Prognosis of Spontaneous Pneumothorax/Pneumomediastinum in Coronavirus Disease 2019: The CoBiF Score

Highlights What are the main findings? Pneumothorax/pneumomediastinum developed without positive pressure ventilation among COVID-19 patients had high fatality. Presence of comorbidity, bilateral pneumothorax, and fever were related with in-hospital mortality among COVID-19 associated spontaneous pneumothorax/pneumomediastinum patients The CoBiF score (Co = comorbidity, Bi = bilateral pneumothorax, F = fever) well-predicted the early mortality of these patients. What is the implication of the main finding? The CoBiF score was validated in multinational cohorts, and it could improve early recognition and treatment of COVID-19 pneumothorax. Abstract Objectives: Pneumothorax and pneumomediastinum are associated with high mortality in invasively ventilated coronavirus disease 2019 (COVID-19) patients; however, the mortality rates among non-intubated patients remain unknown. We aimed to analyze the clinical features of COVID-19-associated pneumothorax/pneumomediastinum in non-intubated patients and identify risk factors for mortality. Methods: We searched PubMed Scopus and Embase from January 2020 to December 2021. We performed a pooled analysis of 151 patients with no invasive mechanical ventilation history from 17 case series and 87 case reports. Subsequently, we developed a novel scoring system to predict in-hospital mortality; the system was further validated in multinational cohorts from ten countries (n = 133). Results: Clinical scenarios included pneumothorax/pneumomediastinum at presentation (n = 68), pneumothorax/pneumomediastinum onset during hospitalization (n = 65), and pneumothorax/pneumomediastinum development after recent COVID-19 treatment (n = 18). Significant differences were not observed in clinical outcomes between patients with pneumomediastinum and pneumothorax (±pneumomediastinum). The overall mortality rate of pneumothorax/pneumomediastinum was 23.2%. Risk factor analysis revealed that comorbidities bilateral pneumothorax and fever at pneumothorax/pneumomediastinum presentation were predictors for mortality. In the new scoring system, i.e., the CoBiF system, the area under the curve which was used to assess the predictability of mortality was 0.887. External validation results were also promising (area under the curve: 0.709). Conclusions: The presence of comorbidity bilateral pneumothorax and fever on presentation are significantly associated with poor prognosis in COVID-19 patients with spontaneous pneumothorax/pneumomediastinum. The CoBiF score can predict mortality in clinical settings as well as simplify the identification and appropriate management of patients at high risk.


Introduction
During the coronavirus disease 2019 (COVID- 19) pandemic, complications such as pneumothorax (PNx) and pneumomediastinum (PMEx) frequently occurred, both independently and more commonly in conjunction with each other [1]. PNx, which was observed previously in severe acute respiratory syndrome coronavirus infection and the Middle East respiratory syndrome, is significantly associated with poor prognosis [2,3]. PNx was also observed among patients with COVID-19, with an estimated incidence of 0.56-4.2%, which increased among patients requiring intensive care [1,4,5]. PNx and PMEx are adverse predictors of mortality, especially among critically ill patients with COVID-19 [6,7]. In particular, substantially high mortality (13.8-63.0%) from PNx/PMEx has been reported in patients with COVID-19, which is worse than that for PNx/PMEx arising from other respiratory etiologies [5,8].
Despite this correlation between PNx/PMEx and COVID-19, PNx/PMEx occurs after the implementation of invasive mechanical ventilation (IMV) in most cases, implying the possibility of adverse effects attributable to barotrauma or complications of critical care. Furthermore, due to its rarity, the existing literature on spontaneous PNx/PMEx in patients who did not receive IMV mainly includes case reports/series and small retrospective studies. Therefore, the true clinical features of COVID-19-associated spontaneous PNx/PMEx remain relatively unknown. Moreover, despite the high associated mortality, to our knowledge, no scoring system for in-hospital mortality in such patients exists.
Therefore, we aimed to comprehensively review spontaneous PNx/PMEx in COVID-19 patients by pooling individual patient data from previous case reports and case series and developing a novel scoring system to predict in-hospital mortality. Additionally, we evaluated multinational cohorts in collaboration with the International COVID-19 Pneumothorax Working Group (ICP-WG) and externally validated our new scoring system using this external dataset.

Search Strategy and Selection Criteria
This systematic review was registered with PROSPERO (CRD42022295621) and performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols (PRISMA-P, Supplementary Table S1).
Articles on COVID-19 patients with PNx and/or PMEx who had not received IMV prior to PNx/PMEx onset, articles on patients with PNx/PMEx without a history of IMV who had been treated for COVID-19 within the past 3 months, and case reports and case series analyzing sufficient individual patient data were included. Articles reporting PNx/PMEx that developed after IMV or an unclear temporal relationship between the events and those on patients aged <14 years were excluded. Review articles, abstracts, letters to the editor, and articles that did not contain sufficient information on patient characteristics or outcomes were also excluded.
Two investigators (WW and VK) searched PubMed/Medline, Embase, and Scopus up to 21 December 2021. The search terms employed are described in detail in Supplementary Table S2. Discrepancies regarding the inclusion/exclusion of studies were discussed and resolved by consensus among four investigators (WW, VK, JIS, and SL). The initial search yielded 303 studies after the elimination of duplicates. After reviewing the abstracts and full texts of these articles, we identified 104 studies (87 case reports and 17 case series) that met the inclusion criteria (Supplementary Table S3). The PRISMA flow diagram of the selection process is depicted in Supplementary Figure S1.
We extracted data on patient demographics, clinical characteristics of PNx/PMEx, COVID-19-associated treatments, radiologic findings, treatments, and clinical outcomes from each eligible case report and case series. Since few studies reported the laboratory findings at presentation, which were not consistent between studies, they were not included in this analysis.

Data Collection
We recorded the first author, publication year, and country of origin for each eligible case report or case series and collected information on demographic and clinical characteristics, including sex, age, comorbidities, smoking history, COVID-19-specific medical treatments, symptoms, location of PNx/PMEx, existence of tension PNx or subcutaneous emphysema, chest computed tomography (CT) findings, treatments, duration of hospitalization, intensive care unit (ICU) admission, outcomes, and mortality.

Statistical Analysis
Data on continuous variables (age, oxygen saturation, and duration of hospitalization) are presented as median and interquartile ranges after determining the normality of the distribution. Differences in these variables were compared using the Mann-Whitney U test. Fisher's exact test was performed to compare categorical variables [9]. We performed logistic regression to determine the predictive factors for in-hospital mortality; logistic regression was considered suitable for analysis as all deaths occurred within 2 months of hospitalization. Variables with p-values < 0.05 in univariate analysis were entered into multivariate analysis, and the variables were selected by backward elimination with a two-tailed p-value of <0.05. The scoring system for in-hospital mortality was devised based on significant factors as follows:

Step 1: Development and Internal Validation of the CoBiF Scoring System
Significant variables for in-hospital mortality (p < 0.05) in the multivariate logistic regression analysis were used to create the CoBiF scoring system. This scoring system was constructed using binary variables to ensure application ease. The score for each variable was measured based on its odds ratio (OR) and regression coefficient. Two variables (comorbidities and bilateral PNx) had similar magnitude for predicting the outcome (score for each: 1), whereas a third variable (fever at PNx/PMEx presentation) had a greater weight (score: 2). The summation of the three scores was used for the prediction model (score 0-4). If the value of any of these three parameters was missing, they were not included in the measurement of predictability. The discriminative power of the CoBiF model was assessed by plotting a receiver operating characteristic curve and calculating the area under the curve (AUC). Bias-corrected AUC was measured for internal validation, and the Mantel-Haenszel chi-square test was performed for calibration.

Step 2: External Validation
We developed an independent dataset with the same inclusion/exclusion criteria for external validation. First, authors who published retrospective studies in 2022 were approached, and their data were included [10][11][12][13]. Second, authors who reported COVID-19-related PNx/PMEx were contacted and encouraged to share data. This was performed by the ICP-WG. Finally, we gathered approved information on 133 patients from 10 countries. Thereafter, the performance of the CoBiF scoring model was assessed by computing the AUC.
All statistical analyses were performed using the Statistical Package for the Social Sciences (SPSS) for Windows version 25.0 (SPSS Inc., IBM Corporation, Armonk, NY, USA) and R version 4·0·4 (R Core Team, Vienna, Austria) and were supervised by a medical statistician (HSL).

Comparison between Pneumomediastinum and Pneumothorax ± Pneumomediastinum
There were no differences in age, sex, comorbidity, clinical manifestations, COVID-19-associated treatments, and radiological findings between the PMEx and PNx ± PMEx groups, although obesity was more common in the PMEx group (Table 1). Due to diseasespecific characteristics, most patients with PMEx were conservatively treated. However, more patients with PMEx than those with PNx received non-invasive ventilation. Overall, disease severity did not differ between the groups with respect to ICU admission, IMV requirement, and mortality.

Clinical Characteristics According to Clinical Scenarios
Supplementary Table S4 describes the patients' clinical scenarios, classified into the following three categories: (1) group A, patients who presented to the hospital with PNx/PMEx (n = 68, 45.0%) and who had COVID-19 symptoms for a median of 7 days prior to the hospital visit; (2) group B, patients who developed PNx/PMEx during inpatient management for COVID-19 (n = 65, 43.1%) at a median of 10.0 days from admission; and (3) group C, patients who underwent re-admission within a median of 16.5 days due to development of PNx/PMEx after discharge following recent COVID-19 treatment (n = 18, 11.9%). There were no remarkable differences in the characteristics of patients who presented with the three clinical scenarios; however, the frequency of symptoms such as fever (p < 0.001), cough (p = 0.001), and mortality (p = 0.087) was higher in groups A and B than in group C.

Patient Characteristics According to Mortality
Overall, the in-hospital mortality rate was 23.2% (35/151). Table 2 demonstrates the differences between survivors and non-survivors. Compared to survivors, non-survivors were older (median 52.5 versus 67.0 years, p < 0.001) and had a frequency of comorbidities (p < 0.001), such as obesity (p = 0.007) and hypertension (p = 0.008). At presentation, the frequency of fever (p < 0.001), cough (p = 0.032), dyspnea (p = 0.026), and symptoms other than chest pain (p = 0.005) were higher in patients with adverse outcomes than in those with non-adverse outcomes. Non-survivors received more COVID-19-specific medical treatments, such as steroids (p = 0.004) and remdesivir (p = 0.037) than survivors. Although there were no differences in radiological findings, the frequency of bilateral PNx (p = 0.029) was higher among non-survivors than among survivors.

The CoBiF Scoring System for in-Hospital Mortality
A novel scoring system was designed to predict in-hospital mortality in patients who developed COVID-19-associated spontaneous PNx/PMEx. Multivariate logistic analyses revealed comorbidities, fever at PNx/PMEx presentation, and bilateral PNx as significant variables. Thus, the scoring system was named CoBiF (Co = comorbidities, Bi = bilateral PNx, F = fever) based on the components of this model. The predicted in-hospital mortality was proportional to the score-1.85%, 8.14%, 29.5%, 66.3%, and 90.3% with a CoBiF score of 0, 1, 2, 3, and 4, respectively (Figure 1). The AUC was 0.887 (95% CI: 0.822-0.951), and the Mantel-Haenszel chi-square test yielded a p-value of <0.0001, signifying a good fit between the model and observed data.

Internal and External Validation of the CoBiF Scoring System
Internal validation was conducted using the bias-corrected AUC (0.818), which demonstrated good predictability. The overall patient characteristics in the internal and external datasets are presented in Table 4. The validation cohort had a worse medical condition in terms of age and comorbidities; the original geographic areas of the two datasets

Internal and External Validation of the CoBiF Scoring System
Internal validation was conducted using the bias-corrected AUC (0.818), which demonstrated good predictability. The overall patient characteristics in the internal and external datasets are presented in Table 4. The validation cohort had a worse medical condition in terms of age and comorbidities; the original geographic areas of the two datasets were notably different. The rates of adverse outcomes, such as ICU admission and mortality, were greater in the external dataset than in the internal dataset. The AUC of the CoBiF scoring system during external validation was 0.709 (95% CI: 0.622-0.796; Figure 2).

Internal and External Validation of the CoBiF Scoring System
Internal validation was conducted using the bias-corrected AUC (0.818), which demonstrated good predictability. The overall patient characteristics in the internal and external datasets are presented in Table 4. The validation cohort had a worse medical condition in terms of age and comorbidities; the original geographic areas of the two datasets were notably different. The rates of adverse outcomes, such as ICU admission and mortality, were greater in the external dataset than in the internal dataset. The AUC of the CoBiF scoring system during external validation was 0.709 (95% CI: 0.622-0.796; Figure  2).   Table 4. The validation cohort had a worse medical condition in terms of age and comorbidities; the original geographic areas of the two datasets were notably different. The rates of adverse outcomes, such as ICU admission and mortality, were greater in the external dataset than in the internal dataset. The AUC of the CoBiF scoring system during external validation was 0.709 (95% CI: 0.622-0.796; Figure  2).

Internal and External Validation of the CoBiF Scoring System
Internal validation was conducted using the bias-corrected AUC (0.818), which demonstrated good predictability. The overall patient characteristics in the internal and external datasets are presented in Table 4. The validation cohort had a worse medical condition in terms of age and comorbidities; the original geographic areas of the two datasets were notably different. The rates of adverse outcomes, such as ICU admission and mortality, were greater in the external dataset than in the internal dataset. The AUC of the CoBiF scoring system during external validation was 0.709 (95% CI: 0.622-0.796; Figure  2).

Discussion
This study described the clinical manifestations, management, and prognosis o spontaneously developed PNx/PMEx among COVID-19 patients based on the current lit erature. This study included observations from various countries and reported real-world

Discussion
This study described the clinical manifestations, management, and prognosis of spontaneously developed PNx/PMEx among COVID-19 patients based on the current literature. This study included observations from various countries and reported real-world data by including institutions with limited resources. We have developed a novel CoBiF scoring system to predict in-hospital mortality among patients with COVID-19 who developed PNx/PMEx without prior IMV. Notably, we incorporated COVID-19 patient data from eight countries and evaluated their disease severity.
Each factor included in the CoBiF scoring system seemed to incorporate the results of previous studies on the prognosis of COVID-19 patients, representing various risk factors for mortality. These factors include age, symptoms (fever, hemoptysis, dyspnea, and loss of consciousness), comorbidities (number of diseases, cancer history, and hypertension), and laboratory findings (D-dimer level, neutrophil-to-lymphocyte ratio, lactate dehydrogenase level, and bilirubin level) [14][15][16]. Fever was considered a poor prognostic sign based on an Iranian national data [17], and it was observed more commonly in COVID-19associated PNx/PMEx than in non-COVID-19-associated PNx [5]. However, the clinical interpretation of fever needs caution since the causes of it could differ according to the patients' population; patients with fever had more underlying medical conditions in this study. As multicollinearity was not observed between comorbidity and fever in the analysis for in-hospital mortality, we could not definitively describe possible other causes for fever. In this study, the presence of fever seemed to be more related to the severity of COVID-19 infection. Even so, clinicians should consider diverse reasons for fever depending on patients' characteristics.
The general predictive factors for COVID-19 also seem to be effective in predicting the prognosis of COVID-19-associated PNx/PMEx because they bear similarities to the risk factors used in the CoBiF score. Since most causes of death in our study were merely suggestive of the detrimental consequences of COVID-19, the specific nature of the relationship between mortality and PNx/PMEx could not be ascertained conclusively. Therefore, the treatment approach for COVID-19-associated PNx/PMEx should be in accordance with general COVID-19 management protocols based on current knowledge.
However, clinicians should focus minutely on bilateral spontaneous PNx, which is rarely reported in other respiratory diseases. Simultaneous bilateral PNx occurs in 1.0-1.6% of all patients with PNx [18][19][20]. These patients have adverse outcomes during hospitalization and long-term mortality [21], necessitating timely intervention [20]. Specifically, PNx is considered a poor prognostic factor, which is observed mainly in patients with underlying lung disease [19]. It is unknown whether bilateral PNx occurs more commonly in COVID-19. However, bilateral pneumonia has been observed in 75-86% of hospitalized patients with COVID-19 [22,23], with the most common CT features being peripherally distributed ground-glass opacities and bilateral lung consolidation. Moreover, the extent and intensity of opacities are suggestive of a poor prognosis [24,25]. Moreover, barotrauma in patients with IMV, i.e., PNx/PMEx, occurs more frequently in COVID-19 than in other acute respiratory distress syndromes [7]. If the vulnerability of COVID-19-infected lungs causes the development of PNx/PMEx, bilateral PNx may indicate the fatal nature of COVID-19. The causal relationship and underlying pathophysiology require further investigation.
The mechanism underlying the pathogenesis of PNx/PMEx in COVID-19 remains obscure, and several hypotheses, including air leakage through the alveolar walls [26], increased vulnerability to PNx/PMEx arising from cyst formation due to severe damage induced by inflammation [27,28], and the Macklin effect as a cause for PMEx, have been postulated [29]. However, recent studies found no difference between the histopathologic findings of COVID-19 and other causes of lung injury [28,30]. Further comparative studies are needed to elucidate the specific mechanism underlying the formation of PNx/PMEx in COVID-19.
Our study has several limitations. First, although we presented data from multiple countries, the heterogeneity in their respective clinical environments and the capacity to deal with the pandemic could have impacted the outcomes. Second, since cases were collected based on the authors' recall and retrospective review of medical records, publication bias could arise from the inclusion of a few patients with poor outcomes. Additionally, this study could not present the laboratory findings of the patients' medical conditions as they were selectively reported by most studies; thus, disease severity in the patients included in this review could not be compared quantitatively. Despite our great efforts to contact the authors of these case reports, we were unable to contact all authors and validate patient data. Moreover, the medical management of COVID-19 in the included cases was highly heterogeneous; some patients were treated with medications that are no longer used for COVID-19 treatment. Therefore, the study findings should be interpreted cautiously in the evolving contemporary context of the COVID-19 pandemic, viz., the different dominant virus strains, vaccine availability, and innate immunity level.

Conclusions
This study is important because it comprehensively reviewed cases of spontaneous PNx/PMEx in COVID-19 and presented a numerically measured prediction model. This novel CoBiF scoring system can be applied in other clinical settings, assist clinicians in identifying patients at high risk of mortality, and facilitate more prompt management. A further improved scoring system can be devised after the accrual of more evidence and research on spontaneous PNx/PMEx in COVID-19.  Informed Consent Statement: Ethical approval was not required for a pooled analysis of published articles. However, the validation dataset consisted of spontaneous COVID-19-associated PNx/PMEx patients who were included in the International COVID-19 Pneumothorax Working Group (ICP-WG). Each hospital had permission from its ethical review board for the collection of observational data, and informed consent was waived due to its retrospective design of the study (IRB No. of the main institution, No. 3-2022-0186).

Data Availability Statement:
The data that support the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgments:
We appreciate the clinicians in the ICP-WG who participated in this project and shared their diverse clinical experiences during the pandemic.

Conflicts of Interest:
The authors declare no conflict of interest.