It’s probably only important in the cumulative though. When we have studies like this for many foods, you could put together a diet that reduces your chance of cancer by 20 or 30%, say.

I don't think that quite transfers, epidemiology is very weak, it only surfaces associates which is a good point to do a interventional trial but that is rarely done. The core problem with these studies is that to isolate variables they have to make a model of that variable in isolation, this relies on both assumptions of the model maker, accuracy of data, and is very vulnerable to p-hacking. Model assumptions that a hamburger and fries counts as meat, but not vegetable (potato) also impact the outcomes.

The large observational food surveys conducted typically have a 1-4 year questionnaire about how many servings of different food someone ate. Once every 4 years leaves lots of room open for forgetfulness.

There is a huge problem with healthy user confounders, people trying to follow all the modern health advice are going to skew results - not because all of the advice is correct, but some of it is. If someone exercises regularly, practices mindfulness, avoids processed foods, avoids meat - Are their improvements due to any single variable, yet on a food survey they get over represented because of these exclusionary behaviors.

We also have multiple different epidemiology studies covering the same topics and getting different results, that probably means we are focusing on the wrong question, it's noisy.

From my reading its far more likely the modern epidemic of chronic disease is caused by the introduction of excessive carbohydrates in processed foods, the novel addition of industrial oils (again processed foods) into the food supply - they account for 30%!!! of the average westerners average calorie intake, exposure to food contaminates from agrochemicals such as pesticides. The metabolic context of people filling out these surveys is a critical part that is being omitted.

In the following graphs notice how the incidence is very high in countries with traditionally low meat consumption like india? This indicates the hypothesis generated from the abstract paper isn't asking the right question.

example graphs

CVD

collapsed inline media

Type 2 Diabetes

collapsed inline media

My point is that you can follow every bit of advice from associative food surveys, but since the wrong questions are being focused on, your outcomes wont be as good as you hope. Quite frankly epidemiology is more about publicity and marketing then being part of the scientific process.

If you haven't read about the Metabolic Theory of Cancer I highly recommend giving it a read. It's a much more compelling model, and explains the surge of cancer since 1900, as well as actionable steps to reduce incidence (reduce sugar and inflammation).