COVID-19 diagnosis prediction in emergency care patients: a machine learning approach

Abstract

The coronavirus disease pandemic has increased the necessity of immediate clinical decisions and effective usage of healthcare resources. Currently, the most validated diagnosis test for COVID-19 (RT-PCR) is in shortage in most developing countries, which may increase infection rates and delay important preventive measures. The objective of this study was to predict the risk of positive COVID-19 diagnosis with machine learning, using as predictors only results from emergency care admission exams. We collected data from 235 adult patients from the Hospital Israelita Albert Einstein in São Paulo, Brazil, from 17 to 30 of March, 2020, of which 102 (43%) received a positive diagnosis of COVID-19 from RT-PCR tests. Five machine learning algorithms (neural networks, random forests, gradient boosting trees, logistic regression and support vector machines) were trained on a random sample of 70% of the patients, and performance was tested on new unseen data (30%). The best predictive performance was obtained by the support vector machines algorithm (AUC: 0.85; Sensitivity: 0.68; Specificity: 0.85; Brier Score: 0.16). The three most important variables for the predictive performance of the algorithm were the number of lymphocytes, leukocytes and eosinophils, respectively. In conclusion, we found that targeted decisions for receiving COVID-19 tests using only routinely-collected data is a promising new area with the use of machine learning algorithms.

A machine learning algorithm to increase COVID-19 inpatient diagnostic capacity

“…Our set of features performed as well as, or better than, the three diagnostic algorithms with the largest number of cases known to us at this time [12,13,16]. A report by Sun et al used epidemiologic, clinical, laboratory and imaging features in their algorithms and reported AUROCs of 0.91 (full model), 0.88 (without epidemiologic features), 0.88 (without imaging features), and 0.65 (with clinical features alone) [12].…”

“…Meng et al reported an AUROC of 0.89 using a different set of features that included activated partial thromboplastin time, triglycerides, uric acid, albumin/globulin, sodium, and calcium [16]. Batista et al developed an algorithm aimed for use in lower resource settings and reported an AUROC of 0.87 in a sparser dataset that only included basic demographics and complete blood cell counts [13]. In fact, our model which incorporated inflammatory markers significantly improved upon this set of features in terms of both AUROC and AUPRC.…”

“…Other models have also been developed based on characteristic imaging changes [15]. We and others were able to demonstrate impressive results in our data silos [12][13][14][15]. Yet, to realize the full potential of machine learning and its applicability to clinical medicine, collaborations from the international community are crucial, both for the sharing of data and for the development and validation of advanced algorithms.…”

“…These changes include elevations in inflammatory markers (ferritin, lactate dehydrogenase [LDH], Creactive protein, among others) and decreases in certain blood cell counts (absolute lymphocyte count) and an increase in the neutrophil to lymphocyte ratio. Since the SARS-CoV-2 epidemic reached pandemic status, research groups developed prediction algorithms applicable to their particular context [12][13][14][15][16]. One of the major limitations of these previous approaches is that the datasets that were used to train and test the approaches were small.…”

A model to predict SARS‐CoV‐2 infection based on the first three‐month surveillance data in Brazil

“…In this study, we identified a set of variables that may help differentiate COVID‐19 cases from other diseases. The model obtained exhibited an excellent AUC in the SP/RJ dataset comparable to more complex tools, including imaging and laboratory tests [11, 17–19]. This is impressive, considering that it is based solely on variables collected by the surveillance system.…”