key: cord-0953682-eze54ei9
authors: Tahir HUYUT, Mehmet; HUYUT, Zübeyir
title: Forecasting of Oxidant/Antioxidant Levels of COVID-19 Patients by using Expert Models with Biomarkers Used in the Diagnosis/Prognosis of COVID-19
date: 2021-09-09
journal: Int Immunopharmacol
DOI: 10.1016/j.intimp.2021.108127
sha: 39012853af931242b9aaca5b1a6a138f33ca34e7
doc_id: 953682
cord_uid: eze54ei9

BACKGROUND: Early detection of oxidant-antioxidant levels and special care in severe patients are important in combating the COVID-19 epidemic. However, this process is costly and time consuming. Therefore, there is a need for faster, reliable and economical methods. METHODS: In this study, antioxidant/oxidant levels of patients were estimated by Expert-models using biomarkers, which are effective in the diagnosis/prognosis of COVID-19 disease. For this purpose, Expert-models were trained and created between the white-blood-cell-count (WBC), lymphocyte-count (LYM), C-reactive-protein (CRP), D-dimer, Ferritin values of 35 patients with COVID-19 and antioxidant/oxidant parameter values of the same patients. Error criteria and R(2) ratio were taken into account for the performance of the models. The validity of the all models was checked by the Box-Jenkis-method. RESULTS: Antioxidant/Oxidant levels were estimated with 95% confidence-coefficient using the values of WBC, LYM, CRP, D-dimer, Ferritin of different 500 patients diagnosed with COVID-19 with the trained models. The error rate of all models was low and the coefficients of determination were sufficient. In the first data set, there was no significant difference between measured antioxidant/oxidant levels and predicted antioxidant/oxidant levels. This result showed that the models are accurate and reliable. In determining antioxidant/oxidant levels, LYM and Ferritin biomarkers had the most effect on models, while WBC and CRP biomarkers had the least effect. The antioxidant/oxidant parameter estimated with the highest accuracy was Native-Thiol divided by Total-Thiol. CONCLUSIONS: The results showed that the antioxidant/oxidant levels of infected patients can be estimated accurately and reliably with LYM, Ferritin, D-dimer, WBC, CRP biomarkers in the COVID-19 outbreak.

Coronavirus disease 2019 (COVID- 19) is an infectious disease caused by coronavirus 2 (SARS-CoV-2), which causes severe acute respiratory syndrome. This disease spreads far beyond the China in a few weeks and reached every part of the world. Real-time reverse transcription polymerase chain reaction (rRT-PCR), which is used in the diagnosis of COVID-19, can give negative results in 30-50% of cases, even if they are infected with the virus [1, 2] . As with immunodiagnostic testing, RT-PCR testing may have difficulties distinguishing between true positive and true negative individuals infected with COVID-19 [3, 4] . Since the test may fail in a significant proportion of suspected and confirmed patients with clinical results, it is prudent not to rely solely on PCR test results and to consider other clinical and molecular evidence [3, 5] . In addition, considering the difficulties in the RT-PCR test results, it was stated that the test should be repeated on more than one sample and the application methodology should be improved in order to increase the overall sensitivity of the test [3] . However, this is a difficult process for the staff and the patient. These difficulties in diagnosing COVID-19 have further increased the importance of routine laboratory results [1, 2] .

Also, while the number of COVID-19 cases is increasing day by day, there is limited information about the effects of hematological and laboratory findings associated with this disease on other parameters such as antioxidant-oxidant balance [1] . In addition, although the emergence of new technologies has accelerated the development of vaccines, there is limited information on the pathophysiology of the COVID-19 virus [6] . Therefore, many studies are still being conducted to determine the place and importance of routine laboratory parameters in COVID-19 [7] [8] [9] [10] .

Covid 19 disease is a multifactorial disease that can affect many parameters in the living body [1] . The studies showed that there was significant increases in proinflammatory cytokine levels in patients infected with covid 19 and a decrease in antioxidant capacity with an increase in oxidative stress [11, 12] .

When a new pathogen such as severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), emerges, all new information can assist in the monitoring and diagnosis of coronavirus disease (COVID-19) [13] . In many previous studies, it has been reported that oxidative stress is associated with the occurrence or prognosis of many diseases such as diabetes, alzheimer, obesity, immune deficiency, cardiovascular diseases and prognosis of viral diseases [14] [15] [16] . In order to reduce the effect of oxidative stress, there are antioxidant enzymes or systems such as superoxide dismutase, catalase and glutathione peroxidase, which provide the main antioxidant defense in the cell [17, 18] . However, in the disease state, oxidative stress increases with mutated antioxidant enzymes, toxins, or decreased consumption of natural antioxidants [19] . Since natural phenolic compounds show antioxidant activity, they play a vital role in suppressing oxidative stress both in the food industry and in the human body [20, 21] . For this reason, it was reported that supplements such as antioxidant vitamin E and C, and also vitamin D were important in patients infected with covid 19 [22, 23] . CRP, Ferritin and D-Dimer parameters among routine biochemistry parameters are especially used as determinants in the prognosis of in Covid 19 [1] . In many studies, it was stated that patients infected with COVID-19 started to increase in proinflammatory cytokines such as TNF-α and IL-1β and oxidants such as lipid peroxidation, disulfide level and oxidative DNA damage [24, 25, 26] . In these studies, it was also stated that patients infected with COVID-19 started to decrease in antioxidants such as superoxide dismutase (SOD), glutathione (GSH), glutathione peroxidase (GPx), total thiol, and native thiol [24, 25, 26] .

However, CRP, Ferritin and D-Dimer, which are among the routine parameters, can be measured easily and cheaply, while proinflammatory cytokine levels, antioxidant and oxidative stress markers can be measured manually by spectrophotometric methods (such as ELISA) by spending more money, effort and time.

Mathematical Models, which provide great advantages in terms of prevention of the time, economy and labor losses, continue to be one of the useful and popular methods in estimating the parameters that may affect the prognosis of many diseases. In this respect, Expert Models and tools (AR-ARMA-ARIMA) are thought to be useful in overcoming difficulties in obtaining values of the antioxidant and oxidant parameters.

Indeed, one study used expert methods to predict key epidemiological parameters and the evolution of COVID-19 [27] . Another study used a generalized logistic growth model, Richards growth model, and a wave growth model to estimate cases reported in Guangdong and Zhejiang, China [28] . Additionally, this study obtained real-time predictions for COVID-19 in China using three phenomenological models [28] . One study used ARIMA expert models to predict confirmed cases of COVID-19 [29] . In addition, different expert models were used in one study for the prediction of avian influenza H5N1 and another study for the prediction of COVID-19 cases [30, 31] .

In terms of predictive ability, ARIMA models are known to perform better than complex structural models [32, 33] . With expert models, one of the population estimation methods, the risk of choosing an individual wrong model is reduced and the hypothesis is generalize [33, 34, 35] .

Depending on the accuracy and diversity of individual models, ARIMA models have excellent generalization performance in predicting the population trend. ARIMA methods, also known as expert models, can make strong predictions using only the historical data of the cases. When the method is created with the right models, it can also be used to estimate the antioxidant-oxidant parameter values of the patients and help the treatment system to serve new patients more quickly and safely.

As far as we know, this manuscript is the first study to estimate antioxidant-oxidant values using routine blood values. In this study, antioxidant-oxidant levels of patients were estimated accurately and reliably with Expert-models using routine biochemistry and hematological laboratory parameters that are effective in the diagnosis-prognosis of COVID-19 disease.

This retrospective single-center study was conducted in accordance with the 1989 Helsinki Declaration. This study was approved by the Ministry of Health and the Ethics Committee. Data matching our criteria were collected from Erzincan Binali Yıldırım University Mengücek Gazi Training and Research Hospital information system between March 2020 and December 2020 and were included in the study. The research only covered people over 18 years old. Laboratory data of the patients were the first blood values measured at the time of first admission to the hospital.

Demographic data and blood values of 535 patients diagnosed with COVID-19 on the specified dates were obtained. The diagnosis of COVID-19 was defined only in cases detected as SARS-CoV-2 by rRT-PCR in nasopharyngeal or oropharyngeal swabs at the dates covered by this study in our hospital. Routine laboratory parameters used in the diagnosis-prognosis of COVID-19 were determined by scanning the literature. Determined biochemical and hematological values were obtained from all patients. The all data sets were obtained digitally from the hospital registry system.

The study consists of two data sets. In the first data set, there were routine blood values (white blood cell count "WBC", lymphocyte count "LYM", C-reactive protein "CRP", D-Dimer, 

Our primary aim in this study is to estimate the antioxidant-oxidant parameter values of the patients in the second data set with ARIMA models trained with the first data set. Our secondary aim is to determine which routine biochemistry or hematological blood values are more effective in predicting antioxidant-oxidant parameters of COVID-19 patients.

IBM SPSS Statistics 25 (Chicago, IL) software was used for statistical analysis. In this study, patients' routine blood values and antioxidant-oxidant values were estimated by the ARIMA model. The ARIMA model, also known as the Box-Jenkins methodology, uses the most recent observations as default values and describes past forecast errors to accurately set the next period [36] . The Box-Jenkins method is a prognostic tool with advanced mathematical and statistical procedures for dealing with complex situations. The method can be easily applied in risk analysis and uncertainty analysis [37, 38] .

Analysis of variance and classical regression will likely be misleading when autocorrelation exists between prediction errors. These data can be analyzed with the ARIMA model [38] . The ARIMA methodology analyzes the stochastic properties of various data series according to a particular philosophy, not the construction of a single or multiple equation model, as, for example, in regression analysis [36, 39, 40, 41, 42] .

Problems such as autocorrelation of lags, linearity problem and residual independence in our data series are solved with ARIMA model and predictions are made with low error rate.

Equations from the ARIMA models were run to estimate the antioxidant-oxidant levels of each patient in the study.

The ARIMA model is defined by parameters p, d or q. Here, p: lagged values of autoregressives; q: parts of the moving average, and d; indicates the number of differences in the data series. ARIMA assumes that the volatility (delays) of stationary data series can be expressed as a linear combination of observed values and errors of current values [35, 43] . If the data series is stationary, the process turns into ARMA (p,q) model and is defined by the following equation.

More details about the mathematical model can be found in [44] . The time series based model was modified for this study as follows. Here , t. the patient's predicted antioxidant-oxidant level; Y, t. the patient's actual antioxidant-oxidant level; c, initial ground level (intercept); ,…, 1 autoregressive (coefficient of each p parameter); ,…, the moving average (coefficient of 1 each q parameter) and , is the amount of residual error in the t-th patient's predicted parameter.

, determines the white noise criterion whose mean is 0 variance . 2 Since the data used in the study were stationary, the minimum difference order and moving average (d, q parameter) was chosen as zero (0) and the AR model (p) was used. White noise which are commonly used error criteria in the literature, were taken into account [46, 47] .

Accordingly, the p parameter were determined by considering the lowest value of the error criteria and the highest value of the R 2 (determination) coefficient. The work flow of the study is summarized in the diagram below (Fig 1) .

The best ARIMA models were obtained for estimating the antioxidant-oxidant levels of COVID-19 patients with the CRP, D-Dimer and Ferritin parameters, which are frequently used in the prognosis of COVID-19 disease (Table 3) . After ensuring the stability of the obtained models, Ljung-Box test was used for autocorrelation control. The validity of all the models appears to be satisfactory according to the Ljung-Box test (p>0.05) ( Table 3 ). In addition, when ACF and PACF graphs were examined, it was seen that autocorrelation or partial autocorrelation coefficients were not significant (p>0.05). In addition, the randomness and independence of the residuals in the prediction were checked by ACF/PACF plots (Fig 2) . Accordingly, none of the residul was significantly different from zero. According to this result, the data series was evaluated as stationary.

When (Table 3) is examined, the most successful prediction was seen in Native Thiol/Total

Thiol model according to all criteria, while the most unsuccessful prediction was seen in Native Thiol model. The algorithm left out of analysis the extreme values that do not fit the model. Errors

were seen at acceptable levels in the all models. However, R 2 (amount of variance explained in the estimation of antioxidant-oxidant levels of the patient group in the first data set) was found to be normal on average. Exponential smoothing was found to be significant in the models obtained to stabilize the lags in the estimation of SOD and GSH levels (p<0.05).

The statistical results of the parameters of the models obtained by ARIMA technique for estimating antioxidant-oxidant levels are given in detail in the (Table 4 ). However, only the parameters that significantly affected the models were reported in the (Table 4 ) (p<0.05).

Considering all prediction models, it was seen that LYM and Ferritin were the most effective parameters. This result was followed by D-Dimer, CRP and WBC parameters, respectively (Table   4 ). It was observed that only D-Dimer was effective in estimating GSH, MDA and GPx levels, and only Ferritin parameter was effective in estimating SOD levels. The LYM parameter was the most effective promoting variable in predicting the disulfide level. In the estimation of the total thiol level, it was observed the WBC parameter had a significant decreasing effect.

It was clearly seen how compatible the models obtained for the prediction with the antioxidant-oxidant levels of the patients in the first data set in Figure 3 . This figure showed that all models provide the fit between the predicted values and the measured values in the first data set. The predicted values showed that the model has a satisfactory level of predictability. The data set of antioxidant-oxidant levels in the first data set did not show a clear upward or downward trend, and these data set showed multiple peaks, most of which were not equally spaced. This result indicates that the data set is stationary for the general trend (Figures 2 and 3) . Confidence limits of the predicted antioxidant-oxidant levels of the second data set were visualized, according to the agreement obtained from the first data set (Fig 3) .

The descriptive statistics and confidence limits of the antioxidant-oxidant parameters of the patients included in the second data set with the models obtained from the study are shown in detail in (Table 5 ). Descriptive statistics were obtained with a 95% confidence coefficient.

In addition, antioxidant-oxidant levels of the patients in the first data set with the models obtained from the study were estimated as SOD: 28 antioxidant-oxidant results of the patients in the first data set. Accordingly, all antioxidant-oxidant prediction results were similar to the measured results and there was no significant difference between them (p > 0.05) (Fig 4) . This showed that the models produced accurate and reliable results.

The estimated antioxidant-oxidant levels and confidence limits of the patient group in the second data set were visualized (Fig 5) . The visuals in Figure 5 consist of two stages. The first step While the predicted antioxidant-oxidant levels for SOD and MDA in men were higher than in the women, GPx and GSH values in woman were higher than in the men. Other antioxidantoxidant levels were not different according to gender (Table 6 ).

It has been reported that oxidative stress plays an important role in the pathogenesis of various diseases [11, 12, 14, 19] . Many studıes showed that the levels of parameters such as MDA,

IL-1B and TNF-alpha associated with oxidative stress in various viral diseases were high compared to the control group while the levels of antioxidan parameters such as GSH, SOD, CAT and GSH-Px were low [24, 25] . Because the measurement of antioxidant-oxidant parameters is a manual process, it is a costly and time consuming endeavor. As in other viral diseases, it is inevitable that antioxidant-oxidant levels in patients with COVID-19 will vary according to healthy people [26] . For this reason, the fact that the data set on antioxidant-oxidant parameters consists of patients diagnosed with COVID-19 has further increased the importance of the study.

In this study, antioxidant-oxidant values of patients in the second data set were estimated at 95% confidence level with ARIMA models trained with the first data set. Next, it was determined which routine biochemistry or hematological blood values were more effective in predicting the antioxidant-oxidant parameters of COVID-19 patients. The best parameters of the prediction model were determined by evaluating the error criteria and the level of determination (R 2 ) [46, 47] . According to our literature review, this study is the first study about the estimation of some antioxidant-oxidant levels with routine blood laboratory findings.

It is known that Expert-ARIMA models perform better than structural models in terms of prediction performance [32, 33] . Expert models based on certain statistical calculations can be used for forward estimation using only the historical data of the cases [33] . For this reason, many studies have been conducted in the future predictions of COVID-19 cases with Expert models and are still being to carried out [48, 49, 50] .

The fact that ARIMA models provide various evaluation possibilities for the predicted dependent variable (for antioxidant-oxidant levels in this study) also reduces the need for a control group in studies [45, 46] . As a matter of fact, the scarcity of the control group in this study did not affect the success in the prediction negatively.

However, most prediction models only take into account the minimization of errors and constitute a single target for the estimation of data series. This may result with overfitting of the population estimation model in data series [35] . However, it is known that ARIMA models have successful generalization performance in estimating the dependent variable [33] . When the ARIMA method was created with the right models, it was emphasized that the method can also be used estimation of different parameters of the patients and it will help the treatment system to provide faster service to new patients [33] .

In this study, the lack of significant difference between the predictive antioxidant-oxidative results obtained for the first patient group and the measured antioxidant-oxidant levels of these patients (Fig 4, p>0 .05) showed that the models produced reliable results. Similarly, in two different studies using Expert models, a linear combination of past and present values and errors in the data set was created and reliable predictive values were obtained [45, 46] .

Considering all prediction models in this study, it was found that LYM and Table 4 ). These results Also, the low number of patients in the first data set prevented the amount of variance explained (R 2 ) in the estimate from being high. However, the variance amounts explained by the models in the estimation of antioxidant-oxidant levels in the first data set were found to be about 60%, and the amount of error was found to be less than 10%. With these results, when the data set of the dependent variable (measured antioxidant-oxidant levels) is augmented, it is expected that the amount of R 2 predicted by the prediction models will definitely increase.

The first data set was conducted on a relatively small population, because limited number of studies have been conducted to measure antioxidant-oxidant levels in COVID-19 patients. For this reason, in order for the amount of variance to be high, it is necessary to make estimation studies in larger data sets. Our data is derived from an electronic recording system that places restrictions on the provision of old information. Retrospective studies naturally lack control of variables. Therefore, prospective cohorts are also needed to validate our study data. In addition, other differential data such as comorbidity of the patients could not be achieved.

In 

How do routine laboratory tests change in coronavirus disease 2019?

Estimating falsenegative detection rate of SARS-CoV-2 by RT-PCR

Recent advances and challenges of RT-PCR tests for the diagnosis of COVID-19. Pathology -Research and Practice

Laboratory testing methods for novel severe acute respiratory syndrome-coronavirus-2 (SARS-CoV-2). Front

Real-time RT-PCR in COVID-19 Detection: Issues Affecting the Results. Expert Revıew of Molecular Dıagnostıcs

Potential SARS-CoV-2 vaccines: Concept, progress, and challenges

Shock index as a predictor of mortality among the Covid-19 patients

The Effectiveness of Blood Routine Parameters and Some Biomarkers as a Potential Diagnostic Tool in the Diagnosis and Prognosis of Covid-19 Disease

Hematologic predictors of mortality in hospitalized patients with COVID-19: a comparative study

Prediction of severe illness due to COVID-19 based on an analysis of initial Fibrinogen to Albumin Ratio and Platelet count

Clinical characteristics of refractory COVID-19 pneumonia in Wuhan, China

SARS-CoV-2 infection pathogenesis is related to oxidative stress as a response to aggression

Acute and post-acute phase of COVID-19: Analyzing expression patterns of miRNA-29a-3p, 146a-3p, 155-5p, and let-7b-3p in PBMC

In vitro antioxidant, antimicrobial, anticholinesterase and antidiabetic activities of Turkish endemic Achillea cucullata (Asteraceae) from ethanol extract

LC-MS/MS analysis, antioxidant and anticholinergic properties of galanga (Alpinia officinarum Hance) rhizomes. Industrial Crops and Products

Anticholinergic and antioxidant activities of usnic acid-An activitystructure insight

Characteristics of resveratrol and serotonin on antioxidant capacity and susceptibility to oxidation of red blood cells in stored human blood in a time-dependent manner

The susceptibility of erythrocytes to oxidation during storage of blood: Effects of melatonin and propofol

Antioxidant activity of food constituents: an overview

Antioxidant and anticholinergic properties of olivetol

Antioxidants and antioxidant methods-An updated overview

Coronavirus disease (COVID-19) and immunity booster green foods: A mini review

Vitamin D high doses supplementation could represent a promising alternative to prevent or treat Covid-19 infection

Revisiting pharmacological potentials of Nigella sativa seed A promising option for COVID-19 prevention and cure

COVID-19, cytokines, inflammation, and spices: How are they related?

Oxygen-ozone immunoceutical therapy in COVID-19 outbreak: facts and figures

Data-based analysis, modelling and forecasting of the COVID-19 outbreak

Real-time forecasts of the COVID-19 epidemic in China from

Forecasting the novel coronavirus COVID-19

Comparison of ARIMA and Random Forest time series models for prediction of avian influenza H5N1 outbreaks

Forecasting of COVID-19 Confirmed Cases in Different Countries with ARIMA Models

Forecasting Irish inflation using ARIMA models

A time series-based statistical approach for outbreak spread forecasting: Application of COVID-19 in Greece

A novel ensemble approach for the forecasting of energy demand based on the artificial bee colony algorithm

Optimal selection of heterogeneous ensemble strategies of time series forecasting with multi-objective programming. Expert Systems With Applications

Introduction to Time Series and Forecasting

Knowledge, business, and innovation: Economies and sustainability of future growth

Arıma model for predıctıng the development of the prıce of gold: european approach. Ekonomicko-manazerske spektrum 2020

Methodologies for forecast modeling for small areas with limited data availability and unique tax structures

Time series analysis using autoregressive integrated moving average (ARIMA) models

Banks´balance sheets and liquidation values: Evidence from real estate collateral

A survey on data mining techniques applied to electricity-related time series forecasting

A hybrid ARIMA and support vector machines model in stock price forecasting

Time series analysis: Forecasting and control. Holden Day

Failed attempts to reduce ınapproprıate laboratory utılızatıon ın an emergency department settıng ın cyprus: lessons learned

Introduction to time series analysis and forecasting

The Success of Restricted Ordination Methods in Data Analysis with Variables at Different Scale Levels

Modeling Nigerian Covid-19 cases: A comparative analysis of models and estimators. Kaos, Solitonlar ve Fraktallar

Modeling and Forecasting for the number of cases of the COVID-19 pandemic with the Curve Estimation Models, the Box-Jenkins and Exponential Smoothing Methods

Evaluation and prediction of COVID-19 in India: A case study of worst hit states

The hemocyte counts as a potential biomarker for predicting disease progression in COVID-19: a retrospective study

Prominent changes in blood coagulation of patients with SARS-CoV-2 infection

Auto Correlation Figure; PACF: Partial Auto Correlation Figure; ARIMA: Autoregressive Integrated Moving Average; MAPE: Mean Absolute Percent Error, MAE: Mean Absolute Error; RMSE: Root Mean Square Error; BIC: Bayesian Information Criterion; R 2 : Determınatıon coeffıcıent

Superoxide dismutase; GPx: Glutathione peroxidase; GSH: Reduced glutathione; MDA: Malondialdehyde; 8-OHdG: 8-hydroxy-2-deoxy guanosine; Min: Minimum; Max: Maximum; St EM: Standart error of mean

Antioxidants/oxidants were estimated by biomarkers used in the prognosis/diagnosis of COVID-19

Blood biomarkers that are effective in predicting antioxidant-oxidant levels were determined

Antioxidants-oxidants were estimated by biomarkers LYM

LYM/Ferritin was most effective and WBC/CRP less effective in estimating antioxidants/oxidants

Expert-methods is an accurate/safe alternative method for estimating antioxidants/oxidants