ABSTRACT
Rationale: Recent advancements in sequencing technologies have led to a substantial increase in the scale and resolution of transcriptomic data. Despite this progress, accessibility to this data, particularly among those who are coming from non-computational backgrounds is limited. To facilitate improved access and exploration of our single-cell RNA sequencing data, we generated several data sharing, mining and dissemination portals to accompany our idiopathic pulmonary fibrosis (IPF), chronic obstructive pulmonary disease (COPD), and lung endothelial cells (Lung EC) cell atlases. Descriptions and links of each website can be found here: https://medicine.yale.edu/lab/kaminski/research/atlas/. Methods: Each interactive data mining website is coded in the R language using the Shiny package and is hosted by Shinyapps.io. Percell expression data for each website is stored on a MySQL database hosted by Amazon Web Services (AWS). Time-associated website engagement statistics and gene query information is collected for each website using a combination of Google Analytics and a gene search table stored on our MySQL database. User exploration of available data is facilitated through several easy-touse visualization tools available on each website. Results: Website usage statistics since the publication of each website shows that 9,772 unique users from 56 countries and five continents have accessed at least one of the three websites. At the time of writing, 300,748 total queries have been made for 15,627 unique genes across the websites. The top five searched genes for the IPF Cell Atlas are CD14, ACE2, ACTA2, IL11 and MUC5B while for the COPD Cell Atlas they are FAM13A, MIRLET7BHG, HHIP, ISM1 and DDT. Finally, the top searched genes for the Lung Endothelial Cell Atlas are BMPR2, PECAM1, EDNRB, APLNR and PROX1. Of note, interaction with the IPF Cell Atlas increased dramatically at the start of the COVID-19 pandemic, with queries for the ACE2 gene, the putative binding receptor for the SARS-CoV-2 virus, increasing substantially at the pandemic's onset in the United States. Conclusions: Usage statistics, gene query information and feedback from users, both within academia and industry, have shown broad engagement with our websites by individuals across computational and non-computational backgrounds. We envision widespread adoption of web-based portals similar to ours will facilitate novel discoveries within these complex datasets and new scientific collaborations.
ABSTRACT
INTRODUCTION: SARS-CoV-2 has been associated with high rates of severe hypoxemic respiratory failure. Severe COVID-19 is characterized by rapid development of acute respiratory distress syndrome (ARDS) requiring mechanical ventilation. ARDS is considered a heterogeneous disorder and the presence of a uniform inciting agent with SARS-CoV-2 allows us to investigate subphenotypes of ARDS. We hypothesized that subphenotypes based on early lung compliance in patients with COVID-ARDS may be associated with disease outcomes including mortality. We sought to test this hypothesis in patients with COVID-ARDS. METHODS: Patients in the Yale New Haven Health System from 3/15/2020 to 5/14/2020 were included if they had a positive SARS-CoV-2 test and required intubation. After exclusion for missing data or transfer from satellite facilities, 140 of patients were included for analysis. Clinical, demographic, ventilator, and laboratory parameters were abstracted from the EMR. To identify ARDS subphenotypes, we implemented unsupervised clustering using a partitioning around medoids (PAM) algorithm on average compliance over the three days following intubation. Clustering was also performed on the NHLBI ALVEOLI cohort for use as comparator. RESULTS: Patients received a median of 6.2 cc/kg of IBW on day 1 and 6.2 on day 3. Plateau pressure was less than 30cm H2O in 81% patients and driving pressure was less than 15 in 69% of patients. Median lung compliances were day 1: 30.4 mL/cm H2O [23.0-36.1];day 2: 28.7 mL/cm H2O [21.6-36.8];and day 3: 29.7 mL/cm H2O [23.2-37.6]. By Berlin criteria, 21% of patients had mild ARDS, 46% moderate, and 26% were severe. 61% of patients were proned. Using PAM, three distinct clusters based on compliance were identified (low [LC], medium [MC], and high [HC]). Median day 1 compliance in HC group was 38.0 mL/cm H2O [33.1-44.2], 29.3 [25.2-32.3] in MC, and 19.5 mL/cm H2O [16.7-22.8] in LC. Compared to the HC group, there were no differences in PEEP, day 1 P/F ratio or tidal volume, and ventilatory ratio. HC group had higher P/F ratio day 3, lower tidal volume day 3, and lower driving pressures. There were no differences in biomarkers, comorbidities, vasopressors, paralytics, or sedatives between groups. However, Kaplan-Meier plot demonstrated higher mortality in the HC group. Cox regression model demonstrated persistence of higher mortality in the HC compared to MC group. These differences were not present in the ALVEOLI cohort. CONCLUSION: In COVID-ARDS, a subphenotype characterized by early high compliance was associated with higher mortality when compared to non-COVIDARDS patients.