https://adventisthealthstudy.org/studies/AHS-2
Participants: 96,000 Adventists living in the U.S. and Canada
- 1 FFQ administered 2-3 years after enrollment
- Ongoing since 2002
- ~23 ish years without a FFQ followup
Since the list of publications from this dataset is quite extensive
Let's see what the good PhD Kay's remarks on Red and Processed Meat and Mortality in a Low Meat Intake Population
- No controls on the population
- Self-selection
- Zero Quartile: Vegan death rate 11.53%
- First Quartile : 11.30%
- Second Quartile: 10.55%
- Third Quartile: 9.33%
- Fourth Quartile (most meat): 9.30%
Dose response, more meat you eat, the less you tend to diet in this group.
However, the study itself reports risk ratios
- Zero Quartile: RR 1
- First Quartile: RR 1.16
- Second Quartile: RR 1.27
- Third Quartile: RR 1.39
- Fourth Quartile (most meat): RR 1.58
So how does a group with the lowest recorded death rate 9.30% have the highest risk ratio of 1.58?
Through statistical adjustments of single variate regressions, they are inputing causative values
- confounder - Risk Adjustment
- smoking - up
- age - down
- alcohol - up
- exercise - down
- gender male - up
- diabetes - up
- bmi - up
- graduate degree - up
Modify the original data, and a 9.30% death rate becomes the most risky 1.58... and a 11.53% death rate becomes the least risky 1.0
Deeper dive on correcting confounders and collinearity.
- Use of multivariate correction assumes non-collinearity
-
- double pump errors
-
- quadruple pump errors
- There are more men in each quartile going up with meat consumption...
- men are more likely to die younger
-
- Therefore the gender and meat consumption factors are colinear, and cannot be corrected using multivariate correction. These variables are not independent.
Since this paper came up today in the general lemmy zeitgeist - https://hackertalks.com/post/15655222
It's worth revisiting here on the carnivore community now this is about cancer and not mortality... but same authors, same dataset, same agenda, same issues.
My opinion:
The actual study Longitudinal associations between vegetarian dietary habits and site-specific cancers in the Adventist Health Study-2 North American cohort
abstract
Background
Associations between vegetarian diets and risk of common cancers are somewhat understood, but such data on medium-frequency cancers are scarce and often imprecise.
Objectives The objectives of this study was to describe multivariable-adjusted associations between different types of vegetarian diets (compared with nonvegetarians) and risk of cancers at different bodily sites.
Methods The Adventist Health Study is a cohort of 95,863 North American Seventh-day Adventists established between 2002 and 2007. These analyses used 79,468 participants initially free of cancer. Baseline dietary data were obtained using a food frequency questionnaire and incident cancers by matching with state and Canadian provincial cancer registries. Hazard ratios (HRs) were estimated using proportional hazard regression. Small amounts of missing data were filled using multiple imputations.
Results Overall cancers, all vegetarians combined compared with nonvegetarians, had HR: 0.88; 95% confidence interval (CI): 0.83,0.93; P < 0.001, and for medium frequency cancers, HR: 0.82; 95% CI: 0.76, 0.89; P < 0.001. Of specific cancers, colorectal (HR: 0.79; 95% CI: 0.66, 0.95; P = 0.011), stomach (HR: 0.55; 95% CI: 0.32, 0.93; P = 0.025), and lymphoproliferative (HR: 0.75; 95% CI: 0.60,0.93; P = 0.010) cancers, were significantly less frequent among vegetarians. A joint test that HR = 1.0 for all vegetarian subtypes compared with nonvegetarians was rejected for cancers of the breast (P = 0.012), lymphoma (P = 0.031), all lymphoproliferative cancers (P = 0.004), prostate cancer (P = 0.030), colorectal cancers (P = 0.023), medium frequency cancers (P < 0.001), and for all cancers combined (P < 0.001).
Conclusions These data indicate a lower risk in vegetarians for all cancers combined, as well as for medium-frequency cancers as a group. Specific cancers with evidence of lower risk are breast, colorectal, prostate, stomach, and lymphoproliferative subtypes. Risk at some other sites may also differ in vegetarians, but statistical power was limited.
I've ranted about my Standards for Nutritional Evidence before, but here are the major issues
- Observational
- Weak Hazard Ratios
- Absolute risk not calculated or published
- Compares only against the Standard American Diet [SAD]
- Cannot inform on cause and effect
- proportional hazard regression is just a fancy way of saying they tried to model out confounders by guessing
- Food Frequency Questionnaire administered asking about 1 year of food administered once, in a study that has been running since 2002 - 23 years! 21 years since a FFQ
