The real-time COVID-HEART model can accurately predict all-cause mortality and cardiac arrest (AM/CA), as well as thromboembolic events in hospitalized patients with COVID-19, according to a study in JACC Advances.
The COVID-HEART predictor was developed and validated in a retrospective cohort study of adults (aged ≥18 years) with SARS-CoV-2 infection admitted as inpatients at 5 hospitals. Patient data were obtained from the retrospective COVID-19 Precision Medicine Analytics Platform Registry (JH-CROWN).
All participants had SARS-CoV-2 infection confirmed by polymerase chain reaction within 14 days before the date of admission or during admission. Patients in the development set for predicting AM/CA were admitted between March 1, 2020, and November 6, 2020. Those in the test set were admitted from November 7, 2020, to January 8, 2021.
Leave-hospital-out validation was conducted after omitting all patients admitted to 1 of the 5 hospitals, repeating the model training and optimization process with use of data from patients admitted to the remaining 4 hospitals, and testing the optimized model with data from patients admitted to the left-out hospital.
A total of 3650 patients were eligible for prediction of AM/CA, of whom 30.1% were assigned to the test set according to the date cutoff. Also, 2650 patients met the eligibility criteria for prediction of thromboembolic events (TEs), of whom 796 (30.0%) were assigned to the test set. Overall, 11.0% of eligible patients had AM/CA, and 1.5% of eligible patients had an imaging-confirmed TE.
The COVID-HEART predictor had area under the receiver operating characteristic curve (AUROC) of 0.918 and 0.771, sensitivities of 0.768 and 0.500, and specificities of 0.903 and 0.879 for the full test set regarding prediction of AM/CA and thromboembolic events, respectively. The mean cross-validation and test AUROCs were 0.917 (95% CI, 0.916-0.919) and 0.923 (95% CI, 0.918-0.927) for prediction of AM/CA and 0.757 (95% CI, 0.751-0.763) and 0.790 (95% CI, 0.756-0.824) for prediction of TEs, respectively.
The mean test AUROC, sensitivity, and specificity for the left-out hospitals for prediction of AM/CA were 0.956 (95% CI, 0.936-0.976), 0.885 (95% CI, 0.838-0.933), and 0.887 (95% CI, 0.843-0.932), respectively. Regarding imaging-confirmed TEs, the mean test AUROC, sensitivity, and specificity for the left-out hospitals were 0.781 (95% CI, 0.642-0.919), 0.453 (95% CI, 0.147-0.760), and 0.863 (95% CI, 0.822-0.904), respectively.
The final COVID-HEART predictor includes 61 features for predicting AM/CA, including routinely and continuously acquired vital signs and basic metabolic tests. It also includes 9 features for predicting TEs.
The requirement for imaging confirmation of TEs is one of several study limitations. In addition, only 35 patients were included in the development set with imaging-confirmed TEs and these outcomes could only be identified per day. Other limitations include the potential for measurement error, inaccurate patient-reported history, and missing data, as well as confounding by indication. Furthermore, only patients who sought care at a hospital were enrolled.
“In its current implementation the predictor can facilitate practical, meaningful change in patient triage and the allocation of resources by providing real-time risk scores for complications occurring commonly in COVID-19 patients,” the researchers wrote. “The COVID-HEART can be retrained to predict additional adverse cardiovascular events including myocardial infarction and arrhythmia. The potential utility of the predictor extends well beyond hospitalized COVID-19 patients, as COVID-HEART could be applied to the prediction of cardiovascular adverse events post-hospital discharge or in patients with chronic COVID syndrome (‘long COVID’).”
Shade JK, Doshi AN, Sung E, et al. Real-time prediction of mortality, cardiac arrest and thromboembolic complications in hospitalized patients with COVID-19. JACC Advances. Published online May 8, 2022. doi:doi.org/10.1016/j.jacadv.2022.100043