Takuya Nishino

Background: Postoperative complications after noncardiac surgery under general anesthesia remain a major cause of morbidity, mortality, and resource-intensive care. Conventional risk assessment tools rely mainly on preoperative information and may not capture perioperative physiological and procedural factors associated with postoperative risk. Objective: This retrospective multicenter observational study aimed to develop and validate electronic health record (EHR)–based machine learning (ML) models for predicting major postoperative complications and to evaluate their discrimination, overall probabilistic performance, calibration, and potential clinical utility. Methods: We analyzed 35,702 adults who underwent noncardiac surgery under general anesthesia at 3 tertiary hospitals between January 2021 and March 2025. Preoperative, intraoperative, and early postoperative variables were extracted from diagnosis procedure combination (DPC) claims, EHRs, and anesthesia information systems. The primary outcome was a 30-day composite of in-hospital death, unplanned reoperation, prolonged mechanical ventilation, new initiation of renal replacement therapy, and postoperative infection. Logistic regression (LR), extreme gradient boosting (XGB), and light gradient boosting machine (LGBM) models were developed using preoperative variables alone or with 7 additional intraoperative and early postoperative variables. Leave-one-facility-out internal-external cross-validation (IECV) was the primary validation framework, with a stratified random 80:20 hold-out analysis as secondary validation. Performance was assessed using the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (PR-AUC), Brier score, scaled Brier score, calibration measures, and decision curve analysis. Results: The primary composite outcome occurred in 3884 of 35,702 patients (10.9%). In pooled IECV, perioperative models consistently outperformed the corresponding preoperative models. AUROC increased from 0.784‐0.791 to 0.810‐0.817, PR-AUC increased from 0.356‐0.361 to 0.405‐0.413, Brier scores decreased from 0.082‐0.083 to 0.078‐0.080, and scaled Brier scores increased from 0.144‐0.151 to 0.175‐0.191. Calibration varied across facilities; pooled calibration slopes for the perioperative models ranged from 0.875 to 0.902, and calibration intercepts ranged from −0.367 to −0.222, indicating overly extreme predictions and overall risk overestimation. Exploratory decision curve analysis generally showed greater net benefit for perioperative models across threshold probabilities of 0.01‐0.30, although the curves overlapped substantially. Secondary and sensitivity analyses were consistent with the primary findings. Conclusions: Adding routinely available intraoperative and early postoperative variables produced modest but consistent improvements in postoperative risk prediction compared with preoperative variables alone. However, facility-specific variation in calibration limits direct transportability of absolute risk estimates. The models may support prioritization for postoperative clinical reassessment, but exploratory decision curve analysis did not establish a specific intervention or triage threshold. Local recalibration and prospective workflow evaluation are required before clinical implementation.

Background: Medical Emergency Team (MET) activation represents acute clinical deterioration requiring rapid decisions regarding escalation of care. Previous post-MET prediction studies have evaluated outcomes including unplanned intensive care unit (ICU) admission and mortality; however, ICU admission is a location-based outcome that may be influenced by institutional factors. Prediction based on treatment intensity rather than location of care has received less attention. Methods: We conducted a retrospective single-center cohort study of adult patients undergoing MET activation on general wards at a tertiary-care hospital in Japan between May 2021 and December 2025. Two outcomes were evaluated: initiation of ICU-level interventions within 24 hours and all-cause 14-day mortality. ICU-level interventions were defined as the initiation of mechanical ventilation, vasopressor or inotropic infusion, renal replacement therapy, or polymyxin B hemoperfusion (PMX). Multivariable logistic regression and Cox proportional hazards models were developed using variables available at the time of MET activation. Model performance was evaluated using discrimination (C-index), calibration, decision curve analysis, and bootstrap-based internal validation. Results: Among 533 patients, 303 (56.8%) required ICU-level interventions and 147 (27.6%) died within 14 days. Lower albumin levels and higher blood urea nitrogen, potassium, and white blood cell count were associated with both outcomes, whereas higher heart rate was associated with 14-day mortality. The model demonstrated moderate discrimination for ICU-level intervention (C-index 0.721, 95% CI 0.660–0.745) and good discrimination for 14-day mortality (C-index 0.784, 95% CI 0.720–0.814). Calibration analyses indicated modest overfitting but generally good agreement between predicted and observed risks. Decision curve analysis demonstrated clinical net benefit across a range of threshold probabilities. Conclusions: In this single-center cohort, treatment escalation and short-term mortality represented related but distinct dimensions of clinical risk. Defining intensive care need based on treatment intensity rather than ICU admission may provide an operational framework for risk stratification at MET activation. External validation is required before clinical implementation.

OBJECTIVES: To determine whether early total bilirubin (T-Bil) trajectories during the first 72 hours after intubation identify trajectory-defined groups with distinct prognostic profiles associated with short-term mortality in mechanically ventilated ICU patients. DESIGN: Multicenter retrospective cohort study with a prespecified day 3 landmark design to mitigate immortal-time bias. SETTING: Mixed medical-surgical ICUs at four tertiary centers. PATIENTS: Adults receiving invasive mechanical ventilation who survived to the day 3 landmark and had serial T-Bil measurements on day 0 and day 3 (± 24 hr). INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Patients were classified according to baseline T-Bil (≤ 1.2 vs. > 1.2 mg/dL) and the direction of change between day 0 and day 3 (change in T-Bil [ΔT-Bil]), yielding four trajectory-defined groups. The primary outcome was 30-day all-cause mortality after the day 3 landmark; 90-day mortality was secondary. Conditional survival was assessed using Kaplan-Meier methods with log-rank tests. Multivariable Cox models adjusted for prespecified covariates available by day 3 (age, sex, body mass index, intubating department, baseline mean arterial pressure, log-transformed aspartate aminotransferase/alanine aminotransferase ratio and platelet count, creatinine, renal replacement therapy, and vasoactive support). Rising T-Bil trajectories were independently associated with higher 30-day mortality, with the strongest association among patients with normal baseline T-Bil who exhibited early increases. In contrast, elevated baseline T-Bil followed by early decline was not associated with excess risk. Findings were consistent in complete-case analyses, in models incorporating day 3 partial Sequential Organ Failure Assessment for severity adjustment, and in analyses of 90-day mortality. Restricted cubic spline analyses demonstrated a nonlinear dose-response relationship between ΔT-Bil and mortality. CONCLUSIONS: Early T-Bil dynamics within 72 hours after intubation were associated with distinct prognostic profiles beyond baseline bilirubin levels and may provide a pragmatic marker for early risk stratification in mechanically ventilated ICU patients.

Background: Hypnotics are frequently prescribed to hospitalized older adults, but comparative evidence on fall risk across hypnotic classes remains inconsistent, partly due to confounding by indication and inadequate consideration of time at risk during hospitalization. Methods: We conducted a retrospective multicenter landmark analysis using administrative claims and electronic medical records from two acute-care hospitals in Japan (2018–2024). A day-7 landmark was defined, including patients aged ≥65 years who remained hospitalized and fall-free through hospital day 7. Sustained hypnotic exposure during hospital days 4–7 (≥2 days) was categorized as no sustained use, benzodiazepines/Z-drugs (BZ/Zs) alone, orexin receptor antagonists or ramelteon (ORA/Ram) alone, or combination therapy. The primary outcome was time to first in-hospital fall during hospital days 8–37. Cox proportional hazards models with multiple imputation were used. Competing-risk analyses and propensity score–matched comparisons between single-class users were conducted as sensitivity analyses. Results: Among 61,663 patients, 3,884 (6.3%) experienced an in-hospital fall after the landmark. Compared with no sustained hypnotic use, adjusted hazard ratios (HRs) for falls were 1.42 (95% CI, 1.24–1.64) for BZ/Zs, 1.46 (95% CI, 1.25–1.70) for ORA/Ram, and 1.42 (95% CI, 1.05–1.91) for combination therapy. Results were consistent in competing-risk analyses. In propensity score–matched analyses restricted to single-class users, fall risk did not differ significantly between BZ/Zs and ORA/Ram (adjusted HR, 0.98; 95% CI, 0.75–1.28). Conclusions: Sustained hypnotic use during hospitalization was associated with a higher incidence of in-hospital falls among older adults. After adjustment for measured clinical factors and treatment selection, no significant difference in fall risk was observed between BZ/Zs and ORA/Ram.

Heart rate (HR) reflects illness severity in critically ill patients, but the prognostic significance of early HR changes is unclear. We aimed to identify HR trajectory phenotypes during the first 24 h of ICU admission and assess their association with short-term mortality. Adults admitted to the intensive care unit (ICU) between January 2018 and December 2024 with ≥ 12 h of HR data were eligible; patients with early discharge/death, mechanical support, or sustained bradycardia were excluded. Hourly HR means from the first 24 h were imputed and clustered via Dynamic Time Warping-based TimeSeriesKMeans; five trajectories were selected using internal validation. Thirty- and 90-day mortality were assessed by multivariable Cox models. A total of 4491 patients were analyzed after applying exclusion criteria. TimeSeriesKMeans identified five early HR trajectory phenotypes: Cluster A (persistently tachycardic), Cluster B (early HR normalization), Cluster C (rising HR), Cluster D (mild tachycardic-normal), and Cluster E (normal HR). Cluster A was associated with a markedly increased risk of 30-day mortality compared with the reference Cluster E (HR 3.21, 95% CI 2.32–4.45; p < 0.001). Cluster C also showed an elevated mortality risk (HR 2.09, 95% CI 1.53–2.85; p < 0.001), whereas clusters characterized by lower or early-normalizing heart-rate trajectories demonstrated more favorable survival profiles. Kaplan-Meier analyses confirmed significant differences in survival across clusters (log-rank p < 0.001). Machine learning-based phenotyping of 24-hour HR trajectories stratified short-term mortality risk and revealed time-dependent patterns potentially reflecting treatment responsiveness. Early adverse heart-rate trajectories, particularly persistent tachycardia, may serve as prognostic signals of physiological stress or poor treatment response during the initial ICU course, supporting early risk stratification and clinical reassessment.

Background: Falls among hospitalized patients are a critical issue that often leads to prolonged hospital stays and increased health care costs. Traditional fall risk assessments typically rely on standardized scoring systems; however, these may fail to capture the complex and multifactorial nature of fall risk factors. Objective: This retrospective observational multicenter study aimed to develop and validate a machine learning–based model to predict in-hospital falls and to evaluate its performance in terms of discrimination and calibration. Methods: We analyzed the data of 83,917 inpatients aged 65 years and older with a hospital stay of at least 3 days. Using Diagnosis Procedure Combination data and laboratory results, we extracted demographic, clinical, functional, and pharmacological variables. Following the selection of 30 key features, 4 predictive models were constructed: logistic regression, extreme gradient boosting, light gradient boosting machine (LGBM), and categorical boosting (CatBoost). The synthetic minority oversampling technique and isotonic regression calibration were applied to improve the prediction quality and address class imbalance. Results: Falls occurred in 2173 (2.6%) patients. CatBoost achieved the highest F1-score (0.189, 95% CI 0.162-0.215) and area under the precision-recall curve (0.112, 95% CI 0.091-0.136), whereas LGBM had the best calibration slope (0.964, 95% CI 0.858-1.070) with good discrimination (F1-score 0.182, 95% CI 0.156-0.209; area under the precision-recall curve 0.094, 95% CI 0.078-0.113). Logistic regression had the lowest discrimination (F1-score 0.120, 95% CI 0.100-0.143). Shapley Additive Explanations analysis consistently identified low albumin, impaired transfer ability, and the use of sedative-hypnotics or diabetes medications as major contributors to fall risk. In incident report analysis (n=435), 49.2% of falls were toileting-related, peaking between 4 and 6 AM, with bedside falls predominating in high or very high risk groups. Conclusions: CatBoost and LGBM offer clinically valuable prediction performance, with CatBoost favored for high-risk patient identification and LGBM for probability-based intervention thresholds. Integrating such models into electronic health records could enable real-time risk scoring and trigger targeted interventions (eg, toileting assistance and mobility support). Future work should incorporate dynamic, time-varying patient data to improve real-time risk prediction.

Background: Multiple long-term conditions (MLTCs) are common among individuals with heart failure (HF); however, the influence of sex on clinical characteristics and prognosis remains unclear. This study investigated sex-related differences in clinical characteristics and prognosis in individuals with HF and a high MLTC burden. Methods: We conducted a multicentre retrospective study of 3004 hospitalised patients with HF who survived to discharge. Comorbid chronic conditions were defined using the Charlson Comorbidity Index (CCI), and MLTC burden was quantified using the age-adjusted CCI. Patients were stratified into high (CCI≥6; n=1514) and low (CCI<6; n=1490) MLTC burden groups based on the median age-adjusted CCI, and comparisons were made by sex. The primary outcome was a composite of all-cause death or HF readmission within 1 year. Results: The prevalence of ≥2 comorbid conditions increased with age, peaking before age 85 and declining slightly thereafter. This trend differed by sex, with a steeper age-related increase observed in men. In the high MLTC burden group, females were older, had higher left ventricular ejection fraction and were prescribed fewer medications than men. Dementia and rheumatologic disease were more common in women. Although no significant sex differences in event-free survival were observed overall in either MLTC group, among individuals aged≥75 years with high MLTC burden, men had a significantly higher risk of adverse events than women (HR: 1.334; 95% CI 1.031 to 1.727). A spline-based three-way interaction analysis (age×sex×MLTC burden) demonstrated that male risk exceeded female risk after age 70, only in the high MLTC burden group. Conclusion: In individuals with HF and a high MLTC burden, age-dependent sex differences in prognosis were evident. These findings highlight the importance of incorporating age-specific and sex-specific approaches into personalised HF care. Trial registration number: UMIN000054854.

BACKGROUND Falls among hospitalized patients are a critical issue that often leads to prolonged hospital stays and increased health care costs. Traditional fall risk assessments typically rely on standardized scoring systems; however, these may fail to capture the complex and multifactorial nature of fall risk factors. OBJECTIVE This retrospective observational multicenter study aimed to develop and validate a machine learning–based model to predict in-hospital falls and to evaluate its performance in terms of discrimination and calibration. METHODS We analyzed the data of 83,917 inpatients aged 65 years and older with a hospital stay of at least 3 days. Using Diagnosis Procedure Combination data and laboratory results, we extracted demographic, clinical, functional, and pharmacological variables. Following the selection of 30 key features, 4 predictive models were constructed: logistic regression, extreme gradient boosting, light gradient boosting machine (LGBM), and categorical boosting (CatBoost). The synthetic minority oversampling technique and isotonic regression calibration were applied to improve the prediction quality and address class imbalance. RESULTS Falls occurred in 2173 (2.6%) patients. CatBoost achieved the highest F1-score (0.189, 95% CI 0.162-0.215) and area under the precision-recall curve (0.112, 95% CI 0.091-0.136), whereas LGBM had the best calibration slope (0.964, 95% CI 0.858-1.070) with good discrimination (F1-score 0.182, 95% CI 0.156-0.209; area under the precision-recall curve 0.094, 95% CI 0.078-0.113). Logistic regression had the lowest discrimination (F1-score 0.120, 95% CI 0.100-0.143). Shapley Additive Explanations analysis consistently identified low albumin, impaired transfer ability, and the use of sedative-hypnotics or diabetes medications as major contributors to fall risk. In incident report analysis (n=435), 49.2% of falls were toileting-related, peaking between 4 and 6 AM, with bedside falls predominating in high or very high risk groups. CONCLUSIONS CatBoost and LGBM offer clinically valuable prediction performance, with CatBoost favored for high-risk patient identification and LGBM for probability-based intervention thresholds. Integrating such models into electronic health records could enable real-time risk scoring and trigger targeted interventions (eg, toileting assistance and mobility support). Future work should incorporate dynamic, time-varying patient data to improve real-time risk prediction.