Dataset Information

Predicting critical state after COVID-19 diagnosis: model development using a large US electronic health record dataset.

ABSTRACT: As the COVID-19 pandemic is challenging healthcare systems worldwide, early identification of patients with a high risk of complication is crucial. We present a prognostic model predicting critical state within 28 days following COVID-19 diagnosis trained on data from US electronic health records (IBM Explorys), including demographics, comorbidities, symptoms, and hospitalization. Out of 15753 COVID-19 patients, 2050 went into critical state or deceased. Non-random train-test splits by time were repeated 100 times and led to a ROC AUC of 0.861 [0.838, 0.883] and a precision-recall AUC of 0.434 [0.414, 0.485] (median and interquartile range). The interpretability analysis confirmed evidence on major risk factors (e.g., older age, higher BMI, male gender, diabetes, and cardiovascular disease) in an efficient way compared to clinical studies, demonstrating the model validity. Such personalized predictions could enable fine-graded risk stratification for optimized care management.

SUBMITTER: Rinderknecht MD

PROVIDER: S-EPMC8292360 | biostudies-literature | 2021 Jul

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Predicting critical state after COVID-19 diagnosis: model development using a large US electronic health record dataset.

Rinderknecht Mike D MD Klopfenstein Yannick Y

NPJ digital medicine 20210720 1

As the COVID-19 pandemic is challenging healthcare systems worldwide, early identification of patients with a high risk of complication is crucial. We present a prognostic model predicting critical state within 28 days following COVID-19 diagnosis trained on data from US electronic health records (IBM Explorys), including demographics, comorbidities, symptoms, and hospitalization. Out of 15753 COVID-19 patients, 2050 went into critical state or deceased. Non-random train-test splits by time were ...[more]

PMID: 34285316

Similar Datasets

Project description:OBJECTIVES:The goal of this study was to use machine learning to more accurately predict survival after echocardiography. BACKGROUND:Predicting patient outcomes (e.g., survival) following echocardiography is primarily based on ejection fraction (EF) and comorbidities. However, there may be significant predictive information within additional echocardiography-derived measurements combined with clinical electronic health record data. METHODS:Mortality was studied in 171,510 unselected patients who underwent 331,317 echocardiograms in a large regional health system. The authors investigated the predictive performance of nonlinear machine learning models compared with that of linear logistic regression models using 3 different inputs: 1) clinical variables, including 90 cardiovascular-relevant International Classification of Diseases, Tenth Revision, codes, and age, sex, height, weight, heart rate, blood pressures, low-density lipoprotein, high-density lipoprotein, and smoking; 2) clinical variables plus physician-reported EF; and 3) clinical variables and EF, plus 57 additional echocardiographic measurements. Missing data were imputed with a multivariate imputation by using a chained equations algorithm (MICE). The authors compared models versus each other and baseline clinical scoring systems by using a mean area under the curve (AUC) over 10 cross-validation folds and across 10 survival durations (6 to 60 months). RESULTS:Machine learning models achieved significantly higher prediction accuracy (all AUC >0.82) over common clinical risk scores (AUC = 0.61 to 0.79), with the nonlinear random forest models outperforming logistic regression (p < 0.01). The random forest model including all echocardiographic measurements yielded the highest prediction accuracy (p < 0.01 across all models and survival durations). Only 10 variables were needed to achieve 96% of the maximum prediction accuracy, with 6 of these variables being derived from echocardiography. Tricuspid regurgitation velocity was more predictive of survival than LVEF. In a subset of studies with complete data for the top 10 variables, multivariate imputation by chained equations yielded slightly reduced predictive accuracies (difference in AUC of 0.003) compared with the original data. CONCLUSIONS:Machine learning can fully utilize large combinations of disparate input variables to predict survival after echocardiography with superior accuracy.

Project description:BackgroundAcute kidney injury (AKI) is a major complication following cardiac surgery that substantially increases morbidity and mortality. Current diagnostic guidelines based on elevated serum creatinine and/or the presence of oliguria potentially delay its diagnosis. We presented a series of models for predicting AKI after cardiac surgery based on electronic health record data.MethodsWe enrolled 1457 adult patients who underwent cardiac surgery at Nanjing First Hospital from January 2017 to June 2019. 193 clinical features, including demographic characteristics, comorbidities and hospital evaluation, laboratory test, medication, and surgical information, were available for each patient. The number of important variables was determined using the sliding windows sequential forward feature selection technique (SWSFS). The following model development methods were introduced: extreme gradient boosting (XGBoost), random forest (RF), deep forest (DF), and logistic regression. Model performance was accessed using the area under the receiver operating characteristic curve (AUROC). We additionally applied SHapley Additive exPlanation (SHAP) values to explain the RF model. AKI was defined according to Kidney Disease Improving Global Outcomes guidelines.ResultsIn the discovery set, SWSFS identified 16 important variables. The top 5 variables in the RF importance matrix plot were central venous pressure, intraoperative urine output, hemoglobin, serum potassium, and lactic dehydrogenase. In the validation set, the DF model exhibited the highest AUROC (0.881, 95% confidence interval [CI] 0.831-0.930), followed by RF (0.872, 95% CI 0.820-0.923) and XGBoost (0.857, 95% CI 0.802-0.912). A nomogram model was constructed based on intraoperative longitudinal features, achieving an AUROC of 0.824 (95% CI 0.763-0.885) in the validation set. The SHAP values successfully illustrated the positive or negative contribution of the 16 variables attributed to the output of the RF model and the individual variable's effect on model prediction.ConclusionsOur study identified 16 important predictors and provided a series of prediction models to enhance risk stratification of AKI after cardiac surgery. These novel predictors might aid in choosing proper preventive and therapeutic strategies in the perioperative management of AKI patients.

Project description:BackgroundElectronic, personalized clinical decision support tools to optimize glycated hemoglobin (HbA1c) screening are lacking. Current screening guidelines are based on simple, categorical rules developed for populations of patients. Although personalized diabetes risk calculators have been created, none are designed to predict current glycemic status using structured data commonly available in electronic health records (EHRs).ObjectiveThe goal of this project was to create a mathematical equation for predicting the probability of current elevations in HbA1c (≥5.7%) among patients with no history of hyperglycemia using readily available variables that will allow integration with EHR systems.MethodsThe reduced model was compared head-to-head with calculators created by Baan and Griffin. Ten-fold cross-validation was used to calculate the bias-adjusted prediction accuracy of the new model. Statistical analyses were performed in R version 3.2.5 (The R Foundation for Statistical Computing) using the rms (Regression Modeling Strategies) package.ResultsThe final model to predict an elevated HbA1c based on 22,635 patient records contained the following variables in order from most to least importance according to their impact on the discriminating accuracy of the model: age, body mass index, random glucose, race, serum non-high-density lipoprotein, serum total cholesterol, estimated glomerular filtration rate, and smoking status. The new model achieved a concordance statistic of 0.77 which was statistically significantly better than prior models. The model appeared to be well calibrated according to a plot of the predicted probabilities versus the prevalence of the outcome at different probabilities.ConclusionsThe calculator created for predicting the probability of having an elevated HbA1c significantly outperformed the existing calculators. The personalized prediction model presented in this paper could improve the efficiency of HbA1c screening initiatives.

Dataset Information

Predicting critical state after COVID-19 diagnosis: model development using a large US electronic health record dataset.

Publications

Predicting critical state after COVID-19 diagnosis: model development using a large US electronic health record dataset.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets