The current study aimed to retrospectively develop and validate ML models based on the most relevant features in determining the risk of COVID-19 mortality derived from extensive literature review coupled with a two-round Delphi survey. For this aim, the J48 decision tree, RF, k-NN, MLP, NB, XGBoost, and LR models were developed using a dataset of laboratory-confirmed COVID-19 hospitalized patients. The experimental results showed that RF had the best performance among the other seven ML techniques with the accuracy of 95.03%, sensitivity of 90.70%, precision of 94.23%, specificity of 95.10%, and ROC around 99.02%. Our results showed that RF, XGBoost, KNN, and MLP models have a good prediction performance, the ROC is all above 96.49%, and their diagnostic efficiency is better than the LR model trained using the same parameters.

Different studies have been evaluating the application of ML techniques in predicting mortality in the patients with COVID-19. Yadaw et al. [30] assessed the performance of four ML algorithms including LR, RF, SVM, and XGBoost using a dataset (n = 3841) for predicting COVID-19 mortality. The model developed with XGBoost happened to be the best model among all the models developed in terms of AUC with 0.91%. In another study [23] a retrospective analysis on the data of 2520 COVID-19 hospitalized patients was conducted. Results of this study showed the model developed by the neural network (NN) yielded better performance and was the best model in terms of AUC with 0.9760% in predicting COVID-19 patient’s physiological deterioration and death among other models developed by logistic regression (LR), SVM, and gradient boosted decision tree. Vaid et al. [41] in their study analyzed data of 4029 confirmed COVID-19 patients from EHRs of five hospitals, and logistic regression with L1 regularization (LASSO) and MLP models was developed via local data and combined data. The federated MLP model (AUC-ROCs of 0.822%) for predicting COVID-19 related mortality and disease severity outperformed the federated LASSO regression model. Other study conducted [42] four ML techniques were trained based on 10,237 patients’ data and, finally, SVM with the sensitivity of 90.7%, specificity of 91.4%, and ROC of 0.963% had the best performance. Moulaei et al. [31] also predicted the mortality of Covid-19 patients based on data mining techniques and concluded that based on ROC (1.00), precision (99.74%), accuracy (99.23%), specificity (99.84%) and sensitivity (98.25%), RF was the best model in predicting mortality. After, the RF, KNN5, MLP, and J48 were the best models, respectively [31]

In the current study, some features such as dyspnea, ICU admission, oxygen therapy (intubation), age, fever, and cough were of the highest importance; on the other hand, alcohol/addiction, platelet count, alanine aminotransferase (ALT), and smoking were of the lowest importance in predicting COVID-19 mortality. However, from the physicians’ point of view, awareness of these factors may be crucial for the success of drug therapy and mortality prediction. But in ML techniques, many of these factors can be ignored from analysis and mortality can be predicted with fewer factors.

Several studies have also reported some important clinical features(predictors) for COVID-19 patient mortality by leveraging a feature analysis technique. The selected features are used as inputs for developing ML-based models for severity, deterioration, and mortality of COVID-19 patient risk analysis. The strongest predictive features included basic data such as age (aged) [11, 17, 28, 30, 43,44,45,46], gender (male) [10, 11, 18, 27, 29, 44, 46], BMI (high) [15,16,17], type of patient encounter (inpatient vs. outpatient) [11, 23, 27, 29], occupation (related to healthcare) [17, 23, 29, 30], clinical symptoms include dyspnea [15, 16, 23, 30, 31, 44, 47], low consciousness [11, 17, 18, 28], dry cough[15, 17, 18, 23, 27, 28, 44] fever [11, 17, 18, 43,44,45, 47], para-clinical indicators consisting of spo2 (decreased) [16, 18, 29, 45, 47], lymphocyte count (low) [10, 23, 27,28,29], platelet count (low) [16, 27,28,29, 47], leukocyte count (raised) [15, 16, 27, 28, 30, 44], neutrophil count (raised) [15, 23, 27, 28, 30, 43, 45], CRP (increased) [15, 29, 30, 45], D dimer (increased) [10, 30, 45], ALT and/or AST (raised) [16, 27, 28, 30, 47], cardiac troponin (increased) [23, 28, 29, 43], and LDH (elevated) [17, 27, 28, 48], and comorbidity conditions associated with poor prognosis including hypertension [28,29,30, 44,45,46], lung disease including chronic obstructive lung disease [11, 16, 27, 28], asthma [16, 18], cardiovascular disease [28,29,30, 43, 45, 47], cancer [11, 44, 47], pneumonia [11, 17, 46,47,48], and chronic renal disease [11, 15, 17, 18, 46]. On the other hand, sore throat [11, 27, 28, 30], myalgia and malaise [11, 29, 30], diarrhea and GI symptoms [23, 44, 45], and headache [11, 17, 47] for clinical manifestation and hemoglobin count [11, 15, 45, 47, 48] as well as mean cell volume (MCV) [16, 17, 28, 44] and hematocrit rate [18, 27,28,29] for the laboratory findings have the least importance for predicting.

Finally, ML can be of great use for the clinicians involved in treating the patients with COVID-19. The proposed algorithms can predict the mortality of the patients with optimum ROC, accuracy, precision, sensitivity, and specificity rates. This prediction can lead to the optimal use of hospital resources in treating the patients with more critical conditions and assisting in providing more qualitative care and reducing medical errors due to fatigue and long working hours in the ICU. Designing a valid predictive model may improve the quality of care and increase the survival rate of the patients. Therefore, predictive models for mortality risk analysis can greatly contribute to identifying high-risk patients and adopting the most effective assistive and treatment care plans. This could lead to decreasing ambiguity by offering quantitative, objective, and evidence-based models for risk stratification, prediction, and eventually episode of the care plan. It offers a better strategy for clinicians to lessen the complications and improve the likelihood of patient survival.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Disclaimer:

This article is autogenerated using RSS feeds and has not been created or edited by OA JF.

Click here for Source link (https://www.biomedcentral.com/)