A JAMA paper was just published that reports results from a large observational cohort study in NYC area. The first author is Eli Rosenberg, and epidemiologist in the Department of Epidemiology and Biostatistics at the University of Albany, NY.
This study was a very well designed cohort study. It covered a group of 25 hospitals that together represent 88% of COVID-19 patients in the greater NYC area. They considered individuals hospitalized between March 15 and 28 to allow enough time for many patients to recover or die. They obtained the patient information for a random sample of 30% of patients from each hospital, also stratified by whether cases were resolved by 3/31, 4/12 or not yet resolved after 4/12. All totaled, this led to 1438 patients, 735 who received HCQ+Az, 271 HCQ alone, 211 Az alone, and 221 received neither drug.
There were major statistically significant differences in the characteristics of the patients in each drug arm,, with more severe patients with poorer prognosis in the HCQ and HCQ+Az arm, with more patients who were male, black, obese, with diabetes, congestive heart failure, high respiratory rate, low blood oxygen, high AST or ALT, or abnormal chest imaging. These differences are why it is not worthwhile to look at unadjusted results, and is why randomized controlled trials are so important for detecting causal treatment effects.
However, this paper does not focus only on unadjusted comparisons -- its primary analysis looks at comparisons after doing covariate adjustment for these factors in the model. After covariate adjustment, there were not statistically significant differences between the groups, but the point estimates still suggested no measurable benefit in time to death for HCQ arm (HR=1.08) or HCQ+Az (1.35) relative to the untreated arm (the Az alone arm had HR=0.56).. One would hope to at least see a trend towards benefit (i.e. HR<1) in the HCQ arms if it was truly helping, but this was not seen here.
Importantly, the rate of cardiac arrest was higher in the HCQ and HCQ+Az arm, and again they performed analyses adjusting for covariates and found that even after adjustment, the HCQ+Az had a statistically significant higher risk of cardiac arrest than the arm with no HCQ or Az.
Overall, this confirms what we have seen in other studies, that when looking across all hospitalized patients, there is not evidence that HCQ or HCQ+Az has benefit in preventing death, and that there is higher risk of cardiac toxicities that must be monitored, which makes it crucial that HCQ only be given under close medical guidance.
The shine is clearly off the HCQ treatment from the initial hype. These results are not promising for those who suggest HCQ should be a front line therapy for hospitalized COVID-19 patients.
However, the randomized studies are still needed, since it is not clear whether the covariate adjustment was sufficient to adjust for the major confounding factors between the treatment arms. There are many ongoing randomized trials involving HCQ, and those results should come in during the coming months and provide more definitive results of its potential benefit and risks.
Comments