Causality is not the admission ticket for an adverse drug reaction report.
Vietnam's official healthcare-facility ADR reporting form asks health professionals to report even when they are not certain that the product caused the reaction or when information is incomplete. That instruction sets the workflow boundary. Suspicion starts a report. Clinical assessment continues alongside it.
This article follows the spontaneous-reporting path for healthcare facilities. Clinical trials, pharmaceutical companies, vaccination systems, and national disease programs have additional or separate safety channels. Hospital procedures also determine local review, approval, and follow-up, so this is not a substitute for a facility's standard operating procedure.
Treat the patient and preserve the timeline
The first obligation is clinical care. A reporting form must not delay assessment, stabilization, treatment, or monitoring. But documentation should begin while the sequence can still be reconstructed.
The National DI and ADR Center's reporting guidance acknowledges that an event during treatment may reflect disease progression, a medicine, an interaction, or another cause. It asks health professionals to describe the reaction clearly, examine its timing in relation to medicine use, review comorbidities, food and concomitant medicines, obtain relevant examinations and tests, and check whether the reaction appears in product information or reliable references.
These activities help assess the event. They do not require the team to prove causation before recording it. The current form preserves that distinction. Patient details, reaction description, seriousness, outcome, suspected medicines, treatment dates, dose, route, dechallenge, rechallenge, and concomitant medicines appear before the facility's causality assessment section.
A useful narrative distinguishes three kinds of information: what staff observed, what the patient or family reported, and what the clinical team inferred. Dates and times should be recorded when available. Unknown fields should remain unknown rather than being completed from assumption.
A suspected report can remain uncertain
The official guidance says to report as early as possible, including when the information collected so far is incomplete. It asks staff to enter as much information as they have, use a separate report for each patient, and add information later when necessary. The form itself distinguishes an initial report from a follow-up.
The guidance issued with Decision 29/QD-BYT also sets national maximum submission times. A serious ADR that is fatal or life-threatening must be sent as soon as possible and no later than seven working days after detection. Other serious ADRs must be sent as soon as possible and no later than 15 working days after detection. Non-serious ADRs may be collected and sent monthly before the fifth day of the following month. A hospital may set a shorter internal handoff, but its local route cannot push submission beyond these national limits.
This matters because missingness is normal at the point of detection. A laboratory result may still be pending. A medicine obtained outside the hospital may not yet have a verified dose. The exact onset time may come from a relative rather than the chart. Delaying the entire report until every field is resolved can lose the chronology and the opportunity for follow-up.
Keep documentation standards high by making uncertainty explicit. The record should show which source supports each material fact, which fields remain unresolved, and who is responsible for obtaining follow-up information.
Uncertainty should remain visible during clinical assessment. Clinicians may consider alternative causes, temporal association, known pharmacology, response after stopping or reducing the medicine, and the evidence available for the suspected reaction. Re-exposure is not a routine information-gathering step and should never be prompted merely to improve a report.
The form provides separate causality options, including unclassified and unclassifiable. This allows the record to represent the state of knowledge instead of forcing confidence that the evidence does not support. A facility can record its assessment and the method used without making that result a gate for submission. The National Center performs further analysis and appraisal after receipt, and patterns may become informative only when reports are considered together.
Local ownership should not become gatekeeping
Vietnam's guidance identifies physicians, dentists, pharmacists, nurses, midwives, and other healthcare providers as potential reporters. Patient and reporter information is treated as confidential.
A hospital may route suspected ADRs through the pharmacy department, general planning department, a pharmacovigilance focal person, or another designated team. That local route should assign ownership for clinical review, form completion, signature, submission, and response tracking. It should not turn reporting into a relay where everyone waits for someone else to begin.
The National DI and ADR Center lists Decision 29/QD-BYT dated 5 January 2022 as the in-force Ministry of Health guidance for ADR monitoring in healthcare facilities. The Center also lists the broader national pharmacovigilance guidance issued under Decision 122/QD-BYT as in force. Facilities still need to verify how those documents are implemented in their own governance and information systems.
Submission is an auditable handoff
The Center's ADR submission page provides the healthcare-facility form and lists four submission routes: post, fax, email, and online reporting. Current contact details are published on that page and on the form.
Before submission, the responsible reviewer should verify patient identity, medicine identity, chronology, attachments, reporter contact details, and the local authorization required to send the record. Health information should not be moved through an unofficial channel merely because it is convenient.
An electronic workflow also needs provenance. It should record who extracted information, who edited the narrative, who approved the report, which version was submitted, and which supporting files went with it. Later additions should remain linked to the initial report rather than silently replacing it.
The same Center page provides separate forms for product-quality problems, medication errors, pharmaceutical businesses, and several national programs. A case may involve more than one safety concern, but an ADR, an error, a quality defect, and treatment failure are not interchangeable labels. The workflow should preserve the correct report type and route.
The workflow continues after receipt
The National DI and ADR Center describes its role as collecting, analyzing, appraising, and aggregating reports for regulators, while also providing feedback to treatment facilities. The section reserved for the Center on the ADR form includes acknowledgment, classification, database entry, review-panel handling, and appraisal results.
Feedback can therefore take several forms. A facility may receive acknowledgment, a request for more information, an appraisal response, urgent communication, or broader medicine-safety information. The Center's 2024 reporting activity summary describes urgent feedback to reporters and facilities as well as routine feedback and ADR information distributed by email and the Center's website.
None of this guarantees an individual causal verdict or a fixed response time for every submission. A hospital should keep the report open for follow-up, add new clinical information, route incoming feedback to the responsible team, and connect relevant findings to local medication-risk management.
The AI boundary is preparation and quality control
AI can help assemble a report without becoming the reporter. It can extract medication names, doses, routes, laboratory results, and dates from the record; build a draft timeline; flag missing fields; retrieve source documents for review; and route the draft to the designated pharmacovigilance team.
Every extracted fact should remain traceable to its source location. A reviewer must be able to correct any field, distinguish observed facts from inferred relationships, and see what the system could not resolve. If the model cannot tell whether a medicine was suspected, concomitant, or used to treat the reaction, it should ask for review rather than guess.
AI must not assign causality as a final decision, determine that a suspicion is not worth reporting, manufacture absent details, sign the form, or submit it autonomously. It must not use a low causality score to close the case. The official reporting threshold is suspicion, including suspicion supported by incomplete information.
Current research does not remove that boundary. A 2026 Pharmaceutical Research study compared biomedical language models with two experts on 150 FAERS and VAERS safety reports. The best configuration reached 64% agreement on the final causality class, with low agreement on critical elements including temporal plausibility, alternative causes, and objective evidence. The authors judged the models suboptimal for reliable ICSR causality assessment.
A separate 2026 Scientific Reports study evaluated a kNN causality model on solicited company reports involving six products. It excluded spontaneous reports and deployed the model as decision support for trained medical reviewers. These studies show plausible uses for triage and oversight. They do not validate autonomous causality decisions or report suppression in Vietnam's spontaneous-reporting pathway.
A safe implementation needs human approval before submission, access control, an audit trail, versioned follow-up, and checks against sending patient data to an unauthorized destination. These are not optional administrative details. They are what allow automation to reduce transcription work without weakening clinical accountability.
The useful outcome is not the number of forms an AI system completes by itself. It is whether healthcare professionals can submit accurate suspected-event information sooner, with fewer missing source details and no increase in fabricated facts, suppressed reports, privacy failures, or overconfident causal claims. That is the boundary a pharmacovigilance assistant must pass before it earns a place in the workflow.