Early mortality in critical illness - A descriptive analysis of patients who died within 24 hours of ICU admission

Development and Validation of a Dynamic Real-Time Risk Prediction Model for Intensive Care Units Patients Based on Longitudinal Irregular Data: Multicenter Retrospective Study

BACKGROUND

Timely and accurate prediction of short-term mortality is critical in intensive care units (ICUs), where patients' conditions change rapidly. Traditional scoring systems, such as the Simplified Acute Physiology Score and Acute Physiology and Chronic Health Evaluation, rely on static variables collected within the first 24 hours of admission and do not account for continuously evolving clinical states. These systems lack real-time adaptability, interpretability, and generalizability. With the increasing availability of high-frequency electronic medical record (EMR) data, machine learning (ML) approaches have emerged as powerful tools to model complex temporal patterns and support dynamic clinical decision-making. However, existing models are often limited by their inability to handle irregular sampling and missing values, and many lack rigorous external validation across institutions.

OBJECTIVE

We aimed to develop a real-time, interpretable risk prediction model that continuously assesses ICU patient mortality using irregular, longitudinal EMR data, with improved performance and generalizability over traditional static scoring systems.

METHODS

A time-aware bidirectional attention-based long short-term memory (TBAL) model was developed using EMR data from the MIMIC-IV (Medical Information Mart for Intensive Care) and eICU Collaborative Research Database (eICU-CRD) databases, comprising 176,344 ICU stays. The model incorporated dynamic variables, including vital signs, laboratory results, and medication data, updated hourly, to perform static and continuous mortality risk assessments. External cross-validation and subgroup sensitivity analyses were conducted to evaluate robustness and fairness. Model performance was assessed using the area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), accuracy, and F1-score. Interpretability was enhanced using integrated gradients to identify key predictors.

RESULTS

For the static 12-hour to 1-day mortality prediction task, the TBAL model achieved AUROCs of 95.9 (95% CI 94.2-97.5) and 93.3 (95% CI 91.5-95.3) and AUPRCs of 48.5 and 21.6 in MIMIC-IV and eICU-CRD, respectively. Accuracy and F1-scores reached 94.1 and 46.7 in MIMIC-IV and 92.2 and 28.1 in eICU-CRD. In dynamic prediction tasks, AUROCs reached 93.6 (95% CI 93.2-93.9) and 91.9 (95% CI 91.6-92.1), with AUPRCs of 41.3 and 50, respectively. The model maintained high recall for positive cases (82.6% and 79.1% in MIMIC-IV and eICU-CRD). Cross-database validation yielded AUROCs of 81.3 and 76.1, confirming generalizability. Subgroup analysis showed stable performance across age, sex, and severity strata, with top predictors including lactate, vasopressor use, and Glasgow Coma Scale score.

CONCLUSIONS

The TBAL model offers a robust, interpretable, and generalizable solution for dynamic real-time mortality risk prediction in ICU patients. Its ability to adapt to irregular temporal patterns and to provide hourly updated predictions positions it as a promising decision-support tool. Future work should validate its utility in prospective clinical trials and investigate its integration into real-world ICU workflows to enhance patient outcomes.

A Clinician's Guide to Understanding Bias in Critical Clinical Prediction Models

This narrative review focuses on the role of clinical prediction models in supporting informed decision-making in critical care, emphasizing their 2 forms: traditional scores and artificial intelligence (AI)-based models. Acknowledging the potential for both types to embed biases, the authors underscore the importance of critical appraisal to increase our trust in models. The authors outline recommendations and critical care examples to manage risk of bias in AI models. The authors advocate for enhanced interdisciplinary training for clinicians, who are encouraged to explore various resources (books, journals, news Web sites, and social media) and events (Datathons) to deepen their understanding of risk of bias.

Opportunities to Improve End-of-Life Care Quality among Patients with Short Terminal Admissions

CONTEXT

Little is known about Veterans who die during a short terminal admission, which renders them ineligible for the Department of Veterans Affairs (VA) Bereaved Family Survey.

OBJECTIVES

We sought to describe this population and identify opportunities to improve end-of-life (EOL) care quality.

METHODS

Retrospective, cohort analysis of Veteran decedents who died in a VA inpatient setting between October 2018-September 2019. Veterans were dichotomized by short (<24 hours) and long (≥24 hours) terminal admissions; sociodemographics, clinical characteristics, VA and non-VA healthcare use, and EOL care quality indicators were compared.

RESULTS

Among 17,033 inpatient decedents, 723 (4%) had short terminal admissions. Patients with short compared to long terminal admissions were less likely to have a VA hospitalization (38% vs. 54%) in the last 90 days of life and were more likely to die in an intensive care (49% vs 21%) or acute care (27% vs 18%) unit. Patients with a short compared to long admission were about half as likely to receive hospice (33% vs 64%) or palliative care (33% vs 69%). Most patients with short admissions (76%) had a life-limiting condition (e.g., cancer, chronic obstructive pulmonary disease) and those with cancer were more likely to receive palliative care compared to those with non-cancer conditions.

CONCLUSION

Veterans with short terminal admissions are less likely to receive hospice or palliative care compared to patients with long terminal admissions. Many patients with short terminal admissions, such as those with life-limiting conditions (especially cancer), receive aspects of high-quality EOL care, however, opportunities for improvement exist.

Real-time machine learning model to predict short-term mortality in critically ill patients: development and international validation

BACKGROUND

A real-time model for predicting short-term mortality in critically ill patients is needed to identify patients at imminent risk. However, the performance of the model needs to be validated in various clinical settings and ethnicities before its clinical application. In this study, we aim to develop an ensemble machine learning model using routinely measured clinical variables at a single academic institution in South Korea.

METHODS

We developed an ensemble model using deep learning and light gradient boosting machine models. Internal validation was performed using the last two years of the internal cohort dataset, collected from a single academic hospital in South Korea between 2007 and 2021. External validation was performed using the full Medical Information Mart for Intensive Care (MIMIC), eICU-Collaborative Research Database (eICU-CRD), and Amsterdam University Medical Center database (AmsterdamUMCdb) data. The area under the receiver operating characteristic curve (AUROC) was calculated and compared to that for the National Early Warning Score (NEWS).

RESULTS

The developed model (iMORS) demonstrated high predictive performance with an internal AUROC of 0.964 (95% confidence interval [CI] 0.963-0.965) and external AUROCs of 0.890 (95% CI 0.889-0.891) for MIMIC, 0.886 (95% CI 0.885-0.887) for eICU-CRD, and 0.870 (95% CI 0.868-0.873) for AmsterdamUMCdb. The model outperformed the NEWS with higher AUROCs in the internal and external validation (0.866 for the internal, 0.746 for MIMIC, 0.798 for eICU-CRD, and 0.819 for AmsterdamUMCdb; p < 0.001).

CONCLUSIONS

Our real-time machine learning model to predict short-term mortality in critically ill patients showed excellent performance in both internal and external validations. This model could be a useful decision-support tool in the intensive care units to assist clinicians.

Factors Affecting the Mortality Rate in Non-COVID-19 Intensive Care Unit Patients During the COVID-19 Pandemic in Cyprus: A Retrospective Cohort Study

INTRODUCTION

Mortality statistics constitute a pivotal element in informing public health policymaking in critical care settings. Mortality rates exhibit temporal variability, and their quantification is susceptible to well-established biases that have been exacerbated in the backdrop of the COVID-19 pandemic. A multitude of factors contribute to the process of patients' outcomes within the intensive care unit (ICU) setting. The primary aim of this study is to compare the mortality rate observed during the initial and subsequent phases of the COVID-19 pandemic in non-COVID-19 patient cohorts. Secondary objectives encompass evaluating the demographic and clinical factors and admission times to the ICU as an independent predictor affecting mortality.

METHODS AND MATERIALS

 A retrospective investigation of the data gathered from 1127 non-COVID-19 patients admitted to an ICU situated in Nicosia, Cyprus between March 2020 and December 2022 was performed. We divided the study period into two distinct timeframes. The first period spanned from the onset of the COVID-19 pandemic up until January 2021, coinciding with the relaxation of COVID-related restrictions. The second period was defined as the period when restrictions were not applied. The time of admission to the ICU is categorized as either off-hours or business hours. We recorded various patient characteristics, including age, gender, Acute Physiology and Chronic Health Evaluation II (APACHE II), Glasgow Coma Scale (GCS), Sequential Organ Failure Assessment (SOFA) scores, hospitalization duration, discharge details, mortality events with precise timestamps and primary diagnosis for admission. Multivariate logistic regression analysis was performed with these characteristics to predict the likelihood of mortality.

RESULTS

 This study included 632 males (56.1%) and 495 females (43.9%). Within the patient cohort, 653 patients (57.9%) were discharged from the ICU, while 474 patients (42.1%) experienced mortality during their ICU stay. No significant correlation was found whether patients were admitted to ICU during the first or second period of the COVID-19 pandemic. There was a significant difference in the comparison of outcomes within the ICU between the off-hours and business hours (p=0.001). A total of 329 of 618 (53.2%) patients admitted in off-hours and 145 of 509 (28.4%) patients admitted in business hours died. Moreover, the mean GCS, APACHE II and SOFA scores were higher in patients admitted during off-hours. APACHE II score (OR: 1.11, 95% CI: 1.06 to 1.15, p<0.01), SOFA (OR: 1.21, 95% CI: 1.10 to 1.31, p<0.01) and GCS (OR: 0.88, 95% CI: 0.84 to 0.92, p<0.01) scores and admission to the ICU in off-hours 2.63 (1.91-3.67) were significantly associated with mortality.

CONCLUSION

 The results of this retrospective cohort analysis have shown that the mortality rate was higher in non-COVID-19 patients admitted to ICU during off-hours compared to those admitted during business hours. However, no significant difference was found in the mortality rate between the admissions during the first and second periods of the COVID-19 pandemic.

Recording early deaths following emergency department visits in inpatient data: An observational study using data of 16 German hospitals

OBJECTIVE

In German hospital emergency departments (EDs), no definite reimbursement rules exist for patients who die within 24 hours after arrival. Our study aimed to assess whether these cases were recorded and billed as inpatient stays. Furthermore, characteristics of patients who die within 24 hours following arrival at the ED were investigated for all ED visits, as well as for the subgroup of ED visits with an ED diagnosis or inpatient principal diagnosis of acute myocardial infarction.

METHODS

This study was part of the INDEED project which aimed to explore utilization and trans-sectoral patterns of care for patients treated in EDs in Germany. The study population includes ED visits of adult patients in 2016 in 16 German hospitals participating in the project. In the data set of combined ED, inpatient, and outpatient treatment information early deaths were classified as patients who died in the ED or in the hospital within 24 hours after arrival. Characteristics of visits followed by early death were analyzed descriptively. Mode of billing as inpatient or outpatient was validated by identifying corresponding billing information using linked inpatient and outpatient data.

RESULTS

In 2016, 454,747 ED visits of adult patients occurred in the participating hospitals and 42.8% resulted in inpatient admission. Among these inpatients 8,317 (4.3%) died during the overall hospital stay, and 1,302 (0.7%) died within 24 hours following arrival. The proportion of early deaths among all deaths in patients with a diagnosis of acute myocardial infarction was higher (27%) compared to the overall patient population (16%). Although all cases of early death were classified as inpatients the corresponding inpatient data was missing in 1.9% of all early deaths and in 3.4% of early deaths with a diagnosis of acute myocardial infarction. Outpatient billing information suggesting that these cases were billed as outpatients, was found in 0.3% of all early deaths and in 0.8 to 1.7% of early deaths with a diagnosis of acute myocardial infarction, respectively.

CONCLUSION

In-hospital mortality might be biased by incomplete recording of early deaths in inpatient data. However, the proportion of patients with early death who were billed as outpatients was marginal in the investigated study population of 16 hospitals. Although the study results are limited by restricted generalizability and subpar data quality, this finding indicates that early deaths might be almost completely recorded in German inpatient data. Nevertheless, data quality should be enhanced by establishing general billing rules for cases with a short treatment duration due to early death.