding and whether those services have sufficient practical capacity.
Existing operational information may already provide a useful starting point. Appointment volumes, waiting lists, diagnostic turnaround times, referral activity and staffing schedules can help identify services operating under pressure.
Clinical leaders should be involved because the appropriateness and urgency of further care cannot be determined through capacity information alone. Healthcare quality and patient-safety teams can help identify risks associated with delays, while operational managers can assess staffing, equipment and appointment availability.
Where follow-up depends on an external organization, that dependency should also be considered.
Technology providers may contribute information about expected AI outputs and performance, but they should not be assumed to know how much additional clinical demand the healthcare organization can accommodate.
A further consideration is the difference between theoretical capacity and capacity actually available for additional patients. Routine workload, existing waiting lists, urgent cases, staff availability and other service requirements all affect how much capacity remains.
For an initial implementation, reviewing the most likely downstream bottleneck may be more useful than developing an extensive theoretical assessment of every possible resource.
As real-world information becomes available, the organization can refine its assumptions and adjust the service.
The objective is to make downstream clinical capacity a visible part of responsible AI implementation without introducing unnecessary administrative burden.
From better detection to better patient care
Clinical AI may help healthcare organizations identify abnormalities, assess risk and direct attention toward patients who require further evaluation.
An additional AI-supported finding may lead to earlier diagnosis and appropriate treatment. It may also generate unnecessary investigations, increase pressure on constrained services or remain unresolved because the patient cannot obtain further care.
The difference is not determined by AI performance alone.
Healthcare organizations need to understand how technology changes demand, whether subsequent services are available and whether patients can move from an initial finding to appropriate assessment and treatment.
For healthcare leaders, the essential question is: If AI helps us identify more patients who need further care, can our healthcare system provide that care safely, appropriately and within a clinically acceptable timeframe?
Answering this requires attention to actual service capacity, professional judgment, referral arrangements, patient access and the potential effects on other patients sharing the same resources.
Responsible implementation should therefore consider downstream capacity before deployment and continue reviewing it as clinical demand changes.
Ultimately, successful healthcare AI is not measured only by how much more it detects or how quickly it produces information. Its real value depends on whether the healthcare system can turn that information into appropriate, timely and safer patient care.
Professional education notice
This article is intended for professional education and knowledge exchange. It does not constitute individualized legal, regulatory, clinical, patient-safety, medical-device, cybersecurity, personal-data-protection, accreditation or quality-management advice.
The terms downstream clinical capacity governance and AI-Driven Clinical Demand and Capacity Readiness Review are used as practical organizational concepts. They are not defined statutory categories or prescribed documentation requirements under Vietnamese law.
The proposed review is not a Ministry of Health framework, mandatory AI capacity assessment, WHO standard or international certification requirement.
Applicable responsibilities depend on the AI system's intended purpose, regulatory status, actual workflow, professional activities, clinical services and relevant legal requirements.
The discussion of clinical capacity, referral demand, waiting times and deployment sequencing represents practical governance guidance. It should not be interpreted as establishing universal staffing ratios, referral thresholds, waiting-time limits or equipment-capacity standards for healthcare organizations in Viet Nam.
The international evidence discussed includes a proof-of-concept simulation, health-system guidance and research using historical service-capacity data. These sources have different purposes and limitations and should not be generalized automatically to the current capacity or expected performance of individual Vietnamese healthcare organizations.
The proposed approach does not replace applicable regulatory authorization, clinical evaluation, professional judgment or existing healthcare quality and patient-safety responsibilities.
References
May P, Brookman-May SD, Garrahy E, von Büren J. Sequencing AI Automation and Data Interoperability in Oncology Using a Scenario-Planning Framework Coupled With Discrete-Event Simulation: Proof-of-Concept Study. Journal of Medical Internet Research. 2026;28. Published 25 May 2026. DOI: 10.2196/92642.
World Health Organization. Guide to Cancer Early Diagnosis. Geneva: World Health Organization; 2017. ISBN: 978-92-4-151194-0.
World Health Organization Regional Office for Europe. Screening Programmes: A Short Guide. Increase Effectiveness, Maximize Benefits and Minimize Harm. Copenhagen: WHO Regional Office for Europe; 2020. ISBN: 978-92-8-905478-2.
Nguyen DV, Pham MD, Hoang TH, Nguyen BT. Radiotherapy and Nuclear Medicine Equipment Shortages in Vietnam: A Critical Gap in Cancer Care. JCO Global Oncology. 2025;11. Published online 27 February 2025. DOI: 10.1200/GO-24-00530.
Ministry of Health of Viet Nam. Thứ trưởng Thường trực Bộ Y tế Vũ Mạnh Hà: Bệnh viện K vận hành máy xạ trị gia tốc thế hệ mới giúp người bệnh được tiếp cận kỹ thuật tiên tiến ngay trong nước. Published 10 March 2026.
National Assembly of Viet Nam. Law No. 134/2025/QH15 on Artificial Intelligence. Adopted 10 December 2025; effective 1 March 2026.
Government of Viet Nam. Decree No. 142/2026/NĐ-CP Detailing a Number of Articles and Measures for Implementation of the Law on Artificial Intelligence. Issued 30 April 2026; effective 1 May 2026.
Prime Minister of Viet Nam. Decision No. 33/2026/QĐ-TTg Promulgating the List of High-Risk Artificial Intelligence Systems. Issued 30 June 2026; effective 15 August 2026.
National Assembly of Viet Nam. Law No. 15/2023/QH15 on Medical Examination and Treatment, as amended. Effective 1 January 2024.
Ministry of Health of Viet Nam. Circular No. 19/2013/TT-BYT Providing Guidance on Quality Management of Medical Examination and Treatment Services in Hospitals. Issued 12 July 2013; effective 15 September 2013.
Government of Viet Nam. Decree No. 98/2021/NĐ-CP on Medical Device Management, as amended.
Ministry of Health of Viet Nam. Circular No. 24/2026/TT-BYT on Determining Risk Levels and Management Measures for Medical-Device Products and Amending Certain Provisions of Circular No. 05/2022/TT-BYT. Issued 30 June 2026; effective 1 July 2026.
National Assembly of Viet Nam. Law No. 91/2025/QH15 on Personal Data Protection. Adopted 26 June 2025; effective 1 January 2026.
Government of Viet Nam. Decree No. 356/2025/NĐ-CP Detailing a Number of Articles and Measures for Implementation of the Law on Personal Data Protection. Issued 31 December 2025; effective 1 January 2026.
World Health Organization. WHO Guideline: Recommendations on Digital Interventions for Health System Strengthening. Geneva: World Health Organization; 2019. ISBN: 978-92-4-155050-5.
World Health Organization. Implementation Playbook: Delivering Impact for Health. Geneva: World Health Organization; 2023. ISBN: 978-92-4-008519-0. WHO online publication dated 19 August 2024.