Novel Scoring Systems to Predict the Need for Oxygenation and ICU Care, and Mortality in Hospitalized COVID-19 Patients: A Risk Stratification Tool

Introduction: A rapid surge in cases during the COVID-19 pandemic can overwhelm any healthcare system. It is imperative to triage patients who would require oxygen and ICU care, and predict mortality. Specific parameters at admission may help in identifying them. Methodology: A prospective observational study was undertaken in a COVID-19 ward of a tertiary care center. All baseline clinical and laboratory data were captured. Patients were followed till death or discharge. Univariable and multivariable logistic regression was used to find predictors of the need for oxygen, need for ICU care, and mortality. Objective scoring systems were developed for the same using the predictors. Results: The study included 209 patients. Disease severity was mild, moderate, and severe in 98 (46.9%), 74 (35.4%), and 37 (17.7%) patients, respectively. The neutrophil-to-lymphocyte ratio (NLR) >4 was a common independent predictor of the need for oxygen (p<0.001), need for ICU transfer (p=0.04), and mortality (p=0.06). Clinical risk scores were developed (10*c-reactive protein (CRP) + 14.8*NLR + 12*urea), (10*aspartate transaminase (AST) + 15.7*NLR + 14.28*CRP), (10*NLR + 10.1*creatinine) which, if ≥14.8, ≥25.7, ≥10.1 predicted need for oxygenation, need for ICU transfer and mortality with a sensitivity and specificity (81.6%, 70%), (73.3%, 75.7%), (61.1%, 75%), respectively. Conclusion: The NLR, CRP, urea, creatinine, and AST are independent predictors in identifying patients with poor outcomes. An objective scoring system can be used at the bedside for appropriate triaging of patients and utilization of resources.


Introduction
The coronavirus disease (COVID-19) pandemic has spread globally causing millions of deaths in multiple waves. India recently witnessed a surge of cases in the third wave of COVID-19 dominated by the Omicron variant [1]. The rapidity with which the cases rise during a pandemic challenges the healthcare system to optimize the available infrastructure, manpower, and resources. The spectrum of COVID-19 disease ranges from asymptomatic cases to severe diseases. Although the proportion of patients with severe disease is much lower, the numbers during a pandemic are sufficiently large enough to overwhelm the existing healthcare systems across the world [2].
Adequate allocation of resources requires appropriate triaging of patients and timely identification of patients who are at risk of developing severe disease and hospitalization. There have been multiple studies that have looked at various clinical and laboratory parameters to determine the same. Comorbidities such as diabetes mellitus (DM), hypertension (HTN), and chronic kidney disease (CKD) as well as laboratory parameters such as lymphopenia, elevated inflammatory markers like c-reactive protein (CRP), and D-dimer are observed in severe disease and supposedly predict early mortality. Multiple such predictors are still under investigation. These predictors of progression to severe disease and mortality can also aid in the selection of patients who require early intensive care unit (ICU) care [3][4][5][6].
There is a marked difference in the nature and intensity of management of critically ill patients and noncritical patients, so improvisations must be made in monitoring such patients outside the ICU [7]. The improvised monitoring should be able to detect any rapid deteriorations in medical conditions that would require escalation of life-support or ICU care. The need for oxygenation, ICU, and the risk of mortality are the three key clinical questions faced by policymakers and clinicians while managing hospitalized COVID-19 patients. The present study was conducted to develop a clinical scoring system to answer these questions using the baseline clinical and biochemical predictors. We expect that the scoring system when used in conjunction with the standard of care, would be able to predict early deterioration and critical events in sick patients and guide policymakers to formulate guidelines to optimize the available resources. The study adds to the evolving scientific literature on predictors of severity and mortality among COVID-19 patients with the help of a novel scoring system.

Materials And Methods
This was a prospective observational study conducted in a dedicated COVID-19 ward of a tertiary care hospital in north India from June 2020 to December 2020. The study population included all admitted patients with confirmed severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection (based on clinical features and positive viral nucleic acid test results on nasal/throat swab samples). Patients were classified as having mild (upper respiratory tract symptoms and/or fever without shortness of breath or hypoxia), moderate (respiratory rate ≥ 24/min, breathlessness or saturation of peripheral oxygen (SpO2) between 90% to 94%) or severe disease (respiratory rate >30/min, breathlessness or SpO2 < 90%) as per institutional guidelines. A sample size of convenience was chosen with as many possible patients being included without any exclusion criteria.
The data of patients were documented in a predesigned proforma. The proforma consisted of three sections. The first one included age and sex. The second section included clinical and co-morbidity data. The clinical data were used to classify the severity of the disease along with mortality data and duration of hospital stay. The third section included laboratory data. The basic laboratory investigations including inflammatory markers were done as per the standard of care decided by the treating team; no additional investigations/interventions were done for the study. All the subjects were followed up till death or discharge from the hospital. The study was undertaken after approval (approval no.IEC-293) by the institute ethics committee of the All India Institute of Medical Sciences (AIIMS), New Delhi.
Statistical analysis was done using Stata 12 software (Stata Corp 2011, College Station, TX, USA). Categorical variables are presented as frequency and percentages. Continuous data are represented as mean/standard deviation or median/interquartile range (IQR). For the purpose of analysis, patients in mild clinical severity were classified as not requiring oxygen and those with moderate and severe disease as requiring oxygen. Patients with severe disease were classified as requiring ICU care. Univariate analysis was done to look for any association between the risk factors and the need for oxygen, ICU care, and mortality. Association was seen using the chi-square/Fischer test for categorical variables. Non-parametric tests including the Mann-Whitney u-test were used to assess the association for continuous variables. Clinical and laboratory correlates with p-value <0.1 in the univariable analysis were included in the multivariable logistic regression analysis. Variables with a p-value <0.1 were considered independent predictors. A predictive score for the need for oxygenation, ICU care, and mortality was derived using coefficients of variables from the regression model. Receiver operating characteristics (ROC) curve analysis was used to develop the cut-off value of the predictive score. Characteristics and performance of score cut-off were reported using sensitivity, and specificity. A p-value <0.05 was considered statistically significant.   Three variables found significant in the multivariable regression model were used to calculate a predictive score: 10*CRP + 14.8*NLR + 12*Urea (estimated regression coefficient of variable = logistic regression coefficient of variable/logistic regression coefficient of CRP*10) ( Table 3). The CRP and urea were assigned a value of "1" if > upper limit of normal (ULN) and "0" if ≤ ULN. The NLR was assigned a value of "1" if >4 and "0" if ≤ 4. Based on ROC curve analysis, the optimal score cut-off for prediction of need for oxygenation in hospitalized COVID-19 patients was found to be ≥14.8 with a sensitivity and specificity of 81.6% and 70%, respectively. The area under the ROC curve was 0.83 (95% confidence interval (CI), 0.75-0.91) ( Figure  1).  Three variables found significant in the multivariable regression model were used to calculate a predictive score: 10*AST+ 15.7*NLR + 14.28*CRP (estimated regression coefficient of variable = logistic regression coefficient of variable/logistic regression coefficient of AST*10). The CRP and AST were assigned a value of "1" if > ULN and "0" if ≤ ULN. The NLR was assigned a value of "1" if >4 and "0" if ≤ 4. Based on the ROC curve analysis, the optimal score cut-off for prediction of need for ICU in hospitalized COVID-19 patients was found to be ≥25.7 with a sensitivity and specificity of 73.3% and 75.7%, respectively. The area under the ROC curve was 0.87 (95% CI, 0.78-0.96) (Figure 2) FIGURE 2: Receiver operator curve (ROC) for scoring system to determine the need for ICU care
Two variables found significant in the multivariable regression model were used to calculate a predictive score: 10*NLR + 10.1*creatinine (estimated regression coefficient of variable = logistic regression coefficient of variable/logistic regression coefficient of NLR*10). Creatinine was assigned a value of "1" if > ULN and "0" if ≤ ULN. The NLR was assigned a value of "1" if >4 and "0" if ≤ 4. Based on ROC curve analysis, the optimal score cut-off for prediction of mortality in hospitalized COVID-19 patients was found to be ≥10.1 with a sensitivity and specificity of 61.1% and 75%, respectively. The area under the ROC curve was 0.73 (95% CI, 0.61-0.83) (Figure 3).

FIGURE 3: Receiver operator curve (ROC) for scoring system to predict mortality Discussion
A rapid surge in COVID-19 cases disrupted routine healthcare services. Policymakers and clinicians were challenged by oxygen shortages and the lack of ICU beds. A large number of sick patients were managed in the ward setting with limited resources. In such conditions, it becomes imperative to optimally utilize the resources for the patients who could get sick during their hospital stay thus requiring intensive monitoring. Through this study, we have identified predictors and developed a scoring system that can be used by clinicians as a bedside tool to objectively assess individual patients and their need for oxygenation, need for ICU transfer, and the risk of mortality.
The current study involved patients admitted to a COVID-19 ward of an apex COVID care center. Mortality was 10% and almost all of these deaths happened in patients with severe disease, which can be ascribed to delays in presenting to the healthcare system as severe cases constituted up to one-fifth of our patients due to a rapid surge in the number of cases. A study in non-ICU patients in the Cleveland Clinic reported a mortality of 6% with 14% requiring ICU transfer and 14% experiencing an increase in oxygen requirement which was similar to our experience [8]. Mortality as high as 62% has been reported in critically ill patients and up to 81% in those requiring mechanical ventilation [9].
Diabetes mellitus was the most common comorbidity among our patients. All of the comorbidities described as predictors in the available literature were also associated with poor outcomes in our patients on univariable analysis, however, none were independent predictors. A meta-analysis has confirmed that older age, HTN, and DM conferred a significantly increased risk of mortality among patients with COVID-19 [3].
In our analysis, biochemical parameters were the strongest predictors of poor outcomes. A recent metaanalysis also revealed that abnormalities in the biomarkers had higher odds of severe disease and greater risk of mortality [10,11]. Among the biochemical predictors in our study, NLR > 4 was consistently an independent predictor of the need for oxygenation, need for ICU transfer, and mortality. The cut-off of >4 was similar to that seen in multiple other studies [12][13][14][15]. A meta-analysis of 13 studies also confirmed the fact that NLR is a good biomarker for predicting disease severity in patients with COVID-19 (area under the curve (AUC)=0.85, sensitivity=0.78, and specificity=0.78), and in predicting mortality as well (AUC=0.90, sensitivity=0.83, and specificity=0.83) [16]. Other parameters including CRP, AST, urea, and creatinine were identified as independent predictors in our study. A meta-analysis of 5350 patients showed that elevated CRP was associated with an increased composite poor outcome (respiratory rate (RR)=1.84) and in the severe COVID-19 (RR=1.41) subgroup. A CRP ⩾10 mg/L has a 51% sensitivity, 88% specificity, and an AUC of 0.84 [17].
The objective scoring system that we developed consistently involved NLR as a component. Among the three scores that we calculated, the equation to predict the need for ICU transfer that included NLR, CRP and AST had the highest discriminatory power with AUC=0.87, followed by the need for oxygenation (AUC=0.83). The equation for mortality had a poorer discriminatory value (AUC=0.73). Multiple studies conducted to determine a scoring system to predict the need for oxygen and mortality have also consistently included NLR and other biomarkers that we have explored [18][19][20][21][22][23].
Our scoring system is in tune with the scoring systems described in the literature. The unique feature of our study is that we have come up with a composite score by taking all the predictors into account. The scoring system can be utilized to risk-stratify patients even in the current and upcoming pandemics. A similar study done at our institute generated a symptom-based clinical scoring system for diagnosis of COVID-19, however, our study looked into poor prognostic markers in hospitalized COVID-19 patients [24].
Our study had a few limitations. First, most of our variables had a non-normal distribution forcing the use of non-parametric tests to assess the association. Second, the exploration of predictors depends on the number of potential predictors studied. Although we have covered most of the variables we may have missed other clinical predictors which may be further explored. Third, being a ward-based study, the data lack comparison to those patients who get admitted directly to ICU. Fourth, triage thresholds for hospitalization vary from center to center thus limiting the generalizability of our study. And fifth, we did not validate our scoring system, hence further studies may be required for internal and external validation.

Conclusions
The present study indicates that the presence of NLR > 4, CRP, urea, creatinine, and AST are independent predictors for escalation of care. Clinical risk scores of ≥14.8, ≥25.7, and ≥10.1, predict the need for oxygenation (sesnsitivity=81.6%, specificity=70%), need for ICU transfer (sensitivity=73.3%, specificity=75.7%), and mortality (sensitivity=61.1%, specificity=75%). An objective scoring system that consistently involves NLR can be used by clinicians at the bedside for predicting adverse outcomes. Intensive monitoring and early triaging of such patients can reduce adverse outcomes.

Additional Information Disclosures
Human subjects: Consent was obtained or waived by all participants in this study. Institute Ethics Committee, All India Institute of Medical Sciences, New Delhi issued approval IEC-293. This study was approved by the Institute Ethics Committee, All India Institute of Medical Sciences, New Delhi. Animal subjects: All authors have confirmed that this study did not involve animal subjects or tissue. Conflicts of interest: In compliance with the ICMJE uniform disclosure form, all authors declare the following: Payment/services info: AIIMS Intramural Project Grant (A-COVID-7). Financial relationships: All authors have declared that they have no financial relationships at present or within the previous three years with any organizations that might have an interest in the submitted work. Other relationships: All authors have declared that there are no other relationships or activities that could appear to have influenced the submitted work.