Multidisciplinary Approach in COVID-19 Pneumonia: A Backward Path by an Expert Team

Background: The novel coronavirus disease 2019 (COVID-19) represents a medical challenge worldwide. COVID-19 pneumonia is an extremely complex disease. The hypothesis of the study was that a multidisciplinary approach involving experienced specialists in diffuse parenchymal lung disease might improve the diagnosis of patients with COVID-19 pneumonia. Methods: Two pulmonologists, two radiologists, and two pathologists reviewed 27 patients who died of severe COVID-19 pneumonia as the main diagnosis made by non-pulmonologists. To evaluate whether the contribution of specialists, individually and/or in combination, might modify the original diagnosis, a three-step virtual process was planned. Pulmonologists, radiologists and pathologists were asked to classify every case into four distinct levels of diagnostic certainty, based on clinical, radiological, and morphological/virologic data obtained from an autoptic lung sample, respectively. The whole lung examination was considered the gold standard for the nal diagnosis. The probability of a correct diagnosis was calculated, and the effectiveness of a multidisciplinary diagnosis was obtained by comparing diagnoses made by experienced pulmonologists with those made by non-pulmonologists. Results: COVID-19 pneumonia was excluded in 2 cases (8%) and was a marginal feature in 3 cases (11%). The probability of a correct diagnosis increased strikingly from an undedicated clinician to an expert specialist, becoming progressively more accurate at different steps. Every single specialist made signicantly more correct diagnoses than any non-pulmonologist. The highest level of accuracy was achieved by the combination of 3 expert specialists. Conclusions: In summary, the dynamic interaction between expert specialists signicantly improves the diagnostic condence and management of patients with COVID-19 pneumonia. for routine histology. Pathological features suggestive of COVID-19 pneumonia (alveolar injury as well as vascular lesions) were quantitatively described using a scoring system, as previously reported [10]. Other associated lesions (neoplasia, infectious diseases, aspiration pneumonia, etc.) were also reported. To conrm the pathological diagnosis of COVID-19 pneumonia, the fragment preserved in RNA later was also processed by real time reverse transciptase-polymerase chain reaction (RT-PCR) for SARS-CoV2 [SARS-CoV-2 (2019-nCoV) Centers for Disease Control and Prevention (CDC) Emergency Use Authorized (EUA) Authorized qPCR probe assay primer/probe mix]. An additional fragment was analyzed by culture, as previously described [11]. In order to dene the levels of certainty for a diagnosis of COVID-19 pneumonia, two expert pathologists (F.C., F.P.) scored all cases independently and blinded to clinical and autopsy data. Based on morphological/virologic evaluation, four distinct levels of diagnostic certainty were dened: 1) Denite COVID-19 pneumonia: all lung samples showing lesions typical of COVID-19 pneumonia (vascular injury and/or diffuse alveolar damage/organizing pneumonia), conrmed SARS-CoV2 lung positivity (both molecular and culture), without other lesions suggestive of alternative diagnoses, 2) Probable COVID-19 pneumonia: lung samples displaying mainly features of COVID-19 pneumonia (+/- lung SARS-CoV2 infection) with other associated lesions (i.e. foci of bacterial infection, 3) Possible COVID-19 pneumonia: lung samples showing only focal changes of COVID-19 pneumonia (+/- lung SARS-CoV2 infection, etc.) with more extensive features consistent with alternative diagnoses (i.e., lung cancer/metastasis, etc.) 4) Non-COVID-19 pneumonia: lung samples not showing any typical lesions, no evidence of SARS-CoV2 infection, and no presence of features consistent with alternative diagnoses. COVID-19 pneumonia: features of COVID-19 pneumonia associated with abnormalities such as pleural effusion, cardiomegaly, or Kerley B lines suggestive of cardiac failure, or lobar consolidation suggestive of bacterial pneumonia; 3) Possible COVID-19 pneumonia: features of COVID-19 pneumonia associated with predominant ndings of alternative diagnoses (e.g., unilateral pulmonary lesions due to lung cancer, pulmonary bilateral metastatic nodules), 4) Non-COVID-19 pneumonia: no typical signs of COVID-19 with features suggestive of alternative diagnoses (e.g., unilateral pulmonary lesions due to lung cancer, reticular changes secondary to interstitial lung disease).


Background
Coronavirus disease  was rst identi ed in Wuhan, China, in December 2019 and is now on its second wave. Genetic sequencing of the virus determined that it is a beta coronavirus named severe acute respiratory syndrome-related coronavirus 2 (SARS-CoV-2) [1]. Although most patients have a favorable prognosis, pneumonia, and severe hypoxemia secondary to SARS-CoV-2 infection can lead to acute respiratory failure (ARF) and death [2]. Elderly male patients with comorbidities such as obesity, hypertension, diabetes, cardiac disease, and neoplasm also have an increased risk for severe disease and death and need distinct management and higher surveillance levels [3][4][5][6]. Management remains suboptimal with high mortality rates, particularly among patients admitted to the intensive care unit (ICU).
The integration of all available data from each patient has proven crucial in the management of diffuse parenchymal lung disease (DPLD). Indeed, the international guidelines suggest health professionals with experience in DPLD be involved in patient diagnosis and management in a multidisciplinary approach to achieve the most con dent diagnosis and optimize treatment [7]. The sudden onset and rapid spread of the COVID-19 pandemic with the high number of infections and deaths have led to a global health emergency. Since this was an unknown disease, the priority has been stemming the infection, which inevitably has limited interaction among experts. As COVID-19 is an extremely complex disease, it would potentially bene t from a multidisciplinary approach even during the pandemic. The hypothesis of this study was therefore that a multidisciplinary approach involving specialists experienced in DPLD (pulmonologists, radiologists, and pathologists) may improve the diagnosis and management of patients with COVID-19 pneumonia, following a decisionmaking approach similar to what is used in DPLD.

Study subjects
The present study was a critical re-evaluation of deceased patients by an expert team of specialists (pulmonologists, radiologists, and pathologists) routinely involved in multidisciplinary meetings of mainly DPLD [8] but also with robust experience in COVID-19 diagnosis and management [9,10]. We retrospectively studied 27 patients who consecutively died from March to May 2020 in our hospital of severe COVID-19 pneumonia as the main diagnosis made by nonpulmonologists (i.e. emergency room clinicians, general practitioners, specialists in infectious diseases, and anaesthesiologists).
At autopsy, the whole lungs were macroscopically examined, and a small sample was taken from the most representative area of lung injury. The sample was in part preserved in RNA later and processed for molecular analyses (see below for molecular processing details) and in part xed in formalin for routine histology. Pathological features suggestive of COVID-19 pneumonia (alveolar injury as well as vascular lesions) were quantitatively described using a scoring system, as previously reported [10]. Other associated lesions (neoplasia, infectious diseases, aspiration pneumonia, etc.) were also reported. To con rm the pathological diagnosis of COVID-19 pneumonia, the fragment preserved in RNA later was also processed by real time reverse transciptase-polymerase chain reaction (RT-PCR) for SARS-CoV2 [SARS-CoV-2 (2019-nCoV) Centers for Disease Control and Prevention (CDC) Emergency Use Authorized (EUA) Authorized qPCR probe assay primer/probe mix]. An additional fragment was analyzed by culture, as previously described [11]. In order to de ne the levels of certainty for a diagnosis of COVID-19 pneumonia, two expert pathologists (F.C., F.P.) scored all cases independently and blinded to clinical and autopsy data. During the autopsy, additional fragments were sampled from both lungs (at least 20 samples for each case) and systematically analyzed, as previously described [12].
Clinical evaluation was performed by two experienced pulmonologists (P.S., E.B.) based on the following data: past and recent medical history including comorbidities, respiratory and systemic signs and symptoms (type and duration) before hospital admission, imaging, laboratory ndings, gas exchange values (FiO2, pO2, and pO2/FiO2) and their changes during hospitalization, and oxygen supplementation. Based on this data, patients were classi ed as follows: 1) De nite COVID-19 pneumonia: clinical ndings typical of COVID-19 such as severe acute respiratory illness (i.e., fever, cough, shortness of breath, hypoxemia) in the absence of an alternative diagnosis that could explain the clinical presentation [13,14]; 2) Probable COVID-19 pneumonia: features of COVID-19 pneumonia associated with ndings suggestive of alternative diagnoses (e.g., pleural effusion, clinical and laboratory ndings in keeping with heart failure, or signs of bacterial pneumonia), 3) Possible COVID-19 pneumonia: features of COVID-19 pneumonia associated with prevalent ndings consistent with alternative etiologies (e.g., lung cancer, pulmonary metastases, pulmonary edema, heart failure), 4) Non-COVID-19 pneumonia: absence of typical signs/symptoms and laboratory ndings of COVID-19 pneumonia in the presence of features consistent with alternative diagnoses (e.g., neoplasm, interstitial lung disease, ischemic heart disease, pulmonary edema).
With regard to the radiological assessment, all chest X-rays and, when available, chest computed tomography (CT) images were assessed by two expert thoracic radiologists (C.G., A.F.). According to the radiological ndings, patients were classi ed as follows: 1) De nite COVID 19 pneumonia: typical ndings of COVID-19 pneumonia, such as bilateral ground-glass opacities and/or consolidations [15], without any signs of alternative diagnoses; 2) Probable COVID-19 pneumonia: features of COVID-19 pneumonia associated with abnormalities such as pleural effusion, cardiomegaly, or Kerley B lines suggestive of cardiac failure, or lobar consolidation suggestive of bacterial pneumonia; 3) Possible COVID-19 pneumonia: features of COVID-19 pneumonia associated with predominant ndings of alternative diagnoses (e.g., unilateral pulmonary lesions due to lung cancer, pulmonary bilateral metastatic nodules), 4) Non-COVID-19 pneumonia: no typical signs of COVID-19 with features suggestive of alternative diagnoses (e.g., unilateral pulmonary lesions due to lung cancer, reticular changes secondary to interstitial lung disease).
Data regarding demographics, smoking history, symptoms, comorbidities, treatment, disease duration, serology, radiological and pathological ndings were included in a dedicated database in REDCap. Informed consent was granted by a relative/legal representative of each deceased patient. The study was approved by the local clinical institutional review Board.

Study design
To evaluate whether the contribution of pulmonologists, radiologists, and pathologists individually and/or in combination, could change the diagnosis originally made by non-pulmonologists, we planned a three-step process, modifying the methodology previously used in the evaluation of patients with DPLD [16].
Brie y, in the rst step, two pulmonologists (P.S., E.B.) and two radiologists (C.G., A.F.) independently reviewed clinical and radiological data for each patient, without pathological data, and recorded their individual diagnoses and con dence levels. In the second step, pulmonologists and radiologists discussed their diagnosis and again recorded their individual or shared (in case of disagreement) con dence level. During the third step, pathologists entered the arena and reported the pathological diagnosis performed on a single lung fragment. The nal diagnosis derived from the whole lung examination and full organ autopsy and was considered the diagnostic gold standard. Virtual meetings via the Zoom platform were set to allow pulmonologists, radiologists, and pathologists to discuss their interpretation with mutual collaboration (Fig. 1).

Analysis
All patients were evaluated by pulmonologists and pathologists. One case was not evaluated by radiologists due to lack of radiological data. Specialist scores (for single specialist and for combinations of different specialists) were compared with the full autopsy diagnosis, that was considered to be the true diagnosis, and were recorded as "correct" or "wrong". Probability of a correct diagnosis (95% con dence interval) was calculated with the method of Wilson using the binconf function of the R package {Hmisc} [17]. To explore the effectiveness of a multidisciplinary diagnosis we compared specialist diagnoses with the non-pulmonologist using a model based on generalized estimating equations (GEE) [18] which expand the application of generalized linear models, providing a framework for analyzing correlated data, especially from repeated measures studies where multiple observations are collected from a speci c sampling unit [19]. In particular, we used a rst-order autoregressive correlation structure and a robust standard error estimation to t our small sample size.
The R package {geepack} was used for the analysis [20]. We exponentiated GEE results to obtain an odds ratio (95% CI) for each specialist (or combination of specialists) on their ability to formulate a correct diagnosis. All analyses and plotting were conducted on R software v.4.0.2 [21]. The full code used for the analysis is available upon request.

Multistep process and interobserver agreement
During the rst step (Individual diagnosis by Radiologists and Pulmonologists), pulmonologists categorized 11 cases as de nite (41%), 9 cases as probable (33%), 3 cases as possible (11%), and 4 as non-COVID-19 pneumonia (15%). Radiologists classi ed 11 patients as de nite (42%), 11 patients as probable (42%), 2 patients as possible (8%) and 2 patients as non-COVID-19 pneumonia (8%). The radiological data of one patient was not available. The overall diagnoses with their corresponding level of con dence are reported in Table 2. During the second step (Discussion between Radiologists and Pulmonologists), a con dent diagnosis was reached in 23 out of 26 cases (88%) with de nite COVID-19 pneumonia in 8 cases (30%). Following discussion, the diagnosis was changed in 6 cases, 3 changes for each specialist group (changes indicated with arrows in Table 2).
The comparison between the diagnosis made by each specialist and the diagnosis made by the team following discussion and with the availability of the autopsy data showed that the ability to formulate a correct diagnosis increased strikingly from a non-pulmonologist to expert specialists, becoming progressively more accurate at different steps ( Table 4). The GEE model showed that every single specialist was able to make a signi cantly more accurate diagnosis than a non-pulmonologist. The highest level of accuracy was achieved by the combination of 3 expert specialists (p = 0.0003) who made diagnoses that were about 35 times more accurate [OR: 35.2 (95%CI: 5.07-244)] than a single non-pulmonologist (Table 5, Fig. 3). Indeed, in only 1 case (4%, case number 10) the multidisciplinary team diagnosis was wrong when compared to the gold standard (full autopsy diagnosis) ( Table 5, Fig. 3). In contrast, the diagnosis formulated by a non-pulmonologist was incorrect in over half of the cases (59%), whereas the diagnosis formulated by a pulmonologist or a thoracic radiologist was not correct in 7 (26%) and 6 cases (23%), respectively. After discussion, radiologists and pulmonologists incorrectly diagnosed ve cases (19%). After a full autopsy and whole lung examination, pathologists misinterpreted two cases (7%).

Discussion
In this study, we showed that the diagnostic accuracy of a multidisciplinary approach involving dedicated DPLD physicians is signi cantly higher than that of non-pulmonologists in a subset of patients infected by SARS-CoV-2 who died at the University Hospital of Padova and who underwent autopsy. We demonstrated that the dynamic interaction among DPLD experts in uenced the level of con dence for the nal diagnosis, which improved step by step. In two cases (8%), the diagnosis of COVID-19-related death was incorrect ( nal diagnosis: no COVID-19 pneumonia), while in three cases (11%), COVID-19 pneumonia was only a marginal feature compared to other pathological lesions. Thus, in 19% of cases the diagnosis was mainly incorrect with consequent inappropriate patient management. This was the case in two patients, one with severe aspiration pneumonia and the other with carcinomatous lymphangitis who would have required different monitoring and management of care. Inappropriate treatment might have impacted on patient survival and outcome.
The global spread of the SARS-CoV-2 infection was quite unexpected, rapidly leading to a worldwide health emergency. As with any pandemic, patient care has been affected by sta ng shortages, a chaotic work environment, and high levels of clinician stress. Clinicians had no choice but to provide care in an extraordinary setting. Moreover, the ICUs rapidly became saturated, and their overcrowding led to the recruitment of non-specialist medical staff, potentially exposing critically ill patients to mismanagement. Based on this distressing experience, COVID-19 health care should be planned adequately during the current second global wave. Today, the challenge is to establish a correct diagnosis taking into consideration several pathological conditions that may mimic and/or overlap with COVID-19 pneumonia, with the aim of optimizing patient management and, consequently, reducing mortality. Although our study consisted of a retrospective analysis (i.e., "a backward path by an expert team"), we believe a multidisciplinary approach involving specialists with experience in DPLD diagnosis and management can be highly bene cial to patient care. The multidisciplinary evaluation has become the diagnostic gold standard for DPLD, as it improves diagnostic con dence and interobserver agreement compared to individual components of the multidisciplinary team in isolation [16,22], as was the case in our study.
An expert team should be involved in patient evaluation at the very time of hospital admission, particularly when patients are fragile and have severe respiratory failure. The chaotic work environment and the stressful conditions of emergency medical staff, which may make a face-to-face multidisciplinary approach nonrealistic, might be successfully overcome by using newer digital technologies. Indeed, during the COVID-19 pandemic, multi-specialist meetings have been suspended and converted into virtual meetings, as occurred in our case.
In the multidisciplinary team of DPLD specialists, radiologists play a key role in that HRCT is largely recognized as a very sensitive and highly speci c tool [22][23][24]. During the early phase of the pandemic, CT was seldom performed in COVID-19-positive patients for safety reasons [25,26]. Although chest X-ray proved to be an accurate and reliable method to assess patients with COVID-19, even allowing the development of dedicated scores [CARE referral score] [26], CT plays a crucial role in recognizing alternative diagnoses, especially in patients with pre-existing pulmonary diseases [27][28][29]. The use of a diagnostic modality other than the gold standard may account for the higher agreement between pathologists and pulmonologists than radiologists in cases of partial agreement. Learning from the di culties encountered in the rst wave of the pandemic, most hospitals worldwide have recently adopted organizational models, which guarantee safe pathways to CT scanners that will surely increase the use of this technique and are expected to have a signi cant impact on the quality of the delivered care [30,31].
In our study, as expected, pathologists showed the highest level of con dence between the rst diagnostic impression on a single lung fragment compared to the nal diagnosis on whole lung examination, with an incorrect diagnosis being made in only two cases. The lung fragments used by pathologists to perform the rst diagnosis were similar in size to those obtained by video-assisted thoracic surgery (VATS) that is suggested to be the gold standard tool for the histological diagnosis of DPLD/ILD [24]. Invasive procedures such as VATS carry a high risk of mortality, particularly in patients with severe respiratory dysfunction and under mechanical ventilation [32]. During the SARS-CoV2 pandemic, invasive diagnostic procedures involving sampling of the lung parenchyma were discouraged. However, given the critically important contribution that pathologists could provide in the diagnosis of COVID19 pneumonia, minimally invasive procedures, such as transbronchial lung biopsy/cryobiopsy could be reconsidered in the diagnostic work-up of COVID-19 pneumonia. This is in line with recent expert recommendations [33] suggesting that bronchoscopy can be safely performed in patients with COVID-19, prioritizing minimization of the risk of viral transmission.
Information coming from a full autopsy of COVID-19 patients with the evaluation of numerous lung samples was considered the gold standard for nal diagnosis in our case series. Data provided by the most recent autopsy studies have been crucial in improving our knowledge of the pathological substrates of COVID-19. Indeed, because of the contribution of autopsy studies, COVID-19 pneumonia is now recognized as a complex disease involving not only the lung parenchyma but also the vascular compartment with features that include vasculitis, angiogenesis, capillaritis, and micro/macrothrombi [10,12,34].
The present study has several limitations. First, the study is monocentric, and the study population is relatively small. However, despite this, we were able to implement a GEE model that is robust and provides reliable results even with small sample sizes. Moreover, this is one of the largest monocentric European case series wherein the same lung sampling methodology and analysis was consistently applied. Declarations