方山 真朱

OBJECTIVES: To operationalize and temporally validate an electronic medical record (EMR)-integrated machine learning system (Big data-driven Evaluation of Survival and Treatment in Acute Illness [BEST-AI]) that generates hourly predictions for multiple ICU outcomes, with emphasis on discrimination, calibration, and workflow integration. DESIGN: Single-center hybrid study with stepwise clinical deployment and forward-in-time temporal validation. SETTING: Thirty-bed tertiary mixed medical-surgical ICU in Japan. PATIENTS: All ICU admissions from August 2017 to March 2025. Exclusions: age younger than 16 years or ICU stay less than 4 hours. Development cohort (n = 11,176; from August 2017 to July 2024) and temporal validation cohort (n = 1,127; from August 2024 to March 2025). INTERVENTIONS: EMR-integrated deployment of BEST-AI providing hourly probabilistic predictions to clinicians within the EMR; no protocolized clinical interventions were mandated. MEASUREMENTS AND MAIN RESULTS: Six prediction tasks (in-hospital mortality, ICU mortality, ICU discharge ≤ 72 hr, intubation ≤ 72 hr, extubation ≤ 72 hr, tracheostomy at ICU discharge) were evaluated. In temporal validation, the area under the receiver operating characteristic curves ranged from 0.856 to 0.960, and the area under the precision-recall curves from 0.302 to 0.786. Decile-based calibration showed overall good agreement; hospital mortality was slightly overestimated at higher predicted probabilities, whereas ICU mortality remained well aligned. The intubation task had comparatively lower discrimination and greater deviation from perfect calibration, consistent with low event counts and heterogeneous timing. A 24-hour landmark sensitivity analysis (one prediction per patient at 24 hr after ICU admission) preserved discrimination and calibration relative to the main analysis, supporting robustness beyond repeated-measures evaluation. The system was successfully maintained with automated hourly updates and EMR-embedded patient- and unit-level visualizations, without prescriptive alerts. CONCLUSIONS: A continuously deployed, EMR-integrated ICU prediction system achieved strong temporal discrimination and generally good calibration. Embedding real-time predictions into routine workflow was feasible, and the system was maintained with automated hourly updates. Prospective multicenter studies are warranted to assess transportability and clinical impact.

OBJECTIVE: The aim of this study was to identify factors that predict admission to the intensive care unit (ICU) after activation of a rapid response system (RRS). METHODS: We conducted a retrospective observational study using data from 12,306 RRS activations recorded in the In-Hospital Emergency Registry in Japan database between November 2017 and September 2023. Patients aged under 18 years, noninpatients, and those who died or were transferred immediately after RRS activation were excluded. The primary outcome was ICU admission after RRS activation. Predictive factors were identified using multivariable logistic regression models: Model 1 included all available data, while model 2 was restricted to data available at the time of RRS activation. RESULTS: We analysed data from 8532 patients; 2298 (26.9%) were admitted to the ICU following RRS activation. Significant factors of ICU admission in model 1 included weekend activation (odds ratio [OR] = 1.17; 95% confidence interval [CI] = 1.02, 1.34), oxygen administration prior to activation (OR = 1.23; 95% CI = 1.08, 1.4), ICU discharge within 72 h before the index event (OR = 1.65; 95% CI = 1.28, 2.11), physician-initiated activation (OR = 2.16; 95% CI = 1.87, 2.50), and multiple abnormal vital signs. Model 2, which was limited to information available at the time of RRS activation, identified a similar pattern of associations. CONCLUSION: This study identified several important factors associated with ICU admission following RRS activation. These findings may support improved clinical decision-making regarding ICU transfers and provide a foundation for future work to develop and validate prediction models tailored to this setting.

BACKGROUND: The Sequential Organ Failure Assessment (SOFA)-2 score was developed to better reflect contemporary critical care practice by incorporating modern organ support modalities and updated thresholds based on recent data. However, the generalizability of this framework to intensive care unit (ICU) populations beyond the development cohort, particularly across organ support subgroups and major disease categories, remains uncertain. We aimed to evaluate the external validity of SOFA-2 using the OneICU database, a large Japanese critical care database with comprehensive domain-level data. METHODS: Adult ICU stays between February 2013 and August 2025 were included and classified into two cohorts: those with complete SOFA-1 and SOFA-2 component data on the day of ICU admission, and those with complete SOFA-2 data on that day. Discriminatory performance for ICU mortality was evaluated using the area under the receiver operating characteristic curve (AUROC) and compared between SOFA-1 and SOFA-2 using the DeLong test. Subgroup analyses were performed by major organ support device use and across disease categories. RESULTS: Among 152,883 eligible ICU stays, 67,116 had complete SOFA-1 and SOFA-2 data, and 121,443 had complete SOFA-2 data. SOFA-2 showed a slightly higher AUROC for ICU mortality than SOFA-1 (0.859 vs. 0.853; p < 0.001), although the absolute difference was small. Across subgroups defined by mechanical circulatory support use, SOFA-2 showed higher discrimination than SOFA-1. Discrimination was similar in other device-defined subgroups and in patients readmitted to the ICU. SOFA-2 also demonstrated good discrimination across major diagnostic groups. CONCLUSIONS: SOFA-2 showed similar discrimination for ICU mortality compared with SOFA-1 and maintained broadly comparable performance across clinically relevant subgroups, supporting its applicability for early severity assessment in heterogeneous ICU populations. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s13054-026-06020-x.