Faculty Bibliography

OBJECTIVES/OBJECTIVE:To evaluate the impact of ICU surge on mortality and to explore clinical and sociodemographic predictors of mortality. DESIGN/METHODS:Retrospective cohort analysis. SETTING/METHODS:NYC Health + Hospitals ICUs. PATIENTS/METHODS:Adult ICU patients with coronavirus disease 2019 admitted between March 24, and May 12, 2020. INTERVENTIONS/METHODS:None. MEASUREMENTS AND MAIN RESULTS/RESULTS:Hospitals reported surge levels daily. Uni- and multivariable analyses were conducted to assess factors impacting in-hospital mortality. Mortality in Hispanic patients was higher for high/very high surge compared with low/medium surge (69.6% vs 56.4%; p = 0.0011). Patients 65 years old and older had similar mortality across surge levels. Mortality decreased from high/very high surge to low/medium surge in, patients 18-44 years old and 45-64 (18-44 yr: 46.4% vs 27.3%; p = 0.0017 and 45-64 yr: 64.9% vs 53.2%; p = 0.002), and for medium, high, and very high poverty neighborhoods (medium: 69.5% vs 60.7%; p = 0.019 and high: 71.2% vs 59.7%; p = 0.0078 and very high: 66.6% vs 50.7%; p = 0.0003). In the multivariable model high surge (high/very high vs low/medium odds ratio, 1.4; 95% CI, 1.2-1.8), race/ethnicity (Black vs White odds ratio, 1.5; 95% CI, 1.1-2.0 and Asian vs White odds ratio 1.5; 95% CI, 1.0-2.3; other vs White odds ratio 1.5, 95% CI, 1.0-2.3), age (45-64 vs 18-44 odds ratio, 2.0; 95% CI, 1.6-2.5 and 65-74 vs 18-44 odds ratio, 5.1; 95% CI, 3.3-8.0 and 75+ vs 18-44 odds ratio, 6.8; 95% CI, 4.7-10.1), payer type (uninsured vs commercial/other odds ratio, 1.7; 95% CI, 1.2-2.3; medicaid vs commercial/other odds ratio, 1.3; 95% CI, 1.1-1.5), neighborhood poverty (medium vs low odds ratio 1.6, 95% CI, 1.0-2.4 and high vs low odds ratio, 1.8; 95% CI, 1.3-2.5), comorbidities (diabetes odds ratio, 1.6; 95% CI, 1.2-2.0 and asthma odds ratio, 1.4; 95% CI, 1.1-1.8 and heart disease odds ratio, 2.5; 95% CI, 2.0-3.3), and interventions (mechanical ventilation odds ratio, 8.8; 95% CI, 6.1-12.9 and dialysis odds ratio, 3.0; 95% CI, 1.9-4.7) were significant predictors for mortality. CONCLUSIONS:Patients admitted to ICUs with higher surge scores were at greater risk of death. Impact of surge levels on mortality varied across sociodemographic groups.

To explore demographics, comorbidities, transfers, and mortality in critically ill patients with confirmed severe acute respiratory syndrome coronavirus 2.

New York City (NYC) has emerged as the global epicenter for the COVID-19 pandemic. The NYC Public Health System (NYC Health +Hospitals, NYC H + H) was key to the city's response because its vulnerable patient population was disproportionately affected by the disease. As cases rose in the city, NYC H+H carried out plans to greatly expand critical care capacity. Primary ICU spaces were identified and upgraded as needed, while new ICU spaces were created in emergency departments (EDs), procedural areas, and other inpatient units. Patients were transferred between hospitals in order to reduce strain. Critical care staffing was supplemented by temporary recruits, volunteers, and military deployments. Supplies to deliver critical care were monitored closely and obtained as needed to prevent interruptions. An ED action team was formed to ensure that the experience of frontline providers was informing network level decisions. The steps taken by NYC H+H greatly expanded its capacity to provide critical care during an unprecedented surge of COVID-19 cases in NYC. These steps, along with lessons learned, could inform preparations for other health systems during a primary or secondary surge of cases. [Editor's Note: This Fast Track Ahead Of Print article is the accepted version of the manuscript. The final edited version will appear in an upcoming issue of Health Affairs.].

In this work, we utilize a combination of free-text and structured data to build Acute Respiratory Distress Syndrome(ARDS) prediction models and ARDS phenotype clusters. We derived 'Patient Context Vectors' representing patientspecific contextual ARDS risk factors, utilizing deep-learning techniques on ICD and free-text clinical notes data. The Patient Context Vectors were combined with structured data from the first 24 hours of admission, such as vital signs and lab results, to build an ARDS patient prediction model and an ARDS patient mortality prediction model achieving AUC of 90.16 and 81.01 respectively. The ability of Patient Context Vectors to summarize patients' medical history and current conditions is also demonstrated by the automatic clustering of ARDS patients into clinically meaningful phenotypes based on comorbidities, patient history, and presenting conditions. To our knowledge, this is the first study to successfully combine free-text and structured data, without any manual patient risk factor curation, to build real-time ARDS prediction models.