Hospital Mortality

ImportanceArtificial intelligence models that predict same-admission outcomes for hospitalized patients, such as inpatient mortality, often rely on International Classification of Diseases (ICD) diagnostic codes, even when these codes are not finalized until after discharge.ObjectiveTo investigate the extent to which the inclusion of ICD codes as features in predictive models are associated with inflated performance metrics via label leakage (eg, including the code for cardiac arrest into an inpatient mortality prediction model) and assess the prevalence and implications of this practice in existing literature.Design, setting, and participantsThis prognostic study examined publicly available, deidentified inpatient electronic health record data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Patients admitted to an intensive care unit or emergency department at Beth Israel Deaconess Medical Center between January 1, 2008, and December 31, 2019, were included. These data were analyzed between December 18, 2024, and January 14, 2025. A targeted literature review of same-admission prediction models using MIMIC with ICD codes as features was performed between November 20 and 27, 2024.Main outcome and measuresUsing a standard training-validation-test split procedure, prediction models were developed for inpatient mortality (logistic regression, random forest, and XGBoost) using only ICD codes as features. Performance in the test set was analyzed using areas under the receiver operating curve and variable importance. Frequencies of studies using same-admission prediction models using MIMIC with ICD codes were calculated from the targeted literature review.ResultsThe study cohort consisted of 180 640 patients (mean [SD] age at admission, 58.7 [19.2] years; 53.0% female), of whom 8573 (4.7%) died during the admission. The models using ICD codes predicted in-hospital mortality with high performance in the test dataset, with areas under the receiver operating curve of 0.976 (95% CI, 0.973-0.980) (logistic regression), 0.971 (95% CI, 0.967-0.974) (random forest), and 0.973 (95% CI, 0.968-0.977) (XGBoost). The most important ICD codes were subdural hemorrhage (OR, 389.99; 95% CI, 28.79-5283.59), cardiac arrest (OR, 219.58; 95% CI, 159.61-302.08), brain death (OR, 112.78; 95% CI, 13.42-947.70), and encounter for palliative care (OR, 98.04; 95% CI, 83.16-115.58). The literature review found that 37 of 92 studies (40.2%) using MIMIC to predict same-admission outcomes included ICD codes as features, even though both MIMIC publications and documentation clearly state that ICD codes are derived after discharge.Conclusions and relevanceThis prognostic study of the MIMIC-IV database suggests that using ICD codes as features in same-admission prediction models may be a severe methodological flaw associated with inflated performance metrics, rendering models incapable of clinically useful predictions. The literature review found that the practice is common. Addressing this challenge is essential for advancing trustworthy artificial intelligence in health care.

BackgroundThe HALP (Hemoglobin, Albumin, Lymphocyte, and Platelet) score integrates key parameters reflecting nutritional and immune status. However, its prognostic value for in-hospital mortality in sepsis patients remains underexplored.ObjectiveTo investigate the association between HALP score and in-hospital mortality in sepsis patients using two large critical care databases.MethodsWe conducted a retrospective cohort study including adult patients with Sepsis-3 from the eICU Collaborative Research Database (derivation cohort, n = 12,899) and the MIMIC-IV database (validation cohort, n = 3,726). HALP was calculated as (hemoglobin × albumin × lymphocyte count)/platelet count, using first available values upon ICU admission. Restricted cubic spline (RCS) models assessed nonlinear relationships between HALP and mortality. Kaplan-Meier survival curves and multivariable Cox regression models, adjusted for demographics, comorbidities, laboratory values, and Acute Physiology Score III, evaluated survival differences between low- and high-HALP groups. Segmented Cox regression examined associations below and above RCS-derived thresholds.ResultsA total of 16,625 patients were analyzed. RCS analysis demonstrated significant nonlinear associations between HALP and in-hospital mortality in both cohorts (overall P P P = 0.002 in MIMIC-IV), with an inflection point near 12.45. Below this threshold, each unit increase in HALP was associated with a 3% reduction in mortality risk (eICU: HR 0.97, 95% CI 0.95-0.99, P = 0.002; MIMIC-IV: HR 0.97, 95% CI 0.94-0.99, P = 0.008). Kaplan-Meier analyses showed significantly higher survival in the high HALP group (eICU: log-rank P = 0.005, HR 0.882, 95% CI 0.808-0.962; MIMIC-IV: log-rank P P = 0.017; MIMIC-IV: HR 0.85, 95% CI 0.74-0.98, P = 0.028).ConclusionThe HALP score demonstrates robust prognostic value for predicting in-hospital mortality in sepsis patients, with consistent nonlinear relationships validated across two large databases. Its simplicity and reliance on routine laboratory parameters support potential clinical application in sepsis risk stratification.

ImportanceGeriatric syndromes are common in hospitalized older adults and complicate acute care; however, their overall prevalence and cumulative burden remain poorly understood, especially in resource-limited settings.ObjectivesTo measure the prevalence of geriatric syndromes upon hospital admission and examine the independent association between the number of geriatric syndromes and 90-day mortality.Design, setting, and participantsThis cohort study used data from the Creating a Hospital Assessment Network in Geriatrics (CHANGE) study, a multicenter, prospective cohort of 43 hospitals, including 38 in Brazil, 1 in Angola, 1 in Chile, 2 in Colombia, and 1 in Portugal. Consecutive patients aged 65 years or older admitted under geriatric teams between June 1, 2022, and December 31, 2023, were enrolled within 48 hours; patients with terminally illness were excluded. Data were analyzed from February 1 to November 23, 2025.ExposureA standardized comprehensive geriatric assessment captured 14 geriatric syndromes: loneliness, dementia, depressive symptoms, sensory impairment, disability, immobility, incontinence, falls, frailty, malnutrition, pressure ulcers, polypharmacy, potentially inappropriate medications, and delirium. The exposure of interest was the within-patient count of syndromes.Main outcomes and measuresThe primary outcome was 90-day all-cause mortality, ascertained by masked telephone follow-up with verification in medical records or public registries. Prespecified mixed-effects Cox proportional hazards regression were performed.ResultsThe study included 2556 participants (mean [SD] age, 79 [9] years, 1437 female [56.2%]). The median number of geriatric syndromes was 5 (IQR, 3-8). The highest prevalence rates for syndromes were 70.8% (95% CI, 69.1%-72.6%) for disability, 61.7% (95% CI, 59.8%-63.6%) for polypharmacy, 58.2% (95% CI, 56.3%-60.1%) for frailty, and 54.7% (95% CI, 52.8%-56.7%) for sensory impairment. Across categories, the mortality rate rose from 8.4% (95% CI, 6.2%-11.4%) for 0 to 2 syndromes to 12.7% (95% CI, 10.1%-15.7%) for 3 to 4 syndromes, 25.4% (95% CI, 22.2%-29.1%) for 5 to 6 syndromes, 30.4% (95% CI, 26.7%-34.5%) for 7 to 8 syndromes, 39.5% (95% CI, 34.4%-44.8%) for 9 to 10 syndromes, and 47.0% (95% CI, 36.4%-57.9%) for 11 or more syndromes. After adjusting for confounders, each additional geriatric syndrome was associated with an increased risk of mortality (hazard ratio, 1.22 [95% CI, 1.15-1.30), which became increasingly pronounced in older age groups.Conclusions and relevanceThis cohort study found that hospitalized older adults had a median of 5 geriatric syndromes, which were independently and incrementally associated with 90-day mortality. Multidomain assessments should be integrated into standard hospital care to identify and address vulnerabilities that commonly affect older adults with acute illness.