#589：营养科学中的因果推断——Daniel Ibsen博士

你好，欢迎收听Sigma营养广播。

Hello, and welcome to Sigma Nutrition Radio.

这是该播客的第589期。

This is episode 589 of the podcast.

我叫丹尼·莱农。

My name is Danny Lennon.

欢迎来到本节目。

You are very welcome to the show.

今天，我们将深入探讨营养研究中一些较为深层的方面。

Today, we're gonna be looking at some of the aspects that are a bit on the deeper side when it comes to nutrition research.

我们将探讨因果推断的概念，以及我们如何在营养学，特别是营养流行病学中回答因果性问题。

We're gonna dig into the concept of causal inference, how we try and answer causal questions within nutrition and within specifically nutrition epidemiology.

当我们试图回答营养学问题、解读研究时，会遇到一些问题，我们还将探讨如何在未来改进营养科学，以及如何理解那些看似矛盾的研究。

Some of the issues that come up when we're trying to answer questions in nutrition or trying to interpret studies and walking through some of the aspects that we can do better nutrition science going forward and how we can make sense of studies that are seemingly conflicting about a certain topic.

但实际上，如果我们了解所使用的方法、这些方法的优势与劣势，以及这些研究的实际设计方式，就能更好地理解它们——即，这种研究设计是否适合我们试图回答的问题？

But in reality, we can understand them much more if we understand what methods are being used, what are the strengths and weaknesses of those, and how those studies are actually designed, asking, is this an appropriate design for the question we are trying to answer?

为了阐述这个概念，我将与丹尼尔·伊布森博士进行对话，他是一位营养科学领域的流行病学家，其研究结合了严谨的因果推断方法——我们今天将讨论这些方法——以及现实世界中的饮食与心血管代谢疾病研究。

And so to walk through this concept, I'm gonna be talking with doctor Daniel Ibsen, who is an epidemiologist in nutritional science whose work bridges rigorous causal inference methods, which we'll be discussing today, as well as looking at real world diet and cardiometabolic disease research.

他目前是丹麦奥胡斯大学的副教授，其研究范围广泛，从客观饮食生物标志物到2型糖尿病，再到我们今天将讨论的一些营养科学方法。

He's currently an associate professor at Aarhus University in Denmark, and his work has spanned across a whole range of topics all the way from things like objective dietary biomarkers to type two diabetes, and then specifically on some of the nutrition science methods that we'll discuss today.

因此，我们将关注替代模型或流行病学中的因果主张。

So looking at substitution models or causal claims in epidemiology.

希望这能帮助澄清你们在播客以往节目中可能听过的某些术语，或我们经常提及的概念。

And so hopefully, this will help clarify some of the terms you may have heard on previous episodes of the podcast or concepts that we have referred to quite regularly.

事实上，我和艾伦·弗拉纳根博士合作的许多期节目都涉及了我们今天将讨论的一些主题。

And indeed, a number of the episodes that doctor Alan Flanagan and I have done have broached some of the topics we'll discuss today.

因此，希望本期节目不仅能帮助你们温习这些观点，还能更深入地探讨，确保你们真正理解它们。

And so hopefully, this serves as not only a refresher of some of those ideas, but goes a bit deeper into that to make sure that they really make sense.

如果你喜欢探讨营养研究中所使用的方法以及如何解读营养研究，那么本期内容将非常详尽。

And so if you're someone who likes the conversation around methods that are used within nutrition research and how to interpret nutrition studies, This will be quite detailed.

内容可能会比较密集，但如果你坚持听完，相信会有所收获。

It's going to be quite dense, but hopefully, it's going to be worth it if you stick with it.

当然，如果你是 Sigma Nutrition 付费订阅用户，你将获得一套完整的详细学习笔记，帮助你更深入地理解本内容，不仅回顾这些最佳观点，还能了解我们使用的术语以及更详尽的例子说明。

And, of course, if you are a Sigma Nutrition premium subscriber, you have a full set of detailed study notes to help you really get more out of this to not only recap over those best ideas, but to have explanations of some of the terminology we use as well as some of the examples laid out in more detail.

这些内容将在描述框中提供，或登录你的付费仪表板查看。

So that will be available in the description box for you or over on your premium dashboard.

对于其他听众，你可以在当前页面的描述框中找到更多信息，包括指向本集页面的链接，该页面会汇总我们在本次讨论中引用的所有研究。

For everyone else, you can find more information in the description box where you are right now, including a link to the episode page, which will link to all the studies that we reference throughout this particular discussion.

好了，就是这样。

So that is it.

闲话少说，接下来请欣赏我和丹尼尔·伊本博士的对话。

Without further ado, please enjoy this conversation between myself and Doctor Daniel Ibsen.

热烈欢迎丹尼尔·伊本博士做客本播客。

A very big welcome to the podcast to doctor Daniel Ibsen.

非常感谢你抽出时间来和我交谈。

Thank you so much for taking the time to come and talk to me.

嗯。

Yeah.

谢谢你的邀请。

Thanks for inviting.

我多年来一直关注这个播客，非常期待

I've followed the podcast for many years, so look forward to

是的。

Yeah.

正如我之前对你说过的，你的许多工作对我自己在我们今天要讨论的一些话题上的思考产生了深远影响，并真正塑造了我对营养科学中这些关键而基本方面的理解。

And I think as I've mentioned to you, much of your work has been very influential in my own thinking on some of the topics we're gonna discuss to get today and has really shaped how I've came to understand some of these really crucial and fundamental aspects of nutrition science.

我们之前在这个播客中提到过，这些话题有多么重要，因此我认为，随着我们深入探讨，这将帮助人们真正理解它们。

And so we've mentioned on this podcast before how important some of these topics are, and so I think this will really serve to help people really understand that as we go through it.

但在提出这些问题之前，也许你可以先向听众介绍一下你自己：你的学术背景是什么？你的主要研究兴趣是什么？还有其他任何相关的信息吗？

But before I get to any of those questions, maybe just to introduce people to who you are, can you let them know what your academic background is, what your main research interests are, and anything else that might be relevant?

当然可以。

Yes, of course.

我是丹麦奥胡斯大学公共卫生系的副教授，同时也是丹麦奥胡斯Steno糖尿病研究中心的高级研究员。

So I'm an associate professor at the Department of Public Health at Aarhus University in Denmark, and I'm also a senior researcher at the Steno Diabetes Centre in Aarhus in Denmark as well.

奥胡斯是丹麦第二大城市，这里做个参考。

So Aarhus is the second largest city in Denmark just for reference there.

我最初就是在那儿完成营养健康专业的本科学习的。

That's where I first did my bachelor in nutritional health.

然后我去了哥本哈根，在哥本哈根大学攻读人类营养学硕士学位。

Then I went to Copenhagen, to the University of Copenhagen to do a master in human nutrition.

接着我回到奥尔比斯攻读流行病学博士学位。

Then I went back to Orbis to do a PhD in epidemiology.

之后我去了斯德哥尔摩的卡罗林斯卡学院做博士后，然后去剑桥做博士后，还在哥本哈根大学做过一段时间的博士后。

And then I went to the Karolinska Institute up in Stockholm to do a postdoc and then to Cambridge to do a postdoc and then some postdoc time as well at the University of Copenhagen.

现在，我刚刚晋升为副教授。

And yeah, now I've just become an associate professor.

我的研究基础是方法论。

So the foundation of my research, that is methods.

正如我所说，我拥有流行病学博士学位，这正是我真正热衷的领域——方法论。

So that's, as I said, I have a PhD in epidemiology and that's what I'm really passionate about, the methods.

而且它大致分为两个方向。

And it sort of goes in two lines.

一个是利用这些方法进行因果推断，并将这些方法应用于营养科学，因为我认为我们可以从中学习很多东西。

So one is this causal inference using those methods and translate the methods into nutrition science because I think there's a lot we can learn from that.

而且，营养科学中的一个关键问题也是核心问题，即评估膳食摄入量。

And also because one of the key sort of problems in nutrition science is also the core thing is assessing dietary intake.

但我们该如何做到呢？

But how do we do that?

在其他科学领域，他们拥有更客观的测量方法，而我们却更多依赖于主观方法，如食物频率问卷和饮食记录，但我们知道这些方法存在诸多问题。

So in other sciences, they have more objective measures, but we much rely on the more subjective things like food frequency questionnaires and dietary records, but we know there are issues with that.

那么，我们该如何客观地进行测量呢？

So how do we do that objectively?

这两方面中，我们正在研究的一个方向是将它们统一起来，形成一种关于如何开展营养研究的共同理论。

And those two things one of the things that we're also working on is actually unifying those two things into like a common theory of how we can do nutrition studies.

因此，这就是基础。

So that's sort of the foundation.

然后我还大量从事糖尿病缓解方面的研究。

Then I also work a lot with diabetes remission.

因此，我们目前正在开展一项大型干预研究。

So also working on a large intervention study we are setting up now.

实际上就是CBPE，这是全球规模最大、基于饮食和运动的糖尿病缓解干预研究。

Was actually CBPE, so the world's largest intervention on diabetes remission with diet and exercise.

所以这目前是我工作的重要部分。

So that's a big part of my job at the moment.

它占用了我大量的时间。

It takes a lot of my time.

此外，我长期以来一直研究植物性饮食，特别是《柳叶刀》饮食。

Then I've also worked a lot and still do with plant based diets, in particular the Eat Lancet diet.

最近刚刚发布了《柳叶刀》饮食的新版本，但我们之前已对2019年版进行了大量研究，既分析了其与健康结局的关联性，也通过队列研究进行评估，同时再次关注方法学问题。

And now there was just this new version of the Eat Lancet diet, but we've done a lot of work sort of evaluating the 2019 diet, looking at both sort of associations and cohort studies with health outcomes, but also actually, again thinking more about the methods.

因此，将《EATLANCID》参考饮食中的这张单一表格转化为实际应用的方式有很多。

So there are many ways of translating this single table in the EATLANCID reference diet.

所以那篇论文已经被翻译成了七、八、九、十种不同的饮食评分体系，结果发现，你使用的评分体系对健康影响或与健康的关联，以及对气候影响的评估，都会产生巨大差异。

So that paper, it's been translated into seven, eight, nine, ten different diet scores and actually turns out to work with it that the score you use, it makes a big difference in terms of the health effects or the associations with health and also the effects on climate intake, how you evaluate that.

然后，我对这个领域的一个基本问题产生了越来越浓厚的兴趣：当人们开始使用减重药物时，他们的饮食会发生什么变化？

And then I have another sort of growing interest in this field of basic question of what happens to people's diets when they start on weight loss medication.

我们把减重药物看作一种医疗干预，但我认为它更像是一种营养干预。

We see these weight loss medication as like a medical intervention, but I think it's much more nutritional intervention.

我认为，这可能是你所能实施的最强大的营养干预手段之一，但我们并没有真正这样对待它。

It's I think actually one of the most powerful nutrition interventions you can do, but we don't really treat it like that.

而且我们其实并不清楚，不仅仅是总摄入量，连饮食结构本身会如何变化。

And we don't really know what how do they not just the total intake, but the composition.

这种变化会带来哪些长期影响？

How does that change and what are the long term implications for that?

因此，我目前正在开展一些相关数据收集工作。

So I'm also working on some data collection we have ongoing now in that area.

这就是我广泛的研究兴趣，我希望在某个时候能将它们更紧密地整合到我的实际工作中。

So that's sort of my big wide research interest that I, at some point, hope to align more into what I'm doing.

非常有趣，有很多真正令人兴奋且富有意义的研究正在进行。

Fascinating and a lot of really interesting and exciting work going on.

今天，我们将重点关注与方法相关的领域，特别是营养流行病学方面。

Today, of course, we're gonna be focusing on that area that relates to methods and particularly in the area of nutritional epidemiology.

我认为在这个播客中，艾伦和我多次向人们强调，如果你想真正理解营养科学，或者理解为什么有时会出现如此多的困惑，甚至某些结论看似矛盾，唯一的途径就是理解所使用的方法、为何采用这些方法、研究的问题类型，以及能否评估这些方法是否运用得当。

And I think on this podcast, Alan and I have emphasized many times to people that if you really want to understand nutrition science or understand why there's sometimes so much confusion and sometimes things seem to be conflicting, The only real way to get at that is if you understand the methodology being used, why it's being used, the types of questions being answered, and then are able to evaluate if they're being done appropriately.

一旦如此，事情就会变得清晰得多。

And then suddenly things become a lot clearer.

因此，我们希望逐步梳理这些问题，帮助听众从根本上理解这些内容。

And so we're hopefully gonna work through some of that and and help people in the audience get to the bottom of that.

但你曾写过一个观点，即营养科学常常提出错误的问题。

But one of the things that you've written about has been this concept that oftentimes nutrition science asks the wrong question.

你能具体谈谈你所说的这句话是什么意思吗？

Can you maybe just speak to what you mean by this?

我想我进入这个领域的起点是在我的博士期间，当时我的导师大量研究食物替代方法。

I guess my whole way into that started during my PhD where I had a supervisor that worked a lot with these food substitution methods.

那时，这并不常见。

And at that time, it was not that common.

在美国，只有少数人做这些分析，而且它们并不是论文的主要内容，但我是和八月时他们那个团队一起工作的。

There were some few people in The US doing most of these analyses and they were not like the main thing of the paper, but they were the group that I worked with in August.

他们真正认为这才是关键所在。

They were really this is actually the key thing.

原因在于，它提出了一个更清晰的问题。

And the reason for that was because it asks a much more clear question.

因此，这就是我开始思考的出发点，认为这确实是理解任何事物的基础。

And so that was my whole way into to under sort of thinking that this is really the sort of the foundation for understanding anything.

比如，如果我们一开始就没有一个好问题，我们从一开始就偏离了方向。

Like, if we don't have a good question to answer in the beginning, we are sort of going off track already already there.

这就是我的故事，我进入这个领域、关注好问题的起点。

So that's my story, my sort of starting point into this with the good questions.

后来，随着我不断了解这些因果推断方法，我也开始运用它们，更多地以试验的视角思考，从更贴近试验的问题出发。

Then as things come along and I've learned more about these causal inference methods, using that as well, thinking more more in in like trials, like starting with a question that more aligns with the trial.

因为在试验中，事情通常更清晰，尽管仍存在一些不确定性或不明确之处。

Because in a trial, things are much more clear, often at least, that there are also uncertainties or things that are not very clear.

但至少表面上看，人们研究的内容似乎更明确。

But at least it seems on the surface at least that it's more clear what people are investigating.

而当你转向观察性研究时，你面对的是一个群体，长期跟踪，收集一些数据并进行建模，但这些建模究竟意味着什么？

Whereas you go to an observational study where you have a population, you follow over time, you have some data, you do some modeling, but what does this modeling what does it really mean?

我认为，作为流行病学家，反过来想：好吧，让我们从试验的角度出发，以此为框架，如何让分析和观察性数据更接近试验？

I think then flipping it around as an epidemiologist saying: Okay, let's start thinking of a trial and then using that as a frame then how can we make the analysis and the observational data look more like that?

当然，我们无法确保它们完全相同。

Of course, we cannot ensure that it's exactly the same.

干预人们的行为与观察他们的自然习惯之间存在差异，但至少可以更进一步地对齐。

There's a difference between intervening on people and following their natural habits, but at least aligning them even more.

通过这项与不同方法的工作，我们实际上提出了一篇论文，因为我们认为在研究营养学或处理营养学中的因果问题时，有三件关键事情需要注意。

And through that work with these different methods came about actually we then brought up a paper because we sort of thought of three key things that you should be aware of when you study nutrition or you have causal questions in nutrition.

其中一项，我们在那篇论文中称之为不同的东西。

And one of them is in that paper we called them different things.

所以第一点我们称之为组成性。

So the first thing was we called it compositional.

但这实际上关乎替代，因为在饮食研究中，这不像吃药。

But that's really the point about the substitution because in diet studies, it's not like a pill.

有时候你会有补充剂，对吧？

Sometimes you have supplementation, right?

但如果你研究的是完整饮食，你往往无法仅仅关注能量摄入的差异，你还关心其他方面的影响。

But if you work with like whole diets, you cannot really it's often you also you're also interested in the effects beyond the difference in energy intake.

我们知道减重会影响许多因素，因此你希望进行等热量干预。

We know weight loss impacts many things, so you wanna do an isocaloric intervention.

但当你这样做时，你总是会比较吃一种食物与吃另一种食物的区别。

But when you're doing that, you will always compare like eating one thing versus eating another.

而你在试验中就是这样做的。

And you do that in trials.

你会进行比较，没错。

You compare yeah.

你将红肉与植物性肉类替代品进行比较。

You compare red meat to plant based meat alternatives.

你还会与某种东西进行比较。

And you compare with something.

问题是，我有时在论文中看到——实际上经常看到——他们会说：‘我们有一个对照组。’

And the thing is that and I sometimes see this in papers or actually often that it's like, oh, we have a control group.

但在营养学研究中，并不存在所谓的对照组。

But there's no such thing as a control group in nutrition studies.

它总是相对于其他某种东西而言的。

It's always something else.

而这个‘其他东西’是什么，对于结果的解读至关重要。

And what that something else is, that's completely important to the interpretation.

你是将红肉与鱼类比较，还是与植物性肉类替代品比较？

Do you compare the red meat with fish or do you compare with a plant based meat alternative?

根据不同的结果，它们可能会显示出不同的效果。

They may show different things depending on the outcome.

这通常是需要首先注意的一点。

That was sort of the first thing that you need to be aware of.

所以当你提出一个问题时，通常涉及的是某种饮食模式、食物或营养素与另一种东西的比较。

So when you ask a question, it's often this either dietary pattern or food or nutrient compared to something else.

这就是所谓的替代部分。

So that's the substitution part of it.

但我们也讨论了另一个方面。

But then there was another thing we also discussed.

那就是基础饮食，这一点在试验中经常被忽视。

That's this part of the baseline diet, which is actually also often overlooked, especially in trials.

如果你做的是补充试验，比如维生素D就是一个很好的例子。

If you have supplementation trials, I think there are good examples with vitamin D, for instance.

那么，如果你纳入试验的人本身维生素D水平已经足够，再给他们补充更多，也不会有明显改善。

Then the people you get into your trial, well, if they already have sufficient levels of vitamin D, then they're giving them more, then it's not going to improve a lot.

当然，还有其他类似的例子。

Or you can there are other examples as well.

如果你考虑进行一项减少超加工食品摄入的干预，这是一个目前非常热门的例子。

If you think about if you want to do an intervention with lowering ultra processed food intake just as a very topical example at the moment.

如果参与者的超加工食品摄入量本来就很少，那么你就很难期望看到显著的效果。

Well, if the people you get in, they already consume very little ultra processed food, then you cannot really expect to see a big difference in that.

但这并不是人们总是会考虑到的问题。

But it's something that's not always thought about.

所以这是其中的一部分。

So that's one part of it.

效果可能取决于初始的摄入水平。

The effect may depend on the initial intake.

但我们也可以说，好吧，我有一个群体。

But we can also say, Okay, so I have a population.

我给他们实施了一项干预。

I give them an intervention.

此外，基线饮食和你如何定义干预措施，这些因素共同影响你对所获得效果的解读。

And also the baseline diet and how you define the intervention, those things together also impact your interpretation of the effect you get.

如果你在另一个摄入量不同的群体中进行同样的干预，比如他们摄入的红肉多或少，而你希望他们都遵循素食饮食，那么这两个群体中的效果将会不同。

And if you do the same intervention in another population with a different intake, so if they have high or low red meat intake and you want them all to follow a vegetarian diet, then the effects in those two populations, they will be different.

但表面上看，你可能会说：哦，这两项研究中素食饮食的效果不同。

But you could say on the surface, it's like, oh, they find different effects of vegetarian diets in these two studies.

但如果他们的起始点不同，这就解释了很多问题。

But if they're two different starting points, then that explains a lot.

所以，再次强调，考虑基线饮食，在观察性研究中，我们很少这样做，因为我们通常没有多个时间点的数据，因此不会调整研究开始前的饮食状况。

So again, like thinking about the baseline diet and in observational studies, one thing we rarely do because we rarely have multiple time points, we don't adjust for the diet going into the study.

这是另一个需要考虑的因素。

So that's another thing.

然后是最后一点，第三点，我们称之为饮食暴露的多维性。

And then the final thing, the third one, we we call it the multidimensionality of dietary exposures.

但这只是意味着你可以从很多不同的角度来审视饮食。

But it's just that you could look at diet in many different ways.

你可以以不同的方式来切分这块蛋糕。

You sort of cut the cake in different types of pieces.

所以你可以从食物层面来看，比如你摄入了多少克蔬菜或苹果？

So you can either look at it on on the food level, how many grams of vegetables do you consume, or apples?

但你也可以看摄入了多少碳水化合物，但这是两个不同的方面。

But you can also look at how many carbohydrates do you consume, but they're like two different different aspects.

苹果中的碳水化合物和其他来源的碳水化合物可能不同，因此这种方式能捕捉到暴露因素的许多不同子部分，而食物层面则是另一个层次。

They could be they're carbohydrate in apple, but also from other things so that it sort of captures many thing many different subparts of an exposure, whereas the food is a different level.

所以它们就像是食物与营养素之间的关系。

So they're like the food nutrient thing.

而近年来，我们还大量讨论加工问题。

And now more recently, we also talk a lot about processing.

它是一个苹果吗？

So is it an apple?

还是被榨成其他形式的苹果汁？

Is it apple juice that's been squeezed into something else?

但可能差不多。

But maybe similar.

也许总量是一样的，但存在于不同的基质中。

Maybe the total is the same but it's in a different matrix.

所以这是另一个我们需要更明确的方面。

So that's another aspect of it that we may also need to be more clear about.

所以如果你把这些因素结合起来，就可以开始提出更清晰的问题。

So if you sort of put those things together, you can start asking much more clear questions.

因此，它是与这个人群中特定饮食的某种情况相比较，从而更清楚地明确你感兴趣的真正食物或饮食模式。

So it's compared to something in this population with a specific diet and be more clear about what are the actual foods or dietary patterns that you're interested in.

然后我认为我们在理解某些事情上会取得更大的进展。

Then I think we get much further in understanding something.

完美。

Perfect.

也许我们可以开始逐一探讨并拆解这些观点，因为它们是非常深刻的观点，我认为这解释了人们在解读营养学时所做的一切。

And maybe let's start working through those and unpacking those because they're really profound points that I think explain a lot of this anything that people are gonna do in terms of interpreting nutrition.

你所说的第一个观点是，我们总是存在某种比较。

So the first thing you said was this idea of we always really have a comparison here.

而在干预试验的情况下，这一点就更加明显了。

And in the case of maybe an intervention trial, it becomes more obvious.

我们有一个明确的干预措施，或者将其与对照组进行比较。

We have this defined intervention we have, or we're comparing that to a control group.

我们通常会保持等热量状态，以确保考虑了热量摄入，然后观察这两种营养干预方案的不同组成。

We'll have this isocaloric situation normally to make sure we're accounting for calories, and then we can see what the two different compositions of those nutrition interventions are.

但在流行病学中，情况就变得稍微复杂了，因为我们没有进行干预。

It becomes a bit more difficult in epidemiology because we're not doing an intervention.

但正如你所指出的，背后始终存在某种程度的比较。

But as you note, there is always some degree of a comparison going on in the background.

因此，如果我们进行一项考虑总能量摄入的分析，那么我们就是在比较这两种情况。

So if we do an analysis that accounts for total energy intake, then we're comparing these two situations.

但究竟什么与什么进行比较，这才是关键。

But what's being compared to what really matters?

你在这里举了一个非常好的例子，这解释了为什么人们常常看到令人困惑或相互矛盾的结果，比如某种特定食物，红肉对健康是正面、负面还是中性影响？

And you gave a really nice example here, and this explains why people maybe often see confusing or conflicting results, right, of does this particular food, does red meat have a positive, negative, or neutral impact on health?

你会看到有人引用一篇论文说，它并没有任何负面影响，另一篇则说实际上确实有。

And you'll see someone cite one paper that shows, well, it doesn't have any negative impact, One that shows actually it does.

人们可能会束手无策。

People can throw their hands up.

正如你所说，很多时候，其中一个原因，以及其他因素，是我们去查看这些单项研究时，实际的分析方法和比较方式可能存在差异。

Really, as you say, a lot of the times, one of the reasons among others could be that when we go and look at those individual studies, there might be a difference in what is being compared with what when we actually see how the methods were done, what way the analysis was done.

因此，我认为正如你提到的，流行病学中比较的对象并不总是显而易见，但这种比较始终存在。

And so I think this idea of what is being compared in epidemiology, as you mentioned, isn't always as obvious, but is always going on.

对吧？

Right?

但它始终存在。

But it's always there.

没错。

Exactly.

我认为许多人存在一种误解。

And I think there is this misconception from many people.

我注意到他们似乎想要一个孤立的影响。

I see that they sort of want this isolated effect.

比如，这种食物的影响是什么？

Like, what's the effect of this food?

但我们真的不能这样看待它。

But we really cannot look at it like that.

它实际上是被与其他因素比较的，而且我们最近有一篇评论刚被《柳叶刀·行星健康》期刊接受，我认为这是一个非常好的例子，所以我才必须对此发表评论。

It's it is really compared to and actually, we just got accepted a comment in this journal Lancet Planetary Health because and I think this is a really good example and that's why I had to make a comment on this.

事实上，几年前有一项研究发表在EPIC队列中，他们考察了超加工食品的摄入量与不同癌症风险之间的关系。

So actually a few years ago there was published this paper in a range of cohort, in this epic cohort, They looked at intake of ultra processed foods and risk of different cancers.

EPIC队列由欧洲大约九个大型队列组成，具体数量略有不同。

This epic is this cohort with, I think it varies a little bit, but often it's around nine large cohorts from Europe.

因此，他们拥有数十万来自1990年代左右接受测量的人群数据。

And so they have many hundreds of thousands of people from often measured around the 1990s.

他们使用食物频率问卷，并对这些人进行了十到二十年的随访。

They have food frequency questionnaires, and then they follow them up for ten, fifteen, twenty years.

他们研究了这里的癌症发病率。

And they looked at incidents of cancer here.

然后他们使用NOVA分类系统来观察这四个NOVA组别，并分析它们与不同癌症风险的关联。

And then they used this NOVA classification system to look at these four NOVA groups and see how they were associated with risk of different cancers.

他们还进行了替代分析。

They also do a substitution analysis.

所以几年后，显然有一些作者决定就此发表评论。

So then after some years, then apparently some authors decided to write a comment on that.

他们在评论中提出的观点非常有趣，因为他们说：我们实际上不同意这个结论。

And what they argue in that comment is really interesting interesting because they said, well, we actually don't agree with the conclusion.

他们的结论是，与NOVA4（即超加工食品）相比，摄入更多的NOVA1（即最少加工食品）与某些癌症风险降低有关，但并非所有癌症都是如此。

They concluded that eating more of this Nova one, which is minimally processed food, compared to NOVA4, which is ultra processed food, is associated with a lower risk of some cancers, but not all.

他们说，不，这种现象的原因在于，多吃最少加工食品具有孤立的影响。

And they say, no, but the reason for this is because there is an isolated effect of eating more minimally processed foods.

所以，这并不是因为超加工食品的有害作用，而是因为最少加工食品的积极作用。

So it's not because of the ultra processed because of the detriment of the ultra processed food, it's because of the positive effects of minimally processed food.

所以我记不清确切的名称了，但我想那篇评论的标题大概是‘对这些效应的误解’之类的。

So I can't recall the exact name of it, but they were I think the the name of the title of that comment was something like a misinterpretation of these effects.

于是作者们进行了回复，令我感到惊讶的是，他们说：‘哦，没错。’

So then the authors replied back, and I was surprised because they said, oh, yeah.

我们同意，这可能更多是关于有益效应或关联，即NOVA1类食物，也就是这些低加工食品。

We agree that it's maybe more about the beneficial effect or association, these Nova one foods, these mineral processed foods.

那时我正好在丹麦讲授我们的营养流行病学博士课程。

And it was around the time when I was teaching our nutritional epidemiology PhD course in Denmark.

就在某个晚上，我偶然在其中一个平台上看到了它。

And it just in the evening, it popped up on one of these.

你会收到不同期刊的推送通知，我看到有人提到食物替代，感到非常感兴趣。

You get notifications from different journals and I saw somebody with food substitutions and I was intrigued.

然后我看到了那些回应，实在忍不住了，必须写点什么。

And then I saw those responses and then I couldn't hold back like I had to write something.

于是我和一位博士后以及一位非常好的同事商量：‘我们必须澄清这个误解，因为当他们在模型中调整了能量摄入后，这些食物并不存在孤立的影响。’

So together with a postdoc and one of my really good colleagues said: Okay, we have to break down this misconception because there is no isolated effect of these things when they in their models adjust for energy intake.

所以在统计模型中进行调整，这更多是幕后发生的事情。

So adjusting in a statistical model, that's more something that happens behind the scenes.

它是你放入回归模型中的一个变量。

It's a variable you put into a regression.

但它实际上是有意义的。

But it actually means something.

因为如果你从试验的角度来想，它在概念上意味着你所比较的两组应该摄入了相同量的能量，或者能量变化相同。

Because if you think about it in a trial, it's conceptually saying that the two groups you have, they should have consumed the same amount of energy or have the same change in energy.

你在队列数据中也是这么做的。

And you're doing the same in the cohort data.

但这就意味着，当他们只观察更高水平的Nova一类食品摄入量，同时调整总能量摄入时，实际上意味着你从其他来源摄入的能量更低了。

But then that means that when they look at just higher intake of Nova one but adjust for total energy, means that you have it then you have lower energy intake that's coming from something else.

对吧？

Right?

而这很可能混合了其他Nova类别。

And that's probably a mix of these other Nova categories.

所以你不能说这只是因为他们吃了更多的NOVA1。

So you cannot say that it's just because they eat more NOVA1.

这是相对于吃较少其他食物而言，摄入了更多的NOVA1。

It's feeding more NOVA1 compared to eating less of these other foods.

因此，这是一种我们确实需要强调的误解。

And in that way, it's just a misconception that we really had to highlight.

而且实际上，我们还可以更进一步。

And actually just taking it a step further.

我曾与一些其他同事合作，我认为这项研究很快就会以预印本形式发表，我们进行了更系统的综述，考察了一些不同的荟萃分析，深入探究它们是否真的将既调整又未调整总能量摄入的研究合并在一起，因为这些估计值的含义是不同的。

I've worked with some other colleagues and I think this is going to be published as a preprint very soon, that we actually did a more systematic review and looked at some different meta analyses and looking more deeply, do they actually put together studies that both adjust and do not adjust for total energy intake because the estimate, it has different meanings.

所以，调整后，我们比较的是能量摄入相似的人群。

So adjust, we compare people with similar energy intake.

也就是说，一种食物摄入量更高，意味着其他食物摄入量更低。

So higher intake in one food, less intake in something else.

你可以定义这一点，也可以选择忽略它。

You can define that or you can leave it out.

然后，这更像是该人群中饮食的加权平均值。

And then it's sort of more the weighted average of the diet in that population.

或者你不做调整，但这样就是在此基础上额外增加了这种食物的摄入量。

Or you cannot adjust, but then it's higher intake of this on top of what you consume.

所以这是两个不同的问题，两者可能都相关，但却是截然不同的两个问题。

So they're two different questions, both probably relevant, but two different very different questions.

但我们发现，大多数研究在进行荟萃分析时并没有考虑这一点。

But we could see that, like, most studies, they do not take that into account when doing a meta analysis.

例如，你只是直接采用调整最充分的模型。

For instance, you just pull the model with the most adjustment.

当你深入单个研究时，它们隐含地估计了替代效应，但很少明确说明这一点。

And if you go into the single studies, they implicitly estimate substitution effects, but they don't really rarely specify it.

我们稍后一定会回到超加工食品，结合你的发表成果来讨论。

And we'll definitely return to ultra processed foods later on for in relation to your publication.

但你在讨论这些调整对比时提出了一个非常重要的观点，这与你最初说的那句话有关——我们究竟要回答一个什么样的具体问题？

But you made a a really important point here when we think about this comparison of these adjustments, and it relates back to the first thing you said that what is that specific question we're answering?

并不是说其中一个一定比另一个更好，但我们必须清楚这个问题究竟是什么。

And it's not that one is necessarily better than the other, but we need to be aware of what that question is.

根据这项研究的设计以及我们设定的方法和分析，它实际上旨在回答什么问题？

And with the design of this study and how we've set up the methods and the analysis, what question is that actually set out to answer?

因为有时论文的结论声称某件事，但当你仔细查看时，会发现根据这些方法，它其实并没有设计来回答那个问题。

Because sometimes we have conclusions from a paper claiming one thing and then you look at it and you would say, well, based on these methods, it's not set up to answer that question.

它实际上是在回答一个基于这些方法的不同问题。

It's actually answering a different question based on that.

再次，你补充的第二部分是：如果我们试图在一方面尽可能进行最大程度的调整，以尽可能精确地观察某一特定因素的影响，这可能会产生影响。

And again, the second part to that, you added that, well, that could have implications if we're trying to either on one end go with the most adjustment possible to really refine down as much as we can to see the effect of one particular thing.

但当我们这样做时，这些饮食变化在现实世界中的意义自然也会有所不同。

But as we do that, of course, then that has different implications for what some of these dietary changes might mean in the real world.

如果某人以某种方式增加或减少其中一种食物的摄入，就会产生一连串的连锁反应。

If someone adjusts their diet in some way up or down of one of these foods, then it has a whole knock on effect.

因此，这变得相当复杂，正因如此，我们需要理解我们试图设计的特定研究的目标是什么。

So it becomes quite complicated, and hence, because of that, we need to understand what is the goal of a particular study that we're trying to design, I suppose.

是的。

Yeah.

没错。

Exactly.

我经常看到人们想要孤立地研究某一个因素的影响。

It's just a thing that I see so often to want to isolate an effect of one thing.

但在营养学中，这种情况几乎不可能实现。

But it's just rare that you can do that in nutrition.

我认为，当我们进行这些不同争论时，问题往往就出在这里，正如你提到的，对吧？

And I think when we have these different debates, that's often what goes wrong, as you mentioned, right?

我们没有就想要回答的问题达成一致。

That we're not aligned on what is the question we are trying to answer.

然后我们可以在已有的分析中挑选自己偏好的，再就此展开争论。

And then we can pick favorites among the analysis that have been done, and then we can debate that.

显然，所有这些都与因果推断这一普遍概念密切相关，而你对此有过大量论述。

So obviously related to all this is the the general concept of causal inference, which you've written a lot about.

正如我之前向你提到的，这进一步强化了我对这一点的理解。

And as I've mentioned to you, has reinforced my thinking about what that is.

对于那些可能听过这个术语但不太确定，或者甚至第一次听说的人，最好的理解方式是什么？在营养学中，因果推断到底是什么？它的意义何在？你认为有哪些关键点需要我们采用方法来接近它？

For maybe people who have maybe heard that term but aren't quite sure or maybe even are hearing it for the first time, what is the best way to think about what causal inference is in nutrition and the importance or or some of the points that you would make around the need for methods that would get us towards this?

对我来说，因果推断的核心定义是：如果你能置身于两个平行世界，在一个世界里你做一件事，只改变这一件事，然后在另一个世界里做别的事——比如我们谈过的饮食，吃一种食物而不是另一种。

For me, fundamentally, causal inference essentially sort of the definition of a causal effect is if you could have yourself in two parallel worlds and you did one thing in one world and only changed that one thing and then did something else like we talked with diet, ate one food or another.

你可以拥有这两个平行世界，然后观察是否存在差异。

Like you can have your two parallel worlds and then you see if there's a difference.

对吧？

Right?

所以，这关乎干预。

So it's about intervening.

这关乎改变某件事，然后某件事随之发生。

It's about changing something and then something happens.

就像这个问题：如果我这么做而不是那样做，会怎么样？

Like this question, what if I did this instead of that?

某件事会产生什么效果？

What would be the effect of something?

我们经常问他们这个问题。

That's this question we ask them all the time.

对于许多因果效应，或者作为人类，我们都会这样思考。

Think for many causal effects or like, as a human, we think about this.

这种思维方式已经深深植根于我们的思维中。

It's very ingrained in in the way we think.

我们预测：好吧，我按下这个开关，就会发生这件事。

We predict, okay, I push this switch, then this will happen.

在某种意义上，这非常简单。

In a way, it's very simple.

但更困难的是，从方法论的角度来看，我们几乎不可能拥有那个平行宇宙，也无法穿越时间，让自己做一件事，然后回到过去做不同的事，再观察之后的结果。

But the thing that comes more difficult, I think from a more methodological point of view, is that we can rarely have that parallel universe or we can time travel so I can have myself do something and then go back in time and do something different and then see what are the outcomes after that.

因为这更接近个体因果效应。

Because that's more like the individual causal effect.

因此，解决这个问题的方法是说：好吧，我们无法做到那样，但如果我们知道有两个群体，他们在平均上完全相同，而在其中一个群体中我们做某件事，在另一个群体中我们做另一件事，那么会发生什么？

So the way to solve that issue is to say: Okay, we cannot do that, but if we knew if we have two groups that are on average exactly the same, but then in this group we do something and this group we do something else, then what happens?

然后，这两个群体之间的差异，也就是它们平均值的差，如果你将它们相减并发现存在差异，那么你就可以说存在一个因果效应。

And then the difference in those two groups, the average difference in those two groups, if you subtract them from each other and there is a difference, then you would say then there was a causal effect.

简单来说就是如此。

Like very simply put.

但我们要从个体层面推及到平均层面，这需要一些假设。

But the way we go from the other individual to the average, that requires assumptions.

对吧？

Right?

所以天下没有免费的午餐。

So there's no free lunch.

因此，存在一些关键的假设。

So there's there are some key assumptions.

其中一个是，我认为任何稍微接触过研究工作的人都会了解可交换性。

So one of them is I think most people that work a little bit with researchers would know this about exchangeability.

因此，这也常被称为混杂。

So it's also often referred to as confounding.

之所以称为可交换性，是因为如果你进行随机试验，比如有一组人，我们抛硬币。

The reason why it's called exchangeability is that if you for instance were to do a randomised trial so we have a group of people, we flip a coin.

一组人做一件事。

One group they will do one thing.

A组和B组做另一件事。

Group A and Group B will do something else.

我们随机分配人群。

And we divide people by chance.

因此，如果样本足够大，这两组在平均上会非常相似。

So on average those two groups, if they are large enough, would be very similar.

我们可以让他们做些事情，观察并看到效果，对比两组，得出差异，看看是否存在差异。

They could do something and we could watch and see an effect and contrast the two, get the difference, see if there is difference.

但如果我们错误地让A组实际接受了B组的干预，反之亦然，我们仍应得到完全相同的组间差异，因为平均来看，它们是相同的。

But if we by mistake had changed so Group A actually got Group B's intervention and vice versa, We should get exactly the same difference between the two groups because on average, like they are the same.

他们 individually 可能并不完全相同，但平均来看是一样的。

They are not individually maybe not exactly the same, but on average they are the same.

这就是可交换性。

So this is exchangeability.

所以我们本可以得到相同的结果。

So we could have And we would have found the same thing.

这是基本概念。

That's like the fundamental concept.

理论上很简单，但在现实中很难保证，对吧？

It's okay, simple in reality, but really hard to ensure, right?

这是其中一点。

That's one thing.

另一个假设是关于正性条件。

The other assumption is about positivity.

基本上，每个人都有机会接受干预，或者在观察性数据中拥有你想要估计的数据，因为你想要估计素食饮食的效果，但没有人食用素食饮食。

So basically everyone having a chance of getting an intervention or a more observational data, having data of what you want to estimate because you want to estimate the effect of a vegetarian diet, but no one consumes a vegetarian diet.

你没有一个人能代表这种情况。

You don't have anyone sort of representing that.

所以这也是关键。

So that's also key.

在试验中，情况更简单。

In a trial, it's more simple.

你进行随机化，每个人都有机会接受干预。

You do randomization and everyone have a chance to get that.

还有另一个方面，我认为也非常关键，但在营养学中常常被忽视，那就是一致性。

And then there's this other one which I also think is very key, but often also overlooked in nutrition especially is this of consistency.

你可以将这理解为实施相同的干预并获得相同的效果。

So you can think of this as doing the same intervention and getting the same effect.

听起来很简单，但在定义你的干预措施时，它实际上非常重要。

It sounds very simple, but it's really, really important in terms of how you define your intervention.

这里有一个关于减重的例子。

So there's one example with weight loss.

我们可以这么说：好吧，也许减重的效果存在差异，因为你可以通过不同的方式减重，对吧？

Well, we could say: Okay, maybe there's a difference in the effect of weight loss because you can lose weight in different ways, right?

你可以截肢一条腿。

You can amputate a leg.

那是一种减重的方式。

That would be one way of losing weight.

你可以服用减重药物。

You can go on weight loss medication.

但在这两种情况下，你可能看不到相同的代谢益处，对吧？

You probably don't see the same metabolic benefits in those two, right?

但它们的减重幅度可能是相同的。

But they could have lost the same amount of weight.

所以，干预措施的定义非常重要。

So again, the definition of intervention is really important.

在营养学中，我们经常谈论，比如低碳水化合物饮食。

And in nutrition, we often talk about, for instance, a low carb diet.

但你可以用很多种方式来定义低碳饮食。

But you can define a low carb diet in many different ways.

因此，这可能包括更健康和更不健康的低碳饮食。

So that could be more healthy and more unhealthy low carb diets.

同样，如果你想对低碳饮食的整体定义发表看法，那么要真正得出其效果可能会非常非常困难，因为你可能采用一种非常健康的低碳饮食组合，从而对代谢健康产生积极影响。

And again, if you want to say something about the overall definition of low carb diets, then it can be really, really hard to actually get the effect of that because you can have a composition that is really healthy low carb diet and they will see positive benefits on metabolic health.

但你也可以拥有一种非常不合理的低碳饮食组合，其效果恰恰相反。

But you can also have a very poorly composition of a low carb diet that will actually do the opposite.

因此，要笼统地说‘低碳饮食对代谢健康有因果影响’就变得非常困难，对吧？

So then it's really hard to say, okay, low carb diets have a causal effect on a metabolic health to take something very general, right?

但由于你可以用多种方式定义它，你实际上很难找到明确的结论。

But because you can define it in many different days, you cannot really find it.

当然，你可以非常明确地定义它，然后观察到一致的效果。

You can, of course, define that very clearly and then see the same effect.

但通常我们并没有真正这样做。

But often we don't really do that.

再者，回到其他讨论中，当你对人们说起低碳饮食时，他们想到的却是两件不同的事。

And again, going back to other discussions where you say low carb diets to people, but they're thinking about two different things.

所以，这就像营养学中的这些误解或认知偏差。

So again, this is like these misconceptions in nutrition or misunderstandings.

是的。

Yeah.

而且，另一个需要我们对术语保持高度精确的原因，不仅在于研究问题本身，正如你所说，还在于所使用的具体干预措施、具体的暴露因素，以及参与者对我们所讨论内容的明确性和细致性。

And again, another reason for being very specific in our terms and how we're defining things, not only that research question, but as you say, whether that's the particular intervention being used, the particular exposures, the participant having a real degree of specificity and granularity about what we're talking about.

你提到的其中一个非常重要的点是，我认为这在营养学讨论中经常出现：当我们谈论因果关系时，尤其是在饮食与疾病的问题上，我们希望回答一些因果性问题。

One of the really important things that you've touched on, I think it comes across in nutrition discussions all the time, is when we're talking about causality in general, because when we're coming to diet disease questions, we want to answer some of these causal questions.

我认为，这个领域常常受到批评，可能是因为与其他更容易进行严格随机对照试验的更清晰的生物医学领域相比，由于各种原因，我们在一些营养相关问题上并不具备这种优势。

And I think oftentimes there has been criticisms of the field in general because maybe compared to other cleaner biomedical fields where it's easier to do really stringent RCTs to answer some of these questions for a variety of reasons, we don't have that luxury with some of these nutrition related questions.

因此，营养流行病学才如此重要。

Hence why why nutrition epidemiology is so important.

而有时，有些人可能并不真正理解我们如何通过流行病学来解答这些因果性问题。

And then I think sometimes there's a tendency for some people to maybe not really understand how we can get to some of these answers to causal questions through epidemiology.

他们只从这种生物医学的角度出发，或者认为随机对照试验是唯一的方法，如果没有它，我们就只是在观察。

They think solely from this kind of maybe biomedical perspective or maybe that RCTs are the only way to do it and without that we're kind of just looking at observations.

但正如您的大量工作所强调的，营养流行病学有着很大的潜力，取决于我们如何开展研究、如何设计实验，以及该领域未来的发展方向，能够通过营养流行病学回答一些因果性问题。

But as a lot of your work has highlighted that there's some real scope within nutrition epidemiology depending on what way we go about it and how we design studies and kind of the future going forward of the field to be able to answer some causal questions through nutrition epidemiology.

您能否谈谈这方面的一些内容，以及您对营养流行病学如何实际用于回答这类问题的看法？

Could you just speak to some of that and your kind of perspective on how actually nutrition epidemiology can be used to answer these types of questions?

我认为这部分也源于一个讨论：在试验中看到的现象，是否也能在观察性研究中看到。

I think some of this also stems from a discussion on whether you can see something in a trial and also see the same thing in an observational study.

因此，这二者之间存在不一致。

So it's this discordance between the two.

已有若干研究试图做到这一点，但实际上，队列研究的分析结果与干预研究的结果很少一致。

There's been some studies that tried to do that, but actually very rarely the analysis with the cohort study aligns with that of an intervention study.

这又成为一个非常困难的论点，因为它们其实并没有回答同一个问题。

And then it also becomes a very hard argument because they're not really answering again the same question.

因此，推动这一方向的一种方法是我们所采用的‘目标试验’框架。

So one way of moving more towards that is one thing we've done is trying to use this target trial framework.

我一开始稍微提到了这一点，即尝试从一个你原本会通过干预性随机对照试验来回答的问题出发，然后看看我们的观察性数据中有哪些可以尽可能模拟这一试验的数据。

So I touched on it a little bit in the beginning, but trying to start with a question that you define more as an intervention, randomized controlled trial, would have made to answer your question and then see what data in our observational data do we have so we can mimic that as close as possible.

当然，在每一步操作中，你都要说：嗯，这本应像一个随机试验，但我们做了这样的调整、假设或差异。

And then of course for each step you do, say: Ah, it should be like this randomized trial, but we make this adjustment or we make this assumption or we make this difference.

在这些步骤中，你也能更清楚地看到：我们当前做法的主要局限性是什么？

And in each of these steps, you can also more clearly see: So what are the big limitations of what we're doing?

而且这不仅仅是一个方向，即从理想的试验走向数据，你也可以反过来想：嗯，但我们甚至都没有能回答这些问题的数据。

And it's not only that one way, right, going from the ideal trial to the data, then you may also go back and say: Ah, but okay, we don't even have the data to answer these questions.

如果这个试验要能被回答，它应该是什么样的？

And what would that trial then be if it should also be something we can answer?

在我于卡罗林斯卡学院做博士后期间，我构思了这两项研究，试图采用这种方法，并将其与更传统的做法进行对比。

So during my time as a postdoc in Karolinska, I sort of conceptualized these two studies where we tried to use this approach, but also contrast it to more the traditional approach.

我认为这实际上突显了一些非常出色的地方。

And I think that's it actually highlights some some really good things.

我认为，完成两项总体相似的研究并让它们都发表出来其实相当困难，因为人们会说：但这两者不是一样的吗？

I think getting it through to like, doing two overall similar studies and then getting both of them published was actually quite hard because people were like, well, but that's the same thing.

为什么要重复做两次？

Like, why do it twice?

但我认为这实际上非常重要。

But I think it was actually really, really important.

所以我们使用了两个瑞典队列：瑞典男性队列和乳腺筛查队列。

So we used two Swedish cohorts, the Swedish cohort of men and the this mammography cohort.

它们已经被用于许多不同的研究。

So they've been used for many different things.

这些队列的优点是包含了多次饮食测量，虽然不多，但在队列研究中已有少量额外的饮食摄入数据。

The good thing about these cohorts is that they have multiple measures of diet, not a lot, but few extra measures of dietary intake in a cohort study.

但这非常有用，因为我们可以至少观察到事物随时间的变化。

But that's really good because then we can at least see if things change along the way.

这真的很重要。

And that's really important.

首先，我们进行了更传统的队列分析。

So first, we did a more traditional cohort analysis.