Epiphanies 2022: QALYs and healthcare economics, and implications for disabled people.
Leorning Cniht
Shipmate
QALYs (Quality Adjusted Life Years) are a tool used in cost-utility calculations in healthcare, particularly in the UK. The most common way of assessing "quality" is with a thing called EQ-5D.
It provides a way of quantifying the benefit of particular treatment. A simple canonical example would look something like this:
Suppose a person is expected to live for an additional five years, and is assessed to have a 0.5 quality factor for their life. So their current life expectancy is 2.5 QALYs. Suppose a proposed treatment is expected to reduce their pain, increase their mobility, and so increase their quality factor to 0.8, but is not expected to alter their life expectancy in years. But their quality of life went up, so they now expect 4.0 QALYs, so the treatment has improved their life by 1.5 QALYs.
Currently, NICE in the UK deems it worth spending somewhere between £20,000 to £30,000 per QALY. So if this proposed treatment were to cost the NHS less than £30,000 to £45,000 then it would be deemed cost-effective, and so supported and made generally available. If it were to cost more than that, then there would be other places where public funds could be more effectively spent, and the treatment would not be approved.
So far, there's nothing terribly sinister going on. Declaring that a treatment that makes a significant improvement to someone's life is more worthwhile than a treatment that has only a marginal effect on someone's life is pretty unobjectionable.
Problems arise, however, when assigning quality values to people who are disabled, and comparing them to people who are not disabled. The EQ-5D questionnaires ask questions about how mobile you are, how able you are to perform daily self-care without assistance, and so on. A person who uses a wheelchair, and requires assistance to wash and dress is going to score close to zero on those particular metrics. The QALY framework then implicitly assumes that that person's life is worth less. Consider a life-extending treatment that would cost £40,000, extend a person's life by two years, but not alter its quality. A strict application of the QALY methodology would say, for an able person in good health "we've extended their life by 2 years, which is 2 QALYs, and that's worth £40K" whereas it would say for the person in the wheelchair "this person has a quality of life of 0.5, so they've only gained 1 QALY, and that's not worth it, so no treatment for them."
This example dealt with individual people, but the problem persists at a population level. Even if you use the QALY methodology merely to approve or reject particular treatments in general (and once a treatment is approved, anyone can have it), a treatment that is primarily used by disabled people would score worse than one that was primarily used by able-bodied people in the QALY framework.
If your metric is leading you to make decisions that you don't think you should be making, then you should change your metric so that it aligns with what you want to happen. In the case of QALYs, the comparison of treatments within the same population group seems unobjectionable - the problem arises when you compare treatments for different things between different population groups, and one group is assessed to start with a lower quality of life, and so be worth less.
I wonder what happens to the QALY approach if you normalize it differently - so rather than having an absolute scale, where 1 = able-bodied person in perfect health, you have a relative scale, where everyone starts on 1, and you're measuring gains and losses relative to their starting point. It seems that that would eliminate the devaluing of disabled lives, but I haven't thought through what else it does to the calculations.
It provides a way of quantifying the benefit of particular treatment. A simple canonical example would look something like this:
Suppose a person is expected to live for an additional five years, and is assessed to have a 0.5 quality factor for their life. So their current life expectancy is 2.5 QALYs. Suppose a proposed treatment is expected to reduce their pain, increase their mobility, and so increase their quality factor to 0.8, but is not expected to alter their life expectancy in years. But their quality of life went up, so they now expect 4.0 QALYs, so the treatment has improved their life by 1.5 QALYs.
Currently, NICE in the UK deems it worth spending somewhere between £20,000 to £30,000 per QALY. So if this proposed treatment were to cost the NHS less than £30,000 to £45,000 then it would be deemed cost-effective, and so supported and made generally available. If it were to cost more than that, then there would be other places where public funds could be more effectively spent, and the treatment would not be approved.
So far, there's nothing terribly sinister going on. Declaring that a treatment that makes a significant improvement to someone's life is more worthwhile than a treatment that has only a marginal effect on someone's life is pretty unobjectionable.
Problems arise, however, when assigning quality values to people who are disabled, and comparing them to people who are not disabled. The EQ-5D questionnaires ask questions about how mobile you are, how able you are to perform daily self-care without assistance, and so on. A person who uses a wheelchair, and requires assistance to wash and dress is going to score close to zero on those particular metrics. The QALY framework then implicitly assumes that that person's life is worth less. Consider a life-extending treatment that would cost £40,000, extend a person's life by two years, but not alter its quality. A strict application of the QALY methodology would say, for an able person in good health "we've extended their life by 2 years, which is 2 QALYs, and that's worth £40K" whereas it would say for the person in the wheelchair "this person has a quality of life of 0.5, so they've only gained 1 QALY, and that's not worth it, so no treatment for them."
This example dealt with individual people, but the problem persists at a population level. Even if you use the QALY methodology merely to approve or reject particular treatments in general (and once a treatment is approved, anyone can have it), a treatment that is primarily used by disabled people would score worse than one that was primarily used by able-bodied people in the QALY framework.
If your metric is leading you to make decisions that you don't think you should be making, then you should change your metric so that it aligns with what you want to happen. In the case of QALYs, the comparison of treatments within the same population group seems unobjectionable - the problem arises when you compare treatments for different things between different population groups, and one group is assessed to start with a lower quality of life, and so be worth less.
I wonder what happens to the QALY approach if you normalize it differently - so rather than having an absolute scale, where 1 = able-bodied person in perfect health, you have a relative scale, where everyone starts on 1, and you're measuring gains and losses relative to their starting point. It seems that that would eliminate the devaluing of disabled lives, but I haven't thought through what else it does to the calculations.
Comments
"If your metric is leading you to make decisions that you don't think you should be making, then you should change your metric so that it aligns with what you want to happen"
Long ago when dinosaurs roamed and 286 computers ground out optimisation routines in Fortran stolen from a book called 'Numerical Recipes', I was involved in machine learning for a while. Your sentence took me back there, powerfully.
No-one would write a thesis or a journal (or even conference) paper which admits as much, but of course that is what we were doing. Actually, that could be OK if the theologians, ethicists and philosophers sort out what we want to happen, properly, first. Then turning it into an algorithm, and testing it thoroughly for perverse outcomes, might end up being an efficient way to optimise national health problems, one half of which (the cost) is at least already numeric.
Your last paragraph looks like an interesting way to start to dick with the algorthm, having in the previous one started to outline why you want the outcome to be different! I think I agree, but I'm only fit to follow the argument these days, novel contributions no longer being much in my line
It has the cost that it abandons the idea that QALYs are measuring something that is objectively there regardless of who has them but I think that is a plus.
But QALYs are only a small part of the problem of a massive mismatch between health service resources and health demands (demands comprising needs and wants - the two being not necessarily the same thing of course). And then there's the 'inverse care law' and the fact that the roots of so many health problems are various types of poverty and deprivation.
Wouldn't it be interesting to run the same QALY calculation, where the "treatment" is ensuring that the "patient" can afford somewhere warm and dry to live, and decent food and the like?
Mathematics and I are not on speaking terms. Righteous anger and I know each other quite well, but I don't like them at all. They get in the way of judgement and good decisions. My training in theology and law means that I value subjective decisions highly, in spite of their susceptibility to prejudice.
ISTM that when a decision is made solely on the basis of the type of calculation described in the OP, the chance of injustice is high. It strikes me that good decisions are made in consultation between experienced people, all of whom know the strengths and weaknesses of the analytical tools available. Drawing upon experienced people in different fields might ameliorate prejudice. Above all, the calculation must inform the decision, rather than make the decision.