Validation of CORONA VIRUS Emergency Triage Tool (CorVETT) among adults in the emergency department of a low resource setting

Background: In this study we aim to validate corona virus emergency triage tool (CorVETT) for accurate triaging and disposition in adults presenting to the emergency department with suspected corona virus symptoms. Observational prospective cohort study conducted in the emergency department. Methodological framework of Transparent Reporting of multivariable prediction model for individual Prognosis and Diagnosis (TRIPOD) type I was adopted. Algorithm tested consists of four sequential stages; presenting features, vitals, associated features and high-risk conditions. Cumulative score of four stages categorizes patient as COVID or non-COVID and was assigned non isolation or isolation beds. Prediction model for presence of relevance (event) was obtained by dividing data into two parts; training (n = 450) and validation (n = 115). Probability of event was estimated using linear logistic regression with training data. Predictive capacity of model was assessed using a receiver operative curve (ROC) curve through set of validation data. The discriminative capacity was evaluated using area under curve (AUC), estimated by a 95% con�dence interval. P-value less than 0.05 was considered statistically signi�cant. The statistical analysis was performed on "R" (version 3.4.1, 2017) and Statistical Package for Social Science (SPSS, version 21.0, 2016).

Further, ED role in this pandemic has been to evaluate risk of adverse outcomes and need for critical interventions, appropriate disposition to isolation and non-isolation beds and using data to make accurate decisions around admission to hospital or outpatient management (3). Additionally, problem is considerable in dealing with suspected or unsuspected cases where mandatory testing may place additional burden on both patient and family as healthcare is borne by the family in our country (4). Strati cation of critical vs non-critical COVID 19 positive and negative 19 patients has been a challenge as inaccurate triaging may pose risk to patients who don't have the virus in contracting it (5). COVID 19 triage tools used in ED are based either on a score, which assign points to predictors to stipulate risk of adverse outcome, or a rule, which utilizes risk predictors to come to a decision of discharge or hospital admission (5)(6)(7). Disease characteristics and severity is markedly different among children and adults in terms of adverse outcomes and in-hospital admission(8). The increasing burden of adult patients with COVID symptoms and due to severity of illness, we focused on an adult COVID emergency triage tool in this study.
Two currently available tools are: the World Health Organization (WHO) COVID decision making algorithm that recommends admission for severe pneumonia, and the National Institute of Health Care and Excellence (NICE) suggests National Early Warning Score 2 (NEWS2) for predicting severity of disease with good sensitivity but poor speci city (5). Tools that were developed during the in uenza pandemic can be adapted for the ED triage protocol but they have not been validated in our setting and they do not share their approach on accurate disposition to isolation beds (5). The Pandemic Modi ed Early Warning Score (PMEWS) utilizes physiological variables but is not based on screening of symptoms, risk factors and history. The Pandemic Respiratory Infection Emergency System Triage (PRIEST) tool assess severity of COVID at triage but not catering risk strati cation which is a major issue in low resource settings with limited hospital isolation beds (5). In this paper we describe Corona Virus Emergency Triage Tool (CorVETT), an algorithm-based approach with primary objective to correctly allocate suspected COVID 19 patients in timely manner within busy emergency departments to isolation areas to prevent cross contamination.
COVID 19 being a relatively new disease with high fatality and transmissibility, needs a tool to help predict disease early, so that patients can be isolated quickly to prevent its transmission and then have treatment initiated. The primary objective of this study was therefore to validate CorVETT in adult patients presenting with suspected COVID 19 characteristics and positive COVID 19 nasopharyngeal PCR.
Our secondary objective was to predict CorVETT in accurate disposition to Respiratory and Nonrespiratory COVID 19 unit from triage.

Results:
A total of 595 participants with suspected COVID-19, admitted from the ED were included in the study. Of those, 349 (58.7%) were male and 246 (41.3%) females. The mean age was 55.35 ± 17.08 (range 19-96 yrs.), with majority 427 (71.7%) between 36-75 years. Average prediction score of CorVETT was 4.57 ± 3.12. The mean triage vitals were systolic blood pressure, 129.8 ± 21.43 mmHg, respiratory rate 24.94 ± 8.36 breaths per minute, oxygen saturation 89.02 ± 10.33% and heart rate 89.61 ± 21.74 beats per minute respectively. Amongst those enrolled, 203 (34%) were of P1 category (this P1 category is given to patients at the time of triaging who require immediate attention due to life or limb threatening ndings, and this is as per the Emergency Severity index-[ESI] triage tool used in triaging patients in the emergency department). Fever and cough were the commonest major symptoms identi ed. In terms of risk strati cation with one, two or more components along with associated features, 117(70%) patients had respiratory distress and 208(62%) had diabetes in their high-risk pro le. Nasopharyngeal PCR for COVID 19 was positive in 381 and negative in 214. Of those who were COVID 19 positive, 183(30%) were discharged with quarantine instructions and majority 412(69%) were admitted either in ED Respiratory Covid Unit (RCU) or in hospital isolation beds (see Table No. 1). There was an association between CorVETT and high-risk conditions and related features as shown in Table No The COVID 19 positive results were more common in males (62.5%) compared to females (37.5%), OR

Discussion:
There are no standardized COVID triage systems and protocols focusing on ED prognostication of COVID 19 patients in Pakistan. CorVETT provides a standardized approach for evaluating patients in the ED or other clinical settings. Our tool was sensitive enough to accurately screen suspected respiratory and non-respiratory COVID 19 adult patients. It has shown to be both sensitive and speci c in early identi cation of suspected COVID 19 patients and their subsequent disposition to the Respiratory COVID Unit (RCU), our ED's isolation facility. The tool helps in having a planned approach during this pandemic in dealing with patients at the level of triage. Accurate triaging and appropriate allocation of limited resources are essential components as a coordinated response to the pandemic. The tool also offers feasibility to healthcare staff by its simplicity and relevance to the population where it is developed.
Males were in the majority for COVID 19 positive results, in accordance with some other studies (1,4,8). This may be due to their risk of contracting the virus at the workplace or during travel. Majority of our study participants were in the high risk pro le which is expected given the higher likelihood of their immunocompromised state making them more susceptible to the virus. Our ndings are similar to those from studies in other countries (9)(10)(11). Similarly, patients in their younger or middle age had less risks of contracting the virus or developing symptoms. This might be due to their strong immune system as shown in other studies in which prevalence of infection was less in the young and middle age group (10,11). The tool provided a sensible approach of using PPE during the pandemic which was sparse and not readily available in most hospitals. The tool is currently used in one of the largest tertiary care hospitals of the city, serving as a standardized triage tool for the ED. Our algorithm is validated in our ED population. ED's of lower middle income countries (LMICs) due to rapid in ux and fragile healthcare systems may suffer numerous pitfalls like poor triaging, mixing of non-COVID 19 patients with COVID 19 patients, and overuse of personal protective equipment's (PPEs) (12,13), to name just a few. Our algorithm is comprehensive as it utilizes variables gathered from published literature as well as our own local practice. It is not meant to replace clinical gestalt and medical decision making but to augment it.
As the disease is evolving, symptom presentation is therefore nonuniform, creating di culty for healthcare staff in doing a good triage. We still recommend that emergency room physicians should follow strict airborne and contact precautions in this pandemic at the triage.
Majority of the COVID triaging tools that are in use are from developed countries; they do not give enough consideration to risk factors and healthcare dynamics, an issue that is more prevalent in low resource settings (3,5,14). Additionally, the tools rely mostly on laboratory and radiologic ndings which make it di cult for the triage staff due to time constraints and overcrowding, a common occurrence during the pandemic (12). Our study results suggest that CorVETT provides good sensitivity and speci city compared to WHO algorithm or PMEWS (Sensitivity 80% and speci city 95% for PMEWS ≥ 5)(15)and makes it a good current pandemic ED triaging tool.

Limitations:
A major limitation of our study is that it is from just a single healthcare organization, and thus it might miss some of the other presentation characteristics of COVID 19 in adults. Another limitation is that the data was collected on a prede ned form and assessment was done by triage nurses; this may have led to lack of important historical and clinical features that clinicians may be better at picking up, hence leading to a lower estimation of the performance of our triage tool. Further, our tool may be too lengthy, placing additional burden on the triage staff, and possibly resulting in over-crowding in a busy ED. However, considering its impact on prevention through early screening and appropriate disposition, CorVETT may prove bene cial in additional studies enrolling more patients in different EDs across the city and country.

Conclusion:
CorVETT was designed to facilitate timely evaluation of suspected COVID-19 in a step wise approach that optimizes the triage process, lessens unnecessary clinical exposure, and improves patient care and resource allocation in EDs of LMICs. With good sensitivity and speci city, the tool is likely to facilitate ED teams of low resource settings in their response to the ongoing pandemic.

Methods:
Study Design and Site: This was a cohort study conducted in the emergency department of Aga Khan University Hospital, which is a 550 bedded large tertiary care teaching facility located in Karachi, Pakistan. The ED is a 62-bedded facility that receives 60,000 patients annually. In the COVID 19 pandemic we divided our ED into two major areas; COVID and non-COVID. The COVID area was further strati ed into respiratory and nonrespiratory unit based on patients requiring interventions like high ow oxygen, nebulization, noninvasive or invasive mechanical ventilation. The Clinical Decision Unit (CDU), an eight-bedded unit was modi ed into negative pressure unit speci cally for suspected COVID patients having respiratory symptoms and named as RCU. The study was conducted and reported in accordance with TRIPOD recommendations(16). The type of prediction model study that was used for this study was type 1 as per the TRIPOD statement in which same data is used for validation and evaluation of performance(16).

Participants (inclusion & exclusion criteria)
Inclusion Criteria: All adults (age 18 years and above) presenting to the emergency department lter triage and were admitted with suspected COVID-19 in the emergency department and subsequently in the hospital will be included.

Exclusion Criteria:
Patients who are directly admitted in ward and are tested positive prior to presentation /in the emergency department from test done either from AKU or outside.

Sampling Method:
Convenience (Non-probability) Sampling Sample Size: With 95% con dence interval, and 4% margin of error, a minimum of n = 585 assigned patients with suspected COVID 19 patients admitted from Emergency department. Sample size for the study, based on AUC, was calculated using method de ne by Hajian-Tilaki K et al (17). Using Microsoft Excel. Formula using for the sample size calculation is as below; Where, and Data Collection: Data collection was done after following the Standard Operating Procedures which our ED and hospital adopted for conducting research during these special circumstances. A standardized data collection form was developed that included predictor variables incorporated into our triage tool. The data collection form was pilot tested on 10 participants before actual data collection was started. Information about study was provided to ED staff prior to start of the data collection. The data was collected at lter triage that was an area restructured during pandemic to screen patients at triage before being either transferred to normal triage or direct transfer to RCU. The patients at triage were given a unique serial number to prevent those with multiple presentations from being included more than once. The study received Institutional Review Board approval (ERC 2020-5222-11403).
The CorVETT Algorithm: The CorVETT tool consists of four sequential steps: presenting features, vitals, associated features (chest pain, cyanosis, respiratory distress, coma or convulsions) and high-risk conditions, as shown in Figure-1. The score was developed based on above variables and categorized by assigning a number to each variable in presenting complaint, a minimum score of two was needed to get a score of four. Similarly, in associated features a score of two is given to a single associated feature. In high-risk conditions a score of six was given for three or more characteristics. The overall score ranges from zero to 10 with 10 being the highest score. Validity and discriminate parameter for CorVETT calculated include sensitivity, speci city, positive and negative predictive value or related proportions for over and under triage. Sensitivity of the tool was calculated to correctly identify patients with COVID 19 infection and percentage/proportion for individuals to get admission in RCU or in-hospital isolation facility. Speci city was calculated to identify patients which the tool could correctly classify as not needing admission and owing to low triage priority. Additionally, positive predictive value (PPV) was calculated for patients admitted to hospital and those who were triaged to high importance. Negative predictive value (NPV) was calculated for those not admitted to hospital and triaged to low priority.
Statistical Analysis: Qualitative variables were reported as means with standard deviation (SD) or medians with Interquartile Range (IQR) and categorical variables as frequencies and percentages. Mean (SD) or median (IQR) differences in scores were assessed by paired t-test or Wilcoxon signed rank test. Signi cant difference of continuous outcomes with normal distribution was compared using t-test or Mann-Whitney U test as appropriate. The probability of event was estimated using linear logistic regression. Selection of variables was based on complete enumeration algorithm and Akaike Information Criteria (AIC). It was assumed that the optimal model is one that minimizes AIC value. The model obtained was summarized in coe cients and standard errors (SE). AIC values resulting from suppression of variable and odds ratio (OR) was estimated using 95% con dence intervals. In order to evaluate predictive capacity of the model, a ROC analysis was carried out using set of validation data. The discriminating capacity was evaluated using ROC curve AUC estimation by a 95% con dence interval. It was considered as a point of optimal cut that minimizes the function: (1-sensitivity) 2+ (1-speci city) 2. Finally, for predictive rule the parameters of sensitivity, speci city and predictive values were estimated through con dence intervals of 95%. A hypothesis contrast was considered statistically signi cant when corresponding p-value was less than 0.05. Data was analyzed on "R" (version 3. The study received Institutional Review Board approval (ERC 2020-5222-11403). As there was no direct intervention in patient's management, consent exemption was also approved. Ethical review committee of the AgaKhan University hospital Karachi.
The study was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.

Consent for publication
The study was reviewed by the Ethical Review Committee of the AgaKhan University Hospital and was approved as an exemption. (letter attached in related les)

Availability of data and materials
All data generated or analysed during this study are included in this published article [and its supplementary information les].   13.1% ----