Comment by chithanh | Hacker News Reader

Comment by chithanh | Hacker News Reader

Parent Full thread

> Ok, so we already have your standard

Who is "we"?

I don't know what you are going on about. Empirical evidence is one thing, mechanism is another source of knowledge by which we can shape public health policy. While empirical evidence is valid only for the situation in which it was obtained, mechanism is universal.

For example, we mandate wearing seatbelts in cars in the name of public health. It is however not necessary to do seatbelt on/off RCTs with actual people. How we know that this is beneficial: Because physics, verification through crash tests (with dummies), and because we know that seatbelt mandates increase the frequency of people wearing them.

Going back to the original question, it was clearly shown in observational studies that giving children sweetened food is bad: Childhood dietary habits shape lifelong food preferences, and preference for sweet food leads to worse outcomes regarding chronic diseases later in life. This has been shown in lots of research, both in humans and in animal models:

https://doi.org/10.1126/science.adn5421

https://doi.org/10.3390/nu16030428

https://doi.org/10.1093/chemse/bjr050

With randomly adding or substituting sugar with artificial sweeteners there is however no empirical evidence nor known mechanism which supports a public health benefit. In fact the mechanisms we know from animal farming suggest a detrimental effect.

> Who is "we"?

Me and anyone else reading this.

> While empirical evidence is valid only for the situation in which it was obtained, mechanism is universal.

Mechanisms are inferred from empirical evidence, I don't see how you can treat them as two separate categories. For example, in your crash test dummy analogy, verification through crash tests (with dummies) is empirical evidence. Yet under your framework, should we assume that it is valid only for the situation in which it was obtained - only for dummies, not people; in cars pushed towards walls in controlled situations, rather than on public roads?

If you name proxy experiments that support your views (crash tests) as mechanisms and ones that don't (SSB replacement with NNS RCTs) as "empirical evidence is valid only for the situation in which it was obtained" then sure, everything you want to believe is supported by sound science and everything you don't isn't. But the view itself seems to contain a logical contradiction, so you're dead before you've even got off the ground.

I would understand mechanistic evidence in the domain of health science to be in vitro and animal studies. Even if we were to grant that mechanism is universal in this field (which I wouldn't, we frequently see heterogenous results even within the same exposures on the same mouse models, for example), there are thousands of mechanisms that come together to influence the outcomes we actually care about. This is why when we look at translation rates of mechanisms to outcomes in humans we typically see rates below 5% (and is also why pharmaceuticals that work perfectly in animal models barely ever make it to market in humans).

Going back to the evidence you've cited in support of your intervention - the first two (the only ones in humans) are neither looking at NNSs nor an intervention on banning them. So it doesn't meet your own goalpost for "if we introduce a public health policy, then we need to take human behavior and adherence rates into account". In the rationing example, you have an entirely different context - one in which people literally cannot purchase large amounts of sugar. This would not be the case if we were to ban NNS today.

Your third study was in mice which, as discussed, has an incredibly low chance of actually translating into human outcomes. I don’t find “we have evidence in RCTs that NNSs are beneficial but there’s this mouse study that says otherwise so let’s ban them” a convincing argument.

So again, any actual evidence in support of your proposed intervention? How do we know, for example, that banning NNSs won't just lead to higher sugar consumption and adverse outcomes, since we know from RCTs that substituting SSBs for NNSs improves health outcomes? If all those consuming your banned substance now switch to SSBs instead of their NNSs, congratulations, you've just worsened health outcomes.

> Me and anyone else reading this.

In that case, no "we" don't, because I am reading this and I do not agree with this "standard" nor your characterization of what I wrote.

> Mechanisms are inferred from empirical evidence, I don't see how you can treat them as two separate categories. For example, in your crash test dummy analogy, verification through crash tests (with dummies) is empirical evidence.

This is not how it works. Crash tests are used for validation, but the data from crash tests is generally not used to infer mechanism. Physicists don't come up with a new theory of mechanics every time a crash test has an unexpected outcome.

> If you name proxy experiments that support your views (crash tests) as mechanisms and ones that don't (SSB replacement with NNS RCTs) as "empirical evidence is valid only for the situation in which it was obtained" then sure, everything you want to believe is supported by sound science and everything you don't isn't.

I don't think you understand. If you want to support a public health intervention you either have the empirical data with a relevant endpoint,

or you can point to mechanism which bridges the part between the data that you have and the outcome which you want to achieve.

When it comes to pharmaceutics and food additives, our mechanistic understanding is insufficient so we often have to resort to empirical studies on humans, including RCTs (Pfizer's Covid vaccine trial had tens of thousands of participants) and also observational studies at population level. And it is the last part where artificial sweeteners fail to show benefit so far.

When it comes to seat belts, our mechanistic understanding is sufficient so we don't need to resort to empirism. Yes we perform validation but only to check if there are no design oversights in the vehicle nor shortcomings with the simulation software, typically in a low triple-digit number of crash tests. But no humans involved and especially no control arm with humans.

(Well if you ignore the one study on the efficacy of parachutes which was done as RCT https://doi.org/10.1136/bmj.k5094 )

> So it doesn't meet your own goalpost

It does, because again, mechanism is provided. You could say that the study has weak evidence for the mechanism and it works like that perhaps only for sugar. Because that some mechanism is found in mice does not mean it is also found in humans, and it would be a fair point. This is why many species are tested and so far the results held up (testing humans takes too long for obvious reasons).

Sorry, I summarised what I thought was your goalpost earlier in the exact same words and you didn’t correct me, so I made the assumption in subsequent replies. I’m not interested in straw manning your argument, just trying to understand.

I’m going to pass on the crash test dummies bit. You’ve misunderstood the point I was making, but it could be poor communication by me and I think the point is becoming increasingly tangential.

> When it comes to pharmaceutics and food additives, our mechanistic understanding is insufficient so we often have to resort to empirical studies on humans, including RCTs

> It does, because again, mechanism is provided. You could say that the study has weak evidence for the mechanism and it works like that perhaps only for sugar. Because that some mechanism is found in mice does not mean it is also found in humans, and it would be a fair point. This is why many species are tested and so far the results held up (testing humans takes too long for obvious reasons).

So you don’t feel you’re being straw manned again, can I get a clear answer to this: is your argument that if we stack together sufficient numbers of mechanistic animal studies we can be sufficiently confident enough in the translation rate of such studies to humans that we can roll out public health interventions without any evidence of efficacy in human populations?

Reply on news.ycombinator.com