Hantavirus Risk Assessment AI: 50,000 Tests, 3 Cases Found

Clinical Decision Support Production Module May 2026 CC BY 4.0 Real-World Deployed

Three Laboratory-Confirmed Hantavirus Cases Caught in the Prodromal Window Across 50,000 Interpreted Blood Test Reports

A multilingual clinician-facing decision-support module that lives where blood test interpretation already happens. Aligned with CDC and WHO guidance, deployed in more than 75 languages and offered free on every Kantesti plan including the consumer free tier.

📖 ~12 minutes 📅 May 10, 2026 🔗 DOI: 10.6084/m9.figshare.32230290

📝 Published: May 10, 2026 🩺 Medically Reviewed: May 10, 2026 ✅ Production Deployed 🔓 Open Access

This clinical decision support module was engineered by Julian Emirhan Bulut, Senior AI Engineer and CEO of Kantesti Ltd, with clinical oversight from Dr. Thomas Klein, MD, Chief Medical Officer at Kantesti AI. Module reasoning weights and prodromal-signature taxonomy were reviewed by the Kantesti AI Medical Advisory Board.

Lead Engineer & Module Architect

Julian Emirhan Bulut

Senior AI Engineer & CEO, Kantesti Ltd

Julian Emirhan Bulut is the founder and CEO of Kantesti Ltd. He designed the reasoning pipeline, the strict JSON output schema, the per-clinic isolation layer and the multilingual rendering pipeline that ships the Hantavirus Risk Assessment in more than 75 languages. He has been building the Kantesti AI Engine since 2019.

LinkedIn GitHub Kaggle ResearchGate About Kantesti

Clinical Oversight & Medical Review

Thomas Klein, MD

Chief Medical Officer, Kantesti AI · Istanbul Nisantasi University, Department of Hematology

Dr. Thomas Klein is a board-certified clinical hematologist and internist with over 15 years of experience in laboratory medicine, affiliated with the Department of Hematology at Istanbul Nisantasi University. He validated the prodromal laboratory signature weighting against current Hantavirus clinical guidance, reviewed the contributing-factor taxonomy and supervised the framing of the module as decision support rather than a diagnostic device.

ORCID 0009-0009-1490-1321 ResearchGate Google Scholar Academia.edu Medical Advisory Board

Kantesti AI Hantavirus Risk Assessment is a multilingual clinical decision support module deployed inside the Kantesti clinic dashboard. It analyses an interpreted blood test report alongside optional exposure history, symptoms, vitals and Hantavirus-specific serology, then returns a structured 0 to 100 risk score with rationale, contributing factors, red flags and a missing-data declaration. Production deployment between February and May 2026 evaluated 50,000 consecutive interpreted blood test reports and identified three laboratory-confirmed Hantavirus cases by IgM ELISA or RT-PCR, all initially misclassified at presentation.

Module typeClinical Decision Support

Aligned withCDC · WHO

Languages75+

Reports analysed50,000

Confirmed cases3

CostFree on every plan

⚡ Quick Summary Production Module — May 10, 2026

50,000 interpreted blood test reports processed by the module between February and May 2026.
14 high or critical risk classifications (0.028 percent), reflecting calibrated specificity for a globally rare condition.
Three laboratory-confirmed Hantavirus cases by IgM ELISA or RT-PCR, per CDC and WHO recommended diagnostic standards.
All three confirmed cases were initially misclassified at presentation as influenza-like illness, atypical pneumonia or bacterial sepsis.
More than 75 languages with native right-to-left rendering for Arabic, Hebrew and Persian. No English fallback.
Strict JSON output schema with explicit confidence and missing-data declarations. Every score is auditable.
Free on every Kantesti plan including the consumer free tier. Public-health value should not gate behind billing.
Open-access preprint: DOI 10.6084/m9.figshare.32230290 · Mirrored on ResearchGate and Academia.edu under CC BY 4.0.

Why Hantavirus triage is hard and why it matters

Hantaviruses cause two clinically severe zoonotic syndromes. Hantavirus Pulmonary Syndrome (HPS) is reported with case-fatality between 30 and 50 percent in unprepared settings, and Hemorrhagic Fever with Renal Syndrome (HFRS) reaches the tens to hundreds of thousands of cases per year across Eurasia. The hardest moment of clinical care for either syndrome is the earliest one. The prodromal phase looks like influenza, dengue, leptospirosis, severe COVID-19 or bacterial sepsis. By the time the cardiopulmonary or oliguric phase is unmistakable, the therapeutic window has narrowed considerably.

The clinical question we face in practice is not whether a given febrile patient has Hantavirus. It is whether the laboratory signature, the exposure context and the symptom profile together justify ordering Hantavirus serology now rather than reaching for a generic "viral syndrome" label and discharging on supportive care. That second question is where the diagnostic gap sits and where post-mortem case reports keep returning. We see a febrile patient. We see thrombocytopenia. We see a creatinine creep. We do not always remember to ask about rodent exposure, because the question itself is uncommon enough to slip the mind on a busy shift.

The Kantesti AI Hantavirus Risk Assessment exists to keep that question on the table. It is a clinician-facing decision-support module that lives inside the patient record beside the latest blood work. It reads the interpreted blood test report the broader Kantesti pipeline has already produced. It asks the clinician for whatever exposure, symptom and serology context is to hand and it returns a structured 0 to 100 risk score with a written rationale, a list of contributing factors, a list of red flags and an explicit declaration of what data it wishes it had.

As Thomas Klein, MD, I supervised the prodromal-signature weighting and reviewed the contributing-factor taxonomy on this module. The deliberate choice we made is to design for the prodromal window. The value of a triage tool that fires only when the cardiopulmonary phase is already obvious is close to zero. The clinical question is whether the engine can be useful 48 to 96 hours earlier. Our broader Medical Validation hub describes the framework. This page describes its applied result.

HPS versus HFRS at a glance

The two Hantavirus syndromes share a prodromal phase but diverge in dominant organ involvement, geography and reservoir. The table below summarises the differences that matter for triage. A clinician reading this on a febrile patient with rodent exposure should consider both, because early presentations are difficult to distinguish without serology.

HPS — Hantavirus Pulmonary Syndrome Americas · case-fatality 30–50% Sin Nombre virus (USA, Canada), Andes virus (South America). Cardiopulmonary phase with non-cardiogenic pulmonary oedema and shock.

HFRS — Hemorrhagic Fever with Renal Syndrome Eurasia · case-fatality 1–15% by serotype Hantaan, Seoul, Puumala, Dobrava-Belgrade. Five-phase course (febrile, hypotensive, oliguric, diuretic, convalescent). Acute kidney injury and hemorrhage.

Shared prodrome (both syndromes) 1–8 weeks incubation Fever, myalgia, headache, gastrointestinal symptoms. Thrombocytopenia, rising hematocrit, leukocytosis with immunoblasts.

Confirmation method (per CDC and WHO) IgM ELISA · RT-PCR Definitive diagnosis requires serology or molecular testing. Clinical assessment alone is insufficient for either syndrome.

The prodromal laboratory signature

A standard complete blood count and chemistry panel ordered for an undifferentiated febrile illness will, in early Hantavirus, often display a recognisable constellation. None of the findings is pathognomonic in isolation. Together, in the right exposure context, they form a probabilistic signature that pattern-recognition tools are well suited to surface.

Platelets are the most consistent abnormality. Thrombocytopenia, often dropping below 150 × 10⁹/L within the first 72 hours, is reported in 70 to 90 percent of HPS cases at presentation. Hematocrit rises secondary to capillary leak and plasma volume contraction. The white-cell count rises with a left shift and immunoblasts (sometimes called atypical lymphocytes or Hantavirus-associated mononuclear cells) appear on the peripheral smear. Lactate rises with tissue hypoperfusion. Transaminases (AST, ALT) and LDH rise modestly with cellular turnover. Creatinine begins to rise sharply in HFRS and more subtly in HPS.

Platelets Often < 150 × 10⁹/L Most consistent prodromal abnormality, present in 70 to 90 percent of HPS cases

Hematocrit Rising Hemoconcentration secondary to capillary leak and plasma volume contraction

White blood cells Leukocytosis with left shift Immunoblasts (atypical lymphocytes) on peripheral smear

Lactate Elevated Reflects early tissue hypoperfusion before frank shock

AST · ALT · LDH Modestly elevated Cellular turnover, generally not in the hepatitis range

Creatinine Rising Sharp rise in HFRS, subtler in HPS

The clinically reassuring property of this signature is that it sits inside an investigation any clinician will already have ordered. Nothing in the prodromal pattern requires a special test. The challenge is one of pattern recognition under cognitive load. A tired clinician on a busy shift looking at a thrombocytopenia and a mild creatinine creep in a febrile patient may reach for sepsis, viral hepatitis or atypical pneumonia long before reaching for Hantavirus serology. That is exactly the moment a calibrated decision-support layer earns its place.

How the module works, end to end

The module is a single-page sub-application inside the Kantesti clinic dashboard, invoked from the patient profile (with the patient auto-selected via deep-link) or from the dashboard's main navigation. It composes three layers: a clinician-facing UI, a reasoning service that calls a secure cloud-hosted large-language-model endpoint with strict JSON output and a per-clinic per-patient assessment store keyed for safe overwrite on language change.

Inputs

The reasoning is anchored on the interpreted blood test JSON that the broader Kantesti pipeline produces from raw laboratory uploads (PDF, photo or structured laboratory feed). This is the critical engineering decision. The AI is not asked to interpret raw numbers in a vacuum. It is given a structured already-curated report containing laboratory values with reference ranges, units and per-parameter clinical commentary, plus a prose interpretation of the panel. On top of that authoritative anchor the clinician may optionally add four blocks of context.

Exposure history. Rodent contact, cleaning of closed or abandoned spaces, recent travel to endemic regions and days since presumed exposure.
Symptom set. Fever, myalgia, dry cough, dyspnoea, gastrointestinal distress and headache.
Vital signs. Peripheral oxygen saturation (SpO2), systolic blood pressure and heart rate.
Hantavirus-specific testing. IgM (ELISA), IgG and RT-PCR, with an acute-phase status flag where available.

All optional sections are explicitly marked as optional. The design point is that missing data should not block scoring. Instead, the AI is required to declare what it is missing and how that missing data affects its confidence. This honesty requirement is enforced at the schema level rather than negotiated at runtime.

Output schema

Every reasoning call returns a strict JSON payload. The schema is what makes this system trustable in a clinical setting. A reviewer can audit any score by reading its contributing factors and red flags and challenge it where appropriate. The model is required to be explicit about what it does and does not know. Vague hand-waving is not in the contract.

risk_score integer 0 to 100 Calibrated risk that the presentation warrants Hantavirus workup

risk_level low / moderate / high / critical Clinical bucket derived from the score

confidence low / medium / high The model's self-reported certainty given the data on hand

recommended_action string A single actionable next step in the clinician's language

explanation string A clinically-readable rationale tying the score to the data

contributing_factors array Each factor names a finding, marks its direction (↑ or ↓ on risk) and assigns an impact

red_flags array Findings that should escalate management independent of the score

missing_critical_data array Data the model wishes it had, with a brief reason for each

All eight fields are required for every reasoning call. The UI renders each one without ever calling innerHTML on AI-derived strings. Every output token reaches the DOM through textContent or attribute assignment, which closes the obvious cross-site-scripting hole that AI outputs would otherwise open. Allow-listed enum values are validated before rendering. AI prose cannot inject markup.

A sample CRITICAL risk output

The screenshot below shows a real CRITICAL classification produced by the production module. The patient profile is a 26-year-old presenting with rodent exposure, recent cleaning of a closed space, fever, myalgia, cough, dyspnoea and SpO2 of 93 percent. The module returned a risk score of 82 out of 100 with moderate confidence and a recommendation to seek urgent in-person evaluation before pending laboratory results.

🔴 Sample case — anonymised input

Exposure context. Recent cleaning of a closed space with documented rodent activity. Patient reports peridomestic rodent contact in the preceding 14 days.

Symptoms. Fever, myalgia, dry cough, dyspnoea on minimal exertion. Mild gastrointestinal distress.

Vital signs. SpO2 93 percent on room air. Other vitals not reported by the referring clinician.

Laboratory. Interpreted blood test report not yet uploaded at the moment of assessment. Hantavirus-specific serology pending.

📋 Sample output — module response

Risk level. CRITICAL. Score 82 out of 100. Confidence moderate.

Recommended action. Seek urgent in-person medical evaluation now, before test results return.

Explanation. "This patient has significant Hantavirus-compatible exposure history plus fever, myalgia, cough, dyspnoea and oxygen saturation below 94 percent, which raises concern for serious disease. The symptom pattern and low SpO2 are concerning for possible early cardiopulmonary involvement. Risk remains uncertain because key laboratory data and test results are still pending."

Red flags. SpO2 93 percent. Dyspnoea.

Contributing factors. HIGH ↑ rodent exposure · HIGH ↑ cleaning a closed space · HIGH ↑ dyspnoea · HIGH ↑ SpO2 93 percent · MEDIUM ↑ fever and myalgia · MEDIUM ↑ cough · LOW ↑ GI symptoms · LOW ↓ age 26 years.

Missing critical data. Platelet count, hematocrit, white blood cell count, creatinine, AST, ALT, LDH, RT-PCR, IgM and IgG.

The clinical value of the explicit missing-data declaration deserves a moment of attention. The model is not pretending to know more than it does. It returned a CRITICAL classification on exposure context plus vital signs alone and stated openly that the picture would tighten with the pending laboratory work. A clinician reviewing this output knows immediately what data, once available, will change the score upward (a thrombocytopenia confirms the prodromal pattern) or move it downward (a normal IgM and a clean platelet count after 96 hours of symptoms argues against active Hantavirus). This is the architectural answer to the AI hallucination concern: every claim is structured in a way that makes uncertainty visible rather than hidden inside fluent prose.

Fifty thousand interpreted reports, three confirmed cases

Between 1 February and 8 May 2026 the Kantesti platform processed 50,000 consecutive interpreted blood test reports for which the Hantavirus Risk Assessment module was either explicitly invoked by a clinician or implicitly evaluated as part of the integrated risk-flag layer surfaced beside the standard interpretation. Reports originated from 127 countries across the Americas, Europe, the Middle East, Sub-Saharan Africa, South Asia, East Asia and Oceania.

Reports analysed 50,000 Feb to May 2026 · 127 countries

14 High or critical

3 Lab-confirmed

75+ Languages active

0.028% High/critical rate

Risk level Reports (n) Proportion Confirmed

Low46,83293.66%—

Moderate3,1546.31%—

High110.022%1 confirmed

Critical30.006%2 confirmed

The distribution is intentionally heavily skewed toward the low and moderate buckets. Hantavirus is uncommon globally and a triage tool that fired high on every febrile illness would be clinically useless. The module is calibrated to surface a high or critical risk only when the laboratory and clinical signature is genuinely consistent with the syndrome. The corresponding case-finding rate of approximately 6 confirmed Hantavirus infections per 100,000 interpreted reports is consistent with what would be expected of a globally rare condition surveilled through a non-targeted clinical workflow.

A few important caveats sit alongside these numbers. Confirmation status is only known where the partner clinic shared post-laboratory follow-up with the platform. The true number of Hantavirus cases in the cohort may be higher. The module is not designed to discriminate Hantavirus from the other syndromes in the prodromal differential (leptospirosis, dengue, severe COVID-19, atypical pneumonia and sepsis). The relevant clinical question is whether the patient warrants targeted Hantavirus workup, not whether they definitely have Hantavirus. The clinical value of a tool with a low base-rate denominator lies in catching the cases that would otherwise be missed during the prodrome.

What the three confirmed cases were called at presentation

All three laboratory-confirmed Hantavirus cases were initially classified by the presenting provider as one of: influenza-like illness, atypical community-acquired pneumonia or undifferentiated bacterial sepsis. None of the three had Hantavirus on the initial differential. The module elevated Hantavirus into consideration on the basis of the laboratory pattern and, where available, exposure history.

Patient identities, geographic locations, age, sex, occupational exposure and clinical details are withheld in compliance with GDPR and HIPAA-aligned safeguards. We can disclose the following aggregate observations.

Common laboratory signature

All three confirmed cases displayed thrombocytopenia (platelets below the lower reference limit), at least two of (rising hematocrit, transaminase elevation, lactate elevation) and a left-shifted leukocytosis at presentation. These findings together are non-specific in isolation but jointly characteristic of the prodromal phase. The module assigned high or critical risk in each case and recommended urgent laboratory confirmation by IgM ELISA or RT-PCR.

Time to confirmation

Subsequent IgM ELISA or RT-PCR confirmation occurred between 24 and 96 hours after the module first surfaced a high or critical classification, in line with typical regional turnaround for confirmatory Hantavirus testing. In each case the formal Hantavirus workup was initiated as a direct response to the module's flag.

Outcomes

All three patients survived and progressed to clinical recovery. We make no causal claim that the module was responsible for these outcomes. Outcomes in Hantavirus depend on many factors including supportive care quality, severity at presentation and individual host response. The module's contribution was to surface Hantavirus as a differential consideration earlier than the clinician's pre-platform workflow would otherwise have done. Whether that earlier consideration changed disposition is a question we cannot answer from the available data.

As Thomas Klein, MD, I want to be careful here. Three confirmed cases do not constitute a randomised trial. They constitute a real-world signal in production traffic, with all the strengths and weaknesses that real-world data carries. The strength is that this is what happens when the module meets actual clinical workflow on actual patients in 127 countries. The weakness is that we do not have a counterfactual. We cannot say what would have happened to these patients without the module. What we can say is that the module performed exactly as designed: it pulled three Hantavirus presentations out of an unfiltered febrile-illness flow that had pattern-matched to the wrong differential at the bedside, and it did so on the basis of a structured rationale that the reviewing clinician could read and challenge.

More than 75 languages, no English fallback

A clinical decision support tool that only speaks one language is, by definition, an inequitable tool. Hantavirus is endemic across the Americas, Europe and East Asia. The patients who would benefit from earlier triage are not, on average, English-speaking. The module therefore localises into more than 75 languages, with native right-to-left rendering for Arabic, Hebrew and Persian, and with no English fallback anywhere in the rendering pipeline.

Languages currently shipping include English, Turkish, German, French, Spanish, Italian, Portuguese, Arabic, Hebrew, Greek, Polish, Dutch, Russian, Ukrainian, Chinese (Simplified), Chinese (Traditional), Japanese, Korean, Hindi, Bengali, Persian, Thai, Vietnamese, Indonesian, Malay, Tagalog, Swedish, Norwegian, Danish, Finnish, Czech, Slovak, Slovenian, Croatian, Bulgarian, Serbian, Latvian, Estonian, Lithuanian, Hungarian, Romanian, Albanian, Macedonian, Maltese, Icelandic, Irish, Welsh, Basque, Catalan, Galician, Afrikaans, Swahili, Amharic, Yoruba, Zulu, Urdu, Punjabi, Tamil, Telugu, Kannada, Malayalam, Sinhala, Nepali, Marathi, Gujarati, Khmer, Lao, Burmese, Mongolian, Kazakh, Uzbek, Azerbaijani, Armenian, Georgian and Pashto.

Engineering details that make the localisation trustable

No silent English fallback. If a translation key is missing for the active locale the build fails. We do not paper over a hole with English. A complete parallel string set is required for every page in every supported language. This is non-trivial engineering investment and it is the price of treating multilingual clinical AI as a first-class concern rather than as a marketing line.

Locale-aware caching. The Hantavirus assessment store keys on (clinic, patient, report, language) semantics. Switching the active language causes the next assessment to overwrite the prior row rather than serving a stale row in the wrong locale. A clinician can switch from English to German and the next score will arrive in German with no inconsistency.

Compound-word safety. German strings such as "Familien-/Patientengesundheit Risikoanalyse" are notoriously fond of breaking responsive layouts. The CSS uses overflow-wrap: anywhere, hyphens: auto and flex-wrap on plan cards so badges drop neatly to a new line on narrow viewports rather than escaping the card. Finnish, Hungarian and Greek receive the same treatment.

Right-to-left support. Arabic, Hebrew and Persian are detected on language change. The document direction is set. The CSS adapts mirror-aware spacing and icon orientation. Score gauges, contributing-factor arrows and timestamp ordering all respect the active reading direction.

Per-clinic isolation and cascade lifecycle

Healthcare software fails on the day a tenant boundary is crossed. The module's design treats per-clinic isolation and patient-data containment as primary requirements, on the same footing as clinical correctness.

A non-exhaustive list of safeguards in place, described at a level of abstraction that does not invite probing.

Authenticated session required. The module's endpoints are not reachable without a valid clinic session. There is no anonymous mode.
Per-clinic isolation. Every storage operation is keyed on the authenticated clinic identity. A clinician in Clinic A cannot see, list, view, regenerate or delete an assessment belonging to Clinic B by any combination of identifiers.
Path-traversal hardening. All filesystem references derived from request parameters pass component-level validation before canonicalisation, with rejection of separators, null bytes, parent-directory tokens and tilde expansion.
Cross-site-scripting safety on AI output. AI-generated strings are inserted into the DOM exclusively through textContent and attribute assignment, never innerHTML. Allow-listed enum values are validated before rendering.
Parameterised SQL throughout. No string-interpolated SQL touches the assessment store. All writes and reads use bound parameters.
Cascade lifecycle. Deleting a patient, deleting a clinic, regenerating a report or deleting a report all trigger explicit cascade hooks that remove or invalidate the corresponding Hantavirus assessments. No orphaned rows accumulate.
Compliance posture. Kantesti Ltd is GDPR-compliant for European data subjects, HIPAA-aligned for US healthcare partners and operates ISO 27001-aligned controls and SOC 2 Type II security practices. The platform holds CE marking for the European market.
Auditability. Every assessment row carries a hash of the source report, the language it was scored in and timestamps for create and update. A clinical reviewer can reconstruct exactly what input produced what output at what time.

We do not publish the specific filtering rules, request validators, regex patterns or attack vectors used during hardening. Advertising the test set advantages an attacker more than it reassures a clinician.

How clinicians and patients access the module

The Hantavirus Risk Assessment is freely available to every Kantesti user. The public-health value of catching Hantavirus earlier should not be a function of a clinic's billing tier, so the module is enabled on the consumer free tier, the single-report and 6-report bundles, the annual plan and every clinic dashboard plan from day one.

For clinicians

Sign in to the Kantesti clinic dashboard. Open a patient record. Click the Hantavirus Risk Assessment action from the patient profile, or open the dashboard's main navigation and pick the patient from the picker. Optionally fill the exposure, symptom, vital-sign and Hantavirus-specific testing sections. Submit. The module returns a structured score in the active dashboard language with full rationale, contributing factors, red flags and a missing-data declaration. Exports are available as a print-ready PDF for inclusion in the patient chart.

For patients

If you are reading this as a patient or family member with a recent blood test that you are concerned about, you can upload the report to the Kantesti consumer portal at kantesti.net/free-blood-test and request the Hantavirus Risk Assessment. The consumer-facing flow returns an interpretable risk band with explicit instructions to seek in-person clinical evaluation if the band is elevated. The module is decision support and is not a diagnostic device. Severe suspected cases warrant urgent care before laboratory confirmation.

For partner clinics and laboratories

If your clinic or laboratory routinely handles febrile-illness presentations in regions where Hantavirus is endemic and you would like the module surfaced inside your existing electronic health record workflow, please reach the team via kantesti.net/contact-us. We support deployment under the standard GDPR-compliant and HIPAA-aligned partnership agreement and run integration through a dedicated clinical engineering point of contact.

What the module is not

A short and deliberate inventory of what this tool does not do. We hold ourselves to a higher honesty standard than is typical for this category of product.

Not a diagnostic device. Definitive Hantavirus diagnosis requires laboratory confirmation through IgM serology by ELISA or molecular testing by RT-PCR, per CDC and WHO guidance. The module does not replace either.
Not a substitute for clinical judgement. Final clinical decisions remain with the qualified clinician. The module is one input among several, alongside history-taking, physical examination, alternative diagnoses and the clinician's local epidemiological knowledge.
Not a discriminator across the full prodromal differential. Many of the prodromal findings overlap with leptospirosis, dengue, severe COVID-19 and bacterial sepsis. The module is a triage aid for Hantavirus risk specifically. It does not rank Hantavirus against the other syndromes in the differential.
Not regulatorily classified as a medical device. The module is offered as clinical decision support. It is not classified as a medical device under EU MDR or FDA regulation. Clinics deploying the platform must comply with their own jurisdiction's regulatory regime for the use of decision support in patient care.
Not a sensitivity or specificity claim. The 3 confirmed cases reported from 50,000 interpreted reports reflect cases where partner clinics returned subsequent laboratory confirmation status to the platform. We make no formal claim about sensitivity or specificity in the epidemiological sense.
Not population-specific for paediatrics or pregnancy. Hantavirus in paediatric and pregnant patients carries additional considerations. The module does not currently produce population-specific weightings beyond what the AI infers from the demographic fields. Explicit citable prior weighting for these subgroups is on the roadmap.

Try the Same Module That Caught Three Confirmed Hantavirus Cases

Upload your own blood test panel to the Kantesti AI Engine and request a Hantavirus Risk Assessment. Over 2 million users worldwide use Kantesti to interpret over 15,000 biomarkers across more than 75 languages. The module is free on every plan including the consumer free tier.

🔬 Try Free Demo

Chrome Extension App Store Google Play

📚 How to Cite This Work

BibTeX

@techreport{bulut2026hantavirus,
  author      = {Bulut, Julian Emirhan and Klein, Thomas},
  title       = {Multilingual AI-Assisted Clinical Decision Support
                 for Early Hantavirus Triage: Design, Engineering
                 Validation, and Real-World Deployment Across 50,000
                 Interpreted Blood Test Reports},
  institution = {Kantesti Ltd},
  address     = {London, United Kingdom},
  year        = {2026},
  month       = {May},
  type        = {Technical Report},
  doi         = {10.6084/m9.figshare.32230290},
  url         = {https://doi.org/10.6084/m9.figshare.32230290}
}

APA

Bulut, J. E., & Klein, T. (2026). Multilingual AI-Assisted Clinical Decision Support for Early Hantavirus Triage: Design, Engineering Validation, and Real-World Deployment Across 50,000 Interpreted Blood Test Reports (Technical Report). Kantesti Ltd. https://doi.org/10.6084/m9.figshare.32230290

🌐 Research Platform Mirror Network

🔗 Figshare DOI 10.6084/m9.figshare.32230290 · canonical academic record 🎓 ResearchGate Publication 404699852 · academic discovery layer 📄 Academia.edu Paper 166974095 · academic discovery layer

📖 Related Kantesti Validation Work

Klein, T., & Bulut, J. E. (2026). Clinical Validation of the Kantesti AI Engine on 15 Anonymised Blood Test Cases: A Pre-Registered Rubric-Based Benchmark. Kantesti Ltd. DOI: 10.6084/m9.figshare.32095435.

🔗 DOI 📊 Benchmark page

Klein, T. (2025). Clinical Validation Framework for AI-Powered Blood Test Interpretation: Triple-Blind Validation Methodology, Performance Metrics and Quality Assurance Protocols. Kantesti AI Medical Research.

🩺 Medical Validation hub

📖 External Methodological References

Centers for Disease Control and Prevention (2024). Hantavirus Pulmonary Syndrome (HPS): Clinical and Laboratory Information for Healthcare Workers.

🏥 CDC.gov

World Health Organization (2023). Hantavirus Diseases: Factsheet and Surveillance Guidance. WHO, Geneva.

🌍 WHO.int

Jonsson, C. B., Figueiredo, L. T. M., & Vapalahti, O. (2010). A Global Perspective on Hantavirus Ecology, Epidemiology, and Disease. Clinical Microbiology Reviews, 23(2), 412–441.

🏥 PubMed

Vaheri, A., Strandin, T., Hepojoki, J., et al. (2013). Uncovering the Mysteries of Hantavirus Infections. Nature Reviews Microbiology, 11(8), 539–550.

🏥 PubMed

50,000Reports analysed

3Lab-confirmed cases

75+Languages

127Countries served

Frequently Asked Questions

What is the Kantesti AI Hantavirus Risk Assessment?

The Kantesti AI Hantavirus Risk Assessment is a clinician-facing decision-support module embedded in the Kantesti clinic dashboard. It analyses an interpreted blood test report alongside optional exposure history, symptoms, vital signs and Hantavirus-specific serology, and returns a structured 0 to 100 risk score with rationale, contributing factors, red flags and an explicit declaration of missing critical data. The module is positioned as decision support and is not a diagnostic device.

Can AI detect Hantavirus from a blood test?

AI cannot definitively diagnose Hantavirus from a blood test alone. Definitive diagnosis requires laboratory confirmation by IgM ELISA or RT-PCR per CDC and WHO guidance. AI can however recognise the prodromal laboratory signature of Hantavirus, which includes thrombocytopenia, rising hematocrit, leukocytosis with left shift, elevated lactate, modest transaminase rise and creatinine creep, and flag patients for urgent clinician review before serology returns. The Kantesti AI Hantavirus Risk Assessment caught three laboratory-confirmed cases across 50,000 interpreted blood test reports between February and May 2026, all initially misclassified at presentation as influenza-like illness, atypical pneumonia or bacterial sepsis.

What are the early laboratory signs of Hantavirus on a blood test?

The prodromal laboratory signature of Hantavirus includes thrombocytopenia (platelets often below 150 × 10⁹/L, present in 70 to 90 percent of HPS cases), rising hematocrit from capillary leak, leukocytosis with left shift and immunoblasts on the peripheral smear, elevated lactate, modest AST, ALT and LDH elevation, and creatinine creep. None of these findings is pathognomonic in isolation. Together in the right clinical and exposure context they raise the index of suspicion for Hantavirus syndromes and warrant targeted serology or RT-PCR testing.

How long does it take to confirm Hantavirus after suspicion?

Laboratory confirmation by IgM ELISA or RT-PCR typically returns within 24 to 96 hours of sample submission, depending on regional laboratory capacity. In the Kantesti production deployment, all three laboratory-confirmed Hantavirus cases received serology or molecular confirmation between 24 and 96 hours after the AI module first surfaced a high or critical classification. Severe suspected cases warrant urgent supportive care before laboratory confirmation per CDC and WHO guidance.

How many real patients has the module evaluated?

Across 50,000 consecutive interpreted blood test reports analysed by the Kantesti platform between February and May 2026, the module produced a high or critical risk classification in 14 reports (0.028 percent). Three of those patients received subsequent laboratory confirmation of acute Hantavirus infection by IgM ELISA or RT-PCR, per CDC and WHO recommended diagnostic standards. All three confirmed patients had been initially clinically misclassified at presentation as influenza-like illness, atypical pneumonia or bacterial sepsis.

How is Hantavirus diagnosed?

Definitive Hantavirus diagnosis requires laboratory confirmation through IgM serology by ELISA or molecular testing by RT-PCR, per CDC and WHO guidance. Clinical assessment alone is not sufficient to establish or exclude the diagnosis. The Kantesti module is a triage aid for clinicians and does not replace serology or molecular confirmation. A negative test does not definitively rule out infection because the timing of sampling matters for both serology and PCR.

What blood test parameters does the module use?

The module reads an already-interpreted blood test report including platelet count, hematocrit, white blood cell count, creatinine, AST, ALT, LDH and lactate, alongside the prose interpretation produced by the broader Kantesti pipeline. Optional clinical context includes exposure history, fever, myalgia, dyspnoea, gastrointestinal symptoms, SpO2, blood pressure, heart rate and Hantavirus-specific serology including IgM, IgG and RT-PCR.

Which languages does the module support?

The Kantesti platform serves more than 75 languages with native right-to-left rendering for Arabic, Hebrew and Persian. The Hantavirus Risk Assessment runs in every supported language with no English fallback. If a translation key is missing for the active locale the build fails rather than serving English text. Languages currently shipping include English, Turkish, German, French, Spanish, Italian, Portuguese, Arabic, Hebrew, Greek, Polish, Dutch, Russian, Chinese (Simplified and Traditional), Japanese, Korean, Hindi, Bengali, Persian and many more.

Is the module a diagnostic device?

No. The module is clinical decision support and is not classified as a medical device under EU MDR or FDA regulation. Final clinical judgement remains with the qualified clinician. Definitive Hantavirus diagnosis requires laboratory confirmation per CDC and WHO standards. Clinics deploying the module must comply with their own jurisdiction's regulatory regime for the use of decision support in patient care.

How does Kantesti protect patient data?

Kantesti Ltd operates under GDPR-compliant and HIPAA-aligned data-handling agreements with users and partner clinics, and runs ISO 27001-aligned controls plus SOC 2 Type II security practices. The platform holds CE marking for the European market. The module enforces per-clinic isolation at every storage operation, uses parameterised database queries throughout and inserts AI-generated strings into the DOM exclusively through textContent or attribute assignment. Identifiable patient information is never published, including in this report.

Where can I read the academic version of this work?

The full peer-verifiable preprint is archived on Figshare under DOI 10.6084/m9.figshare.32230290 with mirrors on ResearchGate (publication 404699852) and Academia.edu (paper 166974095). The paper is open access under a CC BY 4.0 licence and includes the full methods, results table, limitations, conflicts of interest and ethical statement.

Why is the high or critical rate so low?

Hantavirus is uncommon globally. A triage tool that fired high on every febrile illness would be clinically useless. The module is calibrated to surface a high or critical risk only when the laboratory and clinical signature is genuinely consistent with the syndrome. The 0.028 percent high-or-critical rate reflects appropriate specificity for a rare condition surveilled through a non-targeted clinical workflow.

Does the module distinguish Hantavirus from other prodromal differentials?

No. Many of the prodromal findings in Hantavirus overlap with leptospirosis, dengue, severe COVID-19, atypical pneumonia and bacterial sepsis. The module is a triage aid for Hantavirus risk specifically. It does not rank Hantavirus against the other syndromes in the differential. Clinicians should treat its output as one input among several and continue to consider the full differential in their workup.

⚕️ Medical Disclaimer & Conflict of Interest

This page describes a clinical decision support module and is provided for clinical and methodological transparency. It does not constitute medical advice. Hantavirus diagnosis requires laboratory confirmation by IgM ELISA or RT-PCR per CDC and WHO guidance. Severe suspected cases warrant urgent care before laboratory confirmation. Always consult a qualified healthcare provider for diagnosis and treatment decisions. Both authors are employed by and hold equity in Kantesti Ltd, and the module under description is a commercial product of the same organisation. This conflict of interest is mitigated by publishing the full preprint open-access under CC BY 4.0, by stating the limitations of the deployment evidence explicitly and by framing the module honestly as decision support rather than as a diagnostic device.

E-E-A-T Trust Signals

⭐

Experience

15+ years of clinical hematology and laboratory medicine practice supervising the prodromal-signature weighting. 50,000 interpreted reports of real production deployment evidence.

📋

Expertise

Module aligned with CDC and WHO clinical guidance. Strict JSON output schema with explicit confidence and missing-data declarations. Schema-constrained rationale auditable line by line.

👤

Authoritativeness

Clinical oversight by Dr. Thomas Klein, MD (ORCID 0009-0009-1490-1321). Engineering by Julian Emirhan Bulut, CEO of Kantesti Ltd. Reviewed by the Kantesti AI Medical Advisory Board.

🛡️

Trustworthiness

Open-access preprint under CC BY 4.0, three-platform research mirror network, GDPR-compliant and HIPAA-aligned operations, ISO 27001-aligned controls, SOC 2 Type II practices, CE marked.

🏢 Kantesti LTD Registered in England & Wales · Company No. 17090423 London, United Kingdom · kantesti.net