AITRICS VitalCare

Deep learning scores are the core, and the product also sells a conventional surveillance dashboard. The product page describes AITRICS-VC as AI that predicts, from up to 19 routine EMR values, the risk of major adverse events and sepsis in general wards and of death in the ICU, and the company's research names the models as deep learning.

The same screen also shows patient classification, alarm history, vital signs and conventional NEWS and MEWS scores, and that rules-based part is exactly what the US clearance (K240756) covers on its own. So the models are the engine of the core module and the platform carries other value. For the United States, the company does not say which of the two is being sold.

Advisory by label, with a published method for choosing the alarm threshold. The US clearance states the software is not intended to replace bedside monitors or clinicians' decisions and is indicated solely for the general ward, not the ICU or operating rooms, and the product page says it is for use by doctors and nurses. Scores and alarms go to clinicians with the grounds for each alarm shown; nothing is auto-actioned.

The company's 2026 study sets thresholds by alarm workload (about 10 alarms per 100 patient-days), which gives a hospital a way to pick one. What is missing is what happens when the score and NEWS or MEWS disagree, since both appear on the same screen, and how a false low score is caught. B.

The approach and inputs are named. The product page lists the 19 inputs (systolic and diastolic pressure, pulse, respiration rate, temperature, SpO2, total bilirubin, lactate, pH, sodium, potassium, creatinine, hematocrit, white cell count, bicarbonate, platelets, C-reactive protein, Glasgow Coma Scale and age) and the outputs and horizons (adverse events within six hours, sepsis within four, ICU mortality within six), and the company's research describes deep learning models with development and validation cohorts named. No model version, update practice or change notice is published, and the US-cleared configuration is a different, rules-based function. B.

A structurally short chain: the models are the company's own, drawing up to 19 defined values from the hospital record (six vital signs, eleven blood tests, the Glasgow Coma Scale and age), with no outside model provider. The privacy policy lists the company's own processors and sub-processors, including Amazon Web Services, KT Cloud, Bespin Global and NHN Cloud, for its services generally; which of them, if any, handle hospital data for AITRICS-VC is not stated.

A sizable, company-led literature from Korean hospitals, now including prospective work and alarm burden. Sim and colleagues (Acute and Critical Care, 2025) prospectively validated the major adverse event score on 6,039 ward encounters at Keimyung University Dongsan Hospital: area under the curve 0.918 and precision-recall 0.352 six hours ahead, against 0.797 for NEWS and 0.722 for MEWS, with up to 71 percent fewer false positives per true positive.

Kim and colleagues (BMJ Health and Care Informatics, 2025) prospectively validated the sepsis score on 6,455 patients at the same hospital (0.880, about 68 minutes ahead of the operational diagnosis). A before-and-after study at Presbyterian Medical Center (JMIR Medical Informatics, 2025; 30,785 cases) reported 24.97 percent fewer Code Blue and major events after go-live.

Jang and colleagues (Journal of Clinical Medicine, 2026; Yongin Severance) set an alarm-based threshold giving sensitivity of 0.52 at 9.67 alarms per 100 patient-days, against 0.23 and 0.17 for NEWS and MEWS. The product page reports Korean pivotal trial results (0.961 on 2,953 patients for adverse events; 0.869 on 3,668 for sepsis; 0.975 on 754 for ICU mortality). AITRICS staff lead every one of these studies, each is single-center, and none is from outside Korea. Published research short of independent validation.

General assurances, and the published policy does not reach patient data. The company holds ISO/IEC 27701 privacy management certification alongside ISO 27001 (LRQA, October 2025) and describes de-identification and encryption procedures across its data lifecycle. Its privacy policy, written under Korea's Personal Information Protection Act, covers product inquiries, customer support and recruitment, not the ward data AITRICS-VC processes in hospitals.

Nothing public states how long patient data or scores are kept, whether hospital data is used to retrain the models, or where processing happens for deployments outside Korea. No guardrail testing or safety event process is published. C.

No business associate statement and no privacy document that reaches the product. The published privacy policy, under Korea's Personal Information Protection Act, covers website inquiries, customer support contacts and recruitment; it does not describe the patient data AITRICS-VC processes for hospitals. No US business associate status, agreement or HIPAA position is published, although the company holds a US clearance and has signed with Mayo Clinic Platform. D.

Recognized certifications are named, without the artifacts. In October 2025 the company announced ISO/IEC 27001 information security and ISO/IEC 27701 privacy management certifications from LRQA, reporting an audit of 93 security and 49 privacy controls; it also holds ISO 13485 for medical device quality (2022) and MDSAP (2026), which cover quality rather than security. No certificate, scope statement, trust portal or penetration testing summary is published. B.

Approvals abroad cover the AI; the US clearance covers a rules-based dashboard, and the company calls it FDA clearance for the AI. Korea's MFDS approved AITRICS-VC as a medical device (manufacturing license 22-723), designated it an innovative medical device (No. 27) and admitted it to conditional new health technology assessment; the company also reports registrations in Hong Kong, Vietnam, Indonesia (April 2026) and Malaysia (May 2026) and MDSAP certification.

In the United States, K240756 (23 July 2024; 21 CFR 870.2300, product code PLB, predicate the Capsule Surveillance System) clears AITRICS-VC as software that performs rule-based calculation of NEWS, MEWS and qSOFA and screens ward patients against thresholds; it is indicated solely for the general ward, not the ICU or operating rooms, and the submission states the device does not require clinical data. Nothing in it covers the deep learning deterioration, sepsis or ICU mortality scores.

The company's April 2026 news release nonetheless describes AITRICS-VC as its "AI solution ... which predicts patient deterioration" with approvals in five countries "including FDA clearance in the U.S." D: the cleared label attached to a capability the US clearance does not cover.

Results for one at-risk group and a named ethics process, without a full comparison across groups. The prospective validation at Keimyung reports the major adverse event score separately for obstetrics and gynecology patients, a ward group rarely studied (area under the curve 0.964 against 0.918 overall), and the company's development work reports discrimination held above 0.86 across age and sex categories.

In April 2026 the company announced an AI ethics checklist developed with the Korea Information Society Development Institute, covering human rights, safety and accountability. No results by race or ethnicity, calibration by group, or evaluation outside Korean hospitals are published, and the step from Korean tertiary hospitals to US wards (different case mix, laboratory conventions and staffing) is the transfer this product has not yet been tested on. B.

Error rates at a working threshold are published by the company, with method and denominators. In the Yongin Severance study led by AITRICS staff (Journal of Clinical Medicine, 2026), the major adverse event score at an alarm-based threshold detected 52 percent of deteriorations (sensitivity 0.5244) at 9.67 alarms per 100 patient-days on a common sample of 1,678 events, with specificity above 0.95; the prospective Keimyung study reports false positives per true positive. So a buyer can see that about half of deteriorations are not flagged at that setting. Nothing stands behind the output beyond that: no warranty, indemnity or remediation commitment is published.

Record integration is described generically. The US 510(k) describes the software analyzing patient data sourced from the EHR, and the product page describes connecting to the hospital's internal network to collect EMR data. No record system vendor, interface standard beyond that, marketplace listing or write-back is named. Integration evidently works across Korean hospitals, but nothing tells a US buyer what connecting to Epic, Oracle Health or Meditech involves. C.

Deployment is described in one line and residency is not stated. The product page says the software connects to the hospital's internal network and is viewed on a PC screen; the privacy policy names Amazon Web Services, KT Cloud and Bespin Global as cloud providers for the company's services, without saying whether hospital deployments run on them.

No hosting option, region, tenancy model or residency commitment is published, although the product is registered in Korea, Hong Kong, Vietnam, Indonesia, Malaysia and the United States. A single deployment path with its location implied.

No price. The product is sold to hospitals through an inquiry form, and in Korea it has entered the conditional new health technology assessment program, which bears on how it may be reimbursed there. No rate card, unit of charge or implementation fee is published. C.

Coverage is named, adult and split by setting, with validation in Korean wards. The product covers inpatients 19 and older: adverse event and sepsis scores in general wards and a mortality score in the ICU under the Korean approval, while the US clearance covers only a rules-based general ward dashboard. The ward focus matters because ward patients are measured intermittently rather than continuously.

Validation spans several Korean hospitals (Keimyung Dongsan, Presbyterian Medical Center, Yongin Severance and earlier Yonsei and NHIS Ilsan work), including obstetrics and gynecology wards; none is outside Korea, and no pediatric use is offered. B.