FDA_TrackerJim said:

The gap between trial results and real-world results is consistent and it is not fraud.

Filing a mild objection. Mild because I might be wrong; an objection because nobody has addressed the case that does not fit. I would add the less popular caveat: these trial populations under-represented several groups, older adults and the highest BMI categories among them. The results probably generalise, and "probably" should be stated as an assumption rather than dropped.

nick_newbie said:

I keep finding that the number in the press summary and the number in the paper are not the same number, and the difference is always in the same…

Forest plot interpretation for the the trial evidence meta-analysis: when reading the pooled estimate, pay attention to:

1. Point estimate (HR/RR/OR) — center of the diamond
2. Confidence interval width — precision of the estimate
3. I² statistic — heterogeneity across studies
4. Individual study weights — are results driven by one large trial?
5. Prediction interval — range of plausible true effects in future settings

The the trial evidence meta-analysis shows a pooled RR of 0.80 (95% CI 0.72-0.85), I²=53%. This is a robust and consistent effect.

LibrarianMeg said:

I would add the less popular caveat: these trial populations under-represented several groups, older adults and the highest BMI categories among them.

LibrarianMeg said:

...regarding the trial evidence...

I think this is an underappreciated point. To expand on it with some data:

A recent meta-analysis of 15 RCTs (n=12,300) found that the trial evidence was associated with a clinically meaningful effect size across diverse patient populations[1].

The NNT was 12, which is comparable to antihypertensives for stroke reduction. That's a strong clinical argument for this approach.

Following on from dan_philly — and this may be the naive question:

What did you change at the same time, and can you separate the two now?

Closing the loop on my own question.

Update — my curve sits below the published mean and the explanation is that the trial arm had support I do not have. That was reassuring rather than otherwise.