Saturday, November 20, 2004
Gettin' Older Sittin' Still
The magazine IEEE Spectrum was sitting around in the lunchroom yesterday, and I got to read a fascinating article on the application of engineering theory to human longevity. Neat, neat article, with remarkable implications.
It involves something called "Reliability Theory"... quite simply, the study of the way complicated things break down, in a statistical manner. In this context, aging is in fact used, to describe the increasing probability (over time) that the object will completely fail. The simplest example of aging would be a system with just one triply-redundant component. Each instance has a fixed and constant - nonaging - likelihood of failure, in any given span of time.
There's a chance that, in that span of time, all three will simultaneously fail and the device will die; it's just small. But as time goes by, the odds that one of them will fail become cumulatively larger - and with one failed, the odds of the other two failing simultaneously go up. After a while, the odds of 2/3s of the instances having failed goes up, and the odds of the device failing go up again... they become equal to the odds of a single instance failing, in the extreme.
The interesting thing is the shape of these graphs. Start with a hundredfold-redundant system vs. a merely threefold-redundant one, and the former will naturally have a much longer working life... but if you look near the far end of the curve, the odds of each one failing have converged to the same number. That is, after a hundred years, there will be many more of the hundredfold systems left around than of the threefold systems - but if you have a functioning one that's that old, they're both equally likely to fail in the next six months, regardless of what it started as.
The same thing apparently holds true of countries. The odds of dying in the ages 20-25 is way higher for Nigerians than Swedes... but the odds of dying in the ages 95-100 is basically the same. Also, like the triply-redundant system reduced to just one component, the odds of someone dying between 100 and 101 are about the same as the odds of someone dying between 120 and 121; once you make it to that point, the curve flattens out.
The one discrepancy is that a multiply-redundant mechanical system, after the periods of "break-in" (high initial odds of failure, of being a lemon) and "normal working time", follows a straight-line curve of (log of failure rate) vs. (log of time). Humans have a similar curve, including a sharp rise at the very beginning (infant mortality rates), dropoff to "normal odds" (ages 5-15 or so), and then an aging curve where the odds of death increase... but they go as a straight line on (log of death rate) vs. (time). Not the same graph!
However, the two models do converge if you switch ways of looking at it... because the standard machine-longevity assumption is that all components start out working properly.
The authors then redid the model with a machine that was (say) ten times as redundant, but starts out with a whole bunch of its parts already failed. Similar total lifetime expectations... but the curve is a different shape. It's linear in (log of failure rate) vs. (time), like the human mortality curve.
The implication is that evolution did not give us good engineering, it gave us a jury-rig job; we take a lot of chromosomal and other lasting damage, in the womb and during infancy. So what we live out the rest of our lives on is a bunch of massively-redundant but defect-ridden systems. The mathematical implication of this is very interesting:
If we can reduce the failure rate of systems during this time, life expectancy can increase disproportionately.
Folic acid may turn out to be a good case of this. If your mom took folic acid supplements during pregnancy, this helps prevent some of the defect modes. We do it to reduce the rates of overt damage to the child, as seen after birth... but the acid does so by reducing those pre-birth systems failure rates on the cellular level, so the reduction in "immediate failure" may well be expected to presage a substantial increase in overall longevity, because you're moving toward the machine-longevity model which goes as (log of time)! If you can stop X% of initial defects, this will increase your longevity by more than X%. And if you can stop another Y%, the effect will accelerate further.
Experiments with mice fed antioxidants (which help prevent chemical damage to DNA) during gestation seem to agree with this; they've had some brutally long-lived mice out of this kind of process.
So two lessons out of this one. Primus, we may be able to increase longevity substantially... but those who benefit are likely to be those children as yet unborn. For us, the clock is already mostly set, and all we can do is treat symptoms. Secundus, the critcism of mothers who take "just a little drink", or otherwise do anything which increases the odds of fetal problems even a tiny bit ought to be far fiercer than it is... because the near-term odds of problems we can see are just an indicator. The background processes they result from will also shorten their expected lifespan. Shorten it a LOT.
Mostly, here, I'm preaching to the choir... but the choir has a role in singing to the world, and I'd encourage you all to take a much stronger stance on this sort of thing. Anything that helps (folic acid) or hurts (alcohol, street drugs, some prescriptions) a fetus will have its impact not only on the baby, but - more strongly! - on the lifespan of the man or woman it will become.
It involves something called "Reliability Theory"... quite simply, the study of the way complicated things break down, in a statistical manner. In this context, aging is in fact used, to describe the increasing probability (over time) that the object will completely fail. The simplest example of aging would be a system with just one triply-redundant component. Each instance has a fixed and constant - nonaging - likelihood of failure, in any given span of time.
There's a chance that, in that span of time, all three will simultaneously fail and the device will die; it's just small. But as time goes by, the odds that one of them will fail become cumulatively larger - and with one failed, the odds of the other two failing simultaneously go up. After a while, the odds of 2/3s of the instances having failed goes up, and the odds of the device failing go up again... they become equal to the odds of a single instance failing, in the extreme.
The interesting thing is the shape of these graphs. Start with a hundredfold-redundant system vs. a merely threefold-redundant one, and the former will naturally have a much longer working life... but if you look near the far end of the curve, the odds of each one failing have converged to the same number. That is, after a hundred years, there will be many more of the hundredfold systems left around than of the threefold systems - but if you have a functioning one that's that old, they're both equally likely to fail in the next six months, regardless of what it started as.
The same thing apparently holds true of countries. The odds of dying in the ages 20-25 is way higher for Nigerians than Swedes... but the odds of dying in the ages 95-100 is basically the same. Also, like the triply-redundant system reduced to just one component, the odds of someone dying between 100 and 101 are about the same as the odds of someone dying between 120 and 121; once you make it to that point, the curve flattens out.
The one discrepancy is that a multiply-redundant mechanical system, after the periods of "break-in" (high initial odds of failure, of being a lemon) and "normal working time", follows a straight-line curve of (log of failure rate) vs. (log of time). Humans have a similar curve, including a sharp rise at the very beginning (infant mortality rates), dropoff to "normal odds" (ages 5-15 or so), and then an aging curve where the odds of death increase... but they go as a straight line on (log of death rate) vs. (time). Not the same graph!
However, the two models do converge if you switch ways of looking at it... because the standard machine-longevity assumption is that all components start out working properly.
The authors then redid the model with a machine that was (say) ten times as redundant, but starts out with a whole bunch of its parts already failed. Similar total lifetime expectations... but the curve is a different shape. It's linear in (log of failure rate) vs. (time), like the human mortality curve.
The implication is that evolution did not give us good engineering, it gave us a jury-rig job; we take a lot of chromosomal and other lasting damage, in the womb and during infancy. So what we live out the rest of our lives on is a bunch of massively-redundant but defect-ridden systems. The mathematical implication of this is very interesting:
If we can reduce the failure rate of systems during this time, life expectancy can increase disproportionately.
Folic acid may turn out to be a good case of this. If your mom took folic acid supplements during pregnancy, this helps prevent some of the defect modes. We do it to reduce the rates of overt damage to the child, as seen after birth... but the acid does so by reducing those pre-birth systems failure rates on the cellular level, so the reduction in "immediate failure" may well be expected to presage a substantial increase in overall longevity, because you're moving toward the machine-longevity model which goes as (log of time)! If you can stop X% of initial defects, this will increase your longevity by more than X%. And if you can stop another Y%, the effect will accelerate further.
Experiments with mice fed antioxidants (which help prevent chemical damage to DNA) during gestation seem to agree with this; they've had some brutally long-lived mice out of this kind of process.
So two lessons out of this one. Primus, we may be able to increase longevity substantially... but those who benefit are likely to be those children as yet unborn. For us, the clock is already mostly set, and all we can do is treat symptoms. Secundus, the critcism of mothers who take "just a little drink", or otherwise do anything which increases the odds of fetal problems even a tiny bit ought to be far fiercer than it is... because the near-term odds of problems we can see are just an indicator. The background processes they result from will also shorten their expected lifespan. Shorten it a LOT.
Mostly, here, I'm preaching to the choir... but the choir has a role in singing to the world, and I'd encourage you all to take a much stronger stance on this sort of thing. Anything that helps (folic acid) or hurts (alcohol, street drugs, some prescriptions) a fetus will have its impact not only on the baby, but - more strongly! - on the lifespan of the man or woman it will become.
<< Home