ApoB Does Not Beget Plaque? A Statistical Critique of the KETO-CTA Trial

Описание: We break down the recent controversial paper “Plaque Begets Plaque, ApoB Does Not” with a critical eye on its statistical flaws.

Slides are available here:

https://www.jsdatascience.com/keto_ct...

Study:
https://www.jacc.org/doi/10.1016/j.ja...

This is a statistical critique of the KETO-CTA study — not a diet debate.

We evaluate whether the study’s statistical methods support its claim that “plaque begets plaque, ApoB does not.”

This episode covers:

Why modeling plaque change (ΔNCPV) as the outcome introduces clinical ambiguity, statistical fragility, and mathematical coupling.
How the use of univariable regressions, even though they had multiple available covariates, leads to unreliable, fragile estimates.
Why their use of Bayesian inference, intended to support null associations, gives a sense of confirmatory strength despite being applied to unadjusted, exploratory models, with Bayes factors based on a questionable prior and no sensitivity analysis.

We discuss several additional concerns:

Lack of multiple comparison correction despite ≥14 regressions
Evidence of zero-inflation and heteroscedasticity in plaque outcomes
Potential underpowering for detecting ApoB effects
High inter-individual variability in the LMHR group, not good for average changes but fine for regression modeling?

We also cover the authors' response to a letter to the editor published in the same journal, where external researchers raised some methodological concerns.


#KETOCTA #ApoB #LDLcholesterol #keto #ketodiet #StatisticalCritique #PlaqueProgression

00:00 – Introduction & Motivation
5:03 – Study Background and Design
12:15 – Summary of Findings & Critique Setup
16:21 – ΔNCPV as Outcome Variable
24:51 – Change Scores Combine Error
30:23 – Mathematical Coupling
48:25 – Univariable Linear Models
59:40 – Bayesian Modeling
1:00:00 – 3 Bayesian Inference
1:10:40 – 4 Bayesian Prior Choice
1:20:09 – 5 Bayes Factor Interpretation
1:25:17 – Summary
1:28:33 – Additional Concerns
1:28:48 – No Adjustment for Multiple Comparisons
1:30:06 – Zero-inflation, censoring, heteroscedasticity
1:34:03 – Sensitivity Analysis
1:38:27 – Possibly Underpowered Study
1:41:19 – On the Heterogeneity of LMHR
1:45:08 – Letters to the Editor
1:52:15 – Final Thoughts