Study population and setting
This study estimated changes by month in COVID-19 mortality risk among 5,121 hospitalized patients from a single health system in New York City from March 1 to August 31, 2020. Included patients were 18 years and older, with laboratory-confirmed SARS-CoV-2 infection; 229 hospitalizations (4.4%) were repeated hospital stays from 208 patients. Mortality was defined as death in hospital or discharge to hospice. Data were taken from electronic medical records. A multivariable logistic regression model (with a range of demographic and clinical variables including age, sex, race, BMI, comorbidities, admission oxygen saturation, admission d-dimer concentrations, and admission CRP concentrations) were used to calculate monthly adjusted mortality rates. Average marginal effects for each month were estimated with a second model that included month as a covariate.
Summary of Main Findings
The majority of hospitalizations (53%) occurred from late March to mid-April 2020, and 79% of hospitalizations occurred by the end of April. The median duration of hospital stay for those who died or were discharged to hospice was 8 days. Over time, the median age, the proportion of males, and the proportion with comorbidities decreased. Adjusted mortality declined in each successive month from 25.6% in March to 7.6% in August. The average marginal effect of each month was increasingly negative relative to March, reaching -18.2% in August. Restricting the study population to those with at least 3 days in hospital produced similar results; restricting to those with a principal diagnosis of COVID-19, sepsis, or respiratory disease produced similar but attenuated results.
This straightforward analysis used estimates of average marginal effect, which have intuitive interpretations (i.e., % change in probability of death for a given month). Follow-up was complete on the vast majority of patients.
Changing criteria for hospitalization over time could bias results; for example, if patients with less severe disease were admitted in later months, outcomes would appear more favorable. The clinical covariates in multivariable models may not have sufficiently captured patients’ mortality risk status. It is not clear how covariates measuring age, race/ethnicity, BMI, and smoking history were defined and modeled (e.g., categorical vs. continuous). Comorbidities were assessed via binary indicator variables (e.g., yes vs. no) and models are thus subject to residual confounding. The study population is from a single health system, which limits generalizability– for example, New York’s hospital system was particularly overwhelmed during late March and April, and this burden may have contributed to higher mortality during those months.
Declines in mortality among hospitalized patients with COVID-19 have been observed in different settings, but this is the one of the first studies to adjust for patient characteristics and clinical parameters, providing evidence that the mortality risk decline in New York City is not simply due to a differing mix of patients (e.g., younger with fewer comorbidities).
This review was posted on: 2 November 2020