AI-Based Visensia Index vs SMEWS for Mortality Prediction in 13 High-Performing Japanese Hospitals

Tuesday, March 24, 2026

10:00 AM - 11:00 AM Central Time

Location: Connections Central - RST 10

Toru Yoshida, n/a (he/him/his)

JPN

Disclosure(s): No relevant financial relationship(s) to disclose.
TK

Tetsuya Kimura, N/A

PhD
University Hospital, Kyoto Prefectural University of Medicine

Disclosure information not submitted.
Takaki Naito, MD, MPH, PhD (he/him/his)

Saint Marianna University Hospital, Tokyo, JPN

Disclosure(s): Nihon Kohden Corporation: Speaker/Honoraria (includes speakers bureau, symposia, and expert witness) (Terminated, October 16, 2025)
RT

Rimi Tanii, N/A

MD, Assitant Professor
Department of Emergency and Critical Care Medicine, St.Marianna University School of Medicine

Disclosure information not submitted.
DM

Daiki Morikawa, N/A

MD, Assistant Professor
Department of Emergency and Critical Care Medicine, St.Marianna University School of Medicine

Disclosure information not submitted.
JT

Jumpei Tsukuda, MD, MPH, PhD

MD, Assistant Professor
Department of Emergency and Critical Care Medicine, St.Marianna University School of Medicine, Japan

Disclosure information not submitted.
KM

Kenichirou Morisawa, n/a

MD, PhD, Professor
Department of Emergency and Critical Care Medicine, St.Marianna University School of Medicine

Disclosure information not submitted.
SF

Shigeki Fujitani, MD, PhD, FCCM

MD, PhD, Professor and Head of Depparment
St. Marianna University School of Medicine

Disclosure information not submitted.

Introduction: Various early warning scores (EWS) have been proposed for rapid response systems (RRS). We previously used the St. Marianna Modified Early Warning Score (SMEWS), which was auto-calculated but depended on nurse-entered data, risking underestimation due to missing inputs. To address this, we implemented the Visensia Safety Index (VSI), an AI-based score using five parameters: heart rate, blood pressure, respiratory rate, SpO₂, and body temperature. VSI classifies risk as high (≥3.0), medium (2.1–2.9), low (1.5–2.0), or normal (0.0–1.4), and is calculated if ≥3 values are available. We examined the relationship between SMEWS/VSI and unexpected ward deaths within 24 hours before discharge in 13 Japanese hospitals, including time trends.

Methods: We analyzed data from patients aged ≥15 years admitted to and discharged from general wards in 13 RRS Net Study Group hospitals during Jan–Dec 2019 (early) and Apr 2023–Mar 2024 (late), totaling 433,490 cases. For evaluating VSI’s predictive ability, we included cases with (1) SMEWS 0–4 without any parameter scoring 3, and VSI < 1.5 at 24 hours before discharge, and (2) no missing vital signs at discharge. Outcome was ward death at discharge. These 13 institutions were regarded as high-performing hospitals, with median SMRs of 76 (early) and 83 (late).

Results: We identified 77,667 eligible cases. Missing data decreased in the late period. At 1 hour before discharge, VSI ≥3.0 showed 64.3% sensitivity and 33.0% PPV (early), and 60.3%, 32.2% (late). SMEWS ≥7 showed 41.7%, 54.3% (early) and 39.5%, 55.4% (late). At 4 hours before discharge, VSI ≥3.0 had 46.6%, 35.7% (early) and 42.5%, 30.1% (late); SMEWS ≥7 had 31.2%, 64.6% (early) and 24.2%, 52.5% (late). In multivariate analysis adjusting for age, sex, and disease type, both VSI and SMEWS were independent predictors (P < 0.01). R² values were higher for VSI (0.36 early, 0.38 late) than SMEWS (0.28, 0.30).

Conclusions: In 13 high-performing Japanese hospitals, VSI ≥3.0 predicted ward deaths 1 hour before discharge with higher sensitivity but lower positive predictive value (PPV) than SMEWS.VSI consistently had higher R² values, suggesting better model performance and potential advantage with reduced missing data.