Study of critically ill patients with COVID-19 in New York City.

Cummings and colleagues reported the epidemiology, clinical course, and outcomes of 257 critically ill adults with laboratory-confirmed COVID-19 admitted to two hospitals in New York City. The primary outcome was the rate of in-hospital death, and each patient had at least 28 days of observation. The authors report that, as of April 28, 2020, 101 (39%) of 257 patients had died, 94 (37%) remained hospitalised, four (2%) were transferred to another hospital, and 58 (23%) were discharged alive. Surprisingly, the authors show in figure 1 of their Article a cumulative incidence of in-hospital death of approximately 45% at 28 days. Given the numbers of patients at risk reported below the figure, we have identified that this result is not correct. Apparently, the authors censored the patients discharged alive (n=58) at the day of discharge. This methodological error has led to overestimation of the cumulative incidence of death, and distorted the results of the Cox proportional hazards regression. A fundamental assumption in survival analysis is that censoring should be non-informative—ie, that patients censored have the same survival prospects as those who continue to be followed up. Patients discharged alive should not have been censored; their status should be considered as event-free (ie, alive) throughout the study observation period. This methodological error in the COVID-19 literature is common yet serious. We kindly ask the authors to reanalyse the data, and correctly report the cumulative incidence, and the risk factors of in-hospital mortality, considering the above aspects.