Written and maintained by CASRAI Editorial Board
Last updated
Infermedica is an API-first clinical artificial intelligence platform and certified medical device software (CE-marked under the European Medical Device Regulation [EU MDR]) that provides automated symptom assessment, 5-level acuity triage, pre-visit patient intake, and post-visit follow-up for health systems, payers, and telehealth providers. Alongside peer clinical reasoning engines such as Kahun AI, Infermedica exemplifies deterministic and probabilistic symbolic medical AI—using physician-curated Clinical Knowledge Graphs and Bayesian inference networks rather than unconstrained generative Large Language Models (LLMs) to eliminate diagnostic hallucination in AI symptom checker and tele-triage workflows.
This page describes clinical decision support architectures, APIs, and health informatics standards for informational and research purposes; it is not individual medical advice.
Overview: Symbolic Probabilistic Reasoning vs. Generative LLMs in Clinical Triage
In digital front-door patient navigation and clinical decision support systems (CDSS), routing an acute patient safely requires reproducible, auditable probability calculations. Unconstrained generative LLMs can produce non-deterministic triage dispositions or fabricate clinical citations when phrasing changes slightly. Both Infermedica and Kahun AI address this safety requirement by separating language interface layers from a deterministic core reasoning engine:
- Infermedica Architecture (`infermedica api` / `infermedica triage`): Built on a structured medical knowledge base curated by multi-specialty physicians (requiring multiple independent physician reviews per concept link) that connects cientos of conditions, thousands of symptoms, and demographic/epidemiological risk factors. When a patient or nurse enters initial symptoms, Infermedica’s probabilistic Bayesian network computes conditional disease probabilities and dynamically selects the single most informative follow-up question (maximizing information gain / entropy reduction) until a confident 5-level acuity disposition is reached.
- Kahun AI Architecture (`kahun` / `kahun ai`): Developed as an evidence-based clinical reasoning engine modeled on a physician’s diagnostic workup. Kahun’s proprietary knowledge graph maps puluhan million clinical insights extracted directly from peer-reviewed medical journals and clinical guidelines. Every question Kahun asks during a dynamic patient intake interview—and every differential consideration or suggested lab/imaging workup it presents to the clinician—is hyperlinked directly to the underlying peer-reviewed publication snippet and statistical likelihood ratio.
Types and Platform Comparison: Infermedica vs. Kahun AI and Triage Engines
Health systems and digital health architects selecting an API-driven intake and triage engine evaluate platforms across reasoning methodology, regulatory certification, and output structure:
| Platform / Module | Core Reasoning Architecture | Primary Clinical Output | Terminology & Interoperability Standards | Best For / Key Operational Limits |
|---|---|---|---|---|
| Infermedica API & Triage | Multi-level Bayesian probabilistic network over a physician-validated Clinical Knowledge Graph | 5-tier acuity triage level, ranked condition probabilities, red-flag alerts, and specialist routing | ICD-10, SNOMED CT, LOINC; REST JSON API (/v3/diagnosis, /v3/triage, /v3/parse) & HL7 FHIR R4 adapters |
Best for: Payer/hospital digital front doors, nurse call centers, and white-label telehealth routing. Limit: Scoped to primary/urgent triage rather than longitudinal inpatient management. |
| Infermedica Medical Guidance Platform | End-to-end modular SaaS (Triage, Intake, Follow-Up, Call Center Symptom Checker) | Turnkey patient web widgets, pre-visit HPI summary for clinicians, and automated post-visit check-ins | HL7 FHIR R4 (QuestionnaireResponse, Observation, Condition), SMART on FHIR EHR launch |
Best for: Rapid deployment without building custom frontend UI. Limit: Custom UI styling is more constrained than raw API integration. |
| Kahun AI (`Kahun`) | Peer-reviewed literature knowledge graph calculating dynamic pretest/posttest likelihood ratios | Citation-linked differential diagnosis, suggested diagnostic workup, and structured HPI clinical note | SNOMED CT, ICD-10, LOINC; HL7 FHIR R4 EHR write-back & LLM-assisted conversational wrapper | Best for: Physician-facing diagnostic support, pre-visit clinical intake, and transparent FDA § 520(o) CDS. Limit: Designed for detailed clinical history rather than 60-second call-center routing. |
| Fabric Health & Schmitt-Thompson | Deterministic clinical decision trees (STCC protocols) and conversational pathway routing | Standardized nurse telephone triage dispositions and asynchronous virtual care protocols | HL7 FHIR R4, ClearTriage / Epic CRM integration | Best for: Traditional nurse triage telephone lines requiring strict rule-tree adherence. |
Infermedica API Endpoints, 5-Level Triage Scale, and Technical Workflow
Developers integrating the Infermedica API (v3) into a custom patient portal, mobile application, or nurse triage workstation orchestrate a stateful interview loop across four primary RESTful endpoints:
- Natural Language Symptom Extraction (`POST /v3/parse`):
- Accepts free-text patient input (e.g., “I have had a sharp pain in my lower right abdomen and nausea since this morning”) along with patient age and biological sex, returning matched concept IDs (
s_...for symptoms,p_...for risk factors) with negation and duration attributes.
- Accepts free-text patient input (e.g., “I have had a sharp pain in my lower right abdomen and nausea since this morning”) along with patient age and biological sex, returning matched concept IDs (
- Dynamic Bayesian Question Loop (`POST /v3/diagnosis`):
- Accepts the array of
evidenceitems (each marked"present","absent", or"unknown"), biological sex, age, and optional epidemiological risk factors (geographic region, pregnancy status, hypertension, diabetes). - Returns the next dynamically selected question (
question.type:single,group_single, orgroup_multiple), updated ranked condition probabilities, and a booleanshould_stopflag indicating when sufficient statistical certainty has been achieved.
- Accepts the array of
- Acuity Disposition Calculation (`POST /v3/triage`):
- Once
should_stop: trueis reached (or an acute red-flag symptom is detected), calling/v3/triagereturns one of Infermedica’s five standardized triage acuity levels:emergency_ambulance: Immediate life-threatening presentation requiring emergency medical services (EMS).emergency: Urgent hospital Emergency Department evaluation required.consultation_24: Medical evaluation required within 24 hours (urgent care or same-day clinic).consultation: Non-urgent medical evaluation recommended (scheduled primary or specialty care).self_care: Mild, self-limiting symptoms appropriate for home monitoring and self-care guidance.
- Once
- Reasoning Transparency & Specialist Routing (`POST /v3/explain` & `POST /v3/recommend_specialist`):
/v3/explainreturns the specific supporting evidence (`supporting_evidence`), conflicting evidence (`conflicting_evidence`), and unconfirmed evidence driving a specific condition’s probability, while/v3/recommend_specialistmaps the triage outcome to the appropriate clinical specialty (e.g., Cardiology, Gastroenterology, Otolaryngology).
How to Choose by Clinical Setting: Hospital Front Door, Ambulatory Clinic, or Research
- Health System Digital Front Door & Payer Nurse Triage Lines: Choose Infermedica Triage / API when the primary objective is routing patients across care settings (Emergency Department vs. Urgent Care vs. Telehealth vs. Self-Care) in 12–18 dynamic questions with formal EU MDR Class IIb medical device certification and multi-language localization.
- Ambulatory Pre-Visit Intake & Physician Diagnostic Workup Support: Choose Kahun AI when clinicians require a comprehensive pre-visit History of Present Illness (HPI) note accompanied by transparent, peer-reviewed journal citations and sensitivity/specificity likelihood ratios for every diagnostic consideration—satisfying the transparency criteria of Section 520(o)(1)(E) of the U.S. Federal Food, Drug, and Cosmetic Act (FD&C Act) so the clinician can independently review the basis of every recommendation.
- Clinical Research Cohort Screening: Both engines can be deployed via API in decentralized clinical trials to pre-screen participant symptom clusters against standardized SNOMED CT and ICD-10 inclusion/exclusion criteria prior to coordinator review.
Regulatory Labeling, Quality Management, and Maintenance
- Medical Device Quality & Regulatory Status: Infermedica’s core triage and symptom-checking modules are certified under ISO 13485 (Medical Devices Quality Management Systems) and hold CE marking under the European Union Medical Device Regulation (EU MDR 2017/745, Class IIb). When evaluating any clinical reasoning API, verify the current regulatory classification in the target jurisdiction, HIPAA Business Associate Agreement (BAA) terms, and SOC 2 Type II security attestation.
- Knowledge Graph Maintenance and Epidemiological Updates: Clinical reasoning engines require continuous maintenance to remain safe. Infermedica and Kahun release periodic knowledge-graph updates incorporating revised clinical practice guidelines (e.g., AHA, ESC, NICE), seasonal infectious disease prevalence shifts, and updated ICD-10-CM/SNOMED CT mappings; API integrators must pin or test model version headers (`Model` header in Infermedica API) inside staging environments before promoting updates to production.
Frequently Asked Questions
What is infermedica?
Infermedica is a clinical artificial intelligence company and API platform that uses a physician-curated Clinical Knowledge Graph and Bayesian probabilistic inference engine to conduct dynamic symptom assessments, calculate 5-level acuity triage dispositions, automate pre-visit patient intake, and route patients to the appropriate level of care.
How is infermedica used in clinical research and healthcare?
In healthcare and clinical informatics, Infermedica is embedded via REST APIs or SMART on FHIR widgets into hospital patient portals, health insurance member apps, telehealth platforms, and nurse triage call centers. It translates patient-reported symptoms into standardized SNOMED CT and ICD-10 clinical concepts, flags red-flag emergency presentations, reduces unnecessary Emergency Department visits through 5-tier triage routing, and delivers structured pre-visit intake summaries directly into the clinician’s EHR workflow.
Sources
- Infermedica. Infermedica Developer Portal & API v3 Technical Documentation (`/diagnosis`, `/triage`, `/parse`, `/explain`). https://developer.infermedica.com/
- Kahun Medical. Kahun Evidence-Based Clinical Reasoning Engine & Knowledge Graph Documentation. https://www.kahun.com/
- U.S. Food and Drug Administration (FDA). Clinical Decision Support Software: Guidance for Industry and Food and Drug Administration Staff (FD&C Act § 520(o)). https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-decision-support-software
