key: cord-0790911-ldm6396c
authors: Pagano, Andrew; Finkelstein, Mark; Overbey, Jessica; Steinberger, Sharon; Ellison, Trevor; Manna, Sayan; Toussie, Danielle; Cedillo, Mario; Jacobi, Adam; Gupta, Yogesh S.; Bernheim, Adam; Chung, Michael; Eber, Corey; Fayad, Zahi A.; Concepcion, Jose
title: Portable Chest Radiography as an Exclusionary Test for Adverse Clinical Outcomes During the COVID-19 Pandemic
date: 2021-01-29
journal: Chest
DOI: 10.1016/j.chest.2021.01.053
sha: 9e318e674bac06de256eb3514e231e88724b7445
doc_id: 790911
cord_uid: ldm6396c

Background Chest radiography is often performed in the acute setting to help understand the extent of respiratory disease in patients with COVID-19, but a clearly defined role for negative chest radiographs in assessing patients has not been described. Research Question Is portable chest radiography an effective exclusionary test for future adverse clinical outcomes in patients suspected of COVID-19? Study Design and Methods Charts of consecutive patients suspected of COVID-19 at five emergency departments in New York City between March 19, 2020 through April 23, 2020 were reviewed. Patients were categorized based on absence of findings on initial chest radiography (CXR). The primary outcomes were hospital admission, mechanical ventilation, acute respiratory distress syndrome (ARDS), and mortality. Results 3245 adult patients, 474 (14.6%) with a negative initial CXR, were reviewed. Among all patients, a negative initial CXR is associated with a low probability of future adverse clinical outcomes, with negative likelihood ratios of 0.27 (95% CI 0.23-0.31) for hospital admission, 0.24 (95% CI 0.16-0.37) for mechanical ventilation, 0.19 (95% CI 0.09-0.40) for ARDS, and 0.38 (95% CI 0.29-0.51) for mortality. Among the subset of 955 patients younger than 65 years old and with a duration of symptoms of at least 5 days, no patients with a negative CXR died and the negative likelihood ratios were 0.17 (95% CI 0.12-0.25) for hospital admission, 0.09 (95% CI 0.02-0.36) for mechanical ventilation, and 0.09 (95% CI 0.01-0.64) for ARDS. Interpretation Initial chest radiography in adult patients suspected of COVID-19 is a strong exclusionary test for hospital admission, mechanical ventilation, ARDS, and mortality. The value of chest radiography as an exclusionary test for adverse clinical outcomes is highest among young adults, patients with few comorbidities, and those with a prolonged duration of symptoms.

Initial chest radiography in adult patients suspected of COVID-19 is a strong exclusionary test for hospital admission, mechanical ventilation, ARDS, and mortality. The value of chest radiography as an exclusionary test for adverse clinical outcomes is highest among young adults, patients with few comorbidities, and those with a prolonged duration of symptoms. [2] [3] [4] Given the broad spectrum of disease from asymptomatic infections to death, 5 rapidly and accurately identifying patients who may not require advanced care may provide opportunities to preserve scarce medical resources. COVID-19 prognosis models largely rely on demographic, medical history, and laboratory data with imaging input often restricted to chest computed tomography (CT). 6 Early and continuing limits on the availability and timeliness of real-time reverse transcriptase polymerase chain reaction (RT-PCR) testing, particularly in the United States, has led investigators to explore imaging as a way of augmenting or replacing molecular diagnosis. 7-9 COVID-19 pneumonia can result in a characteristic, time-dependent pattern of pulmonary disease on CT. [10] [11] [12] However, evidence that chest CT findings in COVID-19 are nonspecific has led the United States Centers for Disease Control and Prevention (CDC) and the American College of Radiology to recommend against the use of chest CT or chest radiography (CXR) alone for the diagnosis of COVID-19 infection. 13, 14 Regardless of CT's accuracy as a diagnostic test for COVID-19, it has proven impractical given potential for cross-infection and cumbersome cleaning and isolation protocols. 15, 16 In line with the American College of Radiology, hospitals have largely avoided CT and CXR for diagnosis and favor portable CXR when assessing COVID-19 severity and investigating alternate or superimposed diagnoses . 14 J o u r n a l P r e -p r o o f Given the practical advantages of portable CXR and its widespread usage on initial presentation, a limited but growing body of literature has sought to explore radiography as a prognostic tool in COVID- 19 . In one study, a proposed system for grading CXR severity was found to be an independent predictor of hospital admission and intubation among patients with More recent studies have also used chest radiography to assess COVID-19 severity and predict early intubation, continuous renal replacement therapy, and mortality. [18] [19] [20] [21] However, public health organizations like the CDC and published research have largely sought to characterize the accuracy of CXR as a diagnostic tool for COVID-19 pneumonia, concluding it has a low specificity for diagnosis. 22

Herein we aim to explore the clinical utility of CXR as an exclusionary test for adverse clinical outcomes in patients suspected of COVID-19. We hypothesize that negative chest radiography can be used to rule out severe disease progression. We further believe that such a test would be more accurate later in the course of symptoms and in patients with a lower comorbidity burden.

In this retrospective cohort study we examined the hospital course of consecutive adult patients (≥21 years) with a COVID-19-related encounter diagnosis presenting to five emergency departments (EDs) of a multicenter healthcare system in New York City from March 19, 2020 to April 23, 2020. This study was approved by The Mount Sinai Hospital Institutional Review Board (#IRB-20-03508).

Patients with high suspicion of COVID-19 presenting to the ED were identified via our institution's COVID-19 registry database. All patients had a single-view anteroposterior portable CXR performed at initial presentation. Patients with multiple CXRs performed on separate ED encounters were treated separately (N=78).

Demographic and clinical variables including age, sex, race, body mass index (BMI), comorbidities, RT-PCR results, vital signs, and selected laboratory results were obtained from an institutional COVID-19 registry. Symptomatology was obtained through chart review by observers blinded to CXR results and based on the CDC description of COVID-19. 23 Duration of symptoms was considered important in this study because it reflects length of infection and progression of lung abnormalities on imaging.

Patients were categorized by age using the conventional geriatric cutoff of 65 years. This was thought to be a crucial division for two reasons. First, patients below this age tend to have fewer comorbidities. 24 While certain comorbidities were extracted for this analysis, a key limitation of many studies that rely on compiled datasets is a failure to adequately capture the full spectrum and severity of patient comorbidities. Second, at the time of writing this manuscript, a younger US demographic is increasingly testing positive for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection. 25 J o u r n a l P r e -p r o o f Vital signs were summarized using quick Sequential [Sepsis-related] Organ Failure Assessment (qSOFA) score. 26 Patients were variably tested for COVID-19 infection depending on availability of RT-PCR kits and clinical judgement.

The primary outcomes for this study were hospital admission, mechanical ventilation, ARDS, and death for an 85-day follow-up period. Admissions occurring within fourteen days of CXR were included and merged with initial ED presentation data; this time period allowed admission data from patients who may have been prematurely discharged from the ED while disallowing admission data from patients who may have become COVID-19 infected after discharge. To ensure the usage of data available to point of care providers and that results were not associated with infection acquired during hospitalization, only RT-PCR tests within 24 hours of admission were considered.

All CXRs were interpreted at the time of acquisition by 40 radiologists across five ED sites. CXR reports were categorized by two independent, blinded observers: those without airspace opacities or having up to mild atelectasis or chronic findings ("negative") and those with airspace opacities regardless of severity or pattern of distribution ("positive") (e- Table 1 for examples of report language interpretation, e- Figure 1 for examples of CXR images). Where categorizations differed, a third observer acted as arbiter. This approach allowed for a focused analysis of radiographic reads as they are available to front-line staff.

The Cohen's kappa coefficient and complete concordance were used to assess agreement in scoring between report interpretations. Complete concordance (CC) was defined as the percentage of identical categorizations. Negative predictive values (NPV) and negative likelihood ratios (NLR) were calculated for a positive CXR interpretation, thus allowing for assessment for the utility of a negative CXR interpretation. Comparison was performed between NPVs and NLRs using a test of differences. 27 Bivariate analysis of continuous variables was performed using the Kruskal-Wallis H Test and bivariate analysis of categorical variables was performed using chi-squared test. In order to maximize the number of records for multivariable analysis, missing BMI (N=522, 16.1%) and qSOFA (N=11, 0.3%) was imputed using predictive mean matching using models that included the outcomes of interest, demographic information, and clinical variables. This was done using the mice package. 28 Prior to imputation, data was analyzed to ensure no significant departure from the assumption of missingness at random. Sensitivity analysis was performed by setting missing data to the lowest and highest values and ensuring there was no effect in the results related to the primary outcomes of interests and covariates. These values were then utilized in the multivariable model through multiple imputation according to Rubin's rules. 29 Using the lme4 package, 30 a mixed-effects logistic regression model adjusted for demographics, comorbidities, qSOFA at presentation, time of presentation, and the interaction between time of imaging and the categorization of the imaging as fixed effects was performed. The exponentiated coefficient (EC) is presented for the interaction variable, which presents the magnitude of the effect of the variable and may be interpreted as the ratio by which the odds ratio changes when the interaction is present. Adjusted odds ratios (OR) are presented for other covariates. We included a random effect for the radiologist finalizing the read as it was found to explain a substantial proportion of variability J o u r n a l P r e -p r o o f and improved model fit. A p-value of less than 0.05 (two-tailed) was considered statistically significant. All analysis was completed using R version 3. 6 

Computing, Vienna, Austria). vs. 21.4%; p=0.004) were also associated with a negative CXR. Patients with a negative CXR presented with a shorter duration of symptoms (4 vs 7 days from symptom onset; <0.001) and had lower rates of fever (52.4% vs. 61.6%; p<0.001) and shortness of breath (52.6% vs. 69.2%; p<0.001) at presentation. The proportion of negative initial chest radiographs changes with the days of symptom onset (e- Figure 2 ). This trend is not followed by RT-PCR result (e- Figure 3) and is clearly followed in proportion of deaths (e- Figure 4 ). Statistically significant differences were seen in presenting vital signs and select laboratory results between patients with a negative CXR and those with a positive CXR, furthermore a lower positive qSOFA rate (9.5% vs. 23.4%; p<0.001) was associated with a negative initial CXR.

A total of 2600 (80.1%) patients were admitted, 540 (16.6%) were intubated, 235 (7.2%) developed ARDS, and 764 (23.5%) died. Eight (0.2%) patients had not been discharged within a follow-up period of 85 days.

On review of patients with negative CXRs with adverse outcomes, we found often severe comorbidities often not captured by our extensive list of comorbidities. This increased complexity of the false negative group is somewhat reflected in comparing the number of comorbidities of deceased patients with negative CXR to those who survived with negative CXR (median 2, IQR 0 to 3 v. median 0, IQR 0 to 2, p = 0.01); intubation to no intubation (median 2, IQR 0 to 3 v. median 0, IQR 0 to 2, p = 0.01) ; ARDS to no ARDS (median 2, IQR 1 to 3 v. median 0, IQR 0 to 2, p = 0.05) , and admission to no admission (median 2, IQR 0 to 3 v. median 0, IQR 0 to 1, p < 0.001). All comparisons were significant. J o u r n a l P r e -p r o o f DISCUSSION In our study we found that CXR is a strong exclusionary test for adverse clinical outcomes. Use of CXR as a prognostic exclusionary test has not been previously described although its use as a predictive indicator in the COVID-19 pandemic has received increased attention. Our further observation that the predictive power of a negative CXR increases with increased duration of symptoms (at least 5 days) comports with a well understood time dependent nature of SARS-CoV-2 infection that sees a median time from onset of symptoms to hospital admission in 7 days and ARDS in 9 days. 31 As expected, a negative CXR is least predictive early in disease when airspace opacities are unlikely regardless of a patient's future clinical outcome. In later stages of disease, a negative CXR carries more predictive weight as sufficient time has allowed for the potential development of airspace opacities.

We found portable CXR highly sensitive but non-specific in the prediction of observed clinical outcomes. These results suggest that although a negative CXR can be exclusionary of adverse outcomes, a positive CXR does not share this same prognostic value. This observation can be explained by the study's usage of a broad interpretation of positive CXR as consisting of any lung abnormality outside of mild atelectasis or chronic findings and included airspace opacities regardless of severity or pattern of distribution. Studies that have characterized the relationship between positive CXR and poor clinical outcomes have more rigorously assessed severity of disease in positive CXR either by quantifying lung zone involvement 17, 21 or by employing specific scoring systems that capture disease severity. [18] [19] [20] In general, these studies demonstrate that positive CXR becomes more specific for adverse outcomes with increased severity of disease.

J o u r n a l P r e -p r o o f A prominent example of exclusionary testing that shares similarities with the approach presented here is D-dimer testing in risk assessment for venous thromboembolism. A negative D-dimer test has 99% to 100% NPV and 0.08 to 0.27 NLR among patients with a low pretest probability for disease. 32,33 Even in our study population, which had a very high mortality rate (23.5%) relative to mortality in all confirmed cases in New York State (8%), 34 we observed comparably high negative predictive values of a negative CXR for mechanical ventilation (95%) and ARDS (99%). Perhaps more telling are negative likelihood ratios, a statistic not influenced by disease prevalence, for understanding the value of an exclusionary testing. NLRs less than 0.1 result in large and often conclusive changes from pre-to post-test probability. 35 In our study, a negative CXR among patients who are younger than 65 years old and with a duration of symptoms of at least 5 days, there were very low NLRs for mechanical ventilation (0.09), ARDS (0.09), and death (0).

Our study was guided by the observation that two major factors influence the reliability of CXR as an exclusionary test, the timing of the CXR relative to onset of symptoms and the patient's burden of comorbidities. Attempts to subset patients using number of comorbidities was limited by a finite array of comorbidities extracted in our study and by the binary nature of only capturing the presence of disease without fully describing severity. However, age is correlated with the presence and severity of comorbidities, and we found that when combined with duration of symptoms, CXR became a more powerful exclusionary test. A future direction for research would better measure a patient's intrinsic burden of comorbidities to further improve chest radiography as an exclusionary test.

Within our study population, 14.6% of patients presented with a negative initial CXR, a very low proportion and likely a consequence of regional policies that resulted in only the sickest patients reporting to city EDs during the peak of infections in NYC. Furthermore, our cohort consisted of patients with suspected or presumed COVID-19 infection and not necessarily confirmed diagnoses, those with overt symptomatology in need of prompt care. In less acute settings where patients have either mild symptoms or PCR confirmed infections that are otherwise asymptomatic, we would expect both the proportion of patients with self-limiting COVID-19 and the proportion of negative initial CXR to be much higher. In a study conducted from March 9 to March 24, 2020 in patients presenting to urgent care centers in New York City with confirmed COVID-19, 58% of patients had a negative CXR on presentation. 36 On multivariable analysis adjusting for clinical information available to first-line providers, negative CXRs were found to be independent predictors of decreased rates of admission, mechanical ventilation, and death as well as the additional suggestion of a similar association with ARDS that was not statistically significant. Notably, we found that negative CXR interpretations performed later in the crisis were significantly more predictive of decreased death in comparison to earlier interpretations, suggesting acquired expertise by radiologists in detecting COVID-19 pneumonia over the study period. Other possibilities for this trend such as shifts in demographics or disease severity as well as systemic factors related to medical resource allocation were incompletely analyzed in this study.

Due to the retrospective nature of this study, there were certain intrinsic limitations. First, it is unknown how medical decision-making was affected by limited resources and a desire to control spread of disease within the hospital. RT-PCR testing suffered from moderate sensitivity 37 and early rationing of test-kits led to many untested patients either because of clear presence of disease or absence of severe symptoms. For these reasons all patients with a COVID-19 related encounter and not necessarily a confirmed PCR diagnosis were included in the study; 11.6% had a negative RT-PCR test and 17.2% were not tested. Although this approach maximized the practicality and perhaps translatability of the study, our results are possibly derived from an mixture of unknown additional acute respiratory illnesses. Second, although the analyzed hospital system is the largest in NYC, patients may have sought care outside of our hospitals after first reporting to one of our EDs. Third, a 14-day follow-up period was used to track patients presenting to the ED for possible admission that may have been related to the primary presentation. We could not exclude the possibility of re-infection or the possibility of the patient not being infected in the first encounter and interval infection during the 14-day period.

Furthermore, patients were only followed to discharge, raising the possibility of later related admission or adverse outcome. However, any analysis that would include further readmissions would be confounded by the possibility of secondary causes of adverse outcome that could not be isolated from the initial primary cause of admission for which the patient was recruited into the study.

Chest radiography is an inexpensive and ubiquitous test already performed on most patients with complaints of respiratory disease. Our findings suggest that there is a role for portable CXR in determining the absence of severe COVID-19. In this regard we offer the concise view of a J o u r n a l P r e -p r o o f portable CXR as being either negative or not negative, and suggest clarity from interpreting radiologists in differentiating the two. We believe our conclusions can be expanded outside of the ED to wherever patients suspected of COVID-19 present. In addition, use of CXR as an exclusionary test may prove useful in future respiratory pandemics where resources are limited or accurate, rapid molecular testing is not yet available.

In summary, CXR performed in the ED on patients suspected of COVID-19 is a strong exclusionary test for adverse clinical outcomes particularly when limited to patients under 65 years old and with duration of symptoms of at least 5 days.

The authors wish to thank Lena Marra for organizing institutional review board approval of this study.

The lead author (AP) had full access to all of the data in the study and affirms that this manuscript is an is an honest, accurate, and transparent account of the study being reported. Results: Retrospective cohort study of consecutive patients suspected of SARS-CoV-2 infection at five emergency departments in New York City. Analysis showed initial chest radiography is a strong exclusionary test for future hospitalization, intubation, acute respiratory distress syndrome, and mortality.

Interpretation: Negative initial chest radiography in patients suspected of COVID-19 infection may aid in identifying patients not at risk for adverse clinical outcomes including mortality. 

World Health Organization. WHO Director-General's opening remarks at the media briefing on COVID-19 -11

In Pursuit of PPE

Staffing With Disease-Based Epidemiologic Indices May Reduce Shortage of Intensive Care Unit Staff During the COVID-19 Pandemic

Fair Allocation of Scarce Medical Resources in the Time of Covid-19

Characteristics of and Important Lessons From the Coronavirus Disease 2019 (COVID-19) Outbreak in China: Summary of a Report of 72314 Cases From the Chinese Center for Disease Control and Prevention

Prediction models for diagnosis and prognosis of covid-19 infection: systematic review and critical appraisal

The Role of Chest Imaging in Patient Management During the COVID-19 Pandemic: A Multinational Consensus Statement From the Fleischner Society

Diagnosis of the Coronavirus disease (COVID-19): rRT-PCR or CT?

Artificial intelligence-enabled rapid diagnosis of patients with COVID-19

COVID-19): Relationship to Duration of Infection

Relation Between Chest CT Findings and Clinical Conditions of Coronavirus Disease (COVID-19) Pneumonia: A Multicenter Study

Correlation of Chest CT and RT-PCR Testing for

COVID-19) in China: A Report of 1014 Cases

Interim Clinical Guidance for Management of Patients with Confirmed Coronavirus Disease (COVID-19)

ACR Recommendations for the use of Chest Radiography and Computed Tomography (CT) for Suspected COVID-19 Infection

Statements/Recommendations-for-Chest-Radiography-and-CT-for-Suspected-COVID19-Infection

Radiology Department Preparedness for COVID-19: Radiology Scientific Expert Review Panel

Management of patients with suspected or confirmed COVID-19, in the radiology department

Clinical and Chest Radiography Features Determine Patient Outcomes In Young and Middle Age Adults with COVID-19

Performance of a Severity Score on Admission Chest Radiograph in Predicting Clinical Outcomes in Hospitalized Patients with Coronavirus Disease (COVID-19)

Chest X-ray for predicting mortality and the need for ventilatory support in COVID-19 patients presenting to the emergency department

Chest radiograph at admission predicts early intubation among inpatient COVID-19 patients

Correlation of chest radiography findings with the severity and progression of COVID-19 pneumonia

Frequency and Distribution of Chest Radiographic Findings in Patients Positive for COVID-19

Accessed 8

A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation

COVIDView: A Weekly Surveillance Summary of U.S. COVID-19 Activity

Assessment of Clinical Criteria for Sepsis: For the Third International Consensus Definitions for Sepsis and Septic Shock

Comparisons of predictive values of binary medical diagnostic tests for paired designs

Groothuis-Oudshoorn K. mice: Multivariate Imputation by Chained Equations inR

Multiple Imputation for Nonresponse in Surveys

CXR=chest radiograph; IQR=interquartile range; BMI=body mass index; SARS-CoV-2 RT-PCR=severe acute respiratory syndrome coronavirus 2 reverse transcriptase-polymerase chain reaction; COPD=chronic obstructive pulmonary disease; qSOFA=quick sequential

Board (#IRB-20-03508). Informed patient consent was waived by the ethics committee for this retrospective study.