Pet photos thread v2 — what worked for you?

Answering the narrow version, because the broad one does not have a single answer. Read four things before the headline number. The population, because trial populations are selected and supported in ways that real cohorts are not. The comparator, because "better than placebo" and "better than the current standard" are different claims and get reported identically. The primary endpoint as pre-registered, because a secondary endpoint promoted after the fact is a hypothesis rather than a finding. And the completion rate, because a large effect in the half of participants who finished is a different result from a large effect in everybody enrolled.

Dr.GastroMayo said:

Read four things before the headline number.

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 0.76 (95% CI 0.68-0.91), I²=49%. This is a robust and consistent effect.

Dr.MetabolicMD 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…

This matches mine closely enough to be worth saying so. Relative and absolute effects need reading together. A 20% relative reduction on a high baseline risk is a large absolute benefit; the same relative figure on a low baseline risk is a small one, and press summaries almost always quote the relative number because it is bigger.

Clinical perspective, offered as context rather than as advice.

Bayesian meta-analysis perspective on the trial evidence: traditional frequentist meta-analyses report point estimates and confidence intervals. Bayesian approaches provide probability distributions that are more intuitive for clinical decision-making.

For example: "There is a 98.5% probability that semaglutide 2.4mg produces >10% weight loss vs placebo" is more actionable than "RR 3.4, 95% CI 2.8-4.1, p<0.001."

The the trial evidence evidence is strong under both frameworks, but Bayesian analysis better communicates the degree of certainty for individual patient counseling.