Archives of Academic Emergency Medicine. 2023; 11(1): e15 OR I G I N A L RE S E A RC H Clinical Risk Factors of Need for Intensive Care Unit Ad- mission of COVID-19 Patients; a Cross-sectional Study Farshid Sharifi1, Mohammad Hossain Mehrolhassani2, Milad Ahmadi Gohari1, Ali Karamoozian1,3, Yunes Jahani1,3∗ 1. Modeling in Health Research Center, Institute for Futures Studies in Health, Kerman University of Medical Sciences, Kerman, Iran. 2. Health Services Management Research Center, Institute for Futures Studies in Health, Kerman University of medical sciences, Kerman, Iran. 3. Department of Biostatistics and Epidemiology, School of Public Health, Kerman University of Medical Sciences, Kerman, Iran. Received: October 2022; Accepted: December 2022; Published online: 1 January 2023 Abstract: Introduction: It could be beneficial to accelerate the hospitalization of patients with the identified clinical risk factors of intensive care unit (ICU) admission, in order to control and reduce COVID-19-related mortality. This study aimed to determine the clinical risk factors associated with ICU hospitalization of COVID-19 patients. Methods: The current research was a cross-sectional study. The study recruited 7182 patients who had positive PCR tests between February 23, 2020, and September 7, 2021 and were admitted to Afzalipour Hospital in Kerman, Iran, for at least 24 hours. Their demo- graphic characteristics, underlying diseases, and clinical parameters were collected. In order to analyze the relationship between the studied variables and ICU admission, multiple logistic regression model, classification tree, and support vector machine were used. Results: It was found that 14.7 percent (1056 patients) of the study participants were admit- ted to ICU. The patients’ average age was 51.25±21 years, and 52.8% of them were male. In the study, some factors such as decreasing oxygen saturation level (OR=0.954, 95%CI: 0.944-0.964), age (OR=1.007, 95%CI: 1.004-1.011), respiratory distress (OR=1.658, 95%CI: 1.410-1.951), reduced level of consciousness (OR=2.487, 95%CI: 1.721-3.596), hypertension (OR=1.249, 95%CI: 1.042-1.496), chronic pulmonary disease (OR=1.250, 95%CI: 1.006-1.554), heart diseases (OR=1.250, 95%CI: 1.009-1.548), chronic kidney disease (OR=1.515, 95%CI: 1.111-2.066), cancer (OR=1.682, 95%CI: 1.130-2.505), seizures (OR=3.428, 95%CI: 1.615-7.274), and gender (OR=1.179, 95%CI: 1.028-1.352) were found to significantly affect ICU admissions. Conclusion: As evidenced by the obtained results, blood oxygen saturation level, the patient’s age, and their level of consciousness are crucial for ICU admission. Keywords: COVID-19; intensive care units; logistic models; decision trees; support vector machine Cite this article as: Sharifi F, Mehrolhassani MH, Ahmadi Gohari M, Karamoozian A, Jahani Y. Clinical Risk Factors of Need for Intensive Care Unit Admission of COVID-19 Patients; a Cross-sectional Study. Arch Acad Emerg Med. 2023; 11(1): e15. https://doi.org/10.22037/aaem.v11i1. 1853. 1. Introduction The current COVID-19 pandemic caused by SARS-CoV-2 was first detected in December 2019 in Wuhan, China (1, 2). As the number of new cases of COVID-19 increased unexpect- edly and the disease rapidly spread throughout the world, the World Health Organization declared a Coronavirus pan- demic on March 11, 2020 (3). To date, more than 640 mil- lion cases and 6.61 million deaths have been reported world- ∗Corresponding Author: Yunes Jahani; Modeling in Health Research Cen- ter, Second floor, Institute for Futures Studies in Health Building, Kerman University of Medical Sciences, the beginning of the seven gardens road, Kerman, Iran. Postal code/ P.O. Box: 761-6913555, Telephone number: 00983431325405, Fax Number: 00983432114278, Email: u.jahani@kmu.ac.ir; yonesjahani@yahoo.com, ORCID: https://orcid.org/0000-0002-6808-7101. wide (4). Many COVID-19 patients experience a relatively se- vere illness after a period of mild symptoms, and it is cru- cial to make quick and accurate diagnoses and to provide high-quality care to those who require admission to the in- tensive care unit (ICU) (5, 6). According to previous studies, hospitalization in the ICU plays a very important role in the treatment of COVID-19 patients, as it has been proven to be very effective on reducing mortality among these patients (1). Several studies have reported that gender, age, and underly- ing diseases are associated with severe disease and hospital- ization in an ICU. Among patients with severe disease, acute kidney injury, acute respiratory distress syndrome, and heart diseases are the most commonly reported complications (7- 10). Based on the need and capacity of the medical center in various countries, the rate of COVID-19 patients admitted to This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index F. Sharifi et al. 2 ICU varies from 5% to 32% (11). As a result of the increased patient numbers and the lengthening of hospital stays, the hospital’s capacity to admit new patients is severely limited (12). The severely-limited capacity of the hospital to admit new patients may negatively affect patient care quality and also increase mortality rates (13). By identifying and classify- ing patients with high risks, a greater number of beds can be saved in the ICU. Studies have shown that accelerating the admission of patients with the identified major clinical risk factors for ICU admission can reduce mortality rate. In addi- tion to other nations, Iran has also been affected by COVID- 19. It has been estimated that approximately 7.55 million cases and more than 145,000 deaths have occurred since the first case of COVID-19 was identified on February 18, 2020 (4, 14). It is noteworthy that approximately more than 500,000 people live in the city of Kerman, located in southeastern Iran (15). The city of Kerman is one of those in Iran where many pa- tients require admission to the ICU (16). More than half of the hospital’s capacity for ICU admissions is used during ordi- nary times in Kerman’s hospitals (17). Similar to all hospitals and ICUs throughout the world, Kerman’s hospitals and their ICUs face some limitations during a crisis like the COVID-19 epidemic (18). Identifying significant clinical risk factors in the early stages of this disease is crucial for predicting which patients will require admission to the ICU (19). Based on a patient’s disease symptoms, this study aimed to assist physi- cians in quickly identifying patients who will need ICU ad- mission. 2. Methods 2.1. Study design and settings The present study was a cross-sectional study conducted in Afzalipour Hospital, Kerman, Iran (700 beds), with focus on patients who had been hospitalized with COVID-19 disease (20). The admission rules were defined by the guidelines published by the Iranian Ministry of Health for managing COVID-19 patients and based on a combination of clinical data (1). The required data were collected from electronic medical records of 7182 patients with a positive COVID-19 PCR test, who were admitted to Afzalipour Hospital in Ker- man, Iran, between February 23, 2020, and September 7, 2021, using the census method. Research Ethics Commit- tee of Kerman University of Medical Sciences approved this study (No. IR.KMU.REC.1400.451). 2.2. Participants The present study included all COVID-19 patients who had been hospitalized for more than 24 hours and were diag- nosed using SARS-COV-2 nucleic acid RT-PCR (7, 21). Pa- tients with a negative PCR test for COVID-19 were excluded from the study. Also, hospitalized patients whose test status was unclear and were suspected to having COVID-19 were excluded from the study. The patients were admitted to the ICU directly or after being admitted to the general ward. 2.3. Data collection For this study, the primary data source was electronic medical records from the Medical Care Monitoring Center (MCMC) system, which are frequently used in emergency sit- uations. The patients’ data were accurately recorded in the MCMC system. The data for 10 patients were missing and they were excluded from the study. The dependent variable was receiving ICU care, for patients who were hospitalized either in the ICU or in the general ward (2). In this study, studied variables included age, gender, heart diseases, chronic kidney disease, chronic pulmonary dis- ease, chronic liver disease, diabetes, cancer, hypertension, chronic neurological diseases, blood diseases, immunodefi- ciency (Acquired or congenital), fever, myalgia, cough, chest pain, diarrhea, respiratory distress, vomiting and nausea, headache, loss of consciousness, smell or taste disorders, anorexia, seizures, dizziness, opium abuse, smoking, oxygen therapy, oxygen saturation level, and the time between the onset of symptoms and admission to the hospital. 2.4. Statistical analyses In order to analyze the obtained data, the mean, standard de- viation, frequency, and percentage were calculated. To con- trol the effect of confounding variables, Univariate and mul- tiple logistic regression analyses were run and the effect of each variable was evaluated by adjusting for other variables. At first, univariate regression was performed, and variables with p-values less than 0.2 were considered as important and then added to multiple regression model. Finally, p-values less than 0.05 were used to identify significant variables in the multiple logistic regression model and the backward ap- proach. Additionally, the odds ratios and 95% confidence in- tervals were reported (22, 23). The classification and regression trees (CART) model was used to classify the investigated variables, identify specific ICU admission groups, and evaluate the interaction between the variables (24). A tree model was built with 10-fold cross- validation, with 100 cases in the parent node and 50 subjects in the child node, and 10 maximum levels for the Classifica- tion and Regression Trees model (CART) (25). The misclassi- fication cost for patients who were wrongly admitted to the general ward instead of the ICU was 5 times more than pa- tients admitted to the ICU instead of the general ward. There- fore, it was considered that an effective strategy to solve the imbalanced classification problem is to maximize the sum of sensitivity and specificity (26, 27). Moreover, support vec- This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index 3 Archives of Academic Emergency Medicine. 2023; 11(1): e15 tor machine method was used to investigate and control the non-linear effect of confounding variables and separation of the variables affecting hospitalization in the ICU. Support vector machine method provides a powerful sepa- rator margin among classes using the kernel trick and con- verting the data into another dimension. This study used the radial basis function (RBF) as the kernel, and 10-fold cross-validation was applied to estimate model parameters (28, 29). Afterward, the accuracy of support vector machine model and the importance of each variable regarding ICU ad- mission were determined (30). 2.5. Sensitivity and specificity of models The data in this study were validated using the K-fold valida- tion method. In the K-fold method, the data are divided into K subsets. Besides, each sample is used once for validation and k-1 times for training. This process is repeated K times and all the data are used once for validation. Finally, the average of these K validation times is reported. K was equal to 10 in the present study (31). Accuracy, sensitivity, and specificity of the model was evalu- ated by comparing the prediction of the model regarding pa- tients’ need for ICU hospitalization with the reality (32). In the present study, descriptive analysis, univariate and multiple logistic regression, and Classification Tree model were conducted using SPSS.25 software and analysis of sup- port vector machine was done using R software packages (caret and e1071). 3. Results 3.1. Patient characteristics The current study recruited 7,182 COVID-19 patients with the mean age of 51.25 ± 21.00 (range: 0.02 (7 days)-102) years (52.8% male). The number of the patients admitted to the ICU was 1056 (14.7%). Among the patients, 413 (5.8%) cases were smokers and 898 (12.5%) cases were opium abusers. There were 1211 (16.9%) patients with hypertension, 976 (13.6%) patients with diabetes, 694 (9.7%) patients with heart diseases, and 608 (8.5%) patients with chronic pulmonary disease. The most prevalent signs and symptoms of these pa- tients were respiratory distress with 4952 (69.0%), fever with 3426 (47.7%), and cough with 3603 (50.2%) cases. Oxygen therapy was provided to 5684 (71.9%) of the hospitalized pa- tients. According to this analysis, the patients were hospital- ized at an average oxygen saturation level of 90.51±5.61 per- cent, and the time from the onset of symptoms to hospital- ization was 5.7±3.59 days. Table 1 shows the association be- tween studied clinical variables with need for ICU admission. 3.2. Multiple logistic regression analysis of clini- cal risk factors for ICU admission Table 2 shows the results of multiple regression analysis of independent clinical risk factors of COVID-19 patients’ need for ICU admission. In multiple logistic regression, male cases had 1.179 times the odds of being admitted to the ICU compared to females (OR=1.179, 95%CI:1.028-1.352). In addition, being admitted to the ICU was associated with age (OR=1.007, 95%CI: 1.004-1.011). Patients with heart diseases (OR=1.250, 95%CI=1.009-1.548), chronic pulmonary disease (OR=1.250, 95%CI=1.006-1.554), chronic kidney disease (OR=1.515, 95%CI=1.111-2.066), hyperten- sion (OR=1.249, 95%CI=1.042-1.496), and cancer (OR=1.682, 95%CI=1.130-2.50) had higher odds of being admitted to the ICU. There was a higher chance of admission to the ICU among the patients with signs and symptoms of loss of consciousness (OR=2.487, 95%CI=1.721-3.596), seizures (OR=3.428, 95%CI=1.615-7.274), respiratory distress (OR=1.658, 95%CI=1.410-1.951), and a decrease in oxygen saturation level (OR=0.954, 95%CI=0.944-0.964). Moreover, the patients who had myalgia (OR=0.847, 95%CI=0.733- 0.978), cough (OR=0.827, 95%CI=0.722-0.949), diarrhea (OR=0.727, 95%CI=0.561-0.941), headache (OR=0.708, 95%CI=0.593-0.844), and smell or taste disorders (OR=0.622, 95%CI=0.474-0.816) had a lower chance of being admitted to the ICU. The total sensitivity and specificity of the model were max- imized with a cut point of 0.145 in predicted probability of receiving ICU care. Multiple logistic regression model had 62.4% (95%CI=59-65) accuracy, 59.7% (95%CI=51.2-68.2) sensitivity, and 62.7% (95%CI=58.2-67.3) specificity. 3.3. Classification Tree The estimated tree model depicts a tree with five levels and eight nodes (Figure 1). Using the tree structure, it was determined that oxygen saturation level, age, and level of consciousness were the variables affecting ICU hospitaliza- tion. There were four high-risk groups of patients admitted to the ICU: The first group consisted of the patients with oxygen saturation levels less than 86.5%, the second group consisted of the patients younger than 0.196 years (72 days) whose oxygen saturation was greater than 86.5%, the third group consisted of patients older than 66.5 years with oxygen saturation level greater than 86.5%, and the fourth group consisted of the patients with a lowered level of conscious- ness aged between 0.196 and 66.5 years (Figure 1). A cost of misclassification of 5 led both the sensitivity and specificity of the model to reach their maximum levels. Classification Tree Model had an accuracy of 68.5% (95%CI=66.6-70.4), a sensitivity of 56.7% (95%CI=53.4-59.9), and a specificity of 70.5% (95%CI=68.0-73.0). This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index F. Sharifi et al. 4 Table 1: The association between clinical characteristics of COVID-19 cases with need for ICU admission based on univariate analysis Variables Need for ICU care OR 95%CI P No Yes Age (year) Mean ± SD 50.39±20.47 56.20±23.23 1.014 1.011-1.017 <0.001 Gender Female 2935 (86.6) 454 (13.4) 1 Male 3191 (84.1) 602 (15.9) 1.220 1.069-1.391 0.003 Underlying disease Heart diseases 544 (78.4) 150 (21.6) 1.699 1.399-2.062 <0.001 Chronic kidney disease 197 (75.8) 63 (24.2) 1.909 1.427-2.556 <0.001 Chronic pulmonary disease 472 (77.6) 136 (22.4) 1.771 1.445-2.169 <0.001 Chronic liver disease 67 (81.7) 15 (18.3) 1.303 0.742-2.290 0.357 Diabetes 799 (81.9) 177 (18.1) 1.343 1.124-1.604 0.001 Cancer 112 (75.7) 36 (24.3) 1.895 1.294-2.775 <0.001 Hypertension 970 (80.1) 241 (19.9) 1.572 1.340-1.843 <0.001 Blood diseases 33 (73.3) 12 (26.7) 2.122 1.093-4.122 0.02 Immunodeficiency 22 (75.9) 7 (24.1) 1.851 0.789-4.345 0.157 Chronic neurological diseases 139 (84.8) 25 (15.2) 1.044 0.679-1.607 0.843 Fever 2980 (87.0) 446 (13.0) 0.772 0.676-0.881 <0.001 Myalgia 2428 (87.3) 354 (12.7) 0.768 0.669-0.881 <0.001 Cough 3123 (86.7) 480 (13.3) 0.801 0.703-0.914 0.001 Chest pain 518 (87.1) 77 (12.9) 0.852 0.664-1.092 0.205 Diarrhea 635 (89.4) 75 (10.6) 0.661 0.516-0.848 0.001 Respiratory distress 4117 (83.1) 835 (16.9) 1.844 1.575-2.158 <0.001 Vomiting and nausea 1269 (88.6) 164 (11.4) 0.704 0.589-0.840 <0.001 Headache 1545 (89.3) 185 (10.7) 0.630 0.532-0.764 <0.001 Loss of consciousness 91 (65.0) 49 (35) 3.227 2.226-4.596 <0.001 Smell or taste disorders 568 (89.7) 65 (10.3) 0.642 0.492-0.837 0.001 Seizures 24 (66.7) 12 (33.3) 2.92 1.457-5.862 0.003 Anorexia 2277 (85.8) 376 (14.2) 0.935 0.816-1.017 0.331 Dizziness 857 (83.7) 167 (16.3) 1.155 0.964-1.383 0.118 Opium abuse 739 (82.3) 159 (17.7) 1.292 1.074-1.555 0.007 Smoking 352 (85.2) 61 (14.8) 1.006 0.760-1.331 0.969 Oxygen therapy 4896 (86.1) 788 (13.9) 0.739 0.634-0.860 <0.001 Oxygen saturation (%) Mean ± SD 90.88±5.4 88.4±7.8 0.945 0.936-0.954 <0.001 Symptoms to admission (days) Mean ± SD 5.80±3.49 5.679±4.14 0.990 0.972-1.009 0.296 Data are presented as mean ± standard deviation (SD) or frequency (%). OR: Odds ratio; CI: confidence interval. 3.4. Support vector machine A vector machine model was constructed after cross- validation using the radial basis kernel function and param- eters "sigma=0.0279, cost=2". In order to separate the classes and reduce data error, support vector machine used non- linear variable separation as well as the best margin. As a result of using the vector machine model (Figure 2), the vari- able importance graph indicated that oxygen saturation level had the greatest impact (100%) on the classification of hospi- talized patients in ICUs and general wards followed by age (43.62%), respiratory distress (37.54%), and the decreased level of consciousness (29.72%). Heart diseases, chronic pul- monary disease, and hypertension were found to have a sig- nificant effect on patients with about 18 to 20 percent, while myalgia, fever, vomiting and nausea, oxygen therapy, and chronic kidney disease were found to have a significant im- pact between 9 and 12 percent. In general, the importance of gender, seizures, diabetes, di- arrhea, smell or taste disorders, and cancer variables ranged from 5 to 7 percent, while the importance of other vari- ables was ranked to be lower. There was 99.8% (95%CI=99.6- 100.0) accuracy, 99.6% (95%CI=99.3-100.0) sensitivity, and 99.9% (95%CI=99.7-100.0) specificity for the support vector machine model. 4. Discussion This study aimed to determine the clinical risk factors asso- ciated with COVID-19 patients who require admission to the This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index 5 Archives of Academic Emergency Medicine. 2023; 11(1): e15 Table 2: The association between clinical characteristics of COVID-19 cases with need for ICU admission based on multiple regression anal- ysis Variables OR 95% CI P Age (years) 1.007 1.004-1.011 <0.001 Gender Male 1.179 1.028-1.352 0.018 Underlying disease Heart diseases 1.250 1.009-1.548 0.042 Chronic pulmonary Disease 1.250 1.006-1.554 0.044 Chronic kidney disease 1.515 1.111-2.066 0.009 Cancer 1.682 1.130-2.505 0.010 Hypertension 1.249 1.042-1.496 0.016 Myalgia 0.847 0.733-0.978 0.024 Cough 0.827 0.722-0.949 0.007 Diarrhea 0.727 0.561-0.941 0.015 Respiratory distress 1.658 1.410-1.951 <0.001 Headache 0.708 0.593-0.844 <0.001 Loss of consciousness 2.487 1.721-3.596 <0.001 Smell or taste disorders 0.622 0.474-0.816 0.001 Seizures 3.428 1.615-7.274 0.001 Oxygen saturation (%) 0.954 0.944-0.964 <0.001 Data are presented as mean ± standard deviation (SD) or frequency (%). OR: Odds ratio; CI: confidence interval. ICU. By employing results of statistical methods, this cross- sectional study identified some clinical risk factors associ- ated with ICU admission (33). One of the study’s strengths was the use of univariate and multiple logistic regression, classification tree, and support vector machine to develop a list of clinical risk factors that could be used to distinguish ICU patients from general ward patients. According to the results of this study, amongst the demographic characteris- tics affecting hospitalization in the ICU, gender is an impor- tant factor that is relevant and influential, and it was shown that the majority of patients admitted in the ICU are men. Age is one of the most significant clinical risk factors asso- ciated with ICU hospitalization. As the age of patients in- creases, their chances of being hospitalized in the ICU and experiencing a worsening of their disease’s condition also in- crease. Even with acceptable oxygen saturation levels, new- borns (less than 0.196 years old (72 days)) and the elderly (over 66.5 years old) require ICU care. A meta-analysis of 59 studies involving 36,470 patients concluded that men and patients aged over 70 years old are more likely to be admitted to the ICU compared to other patients (34). This study was consistent with the current study. Since the results of the current study identified underlying diseases such as hypertension, chronic pulmonary disease, heart diseases, chronic kidney disease, and cancer as clin- ical risk factors, it can be said that patients with these un- derlying diseases experience a more severe form of COVID- 19 and require special care. The most important underly- ing disease was hypertension. A meta-analysis of 30 original articles found that high blood pressure is significantly asso- ciated with increased mortality and need for intensive care, which is consistent with the results of this research (35). In some cases, certain signs and symptoms are strongly related to the possibility of admission to the ICU. These factors may worsen a patient’s condition and also increase their need for intensive care. A low oxygen saturation level is considered to be the most significant risk factor in this study, and patients with low oxy- gen saturation levels must be admitted to ICU. Additionally, patients with oxygen saturation below 86.5% require imme- diate admission to the ICU. In a previous study conducted on 641 patients at Stony Brook University Hospital, it was found that patients with an oxygen saturation level of less than 92% are at high risk, and ICU hospitalization was rec- ommended due to the danger of low oxygen saturation lev- els and these findings are consistent with the findings of the present study (5). Seizures, respiratory distress, and de- creased level of consciousness are factors affecting the sever- ity of the disease and the need for hospitalization in an ICU. Headache, myalgia, cough, diarrhea, and smell or taste dis- orders do not increase the severity of the disease, and most patients with these symptoms are admitted to the hospital’s general ward. The effects of diabetes, oxygen therapy, fever, vomiting, and nausea are not significant in multiple logistic regression model; whereas, they are considered to be impor- tant in the vector machine model. The reason for this dif- ference may be related to the fact that vector machines con- sider non-linear relationships between variables, which can be used as a predictor of hospitalization ICUs (36). There is a higher chance of admission to the ICU if the patient This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index F. Sharifi et al. 6 Figure 1: Factors affecting patients’ intensive care unit (ICU) admission according to the classification tree model. Misclassification costs of the patients admitted to the general ward instead of the ICU were five times more than the patients admitted to the ICU instead of the general ward. So, in each node, the tree model predicts that the patient will be admitted to the ICU if the proportion of admissions to the ICU exceeds 16.6%. has an underlying disease like diabetes. A meta-analysis of 33 studies involving 16,003 patients confirmed that diabetes is significantly associated with COVID-19-related mortality and disease severity, which is in agreement with our finding (37). Symptoms such as fever, vomiting, and nausea do not worsen the disease, and patients with these symptoms are mostly ad- This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index 7 Archives of Academic Emergency Medicine. 2023; 11(1): e15 Figure 2: The importance graph of the effective intensive care unit (ICU) admission variables as calculated by the vector machine model. mitted to the general ward. Patients with COVID-19 who re- ceive oxygen therapy are less likely to require admission to an ICU. Physicians can utilize this research results to be provided with a simple and accurate stratification tool that will en- able them to manage patients with COVID-19 and similar diseases in a timely manner (1). In general, the combined re- sults of all the models revealed that a number of factors such as decreasing oxygen saturation level (the risk is higher for patients whose oxygen saturation level is less than 86.5%), aging (66.5 years and older), age under 72 days in infants, respiratory distress, decreased level of consciousness, hyper- tension, chronic pulmonary disease, heart diseases, chronic kidney disease, cancer, diabetes, and seizures are associated with severe type of illness and ICU admission in COVID-19 patients. Additionally, these factors may guide physicians to make a better decision. For instance, if a patient aged 66.5 years or older is recommended to seek medical advice at the early stage of their illness, and if hospitalization is re- quired, physicians should be aware of the high risk of se- vere disease in these groups (34). A combination of the re- sults of the three models and the K-fold cross-validation in- creased the external validity of the study, and the comparison of model accuracy revealed that logistic model accuracy was This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index F. Sharifi et al. 8 62.7 percent, classification tree model accuracy was 64.3%, and support vector machine model accuracy was 99%. Sup- port vector machine model was the most accurate one. For clinical researchers, predicting COVID-19 patients’ hospital- ization in the ICU is essential. Accordingly, it is better to use the support vector machine model, which is more accurate. Researchers should use classification tree if they need a sim- ple algorithm for classifying the hospitalization of COVID- 19 patients. In order to examine the association among the variables and severity of COVID-19 disease and hospitaliza- tion in the ICU, it is recommended to use logistic regression model(33). 5. Limitations The type of treatment and different strains of COVID-19 have changed over time and these may affect receiving or not re- ceiving intensive medical care. These two variables were not addressed in this study. Thus, they may cause bias and re- duce the generalizability of the findings to the community. Accordingly, they need to be addressed in future studies. It is possible that additional confounding factors exist that we have not included in the models. The criteria for admission of patients in ICU of different hospitals are different, which it may affect the generalizability of the result. Additionally, this study only focused on some clinical variables and labo- ratory data of hospitalized patients were unavailable, which it is recommended to consider them for future research. 6. Conclusions The results of our study showed that clinical risk factors such as hypertension, chronic pulmonary disease, heart dis- ease, chronic kidney disease, cancer, diabetes, seizures, and gender were predictors of patients’ admission to the ICU, but oxygen saturation level, increasing age, respiratory dis- tress, and reduced level of consciousness were identified as independent predictors of need for ICU admission among COVID-19 patients. The results of this study can help physi- cians and hospital staff to assign timely special care services to patients with COVID-19. 7. Declarations 7.1. Acknowledgments We would like to express our gratitude to the administration and all personnel of Afzalipour Hospital and Kerman Univer- sity of Medical Sciences for their assistance and cooperation in this undertaking. 7.2. Conflict of interest No potential conflict of interest relevant to this article was re- ported. 7.3. Fundings and supports The study did not receive any funding from any organization. 7.4. Authors’ contribution Study concept and design: YJ. Analysis and interpretation of data: YJ, FS and MA. Data collection: MHM. Writing the orig- inal draft of the manuscript: FS and YJ. All authors reviewed, edited, and approved the final article. References 1. Shayganfar A, Sami R, Sadeghi S, Dehghan M, Khademi N, Rikhtehgaran R, et al. Risk factors associated with in- tensive care unit (ICU) admission and in-hospital death among adults hospitalized with COVID-19: a two-center retrospective observational study in tertiary care hospi- tals. Emerg Radiol. 2021;28(4):691-7. 2. Vanhems P, Gustin MP, Elias C, Henaff L, Dananché C, Grisi B, et al. Factors associated with admission to in- tensive care units in COVID-19 patients in Lyon-France. PLoS One. 2021;16(1):e0243709. 3. Kim L, Garg S, O’Halloran A, Whitaker M, Pham H, Anderson EJ, et al. Risk Factors for Intensive Care Unit Admission and In-hospital Mortality Among Hos- pitalized Adults Identified through the US Coron- avirus Disease 2019 (COVID-19)-Associated Hospitaliza- tion Surveillance Network (COVID-NET). Clin Infect Dis. 2021;72(9):e206-e14. 4. Cairns CB, Niemann JT. Intravenous adenosine in the emergency department management of paroxys- mal supraventricular tachycardia. Ann Emerg Med. 1991;20(7):717-21. 5. Zhao Z, Chen A, Hou W, Graham JM, Li H, Rich- man PS, et al. Prediction model and risk scores of ICU admission and mortality in COVID-19. PLoS One. 2020;15(7):e0236618. 6. Zeng Z, Ma Y, Zeng H, Huang P, Liu W, Jiang M, et al. Simple nomogram based on initial laboratory data for predicting the probability of ICU transfer of COVID-19 patients: Multicenter retrospective study. J Med Virol. 2021;93(1):434-40. 7. Sadeghi A, Eslami P, Dooghaie Moghadam A, Pirsalehi A, Shojaee S, Vahidi M, et al. COVID-19 and ICU admission associated predictive factors in Iranian patients. Caspian J Intern Med. 2020;11(Suppl 1):512-9. 8. Huang Y, Lyu X, Li D, Wang L, Wang Y, Zou W, et al. A cohort study of 676 patients indicates D-dimer is a crit- ical risk factor for the mortality of COVID-19. PLoS One. 2020;15(11):e0242045. 9. Tadic M, Cuspidi C, Sala C. COVID-19 and diabetes: Is there enough evidence? J Clin Hypertens (Greenwich). 2020;22(6):943-8. This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index 9 Archives of Academic Emergency Medicine. 2023; 11(1): e15 10. Frydman GH, Boyer EW, Nazarian RM, Van Cott EM, Piazza G. Coagulation Status and Venous Throm- boembolism Risk in African Americans: A Potential Risk Factor in COVID-19. Clin Appl Thromb Hemost. 2020;26:1076029620943671. 11. Halacli B, Kaya A, Topeli A. Critically-ill COVID-19 pa- tient. Turk J Med Sci. 2020;50(Si-1):585-91. 12. Berger E, Winkelmann J, Eckhardt H, Nimptsch U, Panteli D, Reichebner C, et al. A country-level analy- sis comparing hospital capacity and utilisation during the first COVID-19 wave across Europe. Health Policy. 2022;126(5):373-81. 13. Dongelmans DA, Termorshuizen F, Brinkman S, Bakhshi- Raiez F, Arbous MS, de Lange DW, et al. Characteristics and outcome of COVID-19 patients admitted to the ICU: a nationwide cohort study on the comparison between the first and the consecutive upsurges of the second wave of the COVID-19 pandemic in the Netherlands. Ann In- tensive Care. 2022;12(1):5. 14. Ahmadi Gohari M, Chegeni M, Haghdoost AA, Mirzaee F, White L, Kostoulas P, et al. Excess deaths during the COVID-19 pandemic in Iran. Infect Dis (Lond). 2022;54(12):909-917. 15. Population Stat: World Statistical Data; 2019 [updated 2019-09-05. Available from: https://populationstat.com/iran/kerman. 16. Daily reported of COVID-19: Kerman health sector news; 2022 [Available from: https://zil.ink/kmu. 17. Halpern NA, Goldman DA, Tan KS, Pastores SM. Trends in Critical Care Beds and Use Among Population Groups and Medicare and Medicaid Beneficiaries in the United States: 2000-2010. Crit Care Med. 2016;44(8):1490-9. 18. Moghadas SM, Shoukat A, Fitzpatrick MC, Wells CR, Sah P, Pandey A, et al. Projecting hospital utilization during the COVID-19 outbreaks in the United States. Proc Natl Acad Sci USA . 2020;117(16):9122-6. 19. Gottlieb M, Sansom S, Frankenberger C, Ward E, Hota B. Clinical Course and Factors Associated With Hospital- ization and Critical Illness Among COVID-19 Patients in Chicago, Illinois. Acad Emerg Med. 2020;27(10):963-73. 20. Lashkari M, Yazdi-Feyzabadi V, Mohammadi M, Saberi H, Mehrolhassani MH. Designing a Financial Resource Al- location Model Using Goal Programming Approach: A Case Study of a hospital in Iran. Evid Based Health pol- icy Maneg and Econ. 2018;2(2):70-9. 21. Ko JY, Danielson ML, Town M, Derado G, Greenlund KJ, Kirley PD, et al. Risk Factors for Coronavirus Disease 2019 (COVID-19)-Associated Hospitalization: COVID- 19-Associated Hospitalization Surveillance Network and Behavioral Risk Factor Surveillance System. Clin Infect Dis. 2021;72(11):e695-e703. 22. Kleinbaum DG, Dietz K, Gail M, Klein M, Klein M. Logis- tic regression. New York: Springer; 2002. 23. Jewell NP. Statistics for epidemiology. New York: chap- man and hall/CRC; 2003. 24. Rostami M, Garrusi B, Baneshi MR. A study on the use of bootstrap aggregation methods in estimation of stable parameters. J Biostat Epidemiol. 2016;2(2):104-10. 25. Alkhasawneh MS, Ngah UK, Tay LT, Mat Isa NA, Al-Batah MS. Modeling and testing landslide hazard using deci- sion tree. J Appl Math. 2014;2014:929768. 26. Lu H, Xu Y, Ye M, Yan K, Gao Z, Jin Q. Learning misclas- sification costs for imbalanced classification on gene ex- pression data. J BMC bioinformatics. 2019;20(25):681. 27. Mehrnaz HS, Ali MK. Calculating the best cut off point using logistic regression and neural network on credit scoring problem-A case study of a commercial bank. AFR J BUS MANAG . 2013;7(16):1414-21. 28. James G, Witten D, Hastie T, Tibshirani R. An introduc- tion to statistical learning. New York: Springer; 2013. 29. Edureka: Support Vector Machine In R: Using SVM To Predict Heart Diseases 2020 [updated May 15,2020. Available from: https://www.edureka.co/blog/support- vector-machine-in-r/. 30. Chen R-C, Dewi C, Huang S-W, Caraka RE. Selecting crit- ical features for data classification based on machine learning methods. J Big Data. 2020;7(1):52. 31. Anguita D, Ridella S, Rivieccio F, editors. K-fold general- ization capability assessment for support vector classi- fiers. Proceedings. 2005 IEEE International Joint Confer- ence on Neural Networks, 2005.; 2005: IEEE. 32. Long WJ, Griffith JL, Selker HP, D’agostino RB. A compar- ison of logistic regression to decision-tree induction in a medical domain. Comput Biomed Res. 1993;26(1):74-97. 33. Subudhi S, Verma A, Patel AB, Hardin CC, Khandekar MJ, Lee H, et al. Comparing machine learning algorithms for predicting ICU admission and mortality in COVID-19. NPJ Digit Med. 2021;4(1):87. 34. Pijls BG, Jolani S, Atherley A, Derckx RT, Dijkstra JI, Franssen GH, et al. Demographic risk factors for COVID- 19 infection, severity, ICU admission and death: a meta- analysis of 59 studies. J BMJ open. 2021;11(1):e044640. 35. Pranata R, Lim MA, Huang I, Raharjo SB, Lukito AA. Hypertension is associated with increased mortal- ity and severity of disease in COVID-19 pneumo- nia: A systematic review, meta-analysis and meta- regression. J Renin Angiotensin Aldosterone Syst. 2020;21(2):1470320320926899. 36. Jakkula V. Tutorial on support vector machine (svm). J School of EECS, Washington State University. 2006;37(2.5):3. 37. Kumar A, Arora A, Sharma P, Anikhindi SA, Bansal N, Singla V, et al. Is diabetes mellitus associated with mortal- ity and severity of COVID-19? A meta-analysis. Diabetes This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index F. Sharifi et al. 10 Metab Syndr . 2020;14(4):535-45. This open-access article distributed under the terms of the Creative Commons Attribution NonCommercial 3.0 License (CC BY-NC 3.0). Downloaded from: https://journals.sbmu.ac.ir/aaem/index.php/AAEM/index Introduction Methods Results Discussion Limitations Conclusions Declarations References