Dataset Information

Unstructured clinical notes within the 24 hours since admission predict short, mid & long-term mortality in adult ICU patients.

ABSTRACT: Mortality prediction for intensive care unit (ICU) patients is crucial for improving outcomes and efficient utilization of resources. Accessibility of electronic health records (EHR) has enabled data-driven predictive modeling using machine learning. However, very few studies rely solely on unstructured clinical notes from the EHR for mortality prediction. In this work, we propose a framework to predict short, mid, and long-term mortality in adult ICU patients using unstructured clinical notes from the MIMIC III database, natural language processing (NLP), and machine learning (ML) models. Depending on the statistical description of the patients' length of stay, we define the short-term as 48-hour and 4-day period, the mid-term as 7-day and 10-day period, and the long-term as 15-day and 30-day period after admission. We found that by only using clinical notes within the 24 hours of admission, our framework can achieve a high area under the receiver operating characteristics (AU-ROC) score for short, mid and long-term mortality prediction tasks. The test AU-ROC scores are 0.87, 0.83, 0.83, 0.82, 0.82, and 0.82 for 48-hour, 4-day, 7-day, 10-day, 15-day, and 30-day period mortality prediction, respectively. We also provide a comparative study among three types of feature extraction techniques from NLP: frequency-based technique, fixed embedding-based technique, and dynamic embedding-based technique. Lastly, we provide an interpretation of the NLP-based predictive models using feature-importance scores.

SUBMITTER: Mahbub M

PROVIDER: S-EPMC8735614 | biostudies-literature | 2022

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Unstructured clinical notes within the 24 hours since admission predict short, mid & long-term mortality in adult ICU patients.

Mahbub Maria M Srinivasan Sudarshan S Danciu Ioana I Peluso Alina A Begoli Edmon E Tamang Suzanne S Peterson Gregory D GD

PloS one 20220106 1

Mortality prediction for intensive care unit (ICU) patients is crucial for improving outcomes and efficient utilization of resources. Accessibility of electronic health records (EHR) has enabled data-driven predictive modeling using machine learning. However, very few studies rely solely on unstructured clinical notes from the EHR for mortality prediction. In this work, we propose a framework to predict short, mid, and long-term mortality in adult ICU patients using unstructured clinical notes f ...[more]

PMID: 34990485

Similar Datasets

Project description:IntroductionRapid-response teams (RRTs) are interdisciplinary groups created to rapidly assess and treat patients with unexpected clinical deterioration marked by decline in vital signs. Traditionally emergency department (ED) disposition is partially based on the patients' vital signs (VS) at the time of hospital admission. We aimed to identify which patients will have RRT activation within 12 hours of admission based on their ED VS, and if their outcomes differed.MethodsWe conducted a case-control study of patients presenting from January 2009 to December 2012 to a tertiary ED who subsequently had RRT activations within 12 hours of admission (early RRT activations). The medical records of patients 18 years and older admitted to a non-intensive care unit (ICU) setting were reviewed to obtain VS at the time of ED arrival and departure, age, gender and diagnoses. Controls were matched 1:1 on age, gender, and diagnosis. We evaluated VS using cut points (lowest 10%, middle 80% and highest 10%) based on the distribution of VS for all patients. Our study adheres to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for reporting observational studies.ResultsA total of 948 patients were included (474 cases and 474 controls). Patients who had RRT activations were more likely to be tachycardic (odds ratio [OR] 2.02, 95% CI [1.25-3.27]), tachypneic (OR 2.92, 95% CI [1.73-4.92]), and had lower oxygen saturations (OR 2.25, 95% CI [1.42-3.56]) upon arrival to the ED. Patients who had RRT activations were more likely to be tachycardic at the time of disposition from the ED (OR 2.76, 95% CI [1.65-4.60]), more likely to have extremes of systolic blood pressure (BP) (OR 1.72, 95% CI [1.08-2.72] for low BP and OR 1.82, 95% CI [1.19-2.80] for high BP), higher respiratory rate (OR 4.15, 95% CI [2.44-7.07]) and lower oxygen saturation (OR 2.29, 95% CI [1.43-3.67]). Early RRT activation was associated with increased healthcare utilization and worse outcomes including increased rates of ICU admission within 72 hours (OR 38.49, 95%CI [19.03-77.87]), invasive interventions (OR 5.49, 95%CI [3.82-7.89]), mortality at 72 hours (OR 4.24, 95%CI [1.60-11.24]), and mortality at one month (OR 4.02, 95%CI [2.44-6.62]).ConclusionAfter matching for age, gender and ED diagnosis, we found that patients with an abnormal heart rate, respiratory rate or oxygen saturation at the time of ED arrival or departure are more likely to trigger RRT activation within 12 hours of admission. Early RRT activation was associated with higher mortality at 72 hours and one month, increased rates of invasive intervention and ICU admission. Determining risk factors of early RRT activation is of clinical, operational, and financial importance, as improved medical decision-making regarding disposition would maximize allocation of resources while potentially limiting morbidity and mortality.

Project description:BackgroundThere is no generalizable transcriptomics signature of pediatric acute respiratory distress syndrome. Our goal was to identify a whole blood differential gene expression signature for pediatric acute hypoxemic respiratory failure (AHRF) using transcriptomic microarrays within twenty-four hours of diagnosis. We used publicly available human whole-blood gene expression arrays of a Berlin-defined pediatric acute respiratory distress syndrome (GSE147902) cohort and a sepsis-triggered AHRF (GSE66099) cohort within twenty-four hours of diagnosis and compared those children with a PaO2/FiO2 < 200 to those with a PaO2/FiO2 ≥ 200.ResultsWe used stability selection, a bootstrapping method of 100 simulations using logistic regression as a classifier, to select differentially expressed genes associated with a PaO2/FiO2 < 200 vs. PaO2/FiO2 ≥ 200. The top-ranked genes that contributed to the AHRF signature were selected in each dataset. Genes common to both of the top 1,500 ranked gene lists were selected for pathway analysis. Pathway and network analysis was performed using the Pathway Network Analysis Visualizer (PANEV) and Reactome was used to perform an over-representation gene network analysis of the top-ranked genes common to both cohorts. Changes in metabolic pathways involved in energy balance, fundamental cellular processes such as protein translation, mitochondrial function, oxidative stress, immune signaling, and inflammation are differentially regulated early in pediatric ARDS and sepsis-induced AHRF compared to both healthy controls and to milder acute hypoxemia. Specifically, fundamental pathways related to the severity of hypoxemia emerged and included (1) ribosomal and eukaryotic initiation of factor 2 (eIF2) regulation of protein translation and (2) the nutrient, oxygen, and energy sensing pathway, mTOR, activated via PI3K/AKT signaling.ConclusionsCellular energetics and metabolic pathways are important mechanisms to consider to further our understanding of the heterogeneity and underlying pathobiology of moderate and severe pediatric acute respiratory distress syndrome. Our findings are hypothesis generating and support the study of metabolic pathways and cellular energetics to understand heterogeneity and underlying pathobiology of moderate and severe acute hypoxemic respiratory failure in children.

Project description:ObjectiveThe relationship between the levels of the first 24-h PaCO2 and the prognosis of sepsis-associated encephalopathy (SAE) remains unclear, and the first 24-h optimal target for PaCO2 is currently inconclusive. This study was performed to investigate the correlation between PaCO2 and all-cause mortality for SAE patients, establish a reference range of the initial 24-hour PaCO2 for clinicians in critical care, and explain the possible pathophysiological mechanisms of abnormal PaCO2 levels as a higher mortality risk factor for SAE.MethodsThe baseline information and clinical data of patients were extracted from the fourth edition Medical Information Mart for Intensive Care database (MIMIC-IV 2.0). Multivariate logistic regressions were performed to assess the relationship between PaCO2 and all-cause mortality of SAE. Additionally, restricted cubic splines, Kaplan-Meier Survival analyses, propensity score matching (PSM) analyses, and subgroup analyses were conducted.ResultsA total of 5471 patients were included in our cohort. In the original and matched cohort, multivariate logistic regression analysis showed that normocapnia and mild hypercapnia may be associated with a more favorable prognosis of SAE patients, and survival analysis supported the findings. In addition, a U-shaped association emerged when examining the initial 24-hour PaCO2 levels in relation to 30-day, 60-day, and 90-day mortality using restricted cubic splines, with an average cut-off value of 36.3mmHg (P for nonlinearity<0.05). Below the cut-off value, higher PaCO2 was associated with lower all-cause mortality, while above the cut-off value, higher PaCO2 was associated with higher all-cause mortality. Subsequent subgroup analyses revealed similar results for the subcohort of GCS≤8 compared to the original cohort. Additionally, when examining the subcohort of GCS>8, a L-shaped relationship between PaCO2 and the three clinical endpoints emerged, in contrast to the previously observed U-shaped pattern. The findings from the subcohort of GCS>8 suggested that patients experiencing hypocapnia had a more unfavorable prognosis, which aligns with the results obtained from corresponding multivariate logistic regression analyses.ConclusionThe retrospective study revealed the association between the first 24-h PaCO2 and all-cause mortality risk (30-day, 60-day, and 90-day) for patients with SAE in ICU. The range (35mmHg-50mmHg) of PaCO2 may be the optimal target for patients with SAE in clinical practice.

Dataset Information

Unstructured clinical notes within the 24 hours since admission predict short, mid & long-term mortality in adult ICU patients.

Publications

Unstructured clinical notes within the 24 hours since admission predict short, mid & long-term mortality in adult ICU patients.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets