Hospitals are full of both infected people, and very susceptible people – people whose immune systems are, due to age or illness, very weak. This is a terrible combination but unavoidable. And it is why it is important to try and minimise the spread of infectious diseases such as COVID, but also why it is hard to do this, especially for airborne diseases like COVID. With airborne diseases you can reduce transmission via (good, eg FFP2) mask wearing. But masks are not very comfortable, so a more palatable solution is to improve the quality of the air itself, either by improved ventilation (ie more fresh air), or by filtration. Brock et al. at Cambridge Hospitals recently (2024) published a study of the effect of air filtration on COVID transmission in hospital wards.
The results can be expressed in a few different ways, but from their Table 2, we see that on the control wards (no filtration) there was an infection rate of 0.87 % per day, while on the wards with filtration it was 0.60 % per day. Here an infection rate of 1 % per day means an estimated one chance in a hundred of infection with COVID per patient per day that the patient stays on the ward.
The number of infections are of order 100, so the standard fractional error bars are around +/- 2/(100)1/2 = +/- 0.2*. So with statistical error bars, the estimates are 0.70 to 1.0 % per day on the control wards versus 0.48 to 0.70 % on the wards with filtration. So we see that the estimates overlap, hence the study is inconclusive.
But we can still compare the study with a prediction. Unfortunately the study does not report numbers that quantify the filtration, for example via an estimate of how many times per hour the air in a ward passes through one of the filtration units. We can still estimate that the filtration is likely to very roughly double the rate of removal of aerosol particles in the air. Say, for example, without filtration, the lifetime of particles is perhaps half an hour because ventilation exchanges the complete air in the ward twice an hour, and the filtration also filters the air every half an hour**. Then the concentration of aerosols in the wards with filtration should be roughly half that in the control wards.
The next question is what happens to the probability of infection if you halve the aerosol dose of virus? Unfortunately we don’t know this either. But, using data from the general UK population (so all of us, not just hospital patients) both Ferretti et al.*** and I estimate that the infection probability may be a sublinear function of effective dose. The data seem best fit by a power law with an exponent near one half.
If we apply this here then a halving of dose should reduce the infection risk by a factor of 1/(2)1/2 = 0.71. So an infection risk in the range 0.70 to 1.0% per day decreases to 0.50 to 0.71 % per day. This is pretty close to what Brock et al. observed in the wards with air filters, so the data are consistent with filtration halving the inhaled dose which in turn reduces the risk of infection by about 30%.
As Brock et al. note, bigger studies than theirs are needed to prove how effective filtration is. But it is maybe reassuring that this small study is at least consistent what we would expect, i.e., adding filtration (enough to maybe half the doses inhaled) reduces transmission by very roughly 30%. And as Brock et al. note reductions by amounts of this sort are very worthwhile.
* The factor of 2/(N)1/2 with N the number infected assumes that the N infections are independent events. As Brock et al. note the infection events are not at all uniformly distributed, many of them come in spikes. This may well mean they are not independent, if so the statistical uncertainties will be larger, as then the fractional uncertainty scales as one over the square root of the number of independent events, and then the number of independent events will be lower than the number of patients infected.
** The filtration units were not always kept on, especially towards the end of the study, this is in effect an average.
** That is not quite how Ferretti et al.* and I put it in these two papers, but roughly speaking …
Chance & Necessity 