Crossposted from Probably Definitely.

I was excited to see this paper on the effects of air purifiers on birth weight. There’s evidence that exposure to pollution during gestation can decrease birth weight, which is associated with worse outcomes later on. So if air purifiers clean the air, why not put them in the homes of expectant mothers?

The authors did just that. The setting was the city of Ulaanbaatar, Mongolia (population 1.5m), where air pollution is horrible. Its average PM 2.5 concentration was 75 μg/m³ in 2017. For comparison, in recent years Beijing had an average of 43 μg/m³, downtown Los Angeles (one of the worst places in the US) had 18 μg/m³, and the population-weighted average for the US was 8 μg/m³. The World Health Organization recommends an annual mean of no more than 10 μg/m³.

They enrolled 463 pregnant women at fewer than 18 weeks of gestation and who did not currently own an air purifier. Half of them got what I think is a Coway AP-1512HH or something like it based on the described modifications and the fact that Woongjin-Coway provided the devices:

Two features, an internal PM sensor and light that changes colour based on PM concentration, were disabled to avoid biasing participants' behavior. The units were set to operate only on the second-highest fan setting due to noise at the highest fan setting.

They wrote a companion paper checking if the purifiers actually worked, using a low-cost Dylos particle counter in the homes of participants. They found only a 29% reduction in PM 2.5, which is at the lower end of similar studies reviewed in that paper, which find reductions between 32 and 66%. Compliance was an issue. Participants apparently only used the purifiers for 64% of the study period, and some “reported consistently turning air cleaners off at night to minimize noise.” Despite this, the intervention reduced blood cadmium by 14%.

Back in the births paper, the headline result is that...we basically can’t tell. Babies in the purifier condition were 18g heavier, with a 95% confidence interval of -84g to 120g. We do not learn much from this, since that interval encompasses some pretty large effects.

But, the authors argue, it gets complicated. Unexpectedly, they found that preterm births were way more likely in the purifier group (24 vs. 10 in the control, p=.03)! It seems this was probably a fluke, since the alternative is that air purifiers cause preterm births. When they remove these 34 births, the effect of purifiers becomes slightly positive, indicating an increase of 85g (0.2 standard deviations, 95% CI: 3g-167g). 

Unfortunately, I do not think much credence should be given to this second estimate. Basically what the researchers have done is notice that the outcome has more extremely low values in treatment than control, dropped the extremely low values from both groups, and then concluded that the treatment average is higher than the control average.

Another strike against the findings is that the effects were not any larger for families who reported better compliance:

So their main estimate is essentially zero, but the confidence interval cannot rule out economically meaningful effects. How meaningful? I did some simple cost-benefit analysis assuming we could achieve an 18g increase in birth weight in the US.

In a study of twins in California, Royer finds that a 200g increase in birth weight causes education to increase by 0.04 years, so an 18g increase in birth weight would yield 0.0036 years or about a day of additional education. If the average American works 40 years at $65k/year, and a year of education increases income by 7%, then a year of education adds $4.5k to annual income and 0.0036 years adds $16. Discounting at 5% per year and starting the $16 flow at 18 years after the purifier purchase, this gives about $111 in net present value. 

Another study finds that a 1 pound (454 gram) increase in birth weight decreases medical costs by $2,000, so $79 using the 18g increase. 

These are just two easy-to-measure benefits but they suggest that it would probably be worth it if we could spend $100-200 on an intervention that would increase birth weight by 18g on average. 

I want to see this same RCT done with a much larger sample in the US. I would expect very small effects, since the present study rules out large benefits and took place in an area where pollution abatement should have a higher return. But it seems important to know and the intervention is cheap.

A key question left untouched here is: ignoring this study, what would our best guess be of the effect purifiers on birth weight? We should have good estimates on the effects of purifiers on PM2.5 levels in houses (example). The evidence on how PM2.5 pollution affects birth weight will necessarily be worse since we mainly have observational studies, but I would love to see a careful aggregation of existing estimates. These two numbers might be a better guide for cost-benefit analyses of air purifiers, since the estimates in the present trial were ultimately quite noisy.