I often wish these kinds of things didn't get posted here. This is meant for a technical audience, and "technical" here means you know something about studies like this, not you understand computing technology. I've got degrees in applied math and machine learning and barely understand the kinds of statistics that get employed in studies like this. They don't bother to explain what a proportional hazard model is because they assume readers will know because they assume readers are other scientists doing epidemological cohort studies, not a random lay audience that reads popular news sites.
In any case, the results statement is extremely confusing for a lay audience. In a proportional hazard model, they are saying one group has a 1.57% lesser chance of dying in some infinitesimal near-term time step. Obviously, every person in existence has exactly a 100% chance of death. What this is saying is that, if we measured your state as either alive or dead at fixed time intervals, as those intervals approach 0, the relative risk of death for one group versus another has a fixed proportion such that one risk is greater than another. Clearly, the probability you're going to die in the instaneous near future is not the same throughout life. But, all else being equal, if the only difference between two groups is a difference in heterozygosity, the model assumes one group will always have a higher probability than the other and the extent to which it is higher is fixed throughout life.
We do this because it is statistically tractable, but there is no clear way to translate this into terms any normal person would actually care about, which is how long from birth should I expect to live provided my life consists of some fixed set of conditions. A study like this can't tell us that. It's safe to assume the group with lower risk is going to live longer on average, but how much longer we can't tell from an analysis like this. Otherwise, that's what they would report. These results are not meant to be directly interpretable other than to say "one condition is vaguely better than another."
I often wish these kinds of things didn't get posted here. This is meant for a technical audience, and "technical" here means you know something about studies like this, not you understand computing technology. I've got degrees in applied math and machine learning and barely understand the kinds of statistics that get employed in studies like this. They don't bother to explain what a proportional hazard model is because they assume readers will know because they assume readers are other scientists doing epidemological cohort studies, not a random lay audience that reads popular news sites.
In any case, the results statement is extremely confusing for a lay audience. In a proportional hazard model, they are saying one group has a 1.57% lesser chance of dying in some infinitesimal near-term time step. Obviously, every person in existence has exactly a 100% chance of death. What this is saying is that, if we measured your state as either alive or dead at fixed time intervals, as those intervals approach 0, the relative risk of death for one group versus another has a fixed proportion such that one risk is greater than another. Clearly, the probability you're going to die in the instaneous near future is not the same throughout life. But, all else being equal, if the only difference between two groups is a difference in heterozygosity, the model assumes one group will always have a higher probability than the other and the extent to which it is higher is fixed throughout life.
We do this because it is statistically tractable, but there is no clear way to translate this into terms any normal person would actually care about, which is how long from birth should I expect to live provided my life consists of some fixed set of conditions. A study like this can't tell us that. It's safe to assume the group with lower risk is going to live longer on average, but how much longer we can't tell from an analysis like this. Otherwise, that's what they would report. These results are not meant to be directly interpretable other than to say "one condition is vaguely better than another."