One way to think about the recent meta-analysis paper on the health consequences of eating red meat is to think of red meat as a medicine. Let’s call it Red Meat Therapy, RMT for short, and we can imagine administering RMT to patients.
This will seem weird. First, red meat isn’t usually thought of as a medicine, any more than, say, panang curry is thought of as a medicine, but let’s put that aside for a moment. Second, and in case you didn’t read the paper, it showed (based on meta-analysis of a host of other papers) that eating red meat is likely bad for you. We don’t usually administer non-medicines that are likely bad for you to people and call them medicines, or least we don’t usually do that and not call it quackery.
But bear with me. Because the paper argued that we have such weak evidence against eating red meat is that it’s hard to make a strong recommendation against eating red meat. That’s not the same thing as saying red meat is not not bad for you, let alone that red meat is good for you. We just aren’t sure how bad it is, but it seems at least a little bad. Admittedly, it’s not clear what to do with that information. Lots of things are a little bad — sometimes it seems like most things in life1Like life itself, really. are at least a little bad.
What are we to do with things are a little bad? One thing we can do is ignore it. We do that a lot. We can also put it in practical and quantitative terms, which seems like a non-awful idea.
We have ways of doing that kind of calculation. One way is to use a measure called “number needed to treat” (NNT). It tells you how many patients need to be treated with a particular medication before we expect to see an effect, like, say, a saved life.2It’s a fun calculation. NNH is the inverse of the absolute risk reduction (ARR), which is, in turn, the difference between the rate of an experimental treatment (EER) and that of a control treatment (CER), or ARR = CER – EER. To be specific, if a drug reduces the risk of a bad thing happening from 50% to 40%, the ARR is calculated as 0.5 – 0.4 = 0.1, which gives us an NNT of 1/ARR = 10. You would, in other words, need to treat ten people to expect one to benefit. In the best case the NNT is 1, where everyone who is treated benefits. That mostly doesn’t happen, other than in fake medicines for, like, hair loss.
But you can turn NNT around and calculate the number needed to harm (NNH). That, as the name suggests, is how many people need to have a particular treatment before it’s likely we hurt someone. Granted, that isn’t the way we usually think about therapies, for obvious reasons, but it can cover some interventions.
You can apply that method to our RMT. Pretending red meat is medicine, and taking the base incidence3Around 4.6% of one of the main projected bad consequences of red meat’s excessive consumption (colorectal cancer), and then comparing that to the study-based projected increase in colorectal cancer4This is obviously controversial — hence the meta-analysis we are writing about here — but a mid-point is a roughly 20% increase, putting the incidence at 5.4% or so., we can say, at least approximately, how many people would have to take Red Meat Therapy before we expected an additional case of colorectal cancer.
So, how effective is RMT? Not so good, at least under these assumptions. If we were trying to give one more person colorectal cancer by stuffing them regularly with red meat, we would need to treat around 100 people with RMT. If this were a drug, we would likely call it a failure — it doesn’t do much for most of the people who take it.
To return to the original study, this does help us think more coherently about the paper. It isn’t that red meat isn’t bad for you, as some of the resulting media articles implied5Red meat is back on the menu!. It’s just that the effect size is so small, and the confounds so large, that we can’t say much about red meat’s actual effect on most people, most of the time, in most real-world situations, especially given all the stuff we do to ourselves that aren’t good for us.6There is a separate issue here, which is how seriously to take meta-analysis papers of other papers that themslves have weak conclusions based on their data. I sometimes argue that such meta-analysis papers are the research equivalent of collateralized debt obligations (CDOs): bundles of individually squirrelly things that magically become credible when wrapped together in a neat quantitative package. This, of course, should feel unsettling, and no more defensible in meta-analysis papers than in mortgage securities, but that’s a topic for another day. Humans are complex, mischievous systems, and we shouldn’t be surprised when our bodies conspire to make even the best intentioned nutrition researchers look silly.
Here are some papers worth reading:
- Atrial Fibrillation Ablation: Two Concerning Trends
- Loss of microbial diversity and pathogen domination of the gut microbiota in critically ill patients
- The Influence of Caffeine Expectancies on Simulated Soccer Performance in Recreational Individuals
Science & Technology
- Real time analysis of epidemic data
- Forward and Reverse Parking in a Parking Lot
- Common statistical patterns in urban terrorism
- Background noise disrupts host–parasitoid interactions
- Online bots are evolving faster than expected.
Economics & Finance