ObjectiveRetrospective study of COVID-19 MESHD
positive patients treated at NYU Langone Health (NYULH) to identify clinical markers predictive of disease severity to assist in clinical decision triage and provide additional biological insights into disease progression.
Materials and MethodsClinical activity of 3740 de-identified patients at NYULH between January and August 2020. Models were trained on clinical data during different parts of their hospital stay to predict three clinical outcomes: deceased, ventilated, or admitted to ICU.
ResultsXGBoost model trained on clinical data from the final 24 hours excelled at predicting mortality (AUC=0.92, specificity=86% and sensitivity=85%). Respiration rate was the most important feature, followed by SpO2 and age 75+. Performance of this model to predict the deceased outcome extended 5 days prior with AUC=0.81, specificity=70%, sensitivity=75%. When only using clinical data from the first 24 hours, AUCs of 0.79, 0.80, and 0.77 were obtained for deceased, ventilated, or ICU admitted, respectively. Although respiration rate and SpO2 levels offered the highest feature importance, other canonical markers including diabetic MESHD
history, age and temperature offered minimal gain. When lab values were incorporated, prediction of mortality benefited the most from blood urea nitrogen (BUN) and lactate dehydrogenase (LDH). Features predictive of morbidity included LDH, calcium, glucose, and C-reactive protein HGNC
( CRP HGNC
ConclusionTogether this work summarizes efforts to systematically examine the importance of a wide range of features across different endpoint outcomes and at different hospitalization time points.
BACKGROUND AND SIGNIFICANCEThe first cluster of SARS-CoV-2 was reported in Wuhan, Hubei Province on December 31, 2019. Inciting symptoms remarkably similar to pneumonia MESHD
, the disease quickly traveled around the world, earning its pandemic status by the World Health Organization on March 11, 2020. Although the first wave has since passed for hardest-hit regions such as New York City (NYC) and most of Asia, a resurgence of cases has already been reported in Europe and record new cases tallied in the Midwest and rural United States (US). As of November 12th, the US alone logged its highest tally to date with a 317% growth over the preceding 30 days1. The coronavirus disease MESHD
( COVID-19 MESHD
) is far from seeing the end of its days and there remains a compelling need to prioritize care and resources for patients at elevated risk of morbidity and mortality.
Previous work building machine learning models used patient data from Tongji Hospital2,3 (Wuhan, China), Zhongnan Hospital4 (Wuhan China), Mount Sinai Hospital5 (NYC, US), and NYU Family Health Center6 (NYC, US). Surprisingly, clinical features selected varied widely across studies. For example, while McRae et al.s 2-tiered model6 trained on 701 NYC patients to predict mortality was based on actual age, C-reactive protein HGNC
( CRP HGNC
), procalcitonin, and D-dimer, Yan et al.s model2 trained on 485 patients from Wuhan selected lactate dehydrogenase (LDH), lymphocyte count, and CRP HGNC
as the most predictive for mortality. Variations MESHD
in selected features differed greatly even when trained to predict similar outcomes on data from patients of the same city. Yao et al.s model3 was trained on 137 patients from Wuhan and relied on 28 biomarkers in their final model to predict morbidity. Given the differences among prior models, some of which were driven by domain-specific knowledge, we decided to systematically examine the importance of a wide range of features across different endpoint outcomes and at different hospitalization time points.
This study analyzes retrospective PCR-confirmed COVID-19 MESHD
inpatient data collected at NYU Langone Hospital spanning 1/1/2020 to 8/7/2020 to predict three sets of clinical outcomes: alive vs deceased, ventilated vs not ventilated, or ICU admitted vs not ICU admitted. The clinical information of 3740 patient encounters included demographic data (age, sex, insurance, past diagnosis of diabetes MESHD
, presence of cardiovascular comorbidities), vital signs (SpO2, pulse, respiration rate, temperature, blood pressure), and the 50 most frequently ordered lab tests in our dataset. Models were developed using two methods: logistic regression with feature selection using Least Absolute Shrinkage and Selection Operator7 (LASSO) and gradient tree boosting with XGBoost8. An explainable algorithm, such as logistic regression, provides easy to interpret insights into the features of importance. Conversely, the larger model capacity of XGBoost better handles data complexities to explore the extent that predictive performance can be optimized. Together, these methods ensure a holistic survey that explores the clinical underpinnings of disease etiology and the prospects of building models that are sufficiently competent to be effective decision support tools.