ARDSFlag: An NLP/Machine Learning Algorithm to Visualize and Detect High-Probability ARDS Admissions Independent of Provider Recognition and Billing Codes (preprint)

Acute respiratory distress syndrome (ARDS) is a type of respiratory failure characterized by bilateral pulmonary infiltrates that cannot be explained entirely by cardiogenic pulmonary edema. ARDS is the primary cause of mortality in COVID-19 patients and one of the leading causes of morbidity and mortality in ICUs. Despite its significance and prevalence, the detection of ARDS remains highly variable and inconsistent. In this work, we develop a tool to automate the diagnosis of ARDS based on the Berlin definition to increase the accuracy of ARDS detection using electronic health record (EHR) fields. ARDSFlag applies machine learning (ML) and natural language processing (NLP) techniques to evaluate Berlin criteria by incorporating structured and unstructured data. The output is the ARDS diagnosis, onset time, and severity. We have also developed a visualization that helps clinicians efficiently assess ARDS criteria retrospectively and in real time. The method includes separate text classifiers trained using large training sets to find evidence of bilateral infiltrates in radiology reports (accuracy of 91.9%±0.5%) and heart failure/fluid overload in radiology reports (accuracy 86.1%±0.5%) and echocardiogram notes (accuracy 98.4%±0.3%). A holdout set of 300 cases, which was blindly and independently labeled for ARDS by two groups of clinicians, shows that the algorithm generates an overall accuracy of 89.0%, with a specificity of 91.7%, recall of 80.3%, and precision of 75.0%. Compared with two other ARDS identification methods used in the literature, ARDSFlag shows higher performance in all accuracy measures (an increase of 25.5% in overall accuracy, 6.5% in specificity, 44.2% in recall, 31.7% in precision, and 38.20% in F 1 -score over the best of the two detection methods).

Drivers of Mortality in COVID ARDS Depend on Patient Sub-Type (preprint)

Background: The most common cause of death in people with COVID-19 is acute respiratory distress syndrome (ARDS). ARDS is a heterogeneous syndrome, however subgroups that have been identified among non-COVID-19 ARDS patients do not clearly apply to COVID-19 ARDS patients. Additionally, studies of COVID-19 ARDS have been limited by sample size. Methods: We applied an iterative clustering and machine learning framework to electronic health record data from thousands of hospitalized COVID-19 ARDS patients with the goal of defining and characterizing clinically-relevant COVID-19 ARDS subgroups (phenoclusters). We then applied a supervised model to identify risk factors for hospital mortality for each phenocluster and compared these between phenoclusters and the entire cohort. Findings: Risk factors that predict mortality in the overall cohort of COVID-19 ARDS patients do not necessarily predict mortality in phenoclusters. In fact, some risk factors increase the risk of hospital mortality in some phenoclusters, but decrease in others. Interpretation: These phenocluster-specific risk factors would not have been observed with a single predictive model; therefore, it is critical to: (1) cluster patients based on attributes, and (2) model within a cluster to discover the drivers of mortality.

Trajectories of Hypoxemia & Pulmonary Mechanics of COVID-19 ARDS in the NorthCARDS Dataset (preprint)

Background:  Understanding heterogeneity seen in patients with COVIDARDS and comparing to non-COVIDARDS may inform tailored treatments.  Methods:  A multidisciplinary team of frontline clinicians and data scientists worked to create the Northwell COVIDARDS dataset (NorthCARDS) leveraging over 11,542 COVID-19 hospital admissions. The data was then summarized to examine descriptive differences based on clinically meaningful categories of lung compliance, and to examine trends in oxygenation. Findings:  Of the 1595 COVIDARDS patients in the NorthCARDS dataset, there were 538 (34·6%) who had very low lung compliance (<20ml/cmH 2 O), 982 (63·2%) with low-normal compliance (20-50ml/cmH 2 O), and 34 (2·2%) with high lung compliance (>50ml/cmH 2 O). The very low compliance group had double the median time to intubation compared to the low-normal group (107 hours (IQR 26·3, 238·3) vs. 37·9 hours (IQR 4·8, 90·7)). Overall, 67·5% (n=1049) of the patients died during the hospitalization. In comparison to non-COVIDARDS reports, there were less patients in the high compliance category (2.2%vs.12%, compliance ≥50mL/cmH20), and more patients with P/F ≤ 150 (57·8% vs. 45.6%). No correlation was apparent between lung compliance and P/F ratio. The Oxygenation Index was similar, (11·12(SD 5·67) vs.12·8(SD 10·8)). 1  

STudy of Alteplase for Respiratory Failure in SARS-CoV-2 COVID-19 (STARS): A Vanguard Multicenter, Rapidly Adaptive, Pragmatic, Randomized, Controlled Trial (preprint)

Background: Pulmonary vascular microthrombi have been proposed as a mechanism of COVID19 respiratory failure. We hypothesized that early administration of tissue-plasminogen activator(tPA) followed by therapeutic heparin would improve pulmonary function in these patients.Methods: Adults with COVID-19-induced respiratory failure were randomized May14, 2020-March 3, 2021 in two phases: Phase-1(n=36): control (standard-of-care) vs tPA-Bolus (50mg tPA IV-bolus followed by 7 days of heparin (goal aPTT=60-80s); Phase-2(n=14): control vs tPA-Drip (50 mg of tPA IV-bolus, followed by tPA drip 2mg/hr plus heparin 500U/hour over 24 hours, then heparin to maintain aPTT 60-80s/7 days). The primary outcome was PaO2/FiO2 improvement at 48 hours post-randomization. Secondary outcomes included: PaO2/FiO2 improvement>50% or PaO2/FiO2>=200 at 48hrs(COMPOSITE), ventilator-free days(VFD) and mortality.Findings: Fifty patients were randomized: Phase 1:17 control, 19 tPA-Bolus; Phase 2: 8 control, 6 tPA-Drip. There were no severe bleeding events in intervention groups. In tPA-Bolus patients, PaO2/FiO2 was significantly(p<0.017) higher than baseline at 6 through 168 hours post-randomization; controls experienced no significant improvements. Compared to controls, tPA-Bolus patients showed larger, but non-significant, improvements in PaO2/FiO2 at 48hours[16.9%(-8.3–36.8) vs 29.8%(4.5–88.7),p=0.11], resulting in more patients reaching the COMPOSITE outcome (11.8% vs 47.4%,p=0.03). Controls had less VFD[0.0(0.0–9.0) vs 12.0(0.0–19.0),p=0.11] and higher mortality(41.2% vs 21.1%,p=0.19) than tPA-Bolus patients, although not significantly. tPA-Drip patients did not experience benefit compared to simultaneously enrolled controls.Interpretation: The combination tPA-Bolus+heparin is safe in severe COVID-19 respiratory failure. A Phase 3 study is warranted given promising improvements in oxygenation, VFD, and mortality.Trial Registration: The trial was performed according to the Food and Drug Administration (FDA) Investigational New Drug regulations (IND 149634) and registered with ClinicalTrials.gov (NCT04357730).Funding: This investigator-initiated trial (NCT04357730) was funded by Genentech, Inc.Declaration of Interests: CDB, HBM, EEM, and MBY have patents pending related to both coagulation/fibrinolysis diagnostics and therapeutic fibrinolytics, and are passive co-founders and holds stock options in Thrombo Therapeutics, Inc. HBM and EEM have received grant support from Haemonetics and Instrumentation Laboratories. MBY has previously received a gift of Alteplase (tPA) from Genentech, and owns stock options as a co-founder of Merrimack Pharmaceuticals. CDB, HBM, EEM, JW, NH, DST, AS, and MBY have received research grant funding from Genentech. JW receives consulting fees from Camurus A. B.. All other authors have nothing to disclose.Ethics Approval Statement: All participating trial sites had study approval and oversight from their respective Institutional Review Boards. Due to the nature of the study, which enrolled critically ill patients on mechanical ventilation, informed consent for trial participation was obtained from each patient’s Legally Authorized Representative. An independent Data Safety Monitoring Board (DSMB) oversaw the safety of the trial with mandatory reviews at each interim analysis and for all suspected serious adverse events. Data were stored in a REDCap instrument sponsored by NIH/NCATS Colorado CTSA Grant Number UL1 TR002535.

A Randomized Placebo-Controlled Trial of Sarilumab in Hospitalized Patients with Covid-19 (preprint)

BACKGROUNDSarilumab (anti-interleukin-6 receptor- monoclonal antibody) may attenuate the inflammatory response in Covid-19. METHODSWe performed an adaptive, phase 2/3, randomized, double-blind, placebo-controlled trial of intravenous sarilumab 200 mg or 400 mg in adults hospitalized with Covid-19. The phase 3 primary analysis population (cohort 1) was patients with critical Covid-19 receiving mechanical ventilation (MV) randomized to sarilumab 400 mg or placebo. The primary end point for phase 3 was the proportion of patients with [≥]1-point improvement in clinical status from baseline to day 22. RESULTSFour-hundred fifty-seven (457) and 1365 patients were randomized and treated in phases 2 and 3, respectively. Among phase 3 critical patients receiving MV (n=289; 34.3% on corticosteroids), the proportion with [≥]1-point improvement in clinical status (alive not receiving MV) at day 22 was 43.2% in sarilumab 400 mg and 35.5% in placebo (risk difference [RD] +7.5%; 95% confidence interval [CI], -7.4 to 21.3; P=0.3261), representing a relative risk improvement of 21.7%. Day 29 all-cause mortality was 36.4% in sarilumab 400 mg versus 41.9% in placebo (RD -5.5%; 95% CI, -20.2 to 8.7; relative risk reduction 13.3%). In post hoc analyses pooling phase 2 and 3 critical patients receiving MV, the hazard ratio (HR) for death in sarilumab 400 mg compared with placebo was 0.76 (95% CI, 0.51 to 1.13) overall, improving to 0.49 (95% CI, 0.25 to 0.94) in patients receiving corticosteroids at baseline. CONCLUSIONIn hospitalized patients with Covid-19 receiving MV, numerical benefits with sarilumab did not achieve statistical significance, but benefit may be greater in patients receiving corticosteroids. A larger study is required to confirm this observed numerical benefit. (ClinicalTrials.gov number, NCT04315298)

Trajectories of Hypoxemia & Respiratory System Mechanics of COVID-19 ARDS in the NorthCARDS dataset (preprint)

Rationale The preliminary reports of COVID Acute Respiratory Distress Syndrome (COVIDARDS) suggest the existence of a subset of patients with higher lung compliance despite profound hypoxemia. Understanding heterogeneity seen in patients with COVIDARDS and comparing to non-COVIDARDS may inform tailored treatments. Objectives To describe the trajectories of hypoxemia and respiratory compliance in COVIDARDS and associations with outcomes. Methods A multidisciplinary team of frontline clinicians and data scientists created the Northwell COVIDARDS dataset (NorthCARDS) leveraging over 11,542 COVID-19 hospital admissions. Data was summarized to describe differences based on clinically meaningful categories of lung compliance, and compared to non-COVIDARDS reports. A sophisticated method of extrapolating PaO2 from SpO2, as well estimating FiO2 from non invasive oxygen delivery devices were utilized to create meaningful trends of derived PaO2 to FiO2 (P/F). Measurements and Main Results Of the 1595 COVIDARDS patients in the NorthCARDS dataset, there were 538 (34·6%) who had very low lung compliance (<20ml/cmH2O), 982 (63·2%) with low-normal compliance (20-50ml/cmH2O), and 34 (2·2%) with high lung compliance (>50ml/cmH2O). The very low compliance group had double the median time to intubation compared to the low-normal group (107 hours(IQR 26·3, 238·3) vs. 37·9 hours(IQR 4·8, 90·7)). Oxygenation trends have improved in all groups after a nadir immediately post intubation. The P/F ratio improved from a mean of 109 to 155, with the very low compliance group showing a smaller improvement compared to low compliance group. The derived P/F trends closely correlated with blood gas analysis driven P/F trends, except immediately post intubation were the trends diverge as illustrated in the image. Overall, 67·5% (n=1049) of the patients died during the hospitalization. In comparison to non-COVIDARDS reports, there were less patients in the high compliance category (2.2%vs.12%, compliance ≥50mL/cmH20), and more patients with P/F ≤ 150 (57·8% vs. 45.6%). No correlation was apparent between lung compliance and P/F ratio. The Oxygenation Index was similar, (11·12(SD 5·67)vs.12·8(SD 10·8)). Conclusions Heterogeneity in lung compliance is seen in COVIDARDS, without apparent correlation to degree of hypoxemia. Notably, time to intubation was greater in the very low lung compliance category. Understanding ARDS patient heterogeneity must include consideration of treatment patterns in addition to trajectories of change in patient-level data and demographics.

Comparative Survival Analysis of Immunomodulatory Therapy for COVID-19 'Cytokine Storm': A Retrospective Observational Cohort Study (preprint)

BackgroundCytokine storm is a marker of COVID-19 illness severity and increased mortality. Immunomodulatory treatments have been repurposed to improve mortality outcomes. MethodsWe conducted a retrospective analysis of electronic health records across the Northwell Health system. COVID-19 patients hospitalized between March 1, 2020 and April 15, 2020, were included. Cytokine storm was defined by inflammatory markers: ferritin >700ng/mL, C-reactive protein >30mg/dL, or lactate dehydrogenase >300U/L. Patients were subdivided into six groups -no immunomodulatory treatment (standard of care) and five groups that received either corticosteroids, anti-interleukin 6 (IL-6) antibody (tocilizumab) or anti-IL-1 therapy (anakinra) alone or in combination with corticosteroids. The primary outcome was hospital mortality. ResultsThere were 3,098 patients who met inclusion criteria. The most common comorbidities were hypertension (40-56%), diabetes (32-43%) and cardiovascular disease (2-15%). Patients most frequently met criteria with high lactate dehydrogenase (74.8%) alone, or in combination, followed by ferritin (71.4%) and C-reactive protein (9.4%). More than 80% of patients had an elevated D-dimer. Patients treated with a combination of tocilizumab and corticosteroids (Hazard Ratio [HR]: 0.459, 95% Confidence Interval [CI]: 0.295-0.714; p<0.0001) or corticosteroids alone (HR: 0.696, 95% CI: 0.512-0.946; p=0.01) had improved hospital survival compared to standard of care. Corticosteroids and tocilizumab was associated with increased survival when compared to corticosteroids and anakinra (HR: 0.612, 95% CI: 0.391-0.958; p-value=0.02). ConclusionsWhen compared to standard of care, corticosteroid and tocilizumab used in combination, or corticosteroids alone, was associated with reduced hospital mortality for patients with COVID-19 cytokine storm.