Thanks again for rationally explaining this issue.
The fact that the American public is so slow on the uptake of this issue, as well as other issues, is testimony that this country is on a crap slide to hell.
Gonna put my head out there - having worked with some Mount Sinai folks on research and clinical - they generally have not been a good academic center for a bit.
Many PIs are clout chasing more than doing real research.
Good summary. You're probs right, but let me advocate for the Devil anyways:
You berate the review authors for using their "expert" judgement, yet throughout this essay you tell us that significant results are "marginally" or "barely" significant. I'm afraid it's a binary (even if the cutoff is rather arbitrary), and your personal qualifiers here are irrelevant. Adding these qualifiers to significant results tells me more about your bias than the biological significance of the data.
Speaking of biological significance, we both know that statistical significance and biological significance are completely different things. If a p-value comes out a little high (you call 0.01 "marginal"), then this could be for several reasons. It could be that the effect isn't real (your implication). It could equally be that the samples are too heterogenous or that the sample sizes are too small to pick up the real effect size.
The Swedish study has small confidence intervals ("considerable precision"). Why might that be? Sweden is ethnically homogenous. Autism rates vary by ethnicity/race, with whites being lower than blacks or hispanics. Hypothesis: tylenol might contribute to ASD if administered during pregnancy *given* a certain genetic background. So in a racially diverse study, we get noisy data with large confidence intervals and a "marginally" significant result. In a predominantly white study (Sweden), precision is higher? Happy for you to attack this idea, just struck me as perhaps valid?
Sibling study alternative explanation: tylenol use during pregnancy *or breastfeeding* contributes to ASD (perhaps depending on genetic background, per above). Makes complete sense that circulating tylenol gets passed to kids in breast milk. ASD onset is like aged 4-5, so breastmilk tylenol can feasibly go to town on the little developing brain before this. Sweden has very high rates of breastfeeding. No data on tylenol usage during breastfeeding in these studies but we can imagine a mother who takes it during pregnancy of one child might happen to take it during breastfeeding of the other child at a higher rate? Also the sensitivity analysis for the order of exposed children is tricky. Not sure we can garner much given the increasing rates of diagnosis year-on-year (which the authors acknowledge). Curious to know your thoughts on breastmilk (I mean, in the context of tylenol exposure...). I am either shifting the goalposts here or leaving no stone unturned to explain the data--depending on your perspective!
Also in general a lot of chitchat about causal evidence (I mean beyond just your article here). Your Denmark pseudo experiment is cool, but pretty much everyone I know has stocks of tylenol in their house, so dip in purchases is likely smoothed out for actual usage. Further, the dip in purchase was only 10% (I guess it is quite sizeable but that's just a subjective call). One might imagine that pregnant mothers who think they need tylenol would be more incentivised to get some versus other people. I would imagine that pregnant mothers' tylenol purchasing habits were more robust to regulation changes than the general population.
Your Mendelian randomization stuff is v cool. But perhaps not that relevant. Tylenol could biologically synergise with neuronal function SNPs to promote ASD without showing a genetic footprint in drug pharmacokinetics. Also not sure how much these drug pathway SNPs can do for a tiny developing human versus a full grown human. And wouldn't the SNPs of the mother be more important for systemic clearance of the drug anyways? Foetus pharmacokinetics are probably less influenced by foetus genetics than the genetics of their mother?
Can we all just acknowledge that we have zero causal evidence until an RCT is done, and that HHS should commission an independent RCT to set this stuff to rest? Surely RFK is tired of the endless debating. A well designed RCT would just be such a silver bullet for these issues (same with vaccines and autism tbh). And if you're convinced that intervention X doesn't cause autism and therefore insist that such an RCT would be such a waste of money because the result is obvious, consider the societal benefit of an RFK-approved RCT in persuading half of America of the truth.
"You berate the review authors for using their "expert" judgement, yet throughout this essay you tell us that significant results are "marginally" or "barely" significant."
This is an established term. There is no "yet" to speak of.
"I'm afraid it's a binary (even if the cutoff is rather arbitrary), and your personal qualifiers here are irrelevant."
This isn't true, as the cachet given to the binary is the ultimate source of publication bias.
"It could be that the effect isn't real (your implication)."
You should go read my article on this topic instead of asserting that I've implied things I haven't: https://www.cremieux.xyz/p/ranking-fields-by-p-value-suspiciousness. You should also read my prior statements on this, such as those affirming that the effect is exaggeration of effect sizes when the finding is true, implying that I don't think this automatically makes effects unreal.
"It could equally be that the samples are too heterogenous or that the sample sizes are too small to pick up the real effect size."
This would not place the excessively many results coincidentally right under the significance threshold. It's not a legitimate excuse.
"Sweden is ethnically homogenous."
And yet their variances for many of the traits relevant to this discussion are practically as wide as in America, thus rendering consideration of ethnic homogeneity for the confidence intervals meaningless unless you're willing to make the unwarranted assertion of differences in the causal relationship due to diversity elsewhere.
"Hypothesis: tylenol might contribute to ASD if administered during pregnancy *given* a certain genetic background."
Ruled out at any reasonable effect size by the GWAS results.
"Sibling study alternative explanation: tylenol use during pregnancy *or breastfeeding* contributes to ASD (perhaps depending on genetic background, per above)."
Note the lack of published evidence for postnatal effects and the general finding of ORs around 1.
"Also the sensitivity analysis for the order of exposed children is tricky. Not sure we can garner much given the increasing rates of diagnosis year-on-year (which the authors acknowledge)."
We can see the balance by exposure order very clearly and we can readily reject sizable biases from it.
"so dip in purchases is likely smoothed out for actual usage."
To the tune of *no meaningful effect* on usage? This is hardly believable given known compliance rates for prescription usage and the extremely high rates of prescription-based usage in Scandinavia.
"Your Mendelian randomization stuff is v cool. But perhaps not that relevant. Tylenol could biologically synergise with neuronal function SNPs to promote ASD without showing a genetic footprint in drug pharmacokinetics."
If this happens without any involvement from regulators of Tylenol's metabolism, that would be a scientific miracle. It's not realistic.
"And wouldn't the SNPs of the mother be more important for systemic clearance of the drug anyways?"
You'll find that the SNPs of mothers and the SNPs of their children are correlated at r = 0.50.
"Can we all just acknowledge that we have zero causal evidence until an RCT is done"
No, because we have multiple pieces of causal evidence that are not RCTs.
"and that HHS should commission an independent RCT to set this stuff to rest?"
An RCT for what? You can't ethically use a placebo to break fevers in pregnancy and there are no acetaminophen alternatives, so there is no viable RCT design.
Thanks. Helpful. Perhaps a bit more clarification on a few points:
"Ruled out at any reasonable effect size by the GWAS results."
> We're talking about what here? Which GWAS? I'm suggesting ASD-associated SNPs might synergise with tylenol. Did any of the tylenol studies integrate GWAS??
"If this happens without any involvement from regulators of Tylenol's metabolism, that would be a scientific miracle. It's not realistic."
> Really? Why? Let me spell out how it could work:
- acetaminophen metabolite activates TRPV1 on the surface of cells in the brain (a current MOA hypothesis)
- TRPV1 recycling between endosomes and cell surface is regulated by a sorting nexin encoded by SNX19
- SNX19 SNPs are associated with ASD
- SNP to SNX19 affects it's regulation of TRPV1 membrane trafficking
- the ability of tylenol to affect brain cells, potentiating development of autism, is modulated by an ASD-associated SNP without any need for a tylenol metabolism-associated SNP.
"You'll find that the SNPs of mothers and the SNPs of their children are correlated at r = 0.50."
> Yeah r=0.5 kinda destroys the power to reliably infer maternal clearance rates. Also these tylenol SNPs seem to affect clearance by a few hours at most. So like 4 hours to 2 hours or something. That's not really going to affect the ability of the drug to do stuff to the brain of the foetus that much, especially if it's a neuronal pathway that gets activated within minutes. Seems like that would just be an acute effect that occurs regardless of the half-life of the tylenol?
"An RCT for what? You can't ethically use a placebo to break fevers in pregnancy and there are no acetaminophen alternatives, so there is no viable RCT design."
> So eliminate those who develop fever from the trial. Use a large enough cohort to have enough remaining mothers by the end of pregnancy who did not have tylenol-requiring fever. It's like 10-15% of mothers who get a fever at any point during pregnancy. Just adjust the sample size up by a compensating factor. If you get fever, you're out the study and can take tylenol as you wish. Not only is it viable but it is incredibly simple. No ethical concerns with randomising a group to persistently take tylenol since the null hypothesis is that it doesn't do anything to the foetus. Am I wrong?
"We're talking about what here? Which GWAS? I'm suggesting ASD-associated SNPs might synergise with tylenol. Did any of the tylenol studies integrate GWAS??"
All ASD GWAS done to date.
"Really? Why? Let me spell out how it could work"
Your explanation did not, in fact, spell this out. It sounds like you went to an LLM and asked it for something, and it gave you back something irrelevant.
"Yeah r=0.5 kinda destroys the power to reliably infer maternal clearance rates."
You've missed the point that bias is capped and we can do probabilistic sensitivity analyses with this in mind.
"So eliminate those who develop fever from the trial."
And do what? Recommend people take acetaminophen for no reason? This would be a terribly unjustifiable trial. As a rule, you do not needlessly subject people to possible harms, and especially not pregnant women.
Okay I don’t think you’re trying to understand any of my points. And now you’re into the ad hominems.
For the record, I looked up the top ASD SNPs on Google (sorry for outsourcing to a search engine!). Noticed SNX19 because a colleague during my PhD published a paper on it showing how it mediates trafficking between ER and endolysosomes—so it stuck out to me. A SNP here could feasibly affect any cell surface receptor. I then googled papers on the MOA of acetaminophen. Found one that implicated TRPV1. Googled whether it’s expressed in brain. It is. Works as a cell surface receptor. Put it all together as a back-of-the-napkin bit of biology —> gets accused of using an LLM 😭😭
No ASD GWAS to date has included a compensating factor to try to estimate the tylenol metabolism SNPs of the mother. That we can do this analysis is true, that nobody has ever done this analysis is also true.
“And do what? Recommend people take acetaminophen for no reason?”
> errr yeah? That’s what an RCT is. Would be just like any vaccine trial ever. Besides, you’ve written a bunch of substack articles arguing why tylenol is safe, and now you’re saying “oh we can’t ethically do an RCT because it might not be safe”.
Why are you so closed-minded to alternative explanations of the data, or to suggestions of a trial to generate actual causal data? This interaction has been a little annoying tbh. You’re not as intellectually honest as I’d hoped :/ put down the guard and engage in good faith maybe?
"No ASD GWAS to date has included a compensating factor to try to estimate the tylenol metabolism SNPs of the mother."
OK. You have no idea what you're talking about and you're just throwing words around.
You may go download the results from different ASD GWAS' and, as I already noted, find that there's nothing related to acetaminophen metabolism.
"errr yeah? That’s what an RCT is. Would be just like any vaccine trial ever."
Vaccine trials do not tend to use placebo vaccination after it's known that we have a working vaccine, for ethical reasons.
"Besides, you’ve written a bunch of substack articles arguing why tylenol is safe, and now you’re saying “oh we can’t ethically do an RCT because it might not be safe”"
You visibly did not understand what I've said. That Tylenol is safe does not mean that women should just pop pills for no reason. Tylenol being safe does not mean that it introduces zero risk. It is quite easy to induce hepatoxicity and there can be other routes through which chronic use harms infants.
You need to familiarize yourself with the basics of trial ethics before returning to this conversation, if you decide to. If you visibly do not do that, I'm just not going to respond.
"Why are you so closed-minded"
I'm not close-minded about anything except obvious things like being against assaults and murders. I change my mind when new evidence arrives all the time. There's an extensive online record attesting to this fact and your accusation is just an expression of your being uninformed.
1. But do you really think RFK would do any RCT? When it seems it's easier to go by gut belief and enforce his beliefs?
2. You will actually force pregnant mothers people to take tylenol vs no tylenol - not gonna fly with a lot of subjects. Especially if you are trying to take out the confounding "fever” /
1. I don't know whether he would, I'm just saying he should.
2. Yeah I mean the null hypothesis is that tylenol is fine, so no issue forcing mothers to take it. If anyone gets serious fever then they are removed from the trial instantly (and can then take tylenol if they were in the placebo group)
Thanks again for rationally explaining this issue.
The fact that the American public is so slow on the uptake of this issue, as well as other issues, is testimony that this country is on a crap slide to hell.
Gonna put my head out there - having worked with some Mount Sinai folks on research and clinical - they generally have not been a good academic center for a bit.
Many PIs are clout chasing more than doing real research.
Good summary. You're probs right, but let me advocate for the Devil anyways:
You berate the review authors for using their "expert" judgement, yet throughout this essay you tell us that significant results are "marginally" or "barely" significant. I'm afraid it's a binary (even if the cutoff is rather arbitrary), and your personal qualifiers here are irrelevant. Adding these qualifiers to significant results tells me more about your bias than the biological significance of the data.
Speaking of biological significance, we both know that statistical significance and biological significance are completely different things. If a p-value comes out a little high (you call 0.01 "marginal"), then this could be for several reasons. It could be that the effect isn't real (your implication). It could equally be that the samples are too heterogenous or that the sample sizes are too small to pick up the real effect size.
The Swedish study has small confidence intervals ("considerable precision"). Why might that be? Sweden is ethnically homogenous. Autism rates vary by ethnicity/race, with whites being lower than blacks or hispanics. Hypothesis: tylenol might contribute to ASD if administered during pregnancy *given* a certain genetic background. So in a racially diverse study, we get noisy data with large confidence intervals and a "marginally" significant result. In a predominantly white study (Sweden), precision is higher? Happy for you to attack this idea, just struck me as perhaps valid?
Sibling study alternative explanation: tylenol use during pregnancy *or breastfeeding* contributes to ASD (perhaps depending on genetic background, per above). Makes complete sense that circulating tylenol gets passed to kids in breast milk. ASD onset is like aged 4-5, so breastmilk tylenol can feasibly go to town on the little developing brain before this. Sweden has very high rates of breastfeeding. No data on tylenol usage during breastfeeding in these studies but we can imagine a mother who takes it during pregnancy of one child might happen to take it during breastfeeding of the other child at a higher rate? Also the sensitivity analysis for the order of exposed children is tricky. Not sure we can garner much given the increasing rates of diagnosis year-on-year (which the authors acknowledge). Curious to know your thoughts on breastmilk (I mean, in the context of tylenol exposure...). I am either shifting the goalposts here or leaving no stone unturned to explain the data--depending on your perspective!
Also in general a lot of chitchat about causal evidence (I mean beyond just your article here). Your Denmark pseudo experiment is cool, but pretty much everyone I know has stocks of tylenol in their house, so dip in purchases is likely smoothed out for actual usage. Further, the dip in purchase was only 10% (I guess it is quite sizeable but that's just a subjective call). One might imagine that pregnant mothers who think they need tylenol would be more incentivised to get some versus other people. I would imagine that pregnant mothers' tylenol purchasing habits were more robust to regulation changes than the general population.
Your Mendelian randomization stuff is v cool. But perhaps not that relevant. Tylenol could biologically synergise with neuronal function SNPs to promote ASD without showing a genetic footprint in drug pharmacokinetics. Also not sure how much these drug pathway SNPs can do for a tiny developing human versus a full grown human. And wouldn't the SNPs of the mother be more important for systemic clearance of the drug anyways? Foetus pharmacokinetics are probably less influenced by foetus genetics than the genetics of their mother?
Can we all just acknowledge that we have zero causal evidence until an RCT is done, and that HHS should commission an independent RCT to set this stuff to rest? Surely RFK is tired of the endless debating. A well designed RCT would just be such a silver bullet for these issues (same with vaccines and autism tbh). And if you're convinced that intervention X doesn't cause autism and therefore insist that such an RCT would be such a waste of money because the result is obvious, consider the societal benefit of an RFK-approved RCT in persuading half of America of the truth.
"You berate the review authors for using their "expert" judgement, yet throughout this essay you tell us that significant results are "marginally" or "barely" significant."
This is an established term. There is no "yet" to speak of.
"I'm afraid it's a binary (even if the cutoff is rather arbitrary), and your personal qualifiers here are irrelevant."
This isn't true, as the cachet given to the binary is the ultimate source of publication bias.
"It could be that the effect isn't real (your implication)."
You should go read my article on this topic instead of asserting that I've implied things I haven't: https://www.cremieux.xyz/p/ranking-fields-by-p-value-suspiciousness. You should also read my prior statements on this, such as those affirming that the effect is exaggeration of effect sizes when the finding is true, implying that I don't think this automatically makes effects unreal.
"It could equally be that the samples are too heterogenous or that the sample sizes are too small to pick up the real effect size."
This would not place the excessively many results coincidentally right under the significance threshold. It's not a legitimate excuse.
"Sweden is ethnically homogenous."
And yet their variances for many of the traits relevant to this discussion are practically as wide as in America, thus rendering consideration of ethnic homogeneity for the confidence intervals meaningless unless you're willing to make the unwarranted assertion of differences in the causal relationship due to diversity elsewhere.
"Hypothesis: tylenol might contribute to ASD if administered during pregnancy *given* a certain genetic background."
Ruled out at any reasonable effect size by the GWAS results.
"Sibling study alternative explanation: tylenol use during pregnancy *or breastfeeding* contributes to ASD (perhaps depending on genetic background, per above)."
Note the lack of published evidence for postnatal effects and the general finding of ORs around 1.
"Also the sensitivity analysis for the order of exposed children is tricky. Not sure we can garner much given the increasing rates of diagnosis year-on-year (which the authors acknowledge)."
We can see the balance by exposure order very clearly and we can readily reject sizable biases from it.
"so dip in purchases is likely smoothed out for actual usage."
To the tune of *no meaningful effect* on usage? This is hardly believable given known compliance rates for prescription usage and the extremely high rates of prescription-based usage in Scandinavia.
"Your Mendelian randomization stuff is v cool. But perhaps not that relevant. Tylenol could biologically synergise with neuronal function SNPs to promote ASD without showing a genetic footprint in drug pharmacokinetics."
If this happens without any involvement from regulators of Tylenol's metabolism, that would be a scientific miracle. It's not realistic.
"And wouldn't the SNPs of the mother be more important for systemic clearance of the drug anyways?"
You'll find that the SNPs of mothers and the SNPs of their children are correlated at r = 0.50.
"Can we all just acknowledge that we have zero causal evidence until an RCT is done"
No, because we have multiple pieces of causal evidence that are not RCTs.
"and that HHS should commission an independent RCT to set this stuff to rest?"
An RCT for what? You can't ethically use a placebo to break fevers in pregnancy and there are no acetaminophen alternatives, so there is no viable RCT design.
Thanks. Helpful. Perhaps a bit more clarification on a few points:
"Ruled out at any reasonable effect size by the GWAS results."
> We're talking about what here? Which GWAS? I'm suggesting ASD-associated SNPs might synergise with tylenol. Did any of the tylenol studies integrate GWAS??
"If this happens without any involvement from regulators of Tylenol's metabolism, that would be a scientific miracle. It's not realistic."
> Really? Why? Let me spell out how it could work:
- acetaminophen metabolite activates TRPV1 on the surface of cells in the brain (a current MOA hypothesis)
- TRPV1 recycling between endosomes and cell surface is regulated by a sorting nexin encoded by SNX19
- SNX19 SNPs are associated with ASD
- SNP to SNX19 affects it's regulation of TRPV1 membrane trafficking
- the ability of tylenol to affect brain cells, potentiating development of autism, is modulated by an ASD-associated SNP without any need for a tylenol metabolism-associated SNP.
"You'll find that the SNPs of mothers and the SNPs of their children are correlated at r = 0.50."
> Yeah r=0.5 kinda destroys the power to reliably infer maternal clearance rates. Also these tylenol SNPs seem to affect clearance by a few hours at most. So like 4 hours to 2 hours or something. That's not really going to affect the ability of the drug to do stuff to the brain of the foetus that much, especially if it's a neuronal pathway that gets activated within minutes. Seems like that would just be an acute effect that occurs regardless of the half-life of the tylenol?
"An RCT for what? You can't ethically use a placebo to break fevers in pregnancy and there are no acetaminophen alternatives, so there is no viable RCT design."
> So eliminate those who develop fever from the trial. Use a large enough cohort to have enough remaining mothers by the end of pregnancy who did not have tylenol-requiring fever. It's like 10-15% of mothers who get a fever at any point during pregnancy. Just adjust the sample size up by a compensating factor. If you get fever, you're out the study and can take tylenol as you wish. Not only is it viable but it is incredibly simple. No ethical concerns with randomising a group to persistently take tylenol since the null hypothesis is that it doesn't do anything to the foetus. Am I wrong?
"We're talking about what here? Which GWAS? I'm suggesting ASD-associated SNPs might synergise with tylenol. Did any of the tylenol studies integrate GWAS??"
All ASD GWAS done to date.
"Really? Why? Let me spell out how it could work"
Your explanation did not, in fact, spell this out. It sounds like you went to an LLM and asked it for something, and it gave you back something irrelevant.
"Yeah r=0.5 kinda destroys the power to reliably infer maternal clearance rates."
You've missed the point that bias is capped and we can do probabilistic sensitivity analyses with this in mind.
"So eliminate those who develop fever from the trial."
And do what? Recommend people take acetaminophen for no reason? This would be a terribly unjustifiable trial. As a rule, you do not needlessly subject people to possible harms, and especially not pregnant women.
Okay I don’t think you’re trying to understand any of my points. And now you’re into the ad hominems.
For the record, I looked up the top ASD SNPs on Google (sorry for outsourcing to a search engine!). Noticed SNX19 because a colleague during my PhD published a paper on it showing how it mediates trafficking between ER and endolysosomes—so it stuck out to me. A SNP here could feasibly affect any cell surface receptor. I then googled papers on the MOA of acetaminophen. Found one that implicated TRPV1. Googled whether it’s expressed in brain. It is. Works as a cell surface receptor. Put it all together as a back-of-the-napkin bit of biology —> gets accused of using an LLM 😭😭
No ASD GWAS to date has included a compensating factor to try to estimate the tylenol metabolism SNPs of the mother. That we can do this analysis is true, that nobody has ever done this analysis is also true.
“And do what? Recommend people take acetaminophen for no reason?”
> errr yeah? That’s what an RCT is. Would be just like any vaccine trial ever. Besides, you’ve written a bunch of substack articles arguing why tylenol is safe, and now you’re saying “oh we can’t ethically do an RCT because it might not be safe”.
Why are you so closed-minded to alternative explanations of the data, or to suggestions of a trial to generate actual causal data? This interaction has been a little annoying tbh. You’re not as intellectually honest as I’d hoped :/ put down the guard and engage in good faith maybe?
"No ASD GWAS to date has included a compensating factor to try to estimate the tylenol metabolism SNPs of the mother."
OK. You have no idea what you're talking about and you're just throwing words around.
You may go download the results from different ASD GWAS' and, as I already noted, find that there's nothing related to acetaminophen metabolism.
"errr yeah? That’s what an RCT is. Would be just like any vaccine trial ever."
Vaccine trials do not tend to use placebo vaccination after it's known that we have a working vaccine, for ethical reasons.
"Besides, you’ve written a bunch of substack articles arguing why tylenol is safe, and now you’re saying “oh we can’t ethically do an RCT because it might not be safe”"
You visibly did not understand what I've said. That Tylenol is safe does not mean that women should just pop pills for no reason. Tylenol being safe does not mean that it introduces zero risk. It is quite easy to induce hepatoxicity and there can be other routes through which chronic use harms infants.
You need to familiarize yourself with the basics of trial ethics before returning to this conversation, if you decide to. If you visibly do not do that, I'm just not going to respond.
"Why are you so closed-minded"
I'm not close-minded about anything except obvious things like being against assaults and murders. I change my mind when new evidence arrives all the time. There's an extensive online record attesting to this fact and your accusation is just an expression of your being uninformed.
Aside from all these study/metastudy/pseudostudy
1. But do you really think RFK would do any RCT? When it seems it's easier to go by gut belief and enforce his beliefs?
2. You will actually force pregnant mothers people to take tylenol vs no tylenol - not gonna fly with a lot of subjects. Especially if you are trying to take out the confounding "fever” /
aches / pains.
1. I don't know whether he would, I'm just saying he should.
2. Yeah I mean the null hypothesis is that tylenol is fine, so no issue forcing mothers to take it. If anyone gets serious fever then they are removed from the trial instantly (and can then take tylenol if they were in the placebo group)