Scott Alexander vs. Robin Hanson on medical efficacy; Dan Williams on rational persuasion; Glenn Reynolds on the higher education industry; Joel Kotkin on the rise of the Mean Girls
I have been following the debate between Hanson and Alexander on their Substacks, but I have to say that your comment "What Robin argues is that treatments that work are, in the aggregate, offset by other treatments that cause harm" did more to explain his position than I got from reading him directly
"Take something like colonoscopy. On net, do the benefits outweigh the harms? I am skeptical, because I doubt that the benefits are as large as people have been led to believe."
So you are more worried about a perforated bowel than colon cancer? Really? Colon cancer is major cause of death. How many people die from a perforated bowel? If everyone over 50 got regular colonoscopies would the number be significant then?
I would love to read your explanation of how he is wrong. Maybe adverse selection? The people most likely to get frequent colonoscopies are the people least likely to need it. Can you make that case, or any other? Does your opinion have any basis in scientific method?
COVID was a leading cause of death for two years, and I didn't even think it was worth wearing a mask for. Let alone getting something shoved up my ass.
Not relevant to the links you posted, but a link _for_ Arnold Kling in case it is a substack he is not reading: https://substack.com/home/post/p-144153862?source=queue -- Drew Haugen is training XGBoost to predict whether a pitcher will need Tommy John Surgery. So far he has teased out some factors that appear to be relevant but it is not at 'should I draft this man to my fantasy football team' yet.
If the pitchers who XGBoost says are most likely to need Tommy John Surgery read XGBoost, they are going to adjust their behavior to lower their chances of needing Tommy John Surgery and so by XGBoost being right it will also make itself wrong. Something like an Efficient Information Hypothesis, or a more general "Efficient Prediction Hypothesis". Agents knowing the forecast will neutralize the forecast. "You can't beat the future."
'Adjusting their behaviour to lower the chance of needing TJ surgery' has not been a notable characteristic of pitchers so far. Of course, many of them think that having TJ surgery is a net positive, even with the year long rehab. For the fantasy baseball player, though, there is just a downside.
Speaking of medical efficacy, there was a great article about ECMO machines in the New Yorker a couple days ago that really showcases how incredible medical innovations to come will be. Apparently, we really aren't that far off from a world with Futurama style heads in jars: https://www.newyorker.com/science/annals-of-medicine/how-ecmo-is-redefining-death
A minor point: “treatments that cause harm” can easily be misinterpreted. Some treatments cause harm in a direct, obvious way, as when a patient dies on the operating table while undergoing a knee replacement, or—less dramatically—a diagnostic procedure that yields a negative result slightly damages the patient’s health. But I suspect the more common case is that the treatment’s direct effect is slightly good but not worth the cost, and the alternative use to which the resources would have been put if the treatment had not been done would have had effects not just *good*, but *good for health*. Activities that are not billed as “health care” can still have effects on health.
I am skeptical we have good data on what treatment doesn't "work," much less which cause harm. On this week's episode of Econtalk, Roberts talks about his mother's compression fracture in her back. treatment options include doing nothing, a back brace, and surgery to insert cement into the bone. A study(s) has indicated surgery is no more successful than a placebo. Roberts, his siblings, and even his mother knew this ye they opted for surgery. Why? A doctor said the outcome are better. Given that a placebo isn't an option, he could very well be right. Either way, can one make a factual statement that he is wrong? how does one collect factual data on which interventions cause harm? Color me skeptical we have good data on this. Seems like it can't be done without huge subjectivity. Makes me think of happiness data. Is either reliable?
"I can tell you stories about close relatives who were clearly the victims of Hansonian medicine. They would be better off without the medical interventions that they underwent."
What do your anecdotal cases tell us? Is this because that type of intervention should never be done, the doctor/facility doing it lacked the adequate capability, or in a certain number of cases it simply fails but overall the statistics are good for this intervention?
A thing you are missing vis a vis the colonoscopy issue is that the effects on mortality are not the only important benefit. People want to know whether they have cancer. Finding out they don't have one of the most common cancers will help a lot of people sleep at night. You can argue that people shouldn't be like this but good luck changing their minds.
Bobert and Green aren't leaders of large organizations or chief executives authorized to make major decisions for established institutions. As prominent "politicians in American democracy" they are just media figures, i.e., "actresses". "And the Oscar goes to ..." They are, if anything, "Managers of their own brand", not very distinguishable in principle from online 'influencers', Instagram models or OnlyFans, ahem, "providers" squeezing simps and paypigs for all they're worth, though arguably this is unfair to the OnlyFans freelance pornographers and prostitutes who, unlike these Republican politicians, are actually honest businesswomen and reliably fulfill their promises and deliver what they are paid to put out. The business of those politicians is to be celebrities endorsing the product that is "themselves". Not their actual selves but a not-too-distant-but-still-somewhat-faked role to act out in public, refined, focus-grouped, and stragically crafted as an exaggerated caricature version of a public persona, market-optimized to leverage an audience for as much attention, fame, votes, prospects, and money they'll give up.
That's a lot different from a CEO of an established company who has the power to decide who gets raises, promoted, hired, etc.
agree. There is no real downside for ordering unneeded tests and a large but rare downside for not ordering screening exams. Of course one never calls it "unneeded testing", one can always find a reason to test. Specialty societies are good at coming up with algorithms to justify procedures/testing. The Virginia case received wide attention since it involved an academic center following best practices. "Malpractice" seems to involve bad outcomes followed by poor communications. Agree that the liability system is dysfunctional. The Ga Supreme Court decided that one does not have to prove breech of duty followed by causation prior to damages. They stated a jury could start with damages and not decide whether a breech occurred until after damages had been decided. (case involved jury instructions).
"It takes a long time for people to accept that an industry is dying. .... And there are plenty of young people who want to try becoming college professors."
Are you saying this option is dying? Really? It is true that more people want these jobs in a time where the number of openings is decreasing due to increase adjuncts, fellows, etc. but is that the same as dying? Will colleges still be hiring professors 20 years from now? Will there be any professors losing jobs (excepting small struggling private colleges that close or merge)?
Re: "If your views do not persuade someone on the other side, you can choose to believe that the other side is irrational or that your side has a weak case."
Isn't there another possibility, namely that one's interlocutor has different prior beliefs, which constrain rational persuasion? If I understand correctly, convergence of beliefs via Bayesian "updating" of beliefs would be more elusive if people start from different baseline beliefs.
Per life-year is doing a HUGE amount of work here in this argument.
Important work.
75 year olds, or 65, 55, getting a colonoscopy, and a few of them getting pre-cancer treatment that saves them from cancer saves how many life-years? How much cost?
(cost is payment for treatment, tho decision cost by the patient includes the discomfort & hassle of getting an uncomfortable procedure for little likely personal benefit.)
I don't know the studies, but this is a great metric.
One would think in a competitive marketplace that skepticism of the cost effectiveness of some of these procedures would drive new solutions, like Cologuard or the various cancer detecting blood tests in development.
Sure. But currently the accuracy of those tests is rather poor. Maybe one day that will change but in the mean time we have colonoscopies that are very accurate. Their main downfall (more than complications or cost) is when the signs of cancer start and progress too far in the time between coloscopies.
Respectfully, this is a good example of why AI doesn't often seem useful to me and also how to appear to respond to a question without really engaging it.
You don't actually specify any costs, but sink into percentages.
What's the cost of EVERYONE getting a colonoscopy on schedule. Massive, because EVERYONE is a lot of people getting one every few years.
Plus, I think the AI mis-spoke. Probably should be that "The net COST of US colonoscopy recommendations is estimated to be around 1.64 per 1000 for bleeding and 0.85 per 1000 for perforation". That would be instances of those negative side effects.
So, the cost is the cost of the colonoscopies themselves, plus the cost of the side effects, plus the non-economic costs. Obviously, that's a huge cost.
The benefits you cite are all indirect. A "50-70%" chance of avoiding death by certain types of cancer. Well, that's good, but the cost depends on how many people get those cancers in the first place. Which is quite small.
The NordICC study essentially blew up the rationale for screening colonoscopy.https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5435368/ The AI missed this critical study. The 4 groups recommending colonoscopy all make money from the procedure. The data about how often to have screening colonoscopy itself is weak. 10 years, current recommendations, 5 years, 3 years, every 6 months, there is data to claim any of these. Cost effectiveness data is useless in the US because of plaintiff attorneys. If you do not recommend and document the recommendation you will be sued for negligence and probably lose. PSA tests are known to be essentially useless yet are done because a jury in Virginia decided that following the recommendations to not do a PSA under a certain age was negligent. GI physicians, ASCs, hospitals, and anesthesia make money from recommending screening colonoscopy. No one makes money from telling the patient you do not need it.
It wasn't just "one jury in Virginia" for the PSA testing decision liability question. The Sunaryo paper is a good place to start. It is, however a good example to use to explain the problem with incentives in "defensive medicine".
There were about 20 lawsuits per year (about a third settled, a fifth losing in court, and the remainder winning) going back to 2000, which isn't all that many when you compare to how many prostate cancers and related deaths there are, and for which the total payouts didn't sum to much money at all compared to the overall system costs for tests and treatments. It demonstrates the nature of "malpractice insurance" as being "often not actually about bad practice, but having the "struck-by-lightning"-like bad luck for being sued for good practice but still losing because American medical liability adjudication is utterly dysfunctional as juries simply won't follow the evidence."
But since going through the whole ordeal is excruciating, why bother trying to not order unnecessary tests when no one will crucify you likewise for doing so? I've suggested that if we can't fix any other part of the liability system, we should award doctors a large bonus whenever they get sued for following whatever counts as the expert consensus guidelines and recommendations in their field. They may lose the case, but getting a bonus will encourage them to keep not doing bad tests.
The cited results and recommendations for colonoscopy are an example of falsely restricting the desired endpoint. All cause mortality would be a much better endpoint. Also, none of these studies were based on randomized control trials. See the analysis by Vinay Prasad here: https://youtu.be/SMRS4-ng8T0?si=bAKpDyA35LscHvLi.
I have been following the debate between Hanson and Alexander on their Substacks, but I have to say that your comment "What Robin argues is that treatments that work are, in the aggregate, offset by other treatments that cause harm" did more to explain his position than I got from reading him directly
"Take something like colonoscopy. On net, do the benefits outweigh the harms? I am skeptical, because I doubt that the benefits are as large as people have been led to believe."
So you are more worried about a perforated bowel than colon cancer? Really? Colon cancer is major cause of death. How many people die from a perforated bowel? If everyone over 50 got regular colonoscopies would the number be significant then?
Maybe you should read and listen to this.
https://peterattiamd.com/peter-on-the-importance-of-regular-colonoscopies/
I would love to read your explanation of how he is wrong. Maybe adverse selection? The people most likely to get frequent colonoscopies are the people least likely to need it. Can you make that case, or any other? Does your opinion have any basis in scientific method?
COVID was a leading cause of death for two years, and I didn't even think it was worth wearing a mask for. Let alone getting something shoved up my ass.
Not relevant to the links you posted, but a link _for_ Arnold Kling in case it is a substack he is not reading: https://substack.com/home/post/p-144153862?source=queue -- Drew Haugen is training XGBoost to predict whether a pitcher will need Tommy John Surgery. So far he has teased out some factors that appear to be relevant but it is not at 'should I draft this man to my fantasy football team' yet.
If the pitchers who XGBoost says are most likely to need Tommy John Surgery read XGBoost, they are going to adjust their behavior to lower their chances of needing Tommy John Surgery and so by XGBoost being right it will also make itself wrong. Something like an Efficient Information Hypothesis, or a more general "Efficient Prediction Hypothesis". Agents knowing the forecast will neutralize the forecast. "You can't beat the future."
'Adjusting their behaviour to lower the chance of needing TJ surgery' has not been a notable characteristic of pitchers so far. Of course, many of them think that having TJ surgery is a net positive, even with the year long rehab. For the fantasy baseball player, though, there is just a downside.
Speaking of medical efficacy, there was a great article about ECMO machines in the New Yorker a couple days ago that really showcases how incredible medical innovations to come will be. Apparently, we really aren't that far off from a world with Futurama style heads in jars: https://www.newyorker.com/science/annals-of-medicine/how-ecmo-is-redefining-death
A minor point: “treatments that cause harm” can easily be misinterpreted. Some treatments cause harm in a direct, obvious way, as when a patient dies on the operating table while undergoing a knee replacement, or—less dramatically—a diagnostic procedure that yields a negative result slightly damages the patient’s health. But I suspect the more common case is that the treatment’s direct effect is slightly good but not worth the cost, and the alternative use to which the resources would have been put if the treatment had not been done would have had effects not just *good*, but *good for health*. Activities that are not billed as “health care” can still have effects on health.
I am skeptical we have good data on what treatment doesn't "work," much less which cause harm. On this week's episode of Econtalk, Roberts talks about his mother's compression fracture in her back. treatment options include doing nothing, a back brace, and surgery to insert cement into the bone. A study(s) has indicated surgery is no more successful than a placebo. Roberts, his siblings, and even his mother knew this ye they opted for surgery. Why? A doctor said the outcome are better. Given that a placebo isn't an option, he could very well be right. Either way, can one make a factual statement that he is wrong? how does one collect factual data on which interventions cause harm? Color me skeptical we have good data on this. Seems like it can't be done without huge subjectivity. Makes me think of happiness data. Is either reliable?
"I can tell you stories about close relatives who were clearly the victims of Hansonian medicine. They would be better off without the medical interventions that they underwent."
What do your anecdotal cases tell us? Is this because that type of intervention should never be done, the doctor/facility doing it lacked the adequate capability, or in a certain number of cases it simply fails but overall the statistics are good for this intervention?
A thing you are missing vis a vis the colonoscopy issue is that the effects on mortality are not the only important benefit. People want to know whether they have cancer. Finding out they don't have one of the most common cancers will help a lot of people sleep at night. You can argue that people shouldn't be like this but good luck changing their minds.
Information is worth money as well.
From the Kotkin:
"Conservatives of course have their off-key women like Laura Bobert and Marjorie Taylor Greem"
Lol
Bobert and Green aren't leaders of large organizations or chief executives authorized to make major decisions for established institutions. As prominent "politicians in American democracy" they are just media figures, i.e., "actresses". "And the Oscar goes to ..." They are, if anything, "Managers of their own brand", not very distinguishable in principle from online 'influencers', Instagram models or OnlyFans, ahem, "providers" squeezing simps and paypigs for all they're worth, though arguably this is unfair to the OnlyFans freelance pornographers and prostitutes who, unlike these Republican politicians, are actually honest businesswomen and reliably fulfill their promises and deliver what they are paid to put out. The business of those politicians is to be celebrities endorsing the product that is "themselves". Not their actual selves but a not-too-distant-but-still-somewhat-faked role to act out in public, refined, focus-grouped, and stragically crafted as an exaggerated caricature version of a public persona, market-optimized to leverage an audience for as much attention, fame, votes, prospects, and money they'll give up.
That's a lot different from a CEO of an established company who has the power to decide who gets raises, promoted, hired, etc.
I just thought the double typo was funny, agree of course that they are small fries
agree. There is no real downside for ordering unneeded tests and a large but rare downside for not ordering screening exams. Of course one never calls it "unneeded testing", one can always find a reason to test. Specialty societies are good at coming up with algorithms to justify procedures/testing. The Virginia case received wide attention since it involved an academic center following best practices. "Malpractice" seems to involve bad outcomes followed by poor communications. Agree that the liability system is dysfunctional. The Ga Supreme Court decided that one does not have to prove breech of duty followed by causation prior to damages. They stated a jury could start with damages and not decide whether a breech occurred until after damages had been decided. (case involved jury instructions).
"It takes a long time for people to accept that an industry is dying. .... And there are plenty of young people who want to try becoming college professors."
Are you saying this option is dying? Really? It is true that more people want these jobs in a time where the number of openings is decreasing due to increase adjuncts, fellows, etc. but is that the same as dying? Will colleges still be hiring professors 20 years from now? Will there be any professors losing jobs (excepting small struggling private colleges that close or merge)?
Re: "If your views do not persuade someone on the other side, you can choose to believe that the other side is irrational or that your side has a weak case."
Isn't there another possibility, namely that one's interlocutor has different prior beliefs, which constrain rational persuasion? If I understand correctly, convergence of beliefs via Bayesian "updating" of beliefs would be more elusive if people start from different baseline beliefs.
Calculate the cost per life-year saved.
Per life-year is doing a HUGE amount of work here in this argument.
Important work.
75 year olds, or 65, 55, getting a colonoscopy, and a few of them getting pre-cancer treatment that saves them from cancer saves how many life-years? How much cost?
(cost is payment for treatment, tho decision cost by the patient includes the discomfort & hassle of getting an uncomfortable procedure for little likely personal benefit.)
I don't know the studies, but this is a great metric.
One would think in a competitive marketplace that skepticism of the cost effectiveness of some of these procedures would drive new solutions, like Cologuard or the various cancer detecting blood tests in development.
Sure. But currently the accuracy of those tests is rather poor. Maybe one day that will change but in the mean time we have colonoscopies that are very accurate. Their main downfall (more than complications or cost) is when the signs of cancer start and progress too far in the time between coloscopies.
Respectfully, this is a good example of why AI doesn't often seem useful to me and also how to appear to respond to a question without really engaging it.
You don't actually specify any costs, but sink into percentages.
What's the cost of EVERYONE getting a colonoscopy on schedule. Massive, because EVERYONE is a lot of people getting one every few years.
Plus, I think the AI mis-spoke. Probably should be that "The net COST of US colonoscopy recommendations is estimated to be around 1.64 per 1000 for bleeding and 0.85 per 1000 for perforation". That would be instances of those negative side effects.
So, the cost is the cost of the colonoscopies themselves, plus the cost of the side effects, plus the non-economic costs. Obviously, that's a huge cost.
The benefits you cite are all indirect. A "50-70%" chance of avoiding death by certain types of cancer. Well, that's good, but the cost depends on how many people get those cancers in the first place. Which is quite small.
https://www.health.ny.gov/statistics/cancer/registry/table6/tb6colonnys.htm
So (very) roughly out of 1000 people (just to make a round number), it looks like 1 gets colon cancer. So, performing a 1000 colonoscopies will
1. result in about 1-2 serious adverse side effects
2. cost whatever 1000 colonoscopies cost
3. save the life of 0.5 people who get a colonoscopy
4. not save the life of 0.5 people who get a colonoscopy
It's very hard to imagine values where this pays off well at a societal level.
The NordICC study essentially blew up the rationale for screening colonoscopy.https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5435368/ The AI missed this critical study. The 4 groups recommending colonoscopy all make money from the procedure. The data about how often to have screening colonoscopy itself is weak. 10 years, current recommendations, 5 years, 3 years, every 6 months, there is data to claim any of these. Cost effectiveness data is useless in the US because of plaintiff attorneys. If you do not recommend and document the recommendation you will be sued for negligence and probably lose. PSA tests are known to be essentially useless yet are done because a jury in Virginia decided that following the recommendations to not do a PSA under a certain age was negligent. GI physicians, ASCs, hospitals, and anesthesia make money from recommending screening colonoscopy. No one makes money from telling the patient you do not need it.
It wasn't just "one jury in Virginia" for the PSA testing decision liability question. The Sunaryo paper is a good place to start. It is, however a good example to use to explain the problem with incentives in "defensive medicine".
There were about 20 lawsuits per year (about a third settled, a fifth losing in court, and the remainder winning) going back to 2000, which isn't all that many when you compare to how many prostate cancers and related deaths there are, and for which the total payouts didn't sum to much money at all compared to the overall system costs for tests and treatments. It demonstrates the nature of "malpractice insurance" as being "often not actually about bad practice, but having the "struck-by-lightning"-like bad luck for being sued for good practice but still losing because American medical liability adjudication is utterly dysfunctional as juries simply won't follow the evidence."
But since going through the whole ordeal is excruciating, why bother trying to not order unnecessary tests when no one will crucify you likewise for doing so? I've suggested that if we can't fix any other part of the liability system, we should award doctors a large bonus whenever they get sued for following whatever counts as the expert consensus guidelines and recommendations in their field. They may lose the case, but getting a bonus will encourage them to keep not doing bad tests.
https://peterattiamd.com/colonoscopy-and-acm/
Plus there's the cost of all that toilet paper you wind up using preparing for the colonoscopy.
The cited results and recommendations for colonoscopy are an example of falsely restricting the desired endpoint. All cause mortality would be a much better endpoint. Also, none of these studies were based on randomized control trials. See the analysis by Vinay Prasad here: https://youtu.be/SMRS4-ng8T0?si=bAKpDyA35LscHvLi.
All cause mortality is so noisy, you'll never get a result even for things that we know save lives.