4 ms·

Machine learning finds an improved way to match donor organs with patients

56 points by abrax3141 4y ago

nonameking2026 4y ago
In leu of current events - the first thing that crossed my mind is what the CCP will do with this technology. But it's good to see more real world application of machine learning - feels like the distance from research to application is very slow. "Machine learning radiology" is something I used to monitor as a sign of improvement - since it's a "simple" use case. Too "busy" to google about now - any news on this ?
- VHRanger 4y ago
  It's just a more efficient way to do a form of graph matching[1] on a dynamic graph (one that evolves over time). There wasn't really a pure closed form mathematic solution to the problem like we can on static graphs (Gale-Shapely algorithm for one). You find papers like this one [2] with heavy math and some sort of proposed solution. So this type of problem is a really a good fit for machine learning to propose solutions. 1. https://en.wikipedia.org/wiki/Graph_matching https://en.wikipedia.org/wiki/Graph_matching 2. https://www.semanticscholar.org/paper/Thickness-and-Information-in-Dynamic-Matching-Akbarpour-Li/9de8f77c5d2c31988efac77f1bd29ddd6ddd0cfc https://www.semanticscholar.org/paper/Thickness-and-Informat...
- nonameking2026 4y ago
  For more context: https://en.wikipedia.org/wiki/Organ_harvesting_from_Falun_Gong_practitioners_in_China#Discrepancy_in_known_sources_of_organs https://en.wikipedia.org/wiki/Organ_harvesting_from_Falun_Go...
lr1970 4y ago
As someone doing ML for a living this development scares me. Trying policy to de-bias dataset against prior policy (which is changing with time by the way) is very difficult. And any new policy will effect the outcomes and new datasets going forward. Positive feedback loops in Reinforcement Learning is a huge problem.
- jjcon 4y ago
  If it can give better outcomes wouldn’t it be more unethical not to use it? We should certainly be aware of feedback loops but we can’t let perfection be the enemy of good
  - lr1970 4y ago
    > If it can give better outcomes wouldn’t it be more unethical not to use it? Better outcomes by what metric? If out of two people after the operation one will live 5 years and another 6 years -- which one would you choose? How would you trade 5 years of one's life against the 6 years of another's? Family, number of children, societal impact, race, gender, etc ? How to quantify all that?
    - mwint 4y ago
      Zooming into specific anecdotes is a paralyzing trap in almost all policy discussion. This is a population-level question; you look at population level results and only look at individual cases to search for obvious terrible effects.
      - tpoacher 4y ago
        This is true, but still does not counter the "rough prototypes have a bad habit of becoming end-products" curse the OP is effectively talking about. Not to mention your argument could be trivially used to justify social credit scores etc, which most people would agree is bad.
trident5000 4y ago
"Sorry, our internal black box system determined you are not a candidate at this time. Please try again later." Aside from reinforcement loops being an issue, ML is only as good as the programmer makes it. And we will have no idea how it was constructed and neither will non-tech health professionals. Theres no way a system like this can be described in its entirety without mistake to surgeons when one line of code changes everything. So maybe it can help out human decision but I would not want this to have final say for quite some time.
- abrax3141 4y ago
  The FDA is trying hard to figure out how to validate ML: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device https://www.fda.gov/medical-devices/software-medical-device-...
zzbn00 4y ago
Around 90% of liver disease deaths is from preventable risk factors (alcohol, viral hepatitis and obesity, https://ukhsa.blog.gov.uk/2014/09/29/liver-disease-a-preventable-killer-of-young-adults/ https://ukhsa.blog.gov.uk/2014/09/29/liver-disease-a-prevent... ). We have the tools to fix to these (the main being more even distribution of resources in society) yet we don't.
- ryan93 4y ago
  How would giving people more money decrease obesity and alcoholism? The richer our country has gotten the fatter it has gotten!
  - fredoliveira 4y ago
    I'm sure that if you think about it, you'll realize it is not the very wealthy that are getting more obese or drinking more alcohol.
    - conviencefee999 4y ago
      Not really, when Black Americans and other minority groups that grew up in poverty rise up in economic status they actually get fatter. Not sure where the statistic on it is but it was a really interesting find.
      - zzbn00 4y ago
        "Poverty and childhood obesity: a 21st century role reversal" : https://foodactive.org.uk/poverty-and-childhood-obesity-a-21st-century-role-reversal/ https://foodactive.org.uk/poverty-and-childhood-obesity-a-21...
      - SV_BubbleTime 4y ago
        There are more calories in cheap foods. This has gotten worse with processing, but from where I can see, no obvious solution exists.
    - 4y ago
dmckeon 4y ago
Improvements in organ donation/procurement networks in the USA are needed. "In 2020, 21.3 percent of procured kidneys were not transplanted," per https://www.washingtonpost.com/health/2022/08/03/transplant-deaths-mistakes-senate-finance/ https://www.washingtonpost.com/health/2022/08/03/transplant-... and https://onlinelibrary.wiley.com/doi/epdf/10.1111/ajt.16982 https://onlinelibrary.wiley.com/doi/epdf/10.1111/ajt.16982 We should not let an unknowable and unachievable "perfect" be the enemy of better.
t_mann 4y ago
needs a [2020] tag. Mihaela van der Schaar's research group is highly productive, I'd be surprised if they hadn't published some update to that methodology by now
echelon 4y ago
We could revolutionize not only organ transplants, but all of human health though advances in human cloning. The biggest increase to human health span will be when we start replacing our bodies wholesale as they age. Head transplants onto monoclonal, HLA-non-expressing (antigen free) bodies at age 50 will renew cardiovascular, pulmonary, and immune health (eg new thymuses). It resets almost all of the genetic cancer clocks in our bodies. Coupled with drugs to combat brain tissue waste aggregation and plaque formation, it could see humans through to multiple hundred years life spans. As for ethics, the monoclonal bodies will be grown in a lab, without brains, born out of chimeric pig uteruses and artificially innervated for endocrine regulation. I talk about this a lot in my HN post history. https://news.ycombinator.com/item?id=26673243 https://news.ycombinator.com/item?id=26673243 https://news.ycombinator.com/item?id=28534681 https://news.ycombinator.com/item?id=28534681 https://news.ycombinator.com/item?id=30622820 https://news.ycombinator.com/item?id=30622820 https://news.ycombinator.com/item?id=30407908 https://news.ycombinator.com/item?id=30407908
- abrax3141 4y ago
  But why?
  - echelon 4y ago
    To not die at 80. Do be young as long as you live. To cure people of cancer, heart, lung, liver, kidney, and other diseases. Intractable diseases. Once it becomes normalized, you'll even see people swapping their bodies for ones they prefer. Gender. Increased height, musculature, or VO2max. Transgenic craziness. Why not? Why accept what we have now?
    - abrax3141 4y ago
      Why not: Permitting our children to breathe and eat, maybe? You had your 80 years. Get out of the way.
      - echelon 4y ago
cuteboy19 4y ago
Seems like a difficult problem. Do you let 15 people have a subpar match or give 14 a perfect match and 1 person dies.
- Teever 4y ago
  What's the mortality rate of 15 subpar matches over 5 years vs mortality rate of 14 perfect matches over the same time period? That's your starting point for making this kind of decision.
what-imright 4y ago
[flagged]

56 points by abrax3141 4y ago

nonameking2026 4y ago

In leu of current events - the first thing that crossed my mind is what the CCP will do with this technology. But it's good to see more real world application of machine learning - feels like the distance from research to application is very slow. "Machine learning radiology" is something I used to monitor as a sign of improvement - since it's a "simple" use case. Too "busy" to google about now - any news on this ?

VHRanger 4y ago
It's just a more efficient way to do a form of graph matching[1] on a dynamic graph (one that evolves over time). There wasn't really a pure closed form mathematic solution to the problem like we can on static graphs (Gale-Shapely algorithm for one). You find papers like this one [2] with heavy math and some sort of proposed solution. So this type of problem is a really a good fit for machine learning to propose solutions. 1. https://en.wikipedia.org/wiki/Graph_matching https://en.wikipedia.org/wiki/Graph_matching 2. https://www.semanticscholar.org/paper/Thickness-and-Information-in-Dynamic-Matching-Akbarpour-Li/9de8f77c5d2c31988efac77f1bd29ddd6ddd0cfc https://www.semanticscholar.org/paper/Thickness-and-Informat...
nonameking2026 4y ago
For more context: https://en.wikipedia.org/wiki/Organ_harvesting_from_Falun_Gong_practitioners_in_China#Discrepancy_in_known_sources_of_organs https://en.wikipedia.org/wiki/Organ_harvesting_from_Falun_Go...

lr1970 4y ago

As someone doing ML for a living this development scares me. Trying policy to de-bias dataset against prior policy (which is changing with time by the way) is very difficult. And any new policy will effect the outcomes and new datasets going forward. Positive feedback loops in Reinforcement Learning is a huge problem.

jjcon 4y ago
If it can give better outcomes wouldn’t it be more unethical not to use it? We should certainly be aware of feedback loops but we can’t let perfection be the enemy of good
- lr1970 4y ago
  > If it can give better outcomes wouldn’t it be more unethical not to use it? Better outcomes by what metric? If out of two people after the operation one will live 5 years and another 6 years -- which one would you choose? How would you trade 5 years of one's life against the 6 years of another's? Family, number of children, societal impact, race, gender, etc ? How to quantify all that?
  - mwint 4y ago
    Zooming into specific anecdotes is a paralyzing trap in almost all policy discussion. This is a population-level question; you look at population level results and only look at individual cases to search for obvious terrible effects.
    - tpoacher 4y ago
      This is true, but still does not counter the "rough prototypes have a bad habit of becoming end-products" curse the OP is effectively talking about. Not to mention your argument could be trivially used to justify social credit scores etc, which most people would agree is bad.

trident5000 4y ago

"Sorry, our internal black box system determined you are not a candidate at this time. Please try again later." Aside from reinforcement loops being an issue, ML is only as good as the programmer makes it. And we will have no idea how it was constructed and neither will non-tech health professionals. Theres no way a system like this can be described in its entirety without mistake to surgeons when one line of code changes everything. So maybe it can help out human decision but I would not want this to have final say for quite some time.

abrax3141 4y ago
The FDA is trying hard to figure out how to validate ML: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-software-medical-device https://www.fda.gov/medical-devices/software-medical-device-...

zzbn00 4y ago

Around 90% of liver disease deaths is from preventable risk factors (alcohol, viral hepatitis and obesity, https://ukhsa.blog.gov.uk/2014/09/29/liver-disease-a-preventable-killer-of-young-adults/ https://ukhsa.blog.gov.uk/2014/09/29/liver-disease-a-prevent... ). We have the tools to fix to these (the main being more even distribution of resources in society) yet we don't.

ryan93 4y ago
How would giving people more money decrease obesity and alcoholism? The richer our country has gotten the fatter it has gotten!
- fredoliveira 4y ago
  I'm sure that if you think about it, you'll realize it is not the very wealthy that are getting more obese or drinking more alcohol.
  - conviencefee999 4y ago
    Not really, when Black Americans and other minority groups that grew up in poverty rise up in economic status they actually get fatter. Not sure where the statistic on it is but it was a really interesting find.
    - zzbn00 4y ago
      "Poverty and childhood obesity: a 21st century role reversal" : https://foodactive.org.uk/poverty-and-childhood-obesity-a-21st-century-role-reversal/ https://foodactive.org.uk/poverty-and-childhood-obesity-a-21...
    - SV_BubbleTime 4y ago
      There are more calories in cheap foods. This has gotten worse with processing, but from where I can see, no obvious solution exists.
  - 4y ago

dmckeon 4y ago

Improvements in organ donation/procurement networks in the USA are needed. "In 2020, 21.3 percent of procured kidneys were not transplanted," per https://www.washingtonpost.com/health/2022/08/03/transplant-deaths-mistakes-senate-finance/ https://www.washingtonpost.com/health/2022/08/03/transplant-... and https://onlinelibrary.wiley.com/doi/epdf/10.1111/ajt.16982 https://onlinelibrary.wiley.com/doi/epdf/10.1111/ajt.16982 We should not let an unknowable and unachievable "perfect" be the enemy of better.

t_mann 4y ago

needs a [2020] tag. Mihaela van der Schaar's research group is highly productive, I'd be surprised if they hadn't published some update to that methodology by now

echelon 4y ago

We could revolutionize not only organ transplants, but all of human health though advances in human cloning. The biggest increase to human health span will be when we start replacing our bodies wholesale as they age. Head transplants onto monoclonal, HLA-non-expressing (antigen free) bodies at age 50 will renew cardiovascular, pulmonary, and immune health (eg new thymuses). It resets almost all of the genetic cancer clocks in our bodies. Coupled with drugs to combat brain tissue waste aggregation and plaque formation, it could see humans through to multiple hundred years life spans. As for ethics, the monoclonal bodies will be grown in a lab, without brains, born out of chimeric pig uteruses and artificially innervated for endocrine regulation. I talk about this a lot in my HN post history. https://news.ycombinator.com/item?id=26673243 https://news.ycombinator.com/item?id=26673243 https://news.ycombinator.com/item?id=28534681 https://news.ycombinator.com/item?id=28534681 https://news.ycombinator.com/item?id=30622820 https://news.ycombinator.com/item?id=30622820 https://news.ycombinator.com/item?id=30407908 https://news.ycombinator.com/item?id=30407908

abrax3141 4y ago
But why?
- echelon 4y ago
  To not die at 80. Do be young as long as you live. To cure people of cancer, heart, lung, liver, kidney, and other diseases. Intractable diseases. Once it becomes normalized, you'll even see people swapping their bodies for ones they prefer. Gender. Increased height, musculature, or VO2max. Transgenic craziness. Why not? Why accept what we have now?
  - abrax3141 4y ago
    Why not: Permitting our children to breathe and eat, maybe? You had your 80 years. Get out of the way.
    - echelon 4y ago

cuteboy19 4y ago

Seems like a difficult problem. Do you let 15 people have a subpar match or give 14 a perfect match and 1 person dies.

Teever 4y ago
What's the mortality rate of 15 subpar matches over 5 years vs mortality rate of 14 perfect matches over the same time period? That's your starting point for making this kind of decision.

what-imright 4y ago

[flagged]