西山 聖也

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.

BACKGROUND: The role of arterial blood gas (ABG) testing in the intensive care unit (ICU) remains debated within the "less is more" paradigm. While unnecessary testing may pose risks without benefit, timely ABGs provide critical information in unstable patients. Institutional variation in early ABG utilization and its association with outcomes remains unclear. METHODS: We conducted a multicenter retrospective cohort study using the Japanese Intensive Care PAtient Database (JIPAD) between April 2015 and March 2023. Adult ICU patients with a stay ≥24 h and arterial line placement were included. The standardized number of ABGs (SNABGs) within the first 24 h was calculated as the ratio of observed to expected values, where expectations were derived from a multivariable model adjusting for patient covariates. ICUs were categorized into tertiles according to SNABG utilization. The primary outcome was in-hospital mortality, analyzed using multilevel logistic regression with ICU-level random intercepts. Restricted cubic splines were used to explore non-linear associations. RESULTS: Among 117,546 patients from 87 ICUs, the mean number of ABGs varied widely. After standardization, SNABGs ranged from 0.73-0.90 in the low tertile to 1.09-1.15 in the high tertile. In the multilevel model, SNABG was not significantly associated with in-hospital mortality (adjusted OR 0.942 [95% CI 0.807-1.100] for tertile 2; 0.874 [95% CI 0.751-1.017] for tertile 3). Flexible modeling suggested a non-linear trend toward better outcomes with higher utilization, but confidence intervals included unity. CONCLUSION: Early ABG utilization varied across ICUs, yet was not significantly associated with mortality. Sensitivity analysis suggested a non-linear relationship, with a tendency toward better outcomes at higher utilization. These findings warrant further investigation to clarify the role of early ABG utilization in critical care.

BACKGROUND: The aim of this study was to assess in-hospital mortality and identify its predictors in adult patients with hematological malignancies admitted to intensive care units (ICUs) in Japan. METHODS: We conducted a retrospective cohort study of adult patients with hematological malignancies admitted to ICUs participating in the Japanese Intensive care PAtient Database from 2015 to 2020. The primary outcome was in-hospital mortality. We compared survivors and non-survivors based on their characteristics at ICU admission and ICU treatments. We also assessed the relationship between institutional characteristics and in-hospital mortality. RESULTS: A total of 1,700 patients from 69 institutions were included. In-hospital mortality was 46.2%. The most common reason for ICU admission was respiratory failure (28.2%). Mechanical ventilation and continuous renal replacement therapy were used in 49.0% and 24.6% of patients, respectively. In multivariable logistic regression analysis, a higher in-hospital mortality was independently associated with type of neoplasm, Acute Physiological Assessment and Chronic Health Evaluation III score, invasive mechanical ventilation (OR 1.64, 95% CI 1.30-2.08), noninvasive ventilation (OR 1.71, 95% CI 1.22-2.41), and continuous renal replacement therapy (OR 1.98, 95% CI 1.51-2.61), whereas other patient characteristics (e.g., age, comorbidities, ICU admission source, reason for ICU admission) were not associated. There was also no association between institutional characteristics and in-hospital mortality. CONCLUSIONS: In-hospital mortality of adult patients with hematological malignancies admitted to ICUs remains high. Factors associated with in-hospital mortality in these patients differed from those in the general ICU population. Institutional characteristics were not significantly associated with in-hospital mortality.

BACKGROUND: The optimal strategy for discontinuing arginine vasopressin and norepinephrine in patients recovering from shock remains uncertain. Although prior studies have suggested a higher risk of hypotension when arginine vasopressin is discontinued first, these findings may have been influenced by baseline imbalances and tapering practices. We conducted a retrospective study to evaluate whether the order of discontinuation between arginine vasopressin and norepinephrine was associated with the incidence of hypotension during the recovery phase of shock, with vasopressor end doses converted to norepinephrine equivalents for analysis. METHODS: This was a single-center retrospective cohort study of intensive care unit patients with shock who received both arginine vasopressin and norepinephrine from August 2017 to March 2024. Patients were categorized based on whether arginine vasopressin or norepinephrine was discontinued first. The primary outcome was the incidence of hypotension within 24 h of vasopressor cessation, defined as mean arterial pressure < 60 mmHg requiring a ≥ 25% increase in the remaining vasopressor, reinstitution of the stopped agent, or a bolus of ≥ 500 mL crystalloid or 25 g albumin. Overlap weighting using propensity scores was applied to adjust for baseline imbalances both in the overall cohort and in the septic shock subgroup. Propensity scores were estimated using logistic model, including baseline characteristics, hemodynamic parameters, and vasopressor end doses in norepinephrine equivalents. RESULTS: A total of 524 patients were analyzed, with 293 discontinuing AVP first and 231 discontinuing NE first. In the unadjusted cohorts, hypotension occurred in 19% of the AVP-first group and 26% of the NE-first group. After overlap weighting, all baseline covariates were balanced between the groups, and the incidence of hypotension was not significantly different (19% vs 21%, P = 0.59). In the septic shock subgroup (n = 267), the weighted analysis showed no significant difference in the incidence of hypotension between groups. CONCLUSIONS: In patients recovering from shock who received both arginine vasopressin and norepinephrine, discontinuing arginine vasopressin first was not associated with a higher risk of hypotension.