re: Coronavirus Disease 2019 (COVID-19) ***W.H.O. DECLARES A GLOBAL PANDEMIC***

quote:

---

The control group was basically mask use. What type we don’t know? Could have been medical masks as well.

---

They did try to track their mask usage, and they did a follow-up analysis with those who exclusively used the medical masks (170 from control) or cloth mask (38 from control) and included them with the other groups and found similar results.

However, this seemed like a wasted opportunity since 245, over 53% of the control group used both, but they excluded them from the analysis. Put to me, they could provide useful information, since this provides more variance in the treatment measure instead of just a binary variable. And since we would expected people who had say 50/50 mask usage, to have a higher risk than the 100% medical mask group but a lower risk than a 100% cloth group, including them would be able to test that and provide more informative results. So basically they didn’t actually have a control group in practice to make any meaningful comparison, and they failed to use the unique differences of that control group to provide meaningful information.

Regardless, maybe this is just a difference between the social sciences publications (my area) and my medical publications but I’m frustrated by the lack of details in their modeling. I’m used to seeing a correlation tables of all the relevant variables to see what the relationships are with the outcomes and multiple models, p-values, the model fit, a and a comparison of model fit, when doing a stepwise analysis.

Furthermore, for an experimental model with at least 2 already published studies, their process for testing and including confounding variables is unusually exploratory and the process itself is odd to say the least.

Specifically, they say they only included variables that has a p-value less than 0.25 in a univariate analysis then did a backward elimination with those that met the criteria.


This seems odd for a couple of reasons: first I n an experimental study, there are often variables that one would want included regardless of their significance for theoretical and practical reasons. For example, I’d mask compliance and/or vaccinations were not significant, that still may need to be included because they should have impact in the outcome and their lack of significance may be due to a lack of power and/or limited variation in those variables for this sample.

Second, I don’t understand why they did a univariate analysis of the variables for initial selection since their are multiple stepwise procedure they could have used starting with the full model, like they did after the initial selection. Furthermore, some variables could be non-significant in the univariate analysis but significant when other variables are included. So they may have been useful to include. On the other hand, some significant variables in the univariate analysis but are extremely non-significant in the multivariate analysis because their initial significance was due to the relationship with another, better variable.

The whole modeling process was just odd. And since they didn’t provide any tables and/or documentation to show the process and the relationships, we have no idea if they even had the best model to evaluate the experiment.

Finally, and maybe I missed why this is the case, but I’m perplexed by the difference in sizes of the groups which we’re supposedly randomized. They had 1,607 participants and 3 groups. So we would expect equal sized groups, (or 2 groups with 536 and 1 with 535 1,607 isn’t divisible by 3), or at least just a percentage or so difference given that there may be some logistical issues to make them as precisely equal as possible. However, the control group had 458 (28.5%) and the medical mask group has 580 (36.1%). That’s a difference of 122 participants or a 7.6% sample difference. That’s such a large difference that doesn’t appear to be anything like randomization and/or due to minor logistical placement issues with the ward clusters. Again, maybe I missed it, but that seems something that should have been addressed very clearly.

In summary, they present it as a Randomized-Control study, but they don’t actually have a control and their randomization seems questionable as well. So it may be neither. Ok top of that, their modeling procedures are questionable and non-transparent. Regardless, I’m comfortable with the conclusion that medical masks are better than cloth masks, which I thought was widely known anyways. Furthermore, the differences are evident in a hospital setting, for obvious reasons.

However, this study does not indicate whether cloth masks are better or worse than no mask at all in a HOSPITAL setting, let alone outside of that setting. So anyone using this study to say they aren’t useful for the general public, is not basing it on any evidence in the study.