Keto-CTA Study MANIPULATED Charts!? Statistical Violations!?

Описание: Let's review the data and statistics behind the Keto-CTA study “Plaque Predicts Plaque, ApoB Does Not” (previously titled “Plaque Begets Plaque, ApoB Does Not”). The study followed “lean mass hyper-responders” (people on a ketogenic diet with very high LDL cholesterol) and tracked plaque progression for one year.
I want to answer the question: do the data and the statistical analysis actually support its claims?

Using the plaque data released by the funding organization, I recreated the main figures and refit core models from the paper. Here’s what I found:

• The main plaque figure has a manually relabeled y-axis (0–10% instead of the 0–20% range in the raw data), and the red IQR band, which is supposed to show the middle 50% of patients is simply wrong. This can be confirmed by anyone. Just go to the citizen science foundation's webpage for the data at: https://citizensciencefoundation.org/... , download the Keto-CTA Table – Imaging Metrics file, open in Excel or any spreadsheet program, look at the the column "V2_Percent_Atheroma_Volume", (1 Year PAV in chart 1B) find the highest value. It is 0.20, or 20%. The published chart shows it as 10%.
• Output from key statistical models needed to support some of the paper’s main conclusions is missing.
• All key models rely on change scores and one-factor-at-a-time (univariable) models. Change scores add noise, amplify measurement error, and create artificial relationships with baseline values, while univariable models ignore confounding factors like age, sex, and baseline plaque levels, which can make ApoB appear unimportant.
• Model diagnostics show that these linear models violate important liner model assumptions, which makes the reported p-values and conclusions unreliable.

This is not a pro or anti keto video. I am only evaluating whether this particular study’s statistical methods and reporting can justify the claim that ApoB is not related to plaque progression in this cohort. From a statistical standpoint, they cannot.

Please note that these problems are about how the data were modeled and reported, not that the data are unusable. With more appropriate analysis methods (for example, modeling follow-up plaque while adjusting for baseline plaque and basic covariates, using transformations such as log scales and using approaches that handle non-constant variance), many of the assumption violations could be addressed. Outcomes could also be re-analyzed with models better suited to their distribution, rather than simple linear regression.


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

My Git code repo: https://github.com/SloughJE/keto_cta_...

Keto-CTA Data: https://citizensciencefoundation.org/...

Full podcast:    • KETO-CTA study: Data Published... It's WOR...  

#keto #lmhr #ketodiet #apob #ldl #statistics

0:00 Why Revisit the Keto-CTA Study? Because we have the data.
2:33 Chapter 1: Manually Altered Charts
4:57 Chapter 2: Missing Model Output
6:37 Chapter 3: Statistical Modeling Choices that Obscure Relationships
10:10 Chapter 4: Model Assumption Violations
13:02 Valuable Dataset and Credit for Transparency
13:48 The Statistical Verdict