Hemovigilance.

Tracing transfusion-reaction evidence through explicit surveillance definitions.

Development prototype · October 2026

Where the project stands

The prototype connects a deterministic rules engine to an adaptive interview. Timing and presenting findings establish a starting point; each answer changes which questions can still clarify the assessment.

Evidence, source definitions, and unresolved criteria remain available for review. The engine and interview have passed 70 synthetic implementation tests. Clinical validation has not been performed.

This tool is under development and is not intended for clinical use or patient-care decisions.

A focused walkthrough

Follow the evidence.

  1. Enter timing and select the presenting findings.
  2. Answer targeted questions. Unknown is a valid answer.
  3. Review what was entered and inspect the surveillance assessment.

Change an answer at any point. The interview retains earlier evidence and recalculates which questions remain useful.

Use synthetic information only. The assessment runs in browser memory; entries are not submitted or saved.

Evidence is organized as met, absent or contradicted, and still unknown, then evaluated against versioned surveillance definitions. Each assessment retains its criteria and source so its reasoning can be reviewed.
Make the reasoning visible. The interface exposes the evidence behind a classification and the information still missing.
Project status

Adaptive interview built. Development prototype; not clinically validated.

The informatics problem

Translate source definitions into auditable rules and ask useful questions without forcing certainty or hiding competing reactions.

Questions that respond to the case

A conventional form asks for every finding at once. This interview uses the same rule expressions as the classifier to identify which unanswered questions can still change a result or establish a more specific source-defined level.

An answer can make some questions unnecessary while leaving others relevant. Meeting one reaction definition does not automatically remove another. Users can move backward, edit earlier evidence, or open the full question list.

Surveillance classification has a boundary

The current tool evaluates implemented NHSN hemovigilance definitions. Meeting a surveillance definition does not establish a clinical diagnosis; failing to meet one does not exclude a reaction clinically.

That distinction appears on every reaction result. Clinical differential diagnosis, treatment recommendations, severity, and imputability are outside this version’s scope.

Unknown stays unknown

Missing measurements, unanswered findings, and explicit negative findings are different kinds of evidence. The tool preserves those distinctions, including the difference between an omitted value and a measured zero.

Results show criteria met, criteria absent or contradicted, and criteria still unknown. No probability or confidence score is assigned. Where a source includes categories that the current engine has not implemented, the limitation is displayed alongside the result.

Source definitions, visible provenance

The rule set is grounded in the CDC/NHSN Hemovigilance Module Surveillance Protocol, version 3.0, January 2026. Results identify the source section or table and page. Definitions retained in the protocol’s archive are labeled separately from current reporting categories.

Scope and source material
  • Current reporting definitions implemented: acute hemolytic transfusion reaction, TACO, TRALI, and transfusion-transmitted infection.
  • Previously used classifications retained in the protocol: allergic, delayed hemolytic and delayed serologic reactions, FNHTR, hypotensive reaction, post-transfusion purpura, transfusion-associated dyspnea, and TA-GVHD.

Source version and implemented coverage matter. A result applies only to the rules actually evaluated, and archived classifications do not become current reporting requirements simply because they are available here.

Built for review before validation

The current testing checks specified software behavior using synthetic cases, including overlapping definitions, timing boundaries, missing data, and changes made during an interview. It does not establish diagnostic accuracy or patient benefit.

The next research step is a governed validation pathway with independently reviewed cases. This public prototype accepts no identifiers, requires no account, and has no case-saving workflow.