Dataset Information

Developing machine learning models to personalize care levels among emergency room patients for hospital admission.

ABSTRACT:

Objective

To develop prediction models for intensive care unit (ICU) vs non-ICU level-of-care need within 24 hours of inpatient admission for emergency department (ED) patients using electronic health record data.

Materials and methods

Using records of 41 654 ED visits to a tertiary academic center from 2015 to 2019, we tested 4 algorithms-feed-forward neural networks, regularized regression, random forests, and gradient-boosted trees-to predict ICU vs non-ICU level-of-care within 24 hours and at the 24th hour following admission. Simple-feature models included patient demographics, Emergency Severity Index (ESI), and vital sign summary. Complex-feature models added all vital signs, lab results, and counts of diagnosis, imaging, procedures, medications, and lab orders.

Results

The best-performing model, a gradient-boosted tree using a full feature set, achieved an AUROC of 0.88 (95%CI: 0.87-0.89) and AUPRC of 0.65 (95%CI: 0.63-0.68) for predicting ICU care need within 24 hours of admission. The logistic regression model using ESI achieved an AUROC of 0.67 (95%CI: 0.65-0.70) and AUPRC of 0.37 (95%CI: 0.35-0.40). Using a discrimination threshold, such as 0.6, the positive predictive value, negative predictive value, sensitivity, and specificity were 85%, 89%, 30%, and 99%, respectively. Vital signs were the most important predictors.

Discussion and conclusions

Undertriaging admitted ED patients who subsequently require ICU care is common and associated with poorer outcomes. Machine learning models using readily available electronic health record data predict subsequent need for ICU admission with good discrimination, substantially better than the benchmarking ESI system. The results could be used in a multitiered clinical decision-support system to improve ED triage.

SUBMITTER: Nguyen M

PROVIDER: S-EPMC8510323 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:ObjectiveTo predict hospital admission at the time of ED triage using patient history in addition to information collected at triage.MethodsThis retrospective study included all adult ED visits between March 2014 and July 2017 from one academic and two community emergency rooms that resulted in either admission or discharge. A total of 972 variables were extracted per patient visit. Samples were randomly partitioned into training (80%), validation (10%), and test (10%) sets. We trained a series of nine binary classifiers using logistic regression (LR), gradient boosting (XGBoost), and deep neural networks (DNN) on three dataset types: one using only triage information, one using only patient history, and one using the full set of variables. Next, we tested the potential benefit of additional training samples by training models on increasing fractions of our data. Lastly, variables of importance were identified using information gain as a metric to create a low-dimensional model.ResultsA total of 560,486 patient visits were included in the study, with an overall admission risk of 29.7%. Models trained on triage information yielded a test AUC of 0.87 for LR (95% CI 0.86-0.87), 0.87 for XGBoost (95% CI 0.87-0.88) and 0.87 for DNN (95% CI 0.87-0.88). Models trained on patient history yielded an AUC of 0.86 for LR (95% CI 0.86-0.87), 0.87 for XGBoost (95% CI 0.87-0.87) and 0.87 for DNN (95% CI 0.87-0.88). Models trained on the full set of variables yielded an AUC of 0.91 for LR (95% CI 0.91-0.91), 0.92 for XGBoost (95% CI 0.92-0.93) and 0.92 for DNN (95% CI 0.92-0.92). All algorithms reached maximum performance at 50% of the training set or less. A low-dimensional XGBoost model built on ESI level, outpatient medication counts, demographics, and hospital usage statistics yielded an AUC of 0.91 (95% CI 0.91-0.91).ConclusionMachine learning can robustly predict hospital admission using triage information and patient history. The addition of historical information improves predictive performance significantly compared to using triage information alone, highlighting the need to incorporate these variables into prediction models.

Project description:BackgroundAlthough multiple prediction models have been developed to predict hospital admission to emergency departments (EDs) to address overcrowding and patient safety, only a few studies have examined prediction models for prehospital use. Development of institution-specific prediction models is feasible in this age of data science, provided that predictor-related information is readily collectable.ObjectiveWe aimed to develop a hospital admission prediction model based on patient information that is commonly available during ambulance transport before hospitalization.MethodsPatients transported by ambulance to our ED from April 2018 through March 2019 were enrolled. Candidate predictors were age, sex, chief complaint, vital signs, and patient medical history, all of which were recorded by emergency medical teams during ambulance transport. Patients were divided into two cohorts for derivation (3601/5145, 70.0%) and validation (1544/5145, 30.0%). For statistical models, logistic regression, logistic lasso, random forest, and gradient boosting machine were used. Prediction models were developed in the derivation cohort. Model performance was assessed by area under the receiver operating characteristic curve (AUROC) and association measures in the validation cohort.ResultsOf 5145 patients transported by ambulance, including deaths in the ED and hospital transfers, 2699 (52.5%) required hospital admission. Prediction performance was higher with the addition of predictive factors, attaining the best performance with an AUROC of 0.818 (95% CI 0.792-0.839) with a machine learning model and predictive factors of age, sex, chief complaint, and vital signs. Sensitivity and specificity of this model were 0.744 (95% CI 0.716-0.773) and 0.745 (95% CI 0.709-0.776), respectively.ConclusionsFor patients transferred to EDs, we developed a well-performing hospital admission prediction model based on routinely collected prehospital information including chief complaints.

Project description:Fibrinogen plays a key role in hemostasis and is the first coagulation factor to reach critical levels in massively bleeding trauma patients. Consequently, rapid estimation of plasma fibrinogen (FIB) is essential upon emergency room (ER) admission, but is not part of routine coagulation monitoring in many centers. We investigated the predictive ability of the laboratory parameters hemoglobin (Hb) and base excess (BE) upon admission, as well as the Injury Severity Score (ISS), to estimate FIB in major trauma patients.In this retrospective study, major trauma patients (ISS ?16) with documented FIB analysis upon ER admission were eligible for inclusion. FIB was correlated with Hb, BE and ISS, alone and in combination, using regression analysis.A total of 675 patients were enrolled (median ISS 27). FIB upon admission correlated strongly with Hb, BE and ISS. Multiple regression analysis showed that Hb and BE together predicted FIB (adjusted R2 = 0.46; loge(FIB) = 3.567 + 0.223.Hb - 0.007.Hb2 + 0.044.BE), and predictive strength increased when ISS was included (adjusted R2 = 0.51; loge(FIB) = 4.188 + 0.243.Hb - 0.008.Hb2 + 0.036.BE - 0.031.ISS + 0.0003.ISS2). Of all major trauma patients admitted with Hb <12 g/dL, 74% had low (<200 mg/dL) FIB and 54% had critical (<150 mg/dL) FIB. Of patients admitted with Hb <10 g/dL, 89% had low FIB and 73% had critical FIB. These values increased to 93% and 89%, respectively, among patients with an admission Hb <8 g/dL. Sixty-six percent of patients with only a weakly negative BE (<-2 mmol/L) showed low FIB. Of patients with BE <-6 mmol/L upon admission, 81% had low FIB and 63% had critical FIB. The corresponding values for BE <-10 mmol/L were 89% and 78%, respectively.Upon ER admission, FIB of major trauma patients shows strong correlation with rapidly obtainable, routine laboratory parameters such as Hb and BE. These two parameters might provide an insightful and rapid tool to identify major trauma patients at risk of acquired hypofibrinogenemia. Early calculation of ISS could further increase the ability to predict FIB in these patients. We propose that FIB can be estimated during the initial phase of trauma care based on bedside tests.

Project description:BackgroundDevelopment of emergency department (ED) triage systems that accurately differentiate and prioritize critically ill from stable patients remains challenging. We used machine learning models to predict clinical outcomes, and then compared their performance with that of a conventional approach-the Emergency Severity Index (ESI).MethodsUsing National Hospital and Ambulatory Medical Care Survey (NHAMCS) ED data, from 2007 through 2015, we identified all adult patients (aged ≥ 18 years). In the randomly sampled training set (70%), using routinely available triage data as predictors (e.g., demographics, triage vital signs, chief complaints, comorbidities), we developed four machine learning models: Lasso regression, random forest, gradient boosted decision tree, and deep neural network. As the reference model, we constructed a logistic regression model using the five-level ESI data. The clinical outcomes were critical care (admission to intensive care unit or in-hospital death) and hospitalization (direct hospital admission or transfer). In the test set (the remaining 30%), we measured the predictive performance, including area under the receiver-operating-characteristics curve (AUC) and net benefit (decision curves) for each model.ResultsOf 135,470 eligible ED visits, 2.1% had critical care outcome and 16.2% had hospitalization outcome. In the critical care outcome prediction, all four machine learning models outperformed the reference model (e.g., AUC, 0.86 [95%CI 0.85-0.87] in the deep neural network vs 0.74 [95%CI 0.72-0.75] in the reference model), with less under-triaged patients in ESI triage levels 3 to 5 (urgent to non-urgent). Likewise, in the hospitalization outcome prediction, all machine learning models outperformed the reference model (e.g., AUC, 0.82 [95%CI 0.82-0.83] in the deep neural network vs 0.69 [95%CI 0.68-0.69] in the reference model) with less over-triages in ESI triage levels 1 to 3 (immediate to urgent). In the decision curve analysis, all machine learning models consistently achieved a greater net benefit-a larger number of appropriate triages considering a trade-off with over-triages-across the range of clinical thresholds.ConclusionsCompared to the conventional approach, the machine learning models demonstrated a superior performance to predict critical care and hospitalization outcomes. The application of modern machine learning models may enhance clinicians' triage decision making, thereby achieving better clinical care and optimal resource utilization.

Project description:BackgroundRisk prediction models have been developed to identify those at increased risk for emergency admissions, which could facilitate targeted interventions in primary care to prevent these events.ObjectiveSystematic review of validated risk prediction models for predicting emergency hospital admissions in community-dwelling adults.MethodsA systematic literature review and narrative analysis was conducted. Inclusion criteria were as follows;Populationcommunity-dwelling adults (aged 18 years and above); Risk: risk prediction models, not contingent on an index hospital admission, with a derivation and ?1 validation cohort;Primary outcomeemergency hospital admission (defined as unplanned overnight stay in hospital);Study designretrospective or prospective cohort studies.ResultsOf 18,983 records reviewed, 27 unique risk prediction models met the inclusion criteria. Eleven were developed in the United States, 11 in the United Kingdom, 3 in Italy, 1 in Spain, and 1 in Canada. Nine models were derived using self-report data, and the remainder (n=18) used routine administrative or clinical record data. Total study sample sizes ranged from 96 to 4.7 million participants. Predictor variables most frequently included in models were: (1) named medical diagnoses (n=23); (2) age (n=23); (3) prior emergency admission (n=22); and (4) sex (n=18). Eleven models included nonmedical factors, such as functional status and social supports. Regarding predictive accuracy, models developed using administrative or clinical record data tended to perform better than those developed using self-report data (c statistics 0.63-0.83 vs. 0.61-0.74, respectively). Six models reported c statistics of >0.8, indicating good performance. All 6 included variables for prior health care utilization, multimorbidity or polypharmacy, and named medical diagnoses or prescribed medications. Three predicted admissions regarded as being ambulatory care sensitive.ConclusionsThis study suggests that risk models developed using administrative or clinical record data tend to perform better. In applying a risk prediction model to a new population, careful consideration needs to be given to the purpose of its use and local factors.

Project description:BackgroundMachine learning (ML) algorithms have been trained to early predict critical in-hospital events from COVID-19 using patient data at admission, but little is known on how their performance compares with each other and/or with statistical logistic regression (LR). This prospective multicentre cohort study compares the performance of a LR and five ML models on the contribution of influencing predictors and predictor-to-event relationships on prediction model´s performance.MethodsWe used 25 baseline variables of 490 COVID-19 patients admitted to 8 hospitals in Germany (March-November 2020) to develop and validate (75/25 random-split) 3 linear (L1 and L2 penalty, elastic net [EN]) and 2 non-linear (support vector machine [SVM] with radial kernel, random forest [RF]) ML approaches for predicting critical events defined by intensive care unit transfer, invasive ventilation and/or death (composite end-point: 181 patients). Models were compared for performance (area-under-the-receiver-operating characteristic-curve [AUC], Brier score) and predictor importance (performance-loss metrics, partial-dependence profiles).ResultsModels performed close with a small benefit for LR (utilizing restricted cubic splines for non-linearity) and RF (AUC means: 0.763-0.731 [RF-L1]); Brier scores: 0.184-0.197 [LR-L1]). Top ranked predictor variables (consistently highest importance: C-reactive protein) were largely identical across models, except creatinine, which exhibited marginal (L1, L2, EN, SVM) or high/non-linear effects (LR, RF) on events.ConclusionsAlthough the LR and ML models analysed showed no strong differences in performance and the most influencing predictors for COVID-19-related event prediction, our results indicate a predictive benefit from taking account for non-linear predictor-to-event relationships and effects. Future efforts should focus on leveraging data-driven ML technologies from static towards dynamic modelling solutions that continuously learn and adapt to changes in data environments during the evolving pandemic.Trial registration numberNCT04659187.