Power calculations for cluster-RCT in hospital infection control (PPE for Covid19)

Given the current COVID19 pandemic, having an RCT that could evaluate efficacy of PPE (personal protective equipment) would be of value to determine safety needs of Health care workers.

However, previous sample calculations for RCT for infection control required randomization of hundred of hospitals to detect a noticeable difference, making feasibility of such studies difficult.

@eliowa made me aware of this, as seen in this Jama study

Briefly, using event rates of MRSA infections from previous trials, this simulation study showed that their is a large coefficient of variance between clusters. When using the estimated CV with additional parameters such as (1) daily incidence rate of infection, (2) cluster size, (3) followup period, etc the authors were able to calculate different sample size requirements.

For Example:
“For MRSA acquisition, using the calculated CV of 0.58 and assuming a mean of 7 patients at risk for MRSA acquisition per day per cluster, 50 clusters in total (25 in the intervention group and 25 in the control group) are required to observe a 30% decrease in MRSA acquisition daily rate, assuming a 1-year study with a type I error rate of 0.05 and a type II error rate of 0.20. If the effectiveness of the intervention were 10%, 540 clusters would be needed.”

My question:
The event rate for MRSA noted in prior calculations are much lower (~7 pts/day) then the volume seen currently at hospitals wtih COVID19 patients (20+ pts/day). Given this difference in event rate, would their be a noticeable difference in the CV and sample size requirements for a Hospital Cluster-RCT?

Is this study type, or an alternate study design, feasible to test efficacy of PPE? Or would sample requirements still be too high for a reasonable study?


  1. https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2752354
  2. Jama Supplement (methods to estimate CV)
  3. Buggs Trial
  4. Twitter thread

Raj, it’s unclear to me how “testing the efficacy of PPE” would be in any way helpful. Can you position the sort of studies you envision within a meaningful (and ethical!) decision-analytic framework?


I agree that “testing PPE” honestly doesn’t make much sense with regards to patient care. We should get all docs PPE at all times. The OR of HC workers getting infections compared to regular individuals is very high.

However, their is a large degree of potential spread of infection among HC workers themselves, as opposed to getting infected from patients directly (i.e., in hospital community spread). Certain protocols, such as limiting visitors, etc are accepted and implemented. Other protocols are more uncertain.

Should all hospital doors stay open or automate to avoid touching doorknobs? Should everyone have an assigned computer rather than shared keyboards? Should all workers wear regular masks all day, or N95 masks? Even if we think the latter is best, given resource limitations, would it be ethical to randomize a study looking at both?

Hopefully this clarifies the question. It’s more about infection control protocols within the hospital, as opposed to infection control with patient care (although the lack of appropriate PPE here is truly a devastating failure).


Why not examine the outliers (e.g., hospitals with superb infection control) first? Excellent infection-control policies and procedures will surely have the characteristics of an engineered technology, arising out of rapid prototyping and component-wise experimentation & debugging—all guided by substantive theorizing. (Cf. Platt’s strong inference.) This point about engineering applies quite broadly in medical science:

RCT fever has gotten the better of everyone’s common sense. Biostatistician trialists now reflexively inflate a small point about unmeasured confounding into a vast claim that threatens to swallow medical epistemology whole. Epistemically, the RCT is a desperate measure, and ought to be our last resort!

But biostatisticians may consider that they, too, have the experience of working like engineers—when they write statistical software!

1 Like

I agree with your thoughts. I think we can have sufficiently strong casual Inference based on high quality observational data of successful policies.

Alas, that is not always a winning argument. Many infection control specialist (ID doctors!), question role of confounders (community spread), etc. This is seen partly in the droplet vs airborne debate.

I suppose this is a topic that we will be able to answer I time with retrospective comparisons.

1 Like

No, No, No. Say it ain’t so.


Consider what an epistemically desperate state true equipoise would be. I would argue, however, that in practice ‘equipoise’ has more the status of a desperate lie: