Governing AI-Assisted Surgical Site Infection Surveillance in Viet Nam: Case Definition Verification, Infection Prevention Review and Post-Discharge Monitoring

Governing AI-Assisted Surgical Site Infection Surveillance in Viet Nam: Case Definition Verification, Infection Prevention Review and Post-Discharge Monitoring

IMPLEMENTATION INSIGHTS

Last reviewed: 31 October 2026

A patient undergoes colorectal surgery at a hospital in Viet Nam and is discharged after several days of recovery. There is no documented surgical site infection (SSI) at the time of discharge. Ten days after the operation, however, the patient develops fever and increasing abdominal discomfort and seeks care at another healthcare facility.

The patient receives antibiotics, and further investigations are arranged. Unless information about this encounter is subsequently shared with the original hospital through an appropriate and lawful process, its surgical team and infection prevention staff may never learn what happened.

When the hospital reviews its surgical outcomes, the patient may still appear in its records as having recovered without a postoperative infection.

An artificial intelligence (AI) surveillance system could help identify a possible SSI if relevant information becomes available. It might recognize an unexpected antibiotic prescription, an unplanned readmission, abnormal laboratory findings, or documentation suggesting a postoperative complication.

But identifying a possible infection is not the same as confirming an SSI. The findings may reflect another infection, a noninfectious complication, or incomplete documentation. The patient may require clinical assessment, while the infection prevention and control (IPC) team must determine separately whether the available evidence meets the applicable surveillance definition.

These responsibilities are connected, but they serve different purposes.

For healthcare organizations in Viet Nam, AI-assisted SSI surveillance offers an opportunity to reduce repetitive record reviews, improve the identification of suspected infections, and strengthen the information available for surgical quality improvement.

Its value depends on more than the algorithm's ability to identify suspicious records. Reliable surveillance requires consistent case definitions, accurate surgical procedure information, professional verification, appropriate clinical escalation, and realistic arrangements for identifying infections after discharge.

The central question is whether AI can help the hospital recognize relevant postoperative complications, classify surveillance cases accurately, and use the findings to improve care without creating new opportunities for missed infections.

Why surgical site infection surveillance matters

Surgical site infections remain an important cause of postoperative complications, prolonged hospitalization, readmission, additional treatment, and healthcare costs.

They can occur after many types of surgery, although their frequency and consequences vary according to the procedure, patient characteristics, clinical circumstances, and infection prevention practices.

Some infections involve the skin and subcutaneous tissue around the incision. Others extend into deeper soft tissues or involve an organ or anatomical space manipulated during surgery.

A superficial incisional infection may require wound assessment and appropriate treatment. A deep incisional or organ-space infection can be more serious and may require imaging, drainage, another operation, antimicrobial treatment, or other interventions.

The patient's condition determines the appropriate clinical response.

Surveillance serves a different purpose. It helps healthcare organizations understand how frequently SSIs occur, identify patterns across procedures, evaluate prevention practices, and recognize where improvement may be needed.

The World Health Organization (WHO) identifies healthcare-associated infection surveillance, timely feedback, and the use of findings to guide improvement as important components of effective IPC programs [3].

WHO's global guidelines for preventing SSI also provide evidence-based recommendations covering care before, during, and after surgery [4].

However, surveillance information is useful only when infections are identified and classified consistently.

A decrease in reported SSI rates may reflect better surgical care. It may also result from incomplete follow-up, changes in case definitions, differences in surgical case mix, or failures to identify infections treated outside the hospital.

Introducing AI can make these distinctions harder to recognize.

A new system may identify more infections because it searches records more thoroughly. Alternatively, it may report fewer cases because some patients are no longer selected for professional review.

Neither change, by itself, establishes that surgical care has become safer.

The objective is to produce reliable information that supports appropriate clinical action and meaningful infection prevention improvement.

What AI-assisted SSI surveillance actually does

AI-assisted SSI surveillance uses information generated during and after surgical care to identify patients whose records may contain evidence of postoperative infection.

Depending on the application, the system may analyze surgical procedure details, microbiology results, antibiotic prescriptions, inflammatory markers, readmissions, emergency visits, imaging reports, clinical notes, or postoperative wound photographs.

Machine learning models may identify combinations of information associated with previously confirmed infections. Natural language processing can extract potentially relevant findings from clinical documentation, while image-based models can identify wound photographs that appear concerning.

Other surveillance systems use predefined rules rather than machine learning.

For example, a rule-based system may select a record for review when a patient has an unplanned readmission, a positive microbiological result, or an extended course of postoperative antibiotics.

These approaches may support similar operational goals, but they are not technically identical. A conventional rule-based algorithm should not automatically be described as AI simply because it processes electronic health records.

More importantly, the information used by these systems does not necessarily establish an infection.

Antibiotics may have been prescribed for pneumonia or a urinary tract infection rather than an SSI. Fever may reflect another postoperative complication, while a positive culture requires interpretation according to the specimen, timing, and clinical circumstances.

The opposite problem is equally important. A patient may develop an SSI and receive treatment elsewhere, without the relevant encounter appearing in the original hospital's records.

AI cannot identify clinical information that was never captured or made available to the system.

A surveillance flag should therefore be treated as an indication that a case may require further review, not as automatic confirmation of an SSI.

Where findings suggest an active or potentially serious postoperative complication, clinical assessment should not be delayed while surveillance classification is being completed.

Case definition verification must reflect the Vietnamese surveillance framework

Consistent case definitions are fundamental to reliable SSI surveillance.

Viet Nam's Ministry of Health issued Decision No. 1526/QĐ-BYT on March 24, 2023, promulgating the Guidance on Surgical Site Infection Surveillance [1].

This document provides an important national professional reference for hospitals conducting SSI surveillance.

It addresses surveillance objectives, case definitions, risk factors, selection of surgical procedures, data collection, analysis, reporting, and implementation responsibilities.

The guidance distinguishes superficial incisional, deep incisional, and organ-space SSI according to the anatomical tissues involved and specified evidence of infection.

It also makes an important distinction: the criteria used to identify SSI cases for surveillance purposes may not be identical to those used for clinical diagnosis and treatment.

This principle should remain central when AI is introduced.

An algorithm may identify a patient with redness around the wound, a postoperative antibiotic prescription, or a positive microbiological result. However, none of these findings should automatically establish that the patient meets the relevant SSI surveillance definition.

The evidence must be assessed against the selected criteria.

A clinically diagnosed infection may require treatment even when the available documentation does not establish every element of a particular surveillance definition.

Conversely, classification of an event for surveillance purposes does not independently determine the patient's treatment.

For hospitals in Viet Nam, Decision No. 1526/QĐ-BYT should be considered when designing SSI surveillance procedures and evaluating automated case identification.

International frameworks can provide useful additional guidance, but their definitions should not replace applicable Vietnamese requirements without considering differences in scope and implementation.

An AI system developed using cases classified under one surveillance framework may require further evaluation before its outputs can be reliably interpreted under another.

How the CDC/NHSN framework informs surveillance practice

The US Centers for Disease Control and Prevention's National Healthcare Safety Network (CDC/NHSN) provides an internationally recognized SSI surveillance framework.

Its January 2026 Surgical Site Infection Event protocol distinguishes superficial incisional, deep incisional, and organ-space SSI [2].

A superficial incisional SSI involves the skin and subcutaneous tissue of an incision. A deep incisional SSI involves deeper soft tissues, such as fascia and muscle. An organ-space SSI involves an organ or anatomical space deeper than the incision that was opened or manipulated during the operation.

Each category has defined criteria.

For example, under the NHSN protocol, the diagnosis or treatment of cellulitis alone is insufficient to establish a superficial incisional SSI.

For organ-space SSI, the relevant general criteria must be considered together with the appropriate site-specific infection definition.

These distinctions illustrate why a generic AI probability score cannot replace a documented surveillance assessment.

The NHSN framework also provides detailed rules concerning operative procedure categories, attribution of infections to procedures, surveillance periods, and reporting.

Such standardization can support consistent comparisons and quality improvement.

However, the NHSN protocol is a US surveillance framework. It is not automatically a legally mandatory Vietnamese reporting standard.

Healthcare organizations should determine which definitions and reporting requirements apply to their services and use the selected framework consistently.

Where international comparisons are intended, differences between Vietnamese and international case definitions, procedure classifications, and surveillance methods should be examined before interpreting the results.

The objective is not simply to adopt the most detailed available definition. It is to ensure that surveillance cases are identified consistently, reported appropriately, and interpreted according to the purpose of the program.

Surveillance periods and post-discharge follow-up require careful distinction

An SSI may become apparent only after the patient has left the hospital.

The length of the surveillance period therefore affects which infections are counted and whether results are comparable.

Viet Nam's Decision No. 1526/QĐ-BYT recognizes surveillance periods of 30 or 90 days according to the type of surgery and specifies criteria for superficial incisional, deep incisional, and organ-space SSI [1].

The guidance identifies particular operations for which the 90-day period applies, including selected breast, cardiac, spinal, fracture fixation, joint replacement, and other procedures.

The CDC/NHSN January 2026 protocol also uses 30- and 90-day periods according to operative procedure category and tissue level [2].

Under NHSN, superficial incisional SSI is monitored for 30 days across the relevant procedure categories, while deep incisional and organ-space SSI may be monitored for 30 or 90 days.

For example, colon surgery and cesarean section are included in the NHSN 30-day categories. Cardiac surgery, breast surgery, and hip or knee prosthesis procedures are among the 90-day categories.

The procedure date is counted as day one.

These distinctions have practical consequences for automated surveillance.

A system that reviews every postoperative record for only 30 days may fail to identify infections after procedures requiring a longer surveillance period.

Conversely, applying a 90-day period to every procedure without regard to the selected protocol may create inconsistent case classification.

A surveillance system must therefore know which procedure was performed, when it occurred, and which surveillance rules apply.

However, a surveillance period should not be confused with a deadline for clinical assessment.

A patient developing concerning symptoms outside the defined reporting period may still require appropriate medical care.

The end of an epidemiological surveillance period does not mean that postoperative complications can be ignored.

What Vietnamese guidance says about monitoring after discharge

Post-discharge surveillance deserves particular attention because it is both clinically important and operationally difficult.

Decision No. 1526/QĐ-BYT provides specific guidance for collecting postoperative information after patients leave the hospital [1].

Importantly, the document states that post-discharge SSI surveillance is not mandatory for all healthcare facilities.

For institutions able to allocate the necessary resources, it describes telephone-based follow-up using a structured interview form.

The guidance includes patient education at discharge, communication of relevant warning signs, and contact information so patients can report possible wound complications.

Where the described post-discharge surveillance process is implemented, the guidance provides for telephone contact at weekly intervals and at the relevant 30- or 90-day surveillance endpoint.

It also addresses patients who cannot be contacted.

The guidance describes two additional contact attempts on different days after an unsuccessful first call. Patients who remain unreachable after three attempts, decline participation, or have no available telephone number are recorded as lost to follow-up.

These details are important because they make the completeness of surveillance visible.

A hospital should not assume that a patient without a recorded SSI has remained free of infection when follow-up information is unavailable.

The national guidance also specifies how patients classified as lost to follow-up are handled in the denominator for the described post-discharge SSI rate calculations.

This rule should be applied according to the intended indicator and reporting methodology rather than generalized to every surgical quality measure.

For organizational evaluation, it remains useful to report how many patients were initially eligible, how many completed follow-up, and how many were lost.

Otherwise, excluding patients with incomplete follow-up may produce rates that are difficult to interpret, particularly if loss to follow-up differs between surgical departments or patient groups.

AI may help organize follow-up information, identify incomplete records, and prioritize cases needing review. However, it cannot eliminate the need for a workable process to obtain reliable postoperative information.

What recent studies tell us about AI-assisted SSI surveillance

Recent studies provide encouraging evidence that AI-assisted and automated approaches can reduce the work required to identify possible SSIs.

However, the evidence is stronger for case detection and workload reduction than for improvements in clinical outcomes after routine implementation.

A study by Agostinho and colleagues, published in npj Digital Medicine in October 2025, evaluated machine learning and rule-based approaches using data from 3,931 surgical patients [6].

The investigators focused on deep incisional and organ-space SSIs and examined whether automated screening could reduce the number of records requiring manual surveillance review.

The best-performing machine learning approaches achieved sensitivity of up to 90%, with an AUROC reaching 0.968.

One model reduced the estimated manual review workload by 91.6%, while another achieved a reduction of approximately 90.1%.

These results suggest that automated screening can help IPC professionals concentrate their attention on records more likely to contain evidence of infection.

However, the workload reduction came with an important trade-off: the machine learning models did not identify every SSI in the evaluated population.

The rule-based model achieved higher sensitivity but a smaller workload reduction, approximately 70%.

The authors subsequently published a correction in November 2025, revising several reported performance measures for the rule-based model, including its sensitivity to 100% in the study's validation dataset [7].

The correction did not change the overall conclusions.

Nevertheless, this observed sensitivity should not be interpreted as a guarantee that the rule-based approach would identify every infection in another hospital or patient population.

The study illustrates a question that many IPC teams will face: how much manual work can reasonably be reduced without allowing an unacceptable number of infections to be missed?

The appropriate balance depends on the surveillance objective and the consequences of missing cases.

A system intended to support comprehensive institutional reporting may require different safeguards from one used to identify selected patients for additional review.

Why evidence of implementation matters

A broader systematic review published in the Journal of Hospital Infection in 2026 examined AI and machine learning applications in hospital IPC [8].

The review included 59 studies.

Only nine reported real-world implementation, including six pilot studies and three routine implementations. The remaining 50 focused on model development or validation without demonstrating integration into routine clinical workflows.

The review covered multiple IPC applications, not SSI surveillance alone.

Nevertheless, its findings highlight an important limitation in the evidence: technical model development has progressed faster than research demonstrating sustained operational and clinical benefit.

Implemented studies more frequently reported process measures than standardized infection outcomes. Many also provided limited information about unintended consequences and long-term sustainability.

A model may perform well when evaluated retrospectively but provide limited benefit when used in a hospital where clinical data arrive late, records are incomplete, or IPC staff lack capacity to review its findings.

False-positive alerts may also create additional work.

The practical question is therefore not simply whether a model can classify cases accurately.

It is whether the surveillance service can use its outputs reliably, detect missed infections, maintain professional review, and convert findings into meaningful infection prevention activities.

An accurate model is a useful starting point, not proof that the hospital's IPC system has become more effective.

Evidence from Asia and large surgical datasets

A study published in the American Journal of Infection Control in June 2026 evaluated automated SSI surveillance following coronary artery bypass graft surgery at a tertiary hospital in Thailand [9].

The investigators compared an automated screening approach with the hospital's existing surveillance process.

The selected algorithm achieved sensitivity of 100% and specificity of approximately 91.7% in its retrospective validation dataset, while reducing the estimated manual review workload by 87.9%.

These findings provide a useful regional example of how automation may improve surveillance efficiency.

However, the study involved a defined surgical population at one hospital. The number of infection events was limited, and further evaluation is needed before the results can be generalized.

The observed sensitivity does not establish that the system would detect every SSI during routine use elsewhere.

The study also demonstrates why hospitals should consider simpler automation alongside advanced AI approaches.

A well-designed system using existing clinical information may provide meaningful operational benefits without requiring complex machine learning.

Another study, published in Scientific Reports in August 2026, examined data from 765,962 inpatient surgical procedures performed across 18 hospitals in Denmark between 2016 and 2021 [10].

Using a previously validated natural language processing model, the investigators identified 14,018 superficial SSIs within 30 days of surgery, representing approximately 1.8% of procedures in the dataset.

They also examined associations between identified infections and perioperative antibiotic prophylaxis practices.

The study demonstrates how automated analysis of clinical documentation may support large-scale surveillance and investigation of surgical quality patterns.

However, it was a retrospective proof-of-concept study.

The findings depended on the completeness of electronic records and the performance of the identification model.

Associations involving antibiotic prophylaxis do not independently establish causation. Likewise, identifying a large number of infections through natural language processing does not prove that every eligible infection was captured.

The practical value lies in making it possible to investigate patterns across large surgical populations, provided the underlying information and case identification methods are sufficiently reliable.

AI analysis of postoperative wound photographs introduces new considerations

Many postoperative wound concerns become apparent after discharge.

Digital follow-up services may help patients communicate symptoms, submit photographs, or request advice during recovery.

AI-assisted image analysis is one potential component of these services.

In August 2026, JAMA Surgery published a study evaluating deep learning for identifying wound photographs suggestive of incisional SSI [11].

The investigators developed the model using 4,978 wound images and performed external validation using 407 images from 95 patients.

The model achieved an AUROC of 0.91 during internal evaluation and 0.82 during external validation.

These findings suggest that automated image analysis may help prioritize photographs for professional review.

However, the model assessed whether images appeared suggestive of infection. It did not independently establish an SSI diagnosis or demonstrate that routine deployment reduced postoperative complications.

A photograph may show redness, discharge, swelling, or wound separation. It may not reveal a deep or organ-space infection that produces few visible changes at the skin surface.

Image quality also matters. Lighting, camera characteristics, distance, wound location, dressings, and the patient's ability to take a clear photograph may affect interpretation.

A normal-looking image should therefore not provide reassurance when a patient reports fever, increasing pain, or other concerning symptoms.

Healthcare organizations should also consider what patients reasonably expect when they submit photographs through a digital service.

If a patient sends an image of a concerning wound, the hospital needs to explain whether the service provides active clinical monitoring, when submissions are reviewed, and how urgent concerns should be communicated.

An AI system that identifies a suspicious photograph but has no reliable clinical response pathway may provide little benefit.

Digital postoperative monitoring should be designed around the patient's need for appropriate assessment, not merely around an algorithm's image classification capability.

Clinical escalation and surveillance review require different workflows

An important implementation risk is treating every AI surveillance flag as an item for routine retrospective review.

Some alerts concern historical records that require classification by IPC professionals. Others may indicate an active postoperative complication requiring timely clinical attention.

For example, a surveillance system may identify that a patient was readmitted several weeks after surgery with a documented infection. The IPC team can review the record and determine whether the event meets the selected surveillance criteria.

A different alert may involve a patient who has just reported increasing wound discharge, fever, and clinical deterioration.

In that situation, the immediate question is whether the patient requires medical assessment and treatment.

The case may subsequently be reviewed for surveillance purposes, but the clinical response should not depend on completion of epidemiological classification.

Healthcare organizations should therefore distinguish surveillance review from clinical escalation.

IPC professionals contribute expertise in identifying and classifying infections. Surgeons and other appropriately qualified clinicians remain responsible for clinical decisions within their authorized professional roles.

Where AI-generated information suggests an active complication, the workflow should identify who receives the information, how urgency is assessed, and how the patient can obtain appropriate care.

These arrangements are particularly important for patient-facing applications and wound image monitoring.

Without clear responsibilities, a digital system may create the appearance of continuous clinical oversight even when no clinician has been assigned to respond to important findings.

Separating the two workflows helps ensure that patients receive appropriate attention while surveillance information remains consistently classified.

Infection prevention professionals remain central to case verification

AI-assisted surveillance may change how IPC teams identify records requiring attention, but it does not remove the need for professional assessment.

IPC practitioners bring expertise in surveillance definitions, microbiology, surgical procedures, postoperative complications, and hospital epidemiology.

Their role includes reviewing evidence, identifying incomplete information, and determining whether a potential infection meets the selected surveillance criteria.

An automated flag may be supported by a microbiology result, an operative report documenting an abscess, or a subsequent clinical assessment.

Another may be based only on an antibiotic prescription or diagnosis code that does not establish an SSI.

These situations require different interpretations.

Where documentation is incomplete, consultation with the surgical team, microbiology laboratory, or other clinicians may be appropriate.

Differences in interpretation may also arise between surveillance reviewers. A reliable process should allow these questions to be resolved consistently and retain the evidence supporting important classification decisions.

Professional verification also provides an opportunity to learn from model errors.

Repeated false-positive flags may show that the algorithm is responding to antibiotic use unrelated to SSI.

Missed infections may reveal that important information is contained in free-text notes, external care encounters, or data systems not included in the automated analysis.

These findings can guide improvements in data collection, software configuration, workflow design, and professional training.

Review should therefore be understood as part of surveillance and clinical learning, not simply as a final approval step attached to an automated classification.

Reliable surveillance requires accurate numerator and denominator data

SSI surveillance depends on identifying both the infections that meet the selected definition and the surgical procedures included in the surveillance population.

These provide the numerator and denominator used to calculate reported infection rates.

The numerator includes eligible SSI events according to the chosen protocol. The denominator includes the eligible surgical procedures during the specified reporting period.

Errors in either can distort the results.

If relevant infections are missed, the reported SSI rate may appear artificially low. If eligible procedures are omitted from the denominator, the reported rate may be artificially high.

Incorrect procedure coding, patient identification, operation dates, or surveillance periods may also affect case attribution.

Hospitals should therefore examine how operating room records, electronic medical records, microbiology results, readmission information, and IPC databases are connected.

A model that accurately identifies potential infections cannot produce reliable institutional SSI rates if the underlying procedure data are incomplete.

The reporting framework itself also matters.

NHSN uses defined operative procedure categories and coding arrangements. Hospitals in Viet Nam adopting NHSN concepts should verify how those categories relate to their own clinical and administrative records.

Viet Nam's Decision No. 1526/QĐ-BYT provides its own guidance concerning surgical procedure selection, surveillance forms, infection attribution, and calculation of rates [1].

Where post-discharge surveillance is conducted, its method for handling patients lost to follow-up should also be considered.

Such denominator rules are important for consistent reporting, but they do not eliminate the possibility of bias when patients without complete follow-up differ from those successfully contacted.

Hospitals should therefore report follow-up completeness alongside infection rates when evaluating a program.

Comparisons between hospitals require further caution.

SSI risk varies according to procedure type, patient characteristics, wound contamination, urgency of surgery, and other clinical factors.

Changes in surgical case mix or surveillance intensity may influence reported rates even when the underlying quality of care has not changed.

Crude SSI rates should not be used as straightforward rankings of surgical performance without considering case definitions, denominators, surveillance completeness, and appropriate risk adjustment.

Post-discharge surveillance must reflect where patients receive care

One of the greatest limitations of hospital-based SSI surveillance is incomplete information after discharge.

Patients may seek treatment from private clinics, community healthcare providers, another hospital, or the original surgical team.

These encounters may not appear in the hospital's information system.

A facility relying only on its own electronic records may identify infections resulting in readmission to the same hospital while missing those treated elsewhere.

As a result, surveillance may provide an artificially reassuring picture of surgical outcomes.

Viet Nam's SSI surveillance guidance recognizes patient contact after discharge as a method for collecting information when a facility has the resources to implement it [1].

The NHSN framework similarly recognizes approaches such as reviewing medical records, readmission information, communication with clinicians, and patient-reported findings [2].

Telephone calls, postoperative clinic visits, secure messaging, digital questionnaires, and wound photographs may all contribute useful information when they fit the clinical service.

However, each approach has limitations.

Some patients may have limited digital literacy, unreliable internet access, language difficulties, or insufficient support at home.

Others may not respond to follow-up despite developing a complication.

A digital-only approach could therefore produce uneven surveillance completeness across patient groups.

For healthcare organizations in Viet Nam, post-discharge monitoring should reflect the actual pathways through which surgical patients obtain care.

Where information is lawfully shared between organizations, better communication may improve the identification of infections treated elsewhere.

Where such connections are unavailable, the resulting gaps should be acknowledged when interpreting surveillance findings.

The absence of a reported infection is not necessarily evidence that no infection occurred.

Surveillance findings should lead to infection prevention improvement

Identifying SSI cases is only one part of an effective infection prevention program.

Once infections have been verified, hospitals need to understand what the findings mean for surgical practice.

A cluster of infections associated with a particular procedure may warrant review of operating room processes, patient characteristics, antimicrobial prophylaxis, surgical practices, and postoperative care.

Patterns involving superficial incisional infections may require different investigations from those involving deep or organ-space infections.

However, the first question after a change in reported SSI rates should not automatically be whether clinical performance has improved or deteriorated.

The organization should also consider whether case ascertainment has changed.

After introducing AI, an increase in reported infections may reflect better identification of cases previously missed.

Similarly, a lower rate may reflect reduced surveillance sensitivity if the system excludes some records from professional review.

These effects can make simple before-and-after comparisons misleading.

WHO's IPC guidance emphasizes the use of surveillance findings and timely feedback to support infection prevention activities [3].

Its broader global report also highlights the importance of strengthening IPC systems and the availability of reliable information for healthcare improvement [5].

For AI-assisted SSI surveillance, verified findings should inform clinical review, relevant prevention actions, and subsequent evaluation.

The technology's contribution should be assessed partly by whether it helps professionals recognize patterns and address preventable risks.

Producing more automated reports is not the objective. The objective is to provide sufficiently reliable information to support safer surgical care.

Workforce readiness and practical implementation

A surveillance system may perform well during retrospective validation yet provide limited operational benefit after implementation.

Clinical data may arrive too late, relevant information may be missing, or IPC staff may lack capacity to review the flagged records.

False-positive alerts can also consume considerable professional time.

The positive predictive value of a surveillance alert depends partly on how common the target infection is in the monitored population.

When SSI is relatively uncommon, even a model with favorable sensitivity and specificity may generate many alerts that do not correspond to confirmed infections.

Hospitals should therefore examine how many records require review for each verified case and how much professional time is actually saved.

The workload benefit must be considered alongside the possibility of missed infections.

A 2026 international survey of IPC professionals across WHO regions provides additional implementation context [12].

Of 444 survey responses, 441 were valid for analysis. Among these respondents, 24.5% reported current use of AI or digital technologies in IPC activities, while only 6.8% reported formal training.

Frequently reported barriers included insufficient training and skills, limited funding, and difficulties integrating technology with existing systems.

The voluntary survey should not be interpreted as a representative estimate of AI adoption in Vietnamese hospitals.

Nevertheless, the findings illustrate practical conditions that matter during implementation.

AI surveillance may provide limited value if staff do not understand its outputs, relevant information systems are disconnected, or the hospital cannot maintain the process after initial deployment.

Hospitals should also consider whether improved documentation, better data integration, or simpler rule-based automation could address their main operational problems before adopting more complex AI software.

The Vietnamese infection prevention context

AI-assisted SSI surveillance may be relevant to Vietnamese hospitals providing general surgery, obstetric surgery, orthopedic surgery, cardiovascular surgery, and other operative services.

Its potential value differs according to each organization's clinical and information-management capabilities.

Some hospitals may have structured surgical records, connected laboratory systems, electronic medical records, and established IPC surveillance teams.

Others may depend on a combination of paper documentation, separate databases, manual reporting, and limited access to postoperative information from other healthcare providers.

These differences matter because AI cannot reliably identify information that is absent or inaccessible.

Where surgical records or follow-up information are incomplete, improving data quality may be a more immediate priority than introducing AI.

Viet Nam already has a national professional and regulatory framework for IPC.

Circular No. 16/2018/TT-BYT, issued on July 20, 2018, establishes requirements concerning IPC activities in healthcare facilities [13].

Decision No. 1526/QĐ-BYT, issued on March 24, 2023, provides specific guidance on SSI surveillance [1].

The Ministry of Health subsequently issued Decision No. 38/QĐ-BYT on January 3, 2025, approving the National Action Plan on Infection Prevention and Control in Healthcare Facilities for 2025–2030 [14].

The plan identifies the need to strengthen surveillance, professional capacity, organizational responsibilities, and quality improvement.

These documents provide a relevant foundation for hospitals considering automated surveillance.

An additional development occurred in September 2026, when the Ministry of Health published a draft circular intended to replace Circular No. 16/2018/TT-BYT [15].

The proposal included provisions concerning IPC activities, surveillance data, digital technologies, feedback, and improvement.

However, a consultation draft is not equivalent to a promulgated and effective legal instrument.

Healthcare organizations should verify the current applicable regulations and any subsequent amendments before establishing internal procedures.

The existence of national IPC guidance and action plans should not be interpreted as a general requirement to adopt AI or as authorization for a particular surveillance product.

Vietnamese legal and professional responsibilities

The use of AI-assisted SSI surveillance requires consideration of Vietnamese requirements governing healthcare services, infection prevention and control, artificial intelligence, medical devices, and personal data protection.

The relevant obligations depend on what the software actually does.

A tool used to identify records for retrospective surveillance review may raise different regulatory questions from a system intended to provide patient-specific diagnoses or initiate clinical actions.

These distinctions should be assessed before routine deployment.

Clinical and infection prevention responsibilities

Viet Nam's Law on Medical Examination and Treatment No. 15/2023/QH15, effective January 1, 2024, as amended, provides an important foundation for healthcare services, professional practice, and patient rights [16].

Circular No. 16/2018/TT-BYT and Decision No. 1526/QĐ-BYT provide relevant professional guidance for IPC and SSI surveillance [13,1].

Decision No. 38/QĐ-BYT establishes national priorities for strengthening IPC during 2025–2030 [14].

The Law on Disease Prevention No. 114/2025/QH15 and Decree No. 165/2026/NĐ-CP, both effective July 1, 2026, also form part of the broader legal framework concerning disease prevention and related public health activities [17,18].

Their relevance to a particular hospital surveillance activity should be determined from the provisions applicable to that activity.

These instruments should not automatically be interpreted as establishing separate authorization or mandatory technical requirements for AI-assisted SSI surveillance.

An AI-generated indication of possible infection does not independently establish a clinical diagnosis or expand a practitioner's legally permitted professional scope.

Decisions about patient assessment, treatment, and surveillance classification remain subject to applicable professional requirements and institutional responsibilities.

Artificial intelligence and medical device regulation

Viet Nam's Law on Artificial Intelligence No. 134/2025/QH15 took effect on March 1, 2026 [19].

Decree No. 142/2026/NĐ-CP, effective May 1, 2026, provides implementing provisions concerning AI activities, including risk classification and management [20].

Decision No. 33/2026/QĐ-TTg, effective August 15, 2026, promulgates a list of high-risk AI systems [21].

The legal classification of an SSI surveillance application depends on its intended purpose, actual functions, deployment arrangements, and applicable criteria.

A system used to prioritize records for IPC review may differ from software intended to diagnose infection or recommend patient-specific treatment.

However, a surveillance system should not automatically be considered legally low-risk simply because it is operated by an IPC or quality improvement department.

The absence of an expressly named application from a high-risk AI list does not independently establish exemption from other applicable obligations.

Medical device regulation must be considered separately.

Decree No. 98/2021/NĐ-CP on medical device management, as amended, including by Decrees No. 07/2023/NĐ-CP and No. 04/2025/NĐ-CP, forms part of the relevant framework [22].

Circular No. 24/2026/TT-BYT, effective July 1, 2026, further addresses risk determination and management measures for products classified as medical devices [23].

Software used solely for administrative surveillance may require a different regulatory assessment from software intended to provide a medical diagnosis or patient-specific clinical recommendation.

The regulatory status of a product should therefore be determined from its actual intended medical purpose and applicable definitions, rather than assumed from a general description such as AI surveillance.

International research findings or foreign regulatory authorization do not independently establish compliance with all Vietnamese requirements.

Personal data protection and confidentiality

SSI surveillance may involve sensitive information from surgical records, microbiology systems, medication histories, readmissions, clinical notes, wound photographs, and postoperative communication.

These data may identify a patient directly or through linkage with other records.

Viet Nam's Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP took effect on January 1, 2026 [24,25].

Healthcare organizations should assess the applicable legal grounds and purposes for processing patient information, consent requirements and relevant exceptions, access controls, information security, retention arrangements, and responsibilities of technology providers.

Particular attention is needed where patient information is transferred to external AI platforms or processed through cloud services.

The hospital should understand where the data are processed, who can access them, and whether they may be used beyond the intended clinical or surveillance purpose.

Where personal data are transferred outside Viet Nam, applicable cross-border transfer requirements and relevant procedural obligations or exceptions should be assessed separately.

Secondary use of surgical records or wound photographs for algorithm development, retraining, commercial product improvement, or research requires separate consideration of applicable legal and institutional requirements.

Removing names or identification numbers does not necessarily make the underlying records or photographs fully anonymous.

When patients are asked to submit wound photographs or clinical information through digital follow-up services, they should receive appropriate information about how their submissions will be used and reviewed.

Privacy and information security safeguards are important, but they do not replace the need for an appropriate clinical response to concerning findings.

Evaluating whether AI improves SSI surveillance

Healthcare organizations should define what improvement they expect before introducing AI-assisted SSI surveillance.

The objective may be to reduce manual record review, improve case ascertainment, identify infections earlier, strengthen post-discharge surveillance, or provide more reliable information for infection prevention activities.

These objectives require different measures.

Sensitivity is important because missed infections can undermine surveillance reliability.

Specificity and positive predictive value matter because false-positive flags increase professional review workload.

The proportion of records requiring manual adjudication can help estimate operational efficiency, but it should be interpreted alongside the number and characteristics of infections potentially missed.

A system that reduces manual review by 90% is not necessarily preferable to one that reduces it by 70% if the larger saving comes with unacceptable under-ascertainment.

Time to identification is also relevant, but earlier surveillance recognition does not automatically mean earlier treatment.

If an AI-generated finding is not communicated to the appropriate clinical team when action is needed, the benefit to patient care may be limited.

Post-discharge follow-up completeness is another important consideration.

Hospitals should understand what proportion of eligible surgical patients have sufficiently complete information during the applicable surveillance period.

The proportion of patients lost to follow-up should be monitored and interpreted alongside reported SSI rates.

Reported rates also need appropriate procedure denominators, consistent case definitions, and consideration of surgical case mix.

Changes in surveillance intensity, documentation quality, and follow-up arrangements should be examined before comparing results over time.

Implementation costs should include software integration, data preparation, professional review, training, maintenance, postoperative monitoring, and continuing evaluation.

Savings from reduced record review may be offset by additional work elsewhere in the process.

Ultimately, the hospital needs to determine whether AI produces more reliable surveillance information and helps the IPC team identify meaningful opportunities for improvement.

Evidence that AI reduces surveillance workload does not independently establish that it reduces infection incidence, readmissions, antimicrobial resistance, or other clinical outcomes.

Demonstrating those benefits requires appropriate evaluation methods and adequate follow-up.

These are proposed organizational evaluation considerations, not a mandatory Vietnamese AI SSI surveillance scorecard or an independently validated Digital Medicine Vietnam assessment instrument.

A practical direction for healthcare organizations in Viet Nam

Responsible implementation should begin with an examination of the hospital's existing SSI surveillance process.

Before selecting AI software, the organization should identify which procedures are monitored, how potential infections are currently detected, which surveillance definitions are used, and how information is collected after discharge.

The IPC team, surgical departments, medical records personnel, information technology professionals, and relevant quality and safety leaders should participate in defining the intended purpose.

The hospital should first determine whether its existing information systems provide sufficiently complete and reliable data for automated case identification.

Where appropriate, a period of parallel evaluation can help determine how the proposed system compares with established surveillance practice.

During this evaluation, professionals can examine cases identified by both methods, investigate discrepancies, and determine which types of infections the automated approach may miss.

It is also useful to review a suitable sample of records classified as unlikely to contain an SSI, rather than examining only those flagged as positive.

Otherwise, missed infections may remain hidden.

The hospital should establish how suspected cases are verified, how clinically concerning findings are escalated, and how important decisions are documented.

Where patient-facing postoperative monitoring is included, responsibilities should be clear for reviewing submissions, communicating with patients, and arranging clinical assessment when necessary.

Surveillance findings should be communicated to surgical and quality teams in a form that supports prevention and improvement.

After deployment, changes in documentation practices, patient populations, clinical services, software versions, or model performance should prompt appropriate reassessment.

A reliable manual or alternative surveillance process should remain available when the automated system is unavailable or unsuitable.

Hospitals should not assume that the most technically sophisticated product will provide the greatest benefit.

In some settings, better procedure coding, improved microbiology linkage, standardized surveillance definitions, and more complete post-discharge information may deliver substantial improvements before advanced AI is introduced.

The objective is to strengthen the surveillance process and the actions it supports, not to replace infection prevention professionals with automated classifications.

From AI surveillance to safer surgical care

AI-assisted SSI surveillance offers opportunities to improve how healthcare organizations identify, review, and learn from postoperative infections.

Recent studies show promising reductions in manual surveillance workload and demonstrate how electronic records, natural language processing, and postoperative wound images can help identify cases requiring attention.

However, relatively few AI applications have been evaluated after sustained implementation in routine IPC services.

For healthcare organizations in Viet Nam, responsible use begins with national SSI surveillance guidance, consistent case definitions, accurate surgical procedure data, appropriate professional responsibilities, and realistic arrangements for post-discharge follow-up.

An AI flag is not a confirmed SSI. A surveillance classification is not a substitute for clinical diagnosis. A lower reported infection rate does not necessarily mean that surgical care has improved.

The value of surveillance lies in identifying relevant cases reliably, recognizing when patients need clinical assessment, and using the findings to guide infection prevention activities.

AI may reduce repetitive work and help professionals recognize patterns that would otherwise be difficult to identify. Its contribution to patient safety, however, depends on whether those findings lead to appropriate review, timely feedback, and meaningful improvements in care.

The success of AI-assisted surgical site infection surveillance should not be judged only by how many records an algorithm reviews or how accurately it identifies possible infections. It should be judged by whether the healthcare organization verifies cases reliably, maintains appropriate post-discharge follow-up, and turns surveillance findings into safer surgical care.

Professional Education Notice

This article is provided by Digital Medicine Vietnam (DMVN) for professional education, evidence interpretation, and knowledge exchange. It does not constitute individualized surgical care, infection diagnosis, antimicrobial treatment, infection prevention and control, medical device, legal, regulatory, or patient-safety advice.

The opening clinical scenario is illustrative and does not describe a documented patient case.

The discussion concerns AI-assisted and automated approaches to surgical site infection surveillance, professional case verification, infection prevention review, post-discharge monitoring, and organizational quality improvement. It does not establish the safety, effectiveness, regulatory status, or suitability of any particular commercial, open-source, or internally developed AI system.

AI surveillance flags, clinical diagnoses, and surveillance case classifications have different meanings and should not be treated as interchangeable.

The cited studies must be interpreted according to their populations, designs, algorithms, available data, and limitations. Improvements in case identification or reductions in manual surveillance workload do not independently establish lower infection rates or better patient outcomes.

Viet Nam's Ministry of Health Decision No. 1526/QĐ-BYT provides relevant national professional guidance on SSI surveillance. The CDC/NHSN protocol provides an international reference but should not automatically be treated as a legally mandatory Vietnamese reporting standard or substituted for applicable national requirements.

The WHO guidelines discussed provide evidence-based recommendations and do not supersede Vietnamese laws or institutional responsibilities.

Vietnamese legal and professional documents should be interpreted according to their actual scope, applicable requirements, and effective dates. Regulatory consultation drafts must be distinguished from formally promulgated legal instruments.

The organizational approaches described, including AI-supported case finding, professional verification, post-discharge monitoring, surveillance evaluation, and quality improvement, are proposed considerations for responsible healthcare AI implementation.

They are not mandatory Vietnamese AI SSI surveillance protocols, individual legal compliance determinations, or independently validated DMVN clinical assessment instruments.

Clinical decisions concerning suspected postoperative infection require appropriate professional assessment and must reflect the patient's actual condition and applicable standards of care.

References

1. Ministry of Health of Viet Nam. Decision No. 1526/QĐ-BYT Promulgating Guidance on Surgical Site Infection Surveillance. Issued March 24, 2023; effective upon issuance.

2. Centers for Disease Control and Prevention. National Healthcare Safety Network Patient Safety Component Manual. Chapter 9: Surgical Site Infection (SSI) Event. January 2026.

3. World Health Organization. Guidelines on Core Components of Infection Prevention and Control Programmes at the National and Acute Health Care Facility Level. Geneva: World Health Organization; 2016. ISBN: 978-92-4-154992-9.

4. World Health Organization. Global Guidelines for the Prevention of Surgical Site Infection. Second edition. Geneva: World Health Organization; 2018. ISBN: 978-92-4-155047-5.

5. World Health Organization. Global Report on Infection Prevention and Control 2024. Geneva: World Health Organization; 2024. ISBN: 978-92-4-010398-6.

6. Agostinho A, Chalot E, Teixeira D, et al. Semi-Automated Surveillance of Surgical Site Infections Using Machine Learning and Rule-Based Classification Models. npj Digital Medicine. 2025;8:617. DOI: 10.1038/s41746-025-01989-1.

7. Agostinho A, Chalot E, Teixeira D, et al. Author Correction: Semi-Automated Surveillance of Surgical Site Infections Using Machine Learning and Rule-Based Classification Models. npj Digital Medicine. 2025;8:707. DOI: 10.1038/s41746-025-02153-5.

8. Sirago G, Zotti F, Mele F, et al. Artificial Intelligence for Hospital Infection Prevention and Control: Real-World Implementation, Impact, and the Gap Beyond Model Development. Journal of Hospital Infection. 2026;174:410–427. DOI: 10.1016/j.jhin.2026.04.019.

9. Skuntaniyom S, Techasaensiri C, Muntajit T, Piebpien P. Automated Surveillance System for Surgical Site Infection in Coronary Artery Bypass Graft Surgery in Tertiary Care Hospitals. American Journal of Infection Control. 2026;54(6):620–625. DOI: 10.1016/j.ajic.2025.12.001.

10. Winther S, Millarch AS, Dencker EE, et al. A Proof-of-Concept Study Associating Artificial Intelligence Surveillance of Surgical Site Infections With Antibiotic Prophylaxis From 765,962 Surgeries. Scientific Reports. Published August 3, 2026. DOI: 10.1038/s41598-026-64116-4.

11. Bontekoning N, Huisman H, Segura Cabrera PJ, et al. Detecting Incisional Surgical Site Infections on Wound Images Through Deep Learning. JAMA Surgery. 2026;161(10):1066–1074. DOI: 10.1001/jamasurg.2026.3738.

12. Gastaldi S, Satta G, Tartari E, Mookerjee S, Allegranzi B. Global Adoption and Readiness for Artificial Intelligence and Digital Health Technologies in Infection Prevention and Control: A World Health Organization Global Survey. Journal of Hospital Infection. 2026;173:203–210. DOI: 10.1016/j.jhin.2026.05.002.

13. Ministry of Health of Viet Nam. Circular No. 16/2018/TT-BYT on Infection Prevention and Control in Healthcare Facilities. Issued July 20, 2018; effective October 1, 2018.

14. Ministry of Health of Viet Nam. Decision No. 38/QĐ-BYT Approving the National Action Plan on Infection Prevention and Control in Healthcare Facilities for 2025–2030. Issued January 3, 2025.

15. Government Electronic Newspaper of Viet Nam. Đề xuất quy định mới về kiểm soát nhiễm khuẩn tại cơ sở khám bệnh, chữa bệnh [Proposed New Regulations on Infection Prevention and Control in Healthcare Facilities]. Published September 17, 2026.

16. 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.

17. National Assembly of Viet Nam. Law No. 114/2025/QH15 on Disease Prevention. Adopted December 10, 2025; effective July 1, 2026.

18. Government of Viet Nam. Decree No. 165/2026/NĐ-CP Detailing and Guiding the Implementation of Certain Provisions of the Law on Disease Prevention. Issued May 15, 2026; effective July 1, 2026.

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

20. 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.

21. 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.

22. Government of Viet Nam. Decree No. 98/2021/NĐ-CP on Medical Device Management, as amended, including by Decrees No. 07/2023/NĐ-CP and No. 04/2025/NĐ-CP.

23. Ministry of Health of Viet Nam. Circular No. 24/2026/TT-BYT on Risk Determination and Management Measures for Products Classified as Medical Devices and Amendments to Circular No. 05/2022/TT-BYT. Issued June 30, 2026; effective July 1, 2026.

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

25. 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.