Retrospective database analysis of 1,669 patients in the US showing OR 1.81,
p = 0.01. Confounding by indication is likely.
COVID-19 was determined via PCR+ results, therefore authors include patients asymptomatic for COVID-19, but in the hospital for other reasons. While authors adjust for severity, the method used is very poor. 93.5% of patients are classified as "mild", which is patients with no documented care in a critical care unit within 8 hours of admission. Therefore almost all patients are in the same category, and those in a different category may be due to symptoms unrelated to COVID-19. Lower bias toward male patients in the control group also agrees with the hypothesis that the control group is made up of more people that were in hospital for another reason.
Since the analysis covers the initial period of the pandemic in the USA, it is likely that HCQ was used more often earlier in the analysis period when treatment protocols were considerably worse. It's unclear why the analysis only considers patients up to April 27, when the manuscript was submitted in October.
Sands et al., 1/1/2021, retrospective, database analysis, USA, North America, peer-reviewed, 10 authors.
risk of death, 69.9% higher, RR 1.70, p = 0.01, treatment 101 of 973 (10.4%), control 56 of 696 (8.0%), odds ratio converted to relative risk.
Effect extraction follows pre-specified rules prioritizing more serious
outcomes. For an individual study the most serious outcome may have a smaller
number of events and lower statistical signficance, however this provides the
strongest evidence for the most serious outcomes when combining the results of
many trials.