jrissman
commented
3 years ago This graph from Our World in Data has deaths from outdoor air pollution disaggregated into age buckets, with separate data from every country in the world. It cites IHME’s Global Burden of Disease as its source. IHME's tool looks pretty great, with finer age brackets than Our World in Data is using and output in DALYs as well as deaths. To get values for the impact of outdoor air pollution from IHME, select "Risk" from the "Context" menu, then in the "Risk" menu that appears, check the boxes for "Ambient particulate matter pollution" and "Ambient ozone pollution".
We'll still need to normalize these figures against emissions to get the relationship between emissions and deaths (and perhaps scale the U.S. data to distinguish between PM2.5, NOx, and SOx), but even so, this IHME dataset is probably the best lead we have on air pollution health impacts disaggregated by age bracket for many different countries and gets us most of the way there.
EPS 3.0 moved to a new EPA data source, which provides the number of avoided health incidents of various types (premature deaths, hospital admissions, asthma attacks, etc.) per unit of avoided pollutant emissions (disaggregated by type of pollutant and emissions source) in various future years.
One reason these values change in future years is because the population changes. All else being equal, if there are more people around to be exposed to one ton of pollutant, there will be more health incidents.
The EPA emissions-to-incident multipliers must therefore assume a specific (BAU) population level in each future year.
A policy scenario in the EPS often saves lives (lowers the death rate), which means there will be more people in later years. This means that the EPA multipliers can't be based on whatever the population level happens to be in a policy scenario, since this depends on the user's settings.
However, the death of a person in year X+1 who would have died in year X in the BAU case doesn't increase total deaths across the two years - it only shifts the death by one year. In other words, the death of a person in year X in the BAU case prevents that person from dying in year X+1, so the BAU case may have one fewer death in year X+1 as a result, dampening the apparent effect of the policy package on saving lives (relative to the BAU case) in year X+1. This means that simply scaling up the EPA multipliers based on our estimate of the change in population due to the policy package won't work.
The most correct way to handle this is to calculate an actual number of deaths in each year in the BAU case, a number of deaths each year in the policy case, and subtract the two cases to find the difference in each year caused by the policy package. This means that instead of working in units of "avoided deaths," we work in units of "deaths." Epidemiological concentration-response functions typically specify a change in the BAU death rate per unit concentration of pollutants. The way to handle this in the EPS would be to track the population, with BAU inflows (births, immigration) and outflows (deaths, emigration). Deaths are calculated by multiplying the death rate by the population. The policy package changes the death rate used in the policy case.
Health outcomes other than death are easier to calculate because you don't have the complication of deaths in one year preventing deaths in later years (because the person isn't around to die again). But you still would get more accuracy by working in units of incidents rather than avoided incidents and subtracting the two cases, because that better accounts for different population sizes in the two cases. Failing that, scaling the EPA "avoided incident per ton" multipliers based on population size difference between the BAU case and the Policy case would be acceptable for calculating the change in health outcomes other than death.