Welcome to How to 115

A couple years ago I decided to start keeping a bibliography of health related science news. It started out as “How to Immortality,” but then some researchers claimed the upper limit to the human lifespan is 115 years, so I adjusted down to that more modest goal. Over time my “How to 115” document has ballooned both as new news has come out and as I’ve conducted my own mini-lit reviews on whatever catches my interest. I’ve found it very useful to put new research into context. Several friends have asked for access to my Google doc, so I figure there might be some latent demand for a blog like this. My two-part plan is to make themed posts on past research as well as timely updates as I catch them in the news.

I’m interested in any rigorous, actionable research that in some way can benefit my health or happiness based on my own actions. In rough order of importance I’ll talk about

• all-cause mortality
• mortality from heart disease or cancer
• mortality and complications of type 1 diabetes (which my mother, her sister, her father and I have or had)
• anything else that could happen to my body to make me dead or unhappy

I’m a big fan of The New Health Care column in The New York Times Upshot by Aaron Carroll. He writes very nice little little reviews and offers commentary on big new health-related research. He recently published a book, The Bad Food Bible, with an intro that describes his framework for evaluating the quality of scientific evidence. All people who disagree with this framework (and most people who agree with it) will find my blog quite boring. The hierarchy of research quality is, roughly speaking, as follows:

• Anecdotes: Carroll gives the example of someone who claims “My great-grandmother ate a tablespoon of Tabasco every morning, and she lived to be nearly a hundred.” He says these stories, “have no scientific worth whatsoever.” I’ll go a step further and say that due to selection bias, if someone relies on an anecdote to make their point, I think it’s even less likely to be true than if they had said nothing at all. If they’re talking about something many people are interested in but they’re relying on anecdotal evidence it probably means someone conducted some better form of research and got a null result.
• Observational studies
  • Cross-sectional studies and case-control studies: You ask a lot of people what they do right now, or what they remember doing in the past, and measure their health right now. They’re like a statistically significant version of an anecdote. They can’t tell you which way causality runs.
  • Cohort study: You try to find two groups of people who, on average, differ in exactly one way: diabetics/non-diabetics, vegetarians/non-vegetarins, etc. Then you follow them over time and see what happens to them. It’s not randomized or blinded, so it still can’t tell you which way causality runs.
• Experimental studies
  • Randomized controlled trials: randomly split people between a group that get a treatment and those that don’t. If the treatment is truly the only thing that differs between the groups, then a statistically significant difference in outcomes between the groups tells you that the treatment caused the difference. Common pitfalls of RCT’s include p-hacking (one person tests 100 treatments that actually do nothing and accepts any treatment that has p < .05, then publishes 5 useless treatments), publication bias (100 people test one treatment that does nothing, 95 find no effect and don’t publish, 5 publish with p <= .05, meta-study finds an even lower p value), bad control groups (treatment group given acid, control group given vitamin B, everyone knows if they had an acid trip or not (This only undermines the blinded-ness of the trial. It’s still randomized and controlled, but the causality could be due to a placebo effect.)), and not-actually-random assignment to treatment groups.
  • Meta-study: merge data from multiple RCT’s into something that looks like one gigantic RCT. These can overcome p-hacking and can check for publication bias with a funnel plot, but can’t do much about bad control groups or non-random group assignments..
• Human tests > animal tests > Petri dishes. Tests in Petri dishes often don’t transfer to animals and tests in animals often don’t transfer to humans. They’re useful for generating ideas for what to test in animals and then humans, but so rarely actually transfer that they’re not worth much on their own.
• I will rank the importance of outcomes as follows
  • Useless: process measures with no known connection to human disease. For example, the study discussed here that says diet soda causes changes to the gut microbiome. They try to make this sound scary by saying, “It could be that for some people who responded negatively to the artificial sweetener, the bacteria that got crowded out were helping to keep glucose in check. How it’s happening isn’t clear” (emphasis added). So by their own admission, instead of showing a correlation between diet soda and diabetes they showed a correlation between diet soda and something that has no known connection to diabetes. They may as well show that diet soda correlates with needing to pee later.
    • (Side note about how bad this study was: it had only 7 people and no control group, and detected changes in only 4 of the 7 people’s gut microbiomes. With no control group, they haven’t even proven correlation, let alone causation. Maybe 4 out of 7 people have their gut microbiomes change over that much time anyway.)
    • (Other side note: I’m probably in the top decile of diet soda consumption, so I’m strongly biased to think it’s not bad for me. That said, I’d be extremely interested to see an RCT of diet soda vs. unsweetened bevarages  in humans that showed an effect.)
  • Somewhat useful: process measures that have a known connection to human disease. Changes in blood pressure and cholesterol will show up much faster than changes in heart disease, but what we actually care about is heart disease. Any effect size for a process measure is likely to be much smaller for the actual outcome we care about.
  • Actual outcomes: being dead or sick. The stuff you would know was a problem even without a doctor or a lab test.