Helen MacDonald

But do you think some of those models are less to predict for individuals and more to inform policy decisions, perhaps around escalation of care, looking at how likely you are to benefit from admission to hospital or the use of a particular intervention to try and sort of modify your outcome?

So what's our message from this?

I think is that we have to be very wary of the concept of prediction models perhaps in COVID-19 or maybe Carl suggesting for life in general.

I think the whole issue of face masks and other physical barriers to protect against transmission of COVID-19 just seems to continue with lots of questions.

Should the public wear face masks?

what face masks and other protective equipment to healthcare workers need and which masks or which items are most suitable for which activities.

And there's a piece just out in the BMJ headed by a team of international clinical authors, Manuel Schmidt and Tricia Greenhalgh,

about moving from the evidence on masks or uncertainty around that evidence towards making policy.

And they are talking about public use.

But I think some of the principles that they mentioned in their piece could be useful in considering recommendations for other settings.

So just purely looking at the evidence, they say that there are no trials on face masks and COVID so far, which is maybe not surprising.

But there is evidence from trials and observational studies on masks and other barriers in influenza-like illnesses.

So some indirect evidence and systematic reviews of those, including two new ones that are up on the BMJ's preprint server today.

suggests that there are few clear answers.

That's short-cutting an awful lot of data that's in those studies.

Many of the included studies have got some design limitations.

There are wide confidence intervals and there are varying answers from trials and observational data.

But what I found interesting in...

Trisha's piece is developing the thoughts around some of the other considerations that you might take into account when building recommendations for particular groups of people.

So there are some practical issues.