For those making decisions on whether to use hydroxychloroquine +/- azithromycin based on this new VA data:

https://www.medrxiv.org/content/10.1101/2020.04.16.20065920v1

Please don’t. There are at least 4 major issues that make the data impossible to interpret:

1/

Immortal time bias: Follow-up began on admission date. For those treated, then, follow-up includes time PRIOR TO TREATMENT INITIATION. Thus, HCQ patients had to survive long enough to be treated, and HCQ+AZT patients had to survive perhaps even longer to be treated.

2/

Confounding by indication: Treated patients were clearly sicker than non-treated patients. This would tend to bias in the opposite direction of the immortal time bias. So, we have the battle of competing biases. Do they cancel out? Who knows?

3/

Timing of covariate measurement relative to treatment initiation: It is unclear WHEN time-varying factors like labs were measured relative to treatment. If variables measured after treatment initiation are adjusted for, could bias results in unpredictable ways.

4/

Missing data- There is a ton of missing data (like >50% missing for some labs). It appears that this dealt with using missing data indicators, which is known to be biased.

5/

Maybe the healthcare community should never have started using HCQ+/- Azithro. That would be reasonable because we never had strong evidence that the drugs work.

However, if you are using these drugs, this study is not a reason to change your practice.