Governing AI-Assisted Laboratory Test Ordering in Viet Nam: Clinical Appropriateness, Duplicate Testing and Patient Safety

Governing AI-Assisted Laboratory Test Ordering in Viet Nam: Clinical Appropriateness, Duplicate Testing and Patient Safety

IMPLEMENTATION INSIGHTS

Last reviewed: 21 October 2026

A patient with diabetes and chronic kidney disease returns to an outpatient clinic after a recent hospital admission. Several laboratory tests were performed during that admission, and the results are available in the electronic medical record.

During the consultation, the physician prepares to request additional blood tests. An AI-assisted clinical decision support system identifies recent results and suggests that some investigations may not need to be repeated.

The recommendation appears reasonable. The patient's condition has been relatively stable, and some previous results may still be useful. However, one measurement was obtained before a significant treatment change. Another may need to be repeated because the patient's clinical condition has changed since discharge.

If the physician accepts the AI recommendation without reviewing these circumstances, a necessary investigation could be delayed or omitted.

The opposite problem is also possible. An AI system may recommend a broad panel of laboratory tests even though recent results already provide sufficient information for the clinical decision.

These situations illustrate why appropriate laboratory test ordering involves more than avoiding duplicate requests.

Unnecessary testing can expose patients to repeated specimen collection, misleading findings, additional investigations and avoidable costs. Missing or delaying a necessary test can lead to diagnostic uncertainty, inappropriate treatment or failure to recognize an important complication.

For healthcare organizations in Viet Nam, AI may help clinicians select appropriate investigations, find existing results and recognize situations where repeated testing provides little additional value.

Its usefulness, however, depends on whether it improves clinical decisions rather than simply reduces laboratory utilization.

The central question is how hospitals and clinics can govern AI-assisted laboratory ordering so that patients receive the investigations they need, at the appropriate time, without unnecessary testing or avoidable diagnostic harm.

Laboratory test ordering is a clinical decision

A laboratory test is useful when its result can help answer a relevant clinical question or influence patient management.

A physician may request an investigation to evaluate a suspected diagnosis, assess disease severity, monitor treatment response or identify a complication. The appropriateness of that request depends on the patient's condition, the information already available and the decision the result is expected to support.

The same test may be appropriate in one situation and unnecessary in another.

For example, repeating an electrolyte measurement may be clinically justified when a patient's condition is changing rapidly or when treatment could significantly alter electrolyte concentrations. The same investigation may provide little new information during routine follow-up when the patient remains stable and a recent reliable result is available.

Ordering a test also begins a process that extends beyond the laboratory request itself.

The investigation may require patient preparation, specimen collection, transportation, processing and clinical interpretation. An unexpected result may lead to further testing, specialist consultation or treatment.

These activities are valuable when they contribute to appropriate patient care. They become problematic when testing is unlikely to provide meaningful information or when the investigation is poorly matched to the clinical question.

Diagnostic stewardship seeks to improve the appropriate use of diagnostic testing throughout the care process.

Claeys and colleagues, writing in The Journal of Applied Laboratory Medicine in 2025, described diagnostic stewardship as clinically informed changes to test ordering, processing and reporting intended to improve patient outcomes and reduce harm associated with inappropriate testing and incorrect, delayed or missed diagnoses [1].

Laboratory stewardship is closely related, with particular attention to laboratory utilization and the quality of testing processes.

For this article, the focus is deliberately limited to decisions made before testing begins: whether an investigation is needed, which test should be selected, when it should be performed and whether existing results are sufficient.

The purpose is not to examine how AI interprets laboratory findings after testing or produces diagnostic reports.

It is to understand how healthcare organizations can make laboratory ordering more appropriate, consistent and safe.

Conventional ordering rules and AI recommendations are not the same

Many healthcare organizations already use electronic systems to identify duplicate laboratory requests or tests ordered within predefined intervals.

These functions often rely on conventional clinical decision support rather than artificial intelligence.

For example, a hospital information system may identify a recent laboratory result and display it before the physician confirms a new order. Another system may warn that a test is being requested sooner than an established minimum retesting interval.

Such functions generally apply relatively explicit rules to available information.

AI-assisted applications may evaluate more complex combinations of symptoms, diagnoses, treatment history, previous laboratory findings and clinical documentation.

Some models estimate the likelihood that repeating a test will identify a clinically meaningful change. Others may suggest investigations based on patterns found in a patient's medical record.

Generative AI may also recommend laboratory tests in response to a clinician's description of symptoms or suspected diagnoses.

These technologies present different benefits and risks.

A rule-based system identifying a recent duplicate request has a different function from an AI model recommending that a laboratory investigation is unnecessary.

The latter may require broader interpretation of the patient's clinical situation and can create additional risks when relevant information is missing or misunderstood.

An application that recommends which investigations should be ordered also presents different concerns from one that simply displays the status of an existing request.

Healthcare organizations should therefore define the intended purpose of each system and assess the evidence supporting that specific function.

Evidence showing that conventional duplicate-order alerts can reduce laboratory requests does not automatically establish that machine-learning or generative AI applications make safe patient-specific testing recommendations.

The consequences of an incorrect recommendation also matter. A system that asks a physician to review an earlier result is different from one that prevents a potentially necessary investigation from being ordered.

What recent evidence tells us about laboratory ordering support

Research on computerized clinical decision support for laboratory testing has produced encouraging findings, although evidence concerning patient outcomes remains limited.

In 2021, Zare and colleagues published a systematic review in BMC Medical Informatics and Decision Making examining 16 studies of clinical decision support systems used for laboratory ordering in inpatient hospitals [2].

The review found improvements in several laboratory utilization measures, including reductions in redundant requests and testing costs, as well as improvements in guideline adherence.

However, most included studies used quasi-experimental designs. Evidence concerning patient outcomes, including mortality and hospital length of stay, was insufficient to establish broader clinical benefits.

The findings support the potential value of computerized ordering support but do not demonstrate that reducing laboratory requests necessarily improves patient outcomes.

A systematic review by Esho and colleagues, published in Pathology in 2025, provides a more recent assessment [3].

The researchers identified 592 records and included 11 studies examining the clinical, organizational and financial effects of clinical decision support in laboratory medicine.

Several studies reported improved guideline adherence or reduced inappropriate testing, but findings varied across settings. Workflow outcomes were often insufficiently assessed or less favorable than expected.

Only two included studies used randomized trial designs.

The authors emphasized that differences in technologies, clinical environments and implementation methods limited the generalizability of the available evidence.

More directly relevant to AI, Liang and colleagues published a study in NEJM AI in June 2026 describing SmartAlert, a machine-learning clinical decision support system designed to identify potentially unnecessary repeat complete blood counts [4].

The randomized pilot involved 9,270 hospital admissions across eight acute care units at two hospitals.

Among encounters where the system generated a relevant alert, the mean number of complete blood count results within 52 hours was 1.54 in the intervention group compared with 1.82 in the control group. This represented a relative reduction of approximately 15%.

The researchers did not detect statistically significant differences in selected secondary safety outcomes, including mortality, readmission, ICU transfer and length of stay.

These findings demonstrate that a machine-learning system can influence repeat laboratory ordering in a real clinical workflow.

However, the study evaluated one type of laboratory investigation, one prediction system and a defined hospital environment.

The absence of statistically significant differences in selected safety outcomes should not be interpreted as proof that the intervention cannot cause harm or that comparable reductions would be safe for every laboratory test.

A separate study by Zayed and colleagues, published in Clinical Chemistry and Laboratory Medicine in 2025, examined the performance of a large language model as a laboratory test recommender [5].

The researchers evaluated recommendations across 15 simulated clinical cases involving primary and emergency care.

The model demonstrated precision ranging from 68% to 82%, while recall was substantially lower, ranging from 41% to 51%, compared with expert consensus-derived test recommendations.

This indicates that the model did not identify all investigations included in the reference recommendations. Some of the differences involved tests categorized as conditional, reflecting the uncertainty that can accompany diagnostic decisions.

These findings matter because a system may appear selective and efficient while omitting investigations that are appropriate in the circumstances considered.

However, the study was small and simulation-based. It cannot establish real-world rates of missed testing or diagnostic harm, and its results should not be generalized to every language model.

Together, these studies suggest that digital decision support can improve selected aspects of laboratory utilization, but the evidence for improved clinical outcomes remains less established.

Successful implementation requires evaluating both the tests a system helps avoid and those it may inappropriately discourage or fail to recommend.

A repeated test is not necessarily unnecessary

Duplicate testing is an attractive target for electronic decision support because previous orders and results can often be identified through laboratory information systems.

A physician may request a test that was already performed during the current admission or a recent outpatient consultation.

If the previous result remains suitable for the clinical decision, another investigation may provide little additional value.

However, the existence of an earlier result does not automatically make a new request inappropriate.

The patient's condition may have changed, treatment may have been adjusted or the earlier specimen may have been unsuitable. Repeated testing may also be necessary to determine whether a clinical condition is improving or deteriorating.

An effective ordering system needs to distinguish an unnecessary duplicate from a clinically justified repeat investigation.

The status of the previous test also matters.

A laboratory request may have been entered but the specimen never collected. Another specimen may have been rejected. A result may still be pending, preliminary or subject to correction.

An AI application should not treat these situations as equivalent to a completed investigation with a reliable final result.

Similarly, tests with similar names may not be analytically or clinically interchangeable.

Differences in laboratory methods, specimen requirements or intended clinical uses can mean that two apparently similar requests are actually for different investigations.

A useful system should make the previous test, its status, specimen collection time and result available when generating a warning.

This information helps the physician determine whether another test would answer a new clinical question or merely repeat information already available.

The aim is not to eliminate repeated testing. It is to avoid repetition that provides little clinical value.

Test timing is part of clinical appropriateness

The appropriate interval between laboratory investigations depends on the test and the patient's clinical circumstances.

Some measurements may change rapidly and require repeated assessment over a short period. Others reflect biological processes that generally change more slowly.

Repeating a test too soon may provide little meaningful information. Delaying a necessary investigation may prevent timely recognition of deterioration or treatment complications.

Minimum retesting intervals can help identify laboratory requests that deserve reconsideration.

In 2024, the Canadian Agency for Drugs and Technologies in Health and Choosing Wisely Canada published advisory guidance on minimum retesting intervals for selected laboratory tests [6].

The guidance addresses defined clinical circumstances and emphasizes that these intervals are intended to support appropriate testing decisions.

They do not mean a test must be repeated once the specified interval has elapsed. Nor do they prohibit earlier testing when a clinically justified exception applies.

This distinction is essential when hospitals implement automated ordering controls.

A system may identify that a test was performed recently and recommend deferring another request.

That advice may be appropriate for a stable patient undergoing routine monitoring. It may be unsuitable for a patient who has developed an acute illness or experienced a significant treatment change.

The appropriate interval therefore cannot always be determined from the test name and previous collection date alone.

Healthcare organizations should examine whether AI recommendations consider the purpose of testing and whether relevant clinical exceptions can be recognized.

Where the necessary clinical context is unavailable, an advisory warning may be safer than automatically preventing the order.

The objective is to improve the timing of investigations without imposing rigid intervals that overlook individual patient needs.

The selected test must answer the clinical question

Appropriate laboratory ordering also depends on choosing the investigation most relevant to the suspected diagnosis or treatment decision.

Different tests may address similar clinical questions but vary in diagnostic performance, specimen requirements and interpretation.

An inappropriate investigation may fail to answer the question that prompted the request. It may also produce an unexpected finding that leads to further testing or treatment without improving care.

AI may support test selection by presenting relevant clinical guidance, identifying potentially inappropriate choices or suggesting alternative investigations.

However, the usefulness of those recommendations depends on whether the system correctly interprets the clinical question.

A test commonly associated with a diagnosis is not necessarily indicated for every patient with that diagnosis.

Likewise, an AI application may recommend a broad laboratory panel without explaining why each component is necessary.

This can be problematic when tests are routinely grouped into electronic order sets.

Some panels provide several clinically relevant measurements efficiently. Others may include investigations that add little information in a particular situation.

More testing does not automatically produce a better diagnostic assessment.

Hospitals should therefore examine whether AI recommendations help clinicians select tests that are likely to inform care, rather than simply reproduce common ordering patterns.

A frequently ordered investigation is not necessarily a clinically appropriate investigation.

Underuse deserves as much attention as overuse

Many laboratory stewardship initiatives focus on reducing unnecessary investigations.

However, underuse deserves equal attention because missing or delayed testing can also cause harm.

A patient may need an investigation because of new symptoms, a change in clinical condition or treatment requiring laboratory monitoring.

An AI system may overlook that need when relevant information is missing or when its recommendation relies too heavily on previous results.

For example, a system may discourage repeating a laboratory measurement because an earlier result was within the reference interval.

If the patient's condition has changed since that measurement, the previous result may no longer be sufficient.

Another application may suggest a limited selection of investigations while omitting a test that clinicians consider necessary for the current problem.

These failures are particularly difficult to detect using ordinary laboratory utilization reports.

An unnecessary test leaves an order and usually a result in the information system. A test that should have been ordered but was never requested may leave no comparable record.

Healthcare organizations therefore need methods for identifying potential underuse rather than relying only on the number of laboratory requests prevented.

Selected case reviews may examine whether patients with established clinical indications received appropriate investigations.

Hospitals may also review delayed diagnoses, unexpected deterioration or complications where absent or delayed testing could have contributed.

For AI systems that actively recommend tests, evaluation should examine both unnecessary suggestions and omitted investigations.

A system that reduces laboratory volume while increasing clinically important omissions cannot be considered successful.

The objective is appropriate testing, not simply less testing.

Existing results are useful only when they remain relevant

One potential benefit of AI-assisted laboratory ordering is helping clinicians locate results already available in the patient's record.

Relevant information may be distributed across multiple encounters, clinical departments or laboratory information systems.

A physician may not immediately recognize that a test was performed recently or that an earlier result remains relevant to the current decision.

AI may help retrieve this information and present it when a new order is being considered.

However, finding an existing result does not establish that it is appropriate for the current clinical question.

The system needs to distinguish the specimen collection date from the reporting date. It should also recognize whether the result is preliminary, final, corrected or otherwise qualified.

A test collected before a substantial treatment change may no longer represent the patient's current condition.

An earlier result may also be unreliable because of specimen quality problems or analytical limitations.

Differences between laboratories create another challenge.

The same test name may be associated with different analytical methods, units, reference intervals or specimen requirements. Results from different facilities may be comparable in some circumstances but not others.

In Viet Nam, patients may receive care from several healthcare organizations whose laboratory information systems are not fully connected.

An AI application integrated with one hospital may therefore have access to only part of the patient's testing history.

The absence of a result in the accessible electronic record should not be interpreted as confirmation that the investigation has never been performed.

Conversely, identifying an earlier result should not automatically justify canceling a new request.

The responsible clinician needs to determine whether the existing result is reliable, sufficiently current and suitable for the present clinical decision.

Laboratory professionals should help shape ordering decisions

Laboratory medicine professionals contribute expertise beyond performing tests and reporting results.

They understand analytical methods, specimen requirements, the intended clinical use of investigations and factors that influence interpretation.

This expertise can help healthcare organizations identify frequently requested tests that may provide limited value and circumstances where repeated investigations are appropriate.

Laboratory professionals can collaborate with clinical departments to review ordering patterns, develop relevant guidance and evaluate interventions intended to improve test utilization.

Their participation becomes particularly important when AI applications rely on laboratory codes and historical results.

Two investigations may have similar names while differing in method, specimen type or intended clinical use.

Incorrect mapping between a laboratory information system and an AI application may cause the system to identify a false duplicate or overlook a relevant earlier result.

Laboratory personnel can help verify these relationships and identify where additional information is needed.

They can also contribute to evaluating how AI recommendations affect specimen collection, test availability and laboratory turnaround times.

However, this does not mean every AI-generated laboratory suggestion must receive separate approval from a laboratory specialist.

The appropriate degree of consultation depends on the test, clinical circumstances, professional responsibilities and healthcare facility's operating arrangements.

The goal is to include laboratory expertise where it adds value without introducing unnecessary delays.

Restrictions must not prevent necessary investigations

Electronic ordering systems may use different approaches to managing potentially unnecessary tests.

Some display an advisory message showing the previous result and asking the clinician to reconsider the request. Others require a reason for proceeding.

More restrictive controls may prevent the order from being completed until specified conditions are met.

These approaches have different consequences for clinical care.

An advisory message may be appropriate when clinical circumstances vary substantially and an earlier result does not always eliminate the need for another investigation.

A more restrictive control may be considered where the supporting evidence is strong and the organization has established a safe exception process.

However, rigid restrictions can create patient-safety risks when urgent testing is required.

A previous laboratory result should not make a necessary investigation inaccessible simply because the usual retesting interval has not elapsed.

Healthcare organizations should establish how appropriately authorized clinicians can proceed when the patient's circumstances justify an exception.

Emergency testing should not be delayed by unnecessary administrative steps or additional AI approval requirements.

There is also evidence that repeated electronic alerts may become less effective over time.

In January 2026, Greco and colleagues published a study in Scientific Reports involving 929,808 laboratory requests across a Swiss hospital network [7].

Of these requests, 12.4% triggered an electronic warning because they did not meet locally defined minimum retesting interval criteria.

Following the warnings, 11.9% of the alerted requests were canceled. The researchers also observed differences in alert responses according to professional role and a decline in adherence over time, consistent with alert fatigue.

The study evaluated conventional electronic alerts rather than generative AI. It also did not establish that every canceled test was clinically unnecessary.

Its findings illustrate why healthcare organizations need to assess whether warnings remain useful and whether the level of interruption is proportionate to the clinical risk.

The objective should be to support better laboratory decisions, not maximize the number of warnings or canceled orders.

A practical process for reviewing AI-assisted laboratory orders

Healthcare organizations can establish a structured process for considering AI-assisted laboratory recommendations within existing clinical workflows.

The first consideration is whether the recommendation relates to the correct patient, clinical encounter and question being investigated.

The clinician needs to understand whether the system is suggesting a new investigation, identifying a possible duplicate or recommending that a request be deferred.

The next consideration is the information supporting the recommendation.

This may include symptoms, clinical findings, treatment changes and relevant previous laboratory results.

Where an earlier result is important, the system should make its collection time, status and relevant limitations available.

The clinician then determines whether the proposed investigation is likely to help answer the current clinical question.

If AI suggests canceling a repeated test, the previous result needs to be considered alongside the patient's present condition.

If AI recommends an additional investigation, the clinician should assess whether the test is indicated and whether its result is likely to influence care.

Specialized laboratory questions may warrant consultation with laboratory medicine professionals.

The authorized healthcare professional then makes the ordering decision within the applicable professional framework.

A decision to proceed despite an AI warning may be appropriate when the clinical circumstances justify testing.

Likewise, accepting a recommendation to cancel a request should not automatically be considered appropriate without evaluating whether the investigation was needed.

Where a clinically significant safety concern has influenced the decision, relevant information and necessary follow-up should be documented according to applicable requirements and organizational procedures.

The proposed process is an organizational implementation approach, not a mandatory Vietnamese Ministry of Health AI laboratory ordering protocol or an independently validated decision instrument.

Its purpose is to ensure that AI-generated suggestions are considered as part of clinical decision-making rather than followed automatically.

Ordering decisions must fit the laboratory workflow

A laboratory order begins a process that extends beyond the clinician's request.

The order must be received, the specimen collected when applicable, the investigation performed and the result communicated to the relevant clinical team.

AI-assisted ordering can influence each of these activities.

For example, a system may recommend an investigation that is not available at the facility or cannot be completed within the timeframe needed for clinical management.

Another recommendation may create additional demand for testing that requires specialized equipment or referral to an external laboratory.

The usefulness of an investigation depends partly on whether the result can be obtained while it is still relevant to patient care.

A test performed too late may provide information after the clinical decision it was intended to support has already been made.

Conversely, unnecessary urgent requests may consume resources needed for patients with more pressing clinical requirements.

Healthcare organizations should therefore examine whether AI recommendations are compatible with laboratory capabilities, specimen handling arrangements and expected turnaround times.

This does not mean that necessary testing should be suppressed when the laboratory is busy.

Patient need must remain central.

The hospital may need to improve scheduling, specimen transport or communication between clinical and laboratory teams.

AI-assisted ordering should not create additional demand without considering whether the resulting investigations can be delivered safely and in time.

Evaluate appropriate testing, not simply the number of requests

The number of laboratory tests performed is easy to measure.

However, laboratory volume alone cannot establish whether an AI-assisted ordering system improves care.

A decrease in requests may indicate that unnecessary investigations were avoided. It may also mean that clinically necessary tests were omitted.

An increase in requests may reflect inappropriate overuse or improved recognition of previously unmet diagnostic needs.

The interpretation depends on the patients, investigations and clinical circumstances involved.

Healthcare organizations should therefore evaluate both overuse and underuse.

For repeat testing, potential measures include the proportion of requests judged clinically unnecessary, the frequency of testing outside relevant recommended intervals and the number of orders modified after professional review.

For underuse, organizations may examine whether patients with defined clinical indications received appropriate investigations and whether important tests were delayed or omitted.

The denominator is particularly important.

The proportion of repeat requests canceled is not equivalent to the proportion of inappropriate tests prevented.

Similarly, the number of missing tests identified by AI does not establish how frequently necessary investigations were omitted across the entire patient population.

Selected case reviews may be needed to determine whether recommendations and resulting decisions were clinically appropriate.

Hospitals may also examine the time required to obtain important investigations, diagnostic delays and selected adverse outcomes where missing or inappropriate testing may have contributed.

These outcomes should be interpreted with consideration of patient severity, clinical case mix, baseline practices and other changes occurring during implementation.

Cost and resource utilization remain relevant, but they should not be treated as substitutes for clinical quality.

Evaluation should also consider excessive alerts, unnecessary interruptions, clinically justified overrides and the time required to resolve recommendations.

A broader review by Park, published in Critical Reviews in Clinical Laboratory Sciences in May 2026, emphasized that diagnostic stewardship interventions require continuing monitoring because their effects may diminish when active oversight and feedback are withdrawn [8].

For healthcare organizations, this reinforces the need to assess ordering patterns and clinical outcomes over time rather than rely on a one-time reduction in laboratory volume.

The most useful measure of success is whether the system supports more appropriate testing decisions without introducing new diagnostic safety problems.

Local evaluation must include cases where testing should not be discouraged

Before introducing AI-assisted laboratory ordering into routine care, healthcare organizations should test the system using cases representative of their clinical services.

A hospital may begin with selected frequently repeated investigations for which existing results and clinical indications can be assessed reliably.

An outpatient clinic may focus on routine follow-up testing in patients with stable chronic conditions.

The initial use case should be sufficiently narrow to support meaningful evaluation.

Testing should include circumstances in which recent results are reliable and repeating the investigation provides little additional value.

However, evaluation must also include situations where another test is necessary because the patient's condition has changed or treatment requires monitoring.

The system should be assessed for its ability to distinguish these circumstances.

Another group of cases should examine necessary investigations that have not yet been requested.

For applications designed to suggest laboratory tests, evaluation should assess whether important indications are recognized and whether recommendations contain unnecessary investigations.

For generative AI, particular attention is needed when suggestions depend on clinical narratives or information retrieved from several records.

A system may confidently state that a result is normal or recent even when it has used the wrong clinical encounter, misunderstood the specimen collection date or overlooked a change in the patient's condition.

Where possible, the information supporting the recommendation should be available for verification.

Hospitals should also examine how the application behaves when a previous result is unavailable, a test is pending or laboratory data are incomplete.

An initial noninterventional evaluation may help identify unsuitable recommendations before the technology influences actual orders.

However, successful retrospective testing does not establish that the system will improve outcomes after deployment.

Clinical behavior may change once healthcare professionals begin accepting or rejecting AI recommendations during routine care.

Ongoing evaluation should therefore include tests that AI discouraged, orders canceled following warnings and cases where necessary investigations were not suggested.

The evaluation should be designed to identify both excessive and insufficient testing, with particular attention to situations in which an incorrect recommendation could delay important care.

Vietnamese professional requirements remain central

AI-assisted laboratory ordering should operate within the professional and organizational responsibilities established by Vietnamese healthcare law.

The Law on Medical Examination and Treatment No. 15/2023/QH15, as amended and reflected in Consolidated Document No. 26/VBHN-VPQH dated February 26, 2026, provides the general legal framework for professional practice and healthcare services [9].

For laboratory activities, Circular No. 49/2018/TT-BYT provides relevant requirements concerning the organization and management of medical testing within healthcare facilities [10].

Article 5 addresses the management of testing processes, including specimen collection and receipt.

Under the circumstances governed by this provision, specimen collection and receipt are based on a paper or electronic laboratory request containing the required information and the ordering physician's signature.

These requirements are relevant when AI assists with preparing or reviewing laboratory requests.

An AI-generated suggestion does not independently establish the professional authority required to request an investigation or replace the authorization required under the applicable professional framework.

Circular No. 01/2013/TT-BYT provides a separate framework for laboratory quality management in healthcare facilities [11].

This Circular was amended by Circular No. 25/2026/TT-BYT, issued June 30, 2026. The amendment to laboratory quality management took effect on August 15, 2026 [12].

The amendment revises the provision encouraging healthcare facilities to participate in relevant external quality assessment programs, including recognized international programs.

These requirements concern laboratory quality and reliability. They should not be interpreted as establishing a mandatory AI-assisted laboratory ordering process.

Healthcare organizations must distinguish their existing professional and laboratory obligations from the additional risks associated with a particular AI application.

Viet Nam's Law on Artificial Intelligence No. 134/2025/QH15 took effect on March 1, 2026. Decree No. 142/2026/NĐ-CP, effective May 1, 2026, provides further implementation provisions [13, 14].

Decision No. 33/2026/QĐ-TTg, effective August 15, 2026, establishes the list of high-risk AI systems [15].

Ordinary AI-assisted laboratory ordering is not separately identified among the healthcare-specific entries in that list.

This does not establish that every such application is legally low-risk.

The classification of a particular system depends on its intended purpose, actual functions and applicable regulatory criteria.

Whether software falls within medical device regulation also requires a separate assessment where relevant.

Clinical risk and statutory AI risk classification should not be treated as interchangeable.

Where laboratory information and medical records are processed by AI, healthcare organizations must also consider applicable personal data protection requirements.

The Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP took effect on January 1, 2026 [16, 17].

These requirements may be particularly relevant when patient records or laboratory results are processed through external AI providers or cloud-based systems.

Healthcare organizations should assess the relevant obligations according to the actual clinical activity, technology, information-processing arrangements and applicable professional responsibilities.

Making laboratory ordering support practical in Viet Nam

For healthcare organizations in Viet Nam, the most useful starting point is a clearly identified problem with laboratory ordering.

A hospital may experience frequent unnecessary repeat testing in selected departments. Another may have difficulty making recent results accessible across clinical encounters. A specialty clinic may encounter inconsistent test selection or delays in ordering investigations needed for treatment monitoring.

These problems require different responses.

If unnecessary duplicate ordering is the principal concern, improving access to previous results or introducing a well-designed conventional alert may be sufficient.

A complex AI application is not necessarily the best solution to a problem that can be addressed through a reliable rule or simpler workflow change.

AI may offer additional value when appropriate recommendations depend on more complex patterns involving the patient's clinical condition, treatment history and previous laboratory findings.

However, additional technological complexity should be justified by evidence that it improves decisions.

Clinical departments and laboratory services should collaborate to define the intended use case and identify the information required.

Information technology teams should verify that the application retrieves the correct patient data, test identifiers, results, units and specimen dates.

Quality and patient-safety professionals can help evaluate whether unnecessary testing decreases without introducing diagnostic delays or missed investigations.

Hospitals should also consider limitations in laboratory data exchange between healthcare facilities.

Where relevant information is unavailable, the system should communicate that limitation rather than imply that the accessible record contains the patient's complete testing history.

Implementation should reflect available resources and the clinical risks involved.

A smaller healthcare facility may achieve greater immediate benefit from improving laboratory ordering guidance and access to previous results than from adopting an advanced AI recommendation system.

The appropriate objective is to make testing decisions more clinically relevant, consistent and safe, not simply more automated.

From fewer tests to better diagnostic decisions

AI may help healthcare organizations identify unnecessary laboratory requests, use existing results more effectively and recognize situations where further investigation is warranted.

However, fewer tests do not automatically mean better care.

An unnecessary investigation can expose patients to repeated specimen collection, misleading results, additional procedures and avoidable costs.

A missing or delayed investigation can prevent clinicians from recognizing an important diagnosis, complication or change in treatment response.

Both overuse and underuse can cause harm.

For hospitals and clinics in Viet Nam, responsible AI-assisted laboratory ordering means ensuring that recommendations reflect the clinical question, the patient's current condition and the relevance of information already available.

Clinicians must remain able to request necessary investigations when the circumstances justify them. Laboratory professionals should contribute expertise where test selection, analytical limitations or repeated testing require clarification.

Healthcare organizations should evaluate these systems through the appropriateness of testing decisions, diagnostic safety and patient outcomes rather than laboratory volume alone.

The value of AI-assisted laboratory test ordering is not measured by how many tests a system prevents. It lies in helping healthcare professionals request the investigations patients need, avoid those that add little value and make safer diagnostic decisions.

Professional Education Notice

This article is intended for professional education and knowledge exchange. It does not constitute individualized medical, diagnostic, laboratory, legal, regulatory or patient-safety advice.

The clinical scenarios are illustrative and do not represent documented patient cases.

The proposed approaches to laboratory ordering review, repeat testing, exception handling and evaluation are organizational implementation recommendations. They are not mandatory Vietnamese Ministry of Health AI laboratory ordering procedures or independently validated clinical assessment instruments.

Recommendations concerning laboratory test selection, minimum retesting intervals, duplicate testing and clinical review should not be interpreted as identical statutory obligations for every healthcare facility or patient population.

International diagnostic and laboratory stewardship guidance should not automatically be treated as binding professional requirements in Viet Nam.

Evidence supporting conventional computerized decision support should not automatically be generalized to machine-learning or generative AI applications.

Findings from individual studies, patient populations or healthcare settings should not be assumed to establish effectiveness across all healthcare organizations.

AI risk classification, potential medical device status, personal data protection and applicable incident reporting obligations require separate assessment under the relevant legal frameworks.

References

1. Claeys KC, Coffey KC, Morgan DJ. What Is Diagnostic Stewardship? The Journal of Applied Laboratory Medicine. 2025;10(1):130–139. DOI: 10.1093/jalm/jfae130.

2. Zare S, Meidani Z, Shirdeli M, Nabovati E. Laboratory Test Ordering in Inpatient Hospitals: A Systematic Review on the Effects and Features of Clinical Decision Support Systems. BMC Medical Informatics and Decision Making. 2021;21:20. DOI: 10.1186/s12911-020-01384-8.

3. Esho A, Chong YP, Choy KW, Loh TP. Clinical Decision Support in Laboratory Medicine: A Review of Clinical, Organisational and Financial Impacts with Emphasis on Patient Outcomes. Pathology. 2025;57(3):285–292. DOI: 10.1016/j.pathol.2024.11.005.

4. Liang AS, Amrollahi F, Jiang Y, et al. SmartAlert — Implementing Machine Learning–Driven Clinical Decision Support for Inpatient Laboratory Utilization Reduction. NEJM AI. 2026;3(7). Published June 23, 2026. DOI: 10.1056/aics2501086.

5. Zayed AM, Frans G, Delvaux N. Evaluating Large Language Models as Clinical Laboratory Test Recommenders in Primary and Emergency Care: A Crucial Step in Clinical Decision Making. Clinical Chemistry and Laboratory Medicine. 2025;63(11):2186–2197. DOI: 10.1515/cclm-2025-0647.

6. Canadian Agency for Drugs and Technologies in Health and Choosing Wisely Canada. Advisory Panel Guidance on Minimum Retesting Intervals for Lab Tests: Appropriate Use Recommendation. CADTH Health Technology Review, Report HC0078-001. June 2024.

7. Greco A, Garo ML, Zandonà M, Clivio L, Gabutti L. Professional Role and Hierarchy Shape Adherence to Electronic Alerts for Laboratory Test Ordering. Scientific Reports. 2026;16:6669. DOI: 10.1038/s41598-026-37796-1.

8. Park K. Diagnostic Stewardship Monitoring Systems: Institutional and System-Level Approaches. Critical Reviews in Clinical Laboratory Sciences. Published online May 12, 2026. DOI: 10.1080/10408363.2026.2666326.

9. National Assembly of Viet Nam. Law No. 15/2023/QH15 on Medical Examination and Treatment. Adopted January 9, 2023; effective January 1, 2024, as amended. Consolidated Document No. 26/VBHN-VPQH, issued February 26, 2026.

10. Ministry of Health of Viet Nam. Circular No. 49/2018/TT-BYT Providing Guidance on Laboratory Activities in Medical Examination and Treatment. Issued December 28, 2018; effective March 15, 2019.

11. Ministry of Health of Viet Nam. Circular No. 01/2013/TT-BYT Providing Guidance on Laboratory Quality Management in Healthcare Facilities. Issued January 11, 2013; effective March 15, 2013, as amended.

12. Ministry of Health of Viet Nam. Circular No. 25/2026/TT-BYT Amending and Supplementing Certain Provisions of Circular No. 01/2013/TT-BYT and Other Ministry of Health Circulars. Issued June 30, 2026; generally effective August 15, 2026, with specified provisions effective July 1, 2026.

13. National Assembly of Viet Nam. Law No. 134/2025/QH15 on Artificial Intelligence. Adopted December 10, 2025; effective March 1, 2026.

14. Government of Viet Nam. Decree No. 142/2026/NĐ-CP Detailing Certain Provisions and Measures for Implementing the Law on Artificial Intelligence. Issued April 30, 2026; effective May 1, 2026.

15. Prime Minister of Viet Nam. Decision No. 33/2026/QĐ-TTg Promulgating the List of High-Risk Artificial Intelligence Systems. Issued June 30, 2026; effective August 15, 2026.

16. National Assembly of Viet Nam. Law No. 91/2025/QH15 on Personal Data Protection. Adopted June 26, 2025; effective January 1, 2026.

17. Government of Viet Nam. Decree No. 356/2025/NĐ-CP Detailing Certain Provisions and Measures for Implementing the Law on Personal Data Protection. Issued December 31, 2025; effective January 1, 2026.