本文探讨了激素替代疗法（HRT）在绝经后治疗中的争议。文章回顾了HRT从早期被认为益处多、风险少，到因2002年“妇女健康倡议”（WHI）大型随机对照试验（RCT）的结论而导致使用量急剧下降的历程。然而，作者引用了一本2024年的书籍，该书提出了WHI的结论可能基于“垃圾科学”，存在统计错误、选择性报告等问题，并认为HRT对大多数女性是有益的。文章指出，双方都指责对方存在偏见和科学谬误，使得真相难以辨别。作者认为，HRT的争议是理解医学推断、科学方法和科学传播复杂性的一个典型案例，值得深入探讨。

📊 HRT的早期认知与WHI试验的冲击：早期研究认为HRT对绝经后女性益处显著且风险极低，导致其广泛应用。然而，2002年“妇女健康倡议”（WHI）大型随机对照试验（RCT）的结论推翻了这一观点，揭示了其潜在风险，从而大幅降低了HRT的使用率，并被视为“垃圾科学”的警示案例。

🔬 对WHI试验结论的质疑与反驳：一本2024年的书籍提出，WHI试验的结论可能存在统计错误、p-hacking、过度解读初步结果、选择性报告等问题，并认为其数据集和后续研究不支持其负面结论。该书认为，数据更支持HRT对大多数女性是有净益处的。

⚖️ 科学争议中的偏见与推断困境：在HRT的争论中，无论是HRT的支持者还是反对者，都指责对方存在偏见和科学谬误。这种双方都坚称自己追求真理并引用数据支持结论的情况，使得普通人难以辨别专家中的“垃圾科学家”和“理性堡垒”。

💡 HRT争议作为科学方法论的案例研究：作者提出，HRT的争议是一个极佳的案例，用于探讨科学研究如何迭代地修正我们对真相的认知，不同科学和统计方法的优势与局限性，偏见对结论的影响及其心理社会学根源，以及科学传播中的夸大和修辞手法问题。

🧠 追求科学真理的挑战与反思：文章强调了区分事实陈述与修辞风格的重要性，并反思了“最新的‘揭假’最可能是正确的”这种直觉的局限性。作者建议，在信息传播中应努力消除可能影响判断的修辞手法，并保持对现有认知的“认知谦逊”。

Published on July 21, 2025 4:18 PM GMT

I recently came across a 2024 update on a 2018 book making the still-controversial case that hormone replacement therapy (HRT) after menopause is highly beneficial and that rumors of its risks are unfounded, or at least highly exaggerated.  

According to their narrative (apparently not contested), HRT started with the physiologically reasonable idea that menopausal symptoms could be treated with estrogens (which at the time were extracted from pregnant horse urine). Early observational epidemiological studies reported that taking estrogen during menopause was associated with considerable benefits and minimal risks, leading to its widespread use by menopausal women. Then a huge randomized controlled trial (RCT) by the Women's Health Initiative (WHI) famously overturned that prevailing wisdom, debunking the purported benefits and establishing considerable risks. Publication of those conclusions in 2002 led to a drastic reduction in the use of HRT. This segment of the history of HRT became a stock example of the perils of junk science: the fallacy of drawing conclusions from anecdotes or mistaking correlation with causation; the triumph of RCTs in exposing the real truth; and the curious resistance of people to updating their beliefs in the face of iron-clad evidence to the contrary. 

However, the aforementioned book goes on to say (controversially) that the conclusions of the famous WHI study were subsequently proven to be completely wrong. The bulk of the book is spent detailing the reasons these authors believe the previous "myth debunking"  claims were themselves based on junk science: statistical errors, p-hacking, over-interpretation of preliminary results, selective reporting, making claims on the basis of non-significant results, exaggerating clinical significance due to the base rate fallacy or reporting relative risks instead of absolute risks^[1].  In summary they argue that WHI's report reflected mostly their biases. The book claims that the WHI's own dataset does not support their claims, and numerous subsequent studies consistently reproduce benefits of HRT, and fail to reproduce risks. To their reading, the data squarely favor HRT as net beneficial for most women. Since 2002 the WHI has moderated its claims, but still stands by the main conclusions. Unfortunately, I don't have confidence that the book authors are free of bias, either.  To make matters more complicated, there are accusations of ethical impropriety against both the opponents and proponents of HRT.

When both sides of a debate accuse the other of bias and bad science, it can be hard to tell which experts are the junk scientists and which are the bastions of rationality^[2]. In this case, there are people with serious credentials reaching opposite conclusions, each saying they are only committed to the truth, each citing scientific data which, if true, would seem to support their conclusions. Their arguments are not transparently irrational or unscientific.  If one read either side's account alone, it would sound extremely convincing. I have not had time to dig into the primary literature or formal critiques and rebuttals, but it seems like that would be minimally required to sort this out; diving into the datasets would be even better^[3].

I wondered whether this question had been discussed on LessWrong, and found that it was discussed briefly, over a decade ago^[4]. The discussion elicited interest on two levels: (1) Abstractly, as a case study in how to make rational decisions about medical/health claims in the face of conflicting claims, insufficient evidence, snake-oil-purveyors, etc.; and (2) Concretely, as in "so what's the truth about HRT?"  I take it that the first was the intended topic, but discussion of the second is partly informative in working out the first.  At that time the views here seem to have been mostly aligned with the prevailing narrative that WHI was the voice of reason (HRT is dangerous and bad), and some spoke as if this were an open-and- shut case. 

I nominate this topic for a re-boot as a case study in medical inference, scientific methodology, and science communication.  It already has the status of a paradigm example of bad science. There are over a century of data of various kinds, an extensive discussion of scientific and inferential methodology, with multiple rounds of back-and-forth critique^[5], and yet credible experts are still taking opposite positions on the basis of the same factual evidence. I am not confident that any account of the evidence is unbiased.

This could be an interesting test case for discussing:

The complexity of the process by which scientific research iteratively refines what we take to be true, individually and as a scientific community  The intrinsic strengths and limitations of different scientific and statistical methodsThe potential for bias to influence conclusions, the psychology and sociology behind this tendency, and what steps have been or can be taken to combat itThe incentives for sensationalism in peer-reviewed scientific publication as well as science communication directed at the public.  The idea that vigorous debate (adversarial dialogue) promotes truth-finding due to each side exposing the weaknesses and flaws of the others' arguments (does it?)Epistemic humility: calibrating the confidence with which we hold and communicate our current understanding of facts, especially as intelligent, informed non-experts.
 
What has particularly struck me in reading on this topic is the degree to which statements of fact or logic are often interwoven with rhetorical style that serves to sway opinion by means other than reason. This can be as subtle as writing with a tone of definitive finality or condescension, or as blatant as ridiculing opponents' positions.  I suspect writers are often unaware they are doing this, probably including myself.

My first thought is: it would be beneficial to  regularly edit with an eye to detecting and eliminating such rhetorical devices, so that presentations of evidence and logic can be as epistemically clean as possible. Admittedly this could make the prose dry, as such devices are among the best tools for engaging interest. But striving to write this way could help keep the reader's mind -- and the writer's own thinking -- unbiased as they process information.  On the other hand, those rhetorical devices are useful "tells" about a writer's biases, in the absence of which it would be even harder to detect the potential slant in their presentations of facts. Finally, one would not want to eliminate expression of value judgements or emotions; just be more explicit about when one is doing so, and keep it segregated from fact statements.

 

 

 

 

 

 

1. ^{^}

   As a newcomer to your forum it appears the readership is well-educated on these points, but I don't know if you have standard go-to links for these concepts. LMK if citations or links are wanted here.

2. ^{^}

   I notice a tendency to feel that the most recent "debunking" is most likely to be correct, but on reflection this is clearly not a good criterion.  Inasmuch as each round of a debate takes into consideration and responds to all the evidence and arguments of all previous rounds in an objective, dispassionate way, this would be true. But it's possible any rebuttal is driven in part by attachment to a prior position, leading to a slanted re-assessment, in ways that are not necessarily obvious to a reader.

3. ^{^}

   Would it be feasible for all the data of all the past studies to be de-identified, blinded, and combined into a huge dataset, such that people could analyze it using whatever stratifications or modifiers they wanted to, but commit to their conclusions before knowing which group was the placebo group? Or is that intrinsically impossible?

4. ^{^}

   Examples to this effect on LessWrong include: 
    Dealing with the high quantity of scientific error in medicine 

   Sequence Announcement: Applied Causal Inference 

   Why we need better science, example #6,281 

   and the comment threads thereof

5. ^{^}

   Comment on "Four Layers of Intellectual Conversation" 

Discuss