Predicting Prognosis in COVID-19 Patients using Machine Learning and Readily Available Clinical Data

Abstract

RationalePrognostic tools for aiding in the treatment of hospitalized COVID-19 patients could help improve outcome by identifying patients at higher or lower risk of severe disease.ObjectivesThe study objective was to develop models to stratify patients by risk of severe outcomes during COVID-19 hospitalization using readily available information at hospital admission.MethodsHierarchical ensemble classification models were trained on a set of 229 patients hospitalized with COVID-19 to predict severe outcomes, including ICU admission, development of ARDS, or intubation, using easily attainable attributes including basic patient characteristics, vital signs at admission, and basic lab results collected at time of presentation. Each test stratifies patients into groups of increasing risk. An additional cohort of 330 patients was used for blinded, independent validation. Shapley value analysis evaluated which attributes contributed most to the models’ predictions of risk.Measurements and Main ResultsTest performance was assessed using precision (positive predictive value) and recall (sensitivity) of the final risk groups. All test cut-offs were fixed prior to blinded validation. In both development and validation, the tests achieved precision in the lowest risk groups near or above 0.9. The proportion of patients with severe outcomes significantly increased across increasing risk groups. While the importance of attributes varied by test and patient, CRP, LDH, and D-dimer were often found to be important in the assignment of risk label.ConclusionsRisk of severe outcomes for patients hospitalized with COVID-19 infection can be assessed using machine learning-based models based on attributes routinely collected at hospital admission.

Machine learning models for predicting severe COVID-19 outcomes in hospitals

“…Campbell et al [33] devised hierarchical ensemble classification models for the prediction of several severe events connected with Covid-19 based on laboratory and clinical data available at admission. Due to missing values they removed about 50 percent of the training data and 85 percent of test data.…”

Research progress of artificial intelligence technology in COVID-19

“…Numerous researchers have employed diverse predictive indicators in their investigation of COVID-19 to evaluate the prognosis of the disease's ultimate clinical outcome. These predictive indicators comprise blood samples, electrocardiograms, imaging data (CT, XCR), respiratory parameters, clinical symptoms, and other relevant information [28][29][30][31][32][33][34][35] . The concluding indicators of the study were also varied, with a primary focus on the severity of the illness, the risk of mortality, the actual mortality rate, the length of hospitalization, and other relevant factors.…”

Retrospective Post-Hospitalisation COVID-19 Mortality Risk Assessment of Patients in South Africa

“…Campbel et al (2021) [23] develop predictive models that stratify hospitalised COVID-19 patients by their risk of severe outcomes, such as ICU admission, acute respiratory distress syndrome development, or intubation. The models were designed using hierarchical ensemble classification techniques and trained on a dataset of 229 COVID-19 patients.…”

“…Notably, attributes such as C-reactive protein, lactate dehydrogenase, and D-dimer were frequently identified as significant contributors to the risk assessments. The study concluded that machine learning-based models utilising routinely collected admission data can effectively assess the risk of severe outcomes in COVID-19 patients during hospitalisation [23].…”

“…where n is the number of individuals, t i is the survival time for individual i, Z i is the covariate vector for individual i, δ i is the censoring indicator (δ i = 1 if the event is observed and δ i = 0 if the event is censored), S(t i |Z i , ϑ) is the survival function for individual i, and f (u|ϑ) is the density function of the frailty term u. Estimation of the parameters in the frailty model can be done using maximum likelihood estimation or Bayesian methods [22,23].…”