Vikas I. Parekh

An outbreak of SARS-CoV-2 has led to a global pandemic affecting virtually every country. As of August 31, 2020, globally, there have been approximately 25,500,000 confirmed cases and 850,000 deaths; in the United States (50 states plus District of Columbia), there have been more than 6,000,000 confirmed cases and 183,000 deaths. We propose a Bayesian mixture model to predict and monitor COVID-19 mortality across the United States. The model captures skewed unimodal (prolonged recovery) or multimodal (multiple surges) curves. The results show that across all states, the first peak dates of mortality varied between April 4, 2020 for Alaska and June 18, 2020 for Arkansas. As of August 31, 2020, 31 states had a clear bimodal curve showing a strong second surge. The peak date for a second surge ranged from July 1, 2020 for Virginia to September 12, 2020 for Hawaii. The first peak for the United States occurred about April 16, 2020—dominated by New York and New Jersey—and a second peak on August 6, 2020—dominated by California, Texas, and Florida. Reliable models for predicting the COVID-19 pandemic are essential to informing resource allocation and intervention strategies. A Bayesian mixture model was able to more accurately predict the shape of the mortality curves across the United States than other models, including the timing of multiple peaks. However, given the dynamic nature of the pandemic, it is important that the results be updated regularly to identify and better monitor future waves, and characterize the epidemiology of the pandemic.

Crowding is a major challenge faced by EDs and is associated with poor outcomes. Determine the effect of high ED occupancy on disposition decisions, return ED visits, and hospitalizations. We conducted a retrospective analysis of electronic health records of patients evaluated at an adult, urban, and academic ED over 20 months between the years 2012 and 2014. Using a logistic regression model predicting admission, we obtained estimates of the effect of high occupancy on admission disposition, adjusted for key covariates. We then stratified the analysis based on the presence or absence of high boarder patient counts. Disposition decisions during a high occupancy hour decreased the odds of admission (OR = 0.93, 95% CI: [0.89, 0.98]). Among those who were not admitted, high occupancy was not associated with increased odds of return in the combined (OR = 0.94, 95% CI: [0.87, 1.02]), with-boarders (OR = 0.96, 95% CI: [0.86, 1.09]), and no-boarders samples (OR = 0.93, 95% CI: [0.83, 1.04]). Among those who were not admitted and who did return within 14 days, disposition during a high occupancy hour on the initial ED visit was not associated with a significant increased odds of hospitalization in the combined (OR = 1.04, 95% CI: [0.87, 1.24]), the with-boarders (OR = 1.12, 95% CI: [0.87, 1.44]), and the no-boarders samples (OR = 0.98, 95% CI: [0.77, 1.24]). ED crowding was associated with reduced likelihood of hospitalization without increased likelihood of 2-week return ED visit or hospitalization. Furthermore, high occupancy disposition hours with high boarder patient counts were associated with decreased likelihood of hospitalization.