Nathan Hwang

The Mundane Science of Living Forever

Epistemic Status: timeboxed research, treat as a stepping stone to more comprehensive beliefs. Known uncertainty called out.

Live forever, or die trying!

Previously: Lifestyle interventions to increase longevity @ LessWrong9s of cats.

TLDR?

Yes, Immortality

I wrestled with whether to shoot for a more normal and mundane title, like “In Pursuit of longevity”, but “live a long time!” just doesn’t have the ring that “live forever!” does.

Clarification: I don’t have the Fountain of Youth. I’m relying on the future to do the heavy lifting. Kurzweil’s escape velocity idea is the key idea: we want to live long enough that life expectancy starts increasing more than 1 year per year. Life expectancy is currently stagnant, so we want to live as long as possible to maximize our chances of hitting some sort of transition.

In other words, we need silver bullets to overcome the Gompertz curve, but there are no silver bullets yet, just boring old lead bullets. We’ll have to make our own silver bullet factory, and use the lead bullets to get there.

So, the bulk of this post will be devoted to simply living healthily. A lot of the advice is boring and standard: eat your vegetables, exercise, get enough sleep. However, I wanted to check out the science and see what holds up under (admittedly amateur) scrutiny.

(I’ll be ignoring the painfully obvious things, like not smoking. If you’re smoking, stop smoking[1].)

My process: I timeboxed myself to 20 hours of research, ending in August 2017. First, I looked up the common causes of death and free-form generated possible interventions. Then, I followed the citations in the Lifestyle interventions to increase longevity post and then searched Google Scholar, especially for meta-analyses, and read the studies, evaluating them in a non-rigorous way. I discarded interventions that I wasn’t certain about: for example, Sarah lists some promising drugs and gene therapies but based only on animal studies, where I wanted more certainty. I ended up using 30+ hours, so not everything is exhaustively researched as much as I would like: for example, there was a fair amount of abstract skimming. I did not read every paper I reference end-to-end. On the other hand, many papers were also locked behind paywalls so I couldn’t do much more than that.

This means if you read one of these results and implement it without talking to your doctor about it and bad things happen to you, I will ask you: ARE YOU A SPRING LAMB? WHY THE FUCK ARE YOU DOING THINGS A RANDOM PERSON ON THE INTERNET TOLD YOU TO DO? AND WITHOUT VETTING THOSE THINGS?

Or more concretely: you are a unique butterfly, and no one cares except the medical world. What happens for the faceless statistical masses might not happen for you. I will not cover every single possible interaction and caveat, because that is what those huge medical diagnosis books are for, and I don’t have the knowledge to tell you about the contents of those books. Don’t hurt yourself, ask your doctor.

An example: blood donation

First, I wanted to lead with an example of how the wrong methods can cripple a conclusion and end up with bad results.

Now, blood donation looks like it is very, very good for male health outcomes. From “Blood donation and blood donor mortality after adjustment for a healthy donor effect.” with 1,182,495 participants (N=1,182,495) published in 2015 (note it’s just an abstract, but the abstract has the data we want):

» For each additional annual blood donation, the all-cause mortality RR (relative risk) is 0.925, with a 95% CI (confidence interval) from 0.906 to 0.943[2]. I’ll be summarizing this information as RR = 0.925[0.906, 0.943] throughout the post.

(Unless otherwise stated, in this post an RR measure will refer to all-cause mortality, and X[Y, Z] CI reports will be values followed by 95% confidence intervals. There will also be references to OR (odds ratio) and HR (hazard ratio)).

There’s even a well fleshed out mechanism, where iron ends up oxidizing parts of the cardiovascular system and damaging it, and hence doing regular blood donation removes excess blood iron.

But there are some possible confounders:

  • blood donation carries some of the most stringent health screens most people face, which results in a healthy donor effect,
  • altruism could be correlated with conscientiousness, which might affect health outcomes.

The study cited earlier is observational: they’re looking at existing data gathered in the course of normal donation and studying it to see if there’s an effect. In order to make a blanket recommendation that men should donate blood at some regular interval, what we really want is to isolate the effect of donation by putting people through the normal intake and screening process, and then right before putting the needle in randomize for actually taking the donation or not, or even stick the needle in and not actually draw blood.

(Note that randomization is not strictly better than observational studies: observations can provide insights that randomization would miss[3], and a rigorous RCT might not match real world implementations. Nevertheless, most of the time I want a randomized trial.)

No one had done an RCT (randomized controlled trial) in this fashion, and I expect any such study to have a really hard time passing an ethics board when I get numerous calls to help alleviate emergency blood need at a number of times throughout the year.

However, Quebec noticed that their screening procedures were too strict: a large group of people were being rejected when they were in fact perfectly healthy. The rejection trigger didn’t appear to otherwise correlate with health, so this was about as good a randomized experiment as we were going to get. Their results were reported in “Iron and cardiac ischemia: a natural, quasi-random experiment comparing eligible with disqualified blood donors” (2013, N=63,246):

» Donors vs nondonors, RR = 1.02[0.92, 1.13]

In other words, there was basically no correlation. In fact, in another section of the paper the authors could get the correlation to come back by slicing their data in a way that better matched the healthy donor process.

The usual hallmarks of science laypeople can pick apart aren’t there: the N is large, there’s a large cross-section of the community (no elderly Hispanic women effect) and there’s no way to even fudge our interpretation of the numbers: we’re not beholden to science’s fetish with p=0.05, so failing the 95% CI could be okay if it were definitely leaning in the right direction. But it’s almost exactly in the middle. The effect isn’t there or is so tiny that it’s not worth considering.


So that’s an example of how things can look like great interventions, and then turn out to have basically no effect. With our skeptic hats firmly in place, let’s dive into the rest!

Easy, Effective

Vitamin D

Vitamin D gets the stamp of approval from both Cochrane and Gwern[4]. Lots of big randomized studies have been done with vitamin D supplementation, so the effect size is pretty pinned down.

From “Vitamin D supplementation for prevention of mortality in adults” (2012, N=95,286, Cochrane):

» Supplementation with vitamin D vs none, RR = 0.94[0.91, 0.98]

Another meta-analysis used by Gwern, “Vitamin D with calcium reduces mortality: patient level pooled analysis of 70,528 patients from eight major vitamin D trials” (2012, N=70,528):

» Supplementation with vitamin D vs none, HR = 0.93[0.88, 0.99]

You might think that one side of the CI is pretty bad, since RR = 0.98 means the intervention is almost the same as the control. On the other hand, (1) wait until you read the rest of the post (2) keep in mind that it’s very cheap to supplement vitamin D. Your local drugstore probably has a years worth for $20. In a pinch, more sunlight also works, but if you have darker skin, sunlight is less effective.

If you’re interested, there’s lots of hypothesizing on the mechanisms by which more vitamin D impacts things like cardiovascular health (overview).

(If you want a striking visual example of vitamin D precursors correlating with cancer, there’s a noticable geographic gradient in certain cancer deaths; “An estimate of premature cancer mortality in the U.S. due to inadequate doses of solar ultraviolet-B radiation” (2002) states that some cancers are twice as prevalent in the northern US than the southern. There’s more sun in the south, and sunlight helps synthesize vitamin D. Coincidence?! If you want to, you can see this effect yourself by going to the Cancer Mortality Maps viewer from the National Cancer Institute and taking a look at the bladder, breast, corpus uteri or rectum cancers.)

Difficult, but Effective

Exercise

Exercising is hard work, but it pays off big.

From “Domains of physical activity and all-cause mortality: systematic review and dose–response meta-analysis of cohort studies” (2011, N=unknown subset of 1,338,143[5]):

» Comparing people that get 300 minutes of moderate-vigorous exercise/week vs sedentary populations, RR = 0.74[0.65, 0.85]

Unfortunately, “moderate-vigorous” is pretty vague, and the number of multiple comparisons being made is breathtaking.

MET-h is a unit of energy expenditure roughly equivalent to sitting and doing nothing for an hour. Converting different exercises (or intensities of exercise) to MET-h measures can allow directly comparing/aggregating different exercise data. This also makes it easier to decide exactly what “moderate-vigorous” exercise is, roughly mapping to less than 3 MET/h for light, 3-6 for moderate, and above 6 for vigorous.

With this in mind, we can get a regression seeing how additional MET-hs impact RR. From the previous study (2011, N=unknown subset of 844,026):

» +4 MET-h/day, RR = 0.90[0.87, 0.92] (roughly mapping to 1h of moderate exercise)

» +7 MET-h/day, RR = 0.83[0.79, 0.87] (roughly mapping to 1h vigorous exercise)

There’s a limit, though: exercising for too long, or too hard, will eventually stop providing returns. The same study places the upper limit at around a maximum RR = 0.65 when comparing the highest and lowest activity levels. The Mayo Clinic in “Exercising for Health and Longevity vs Peak Performance: Different Regimens for Different Goals” recommends capping vigorous exercise at 5 hours/week for longevity.

A quick rule of thumb is that each hour of exercise can return 7x time dividends (news article). This sounds great, but do some math: put this return together with the 5 hours/week limit, assume that you’re 20yo and doing the maximum exercise you can until 60, and this works out to adding roughly 8 years to your life (note that the study the rule of thumb is based on (2012) gives a slightly lower average maximum gain, around 7 years). Remember the Gompertz curve? We can huff and puff to get great RRs, and it only helps a bit. Unfortunate.

(While we’re exercising: keep in mind that losing weight isn’t always good: if you’re already at a health weight and start losing weight without intending to, that could be a sign that you’re sick and don’t know it yet (source).)

Other studies I looked at:

Unfortunately, most of these studies are based on surveys, which have the usual problems with self reports. There are some studies based on measuring VO2max more rigorously as a proxy for fitness, except those have tiny Ns, in the tens if they’re lucky (it’s expensive to measure VO2max!).

Diet

Overall, many of these studies are observational and based on self-reports; a few are based on randomized provided food, but the economics dictate they have smaller Ns. I’ve put all the diet-related things together, since in aggregate they are fairly impactful (if difficult to put into practice), but note that some of the subheadings contain less certain results.

Fruit and vegetables

It’s like your childhood authority figures said: eat your vegetables.

From “Fruit and vegetable consumption and mortality from all causes, cardiovascular disease, and cancer: systematic review and dose-response meta-analysis of prospective cohort studies” (2014, N=833,234):

» +1 serving fruit or vegetable/day, HR = 0.95[0.92, 0.98]

Like exercise, fruits/vegetables don’t stack forever either; there’s around a 5 serving/day limit after which effects level off. Still, that adds up to around HR = 0.75, competitive with maximally effective exercise.

Potatoes are a notable exception, having a uniquely high glycemic load among vegetables; this roughly means that your blood sugar will spike after eating potatoes, which seems bad. You can find plenty of debate about whether this is in fact bad[6].

Other reports I looked at:

Red/Processed Meat

You know bacon is bad for you, but… bacon is pretty bad for you.

From “Red Meat and Processed Meat Consumption and All-Cause Mortality: A Meta-Analysis” (2013, N=unknown subset of 1,330,352) effects from both plain red meat (hamburger, steak) and processed red meat (dried, smoked, bacon):

» Highest vs lowest consumption categories[7] for red meat, RR = 1.10[0.98, 1.22]

» Highest vs lowest consumption categories for processed red meat, RR = 1.23[1.17, 1.28]

There isn’t all-cause data I could find on fried foods specifically, but “Intake of fried meat and risk of cancer: A follow-up study in Finland” specifically covers cancer risks (1994, N=9,990):

» Highest vs lowest tetrile fried meat: RR = 1.77[1.11, 2.84]

Note that the confidence intervals are wide: for example, the red meat CI covers 1.0, which is pretty poor (and yet the best all-cause data I could find). If we were strictly following NHST (null hypothesis significance testing), we’d reject this conclusion. However, I’ll begrudgingly accept waggled eyebrows and “trending towards significance”[8].

If you’re paleo, you might not have cause to worry, since you’re probably eating better than most other red meat eaters, but I have no data for your specific situation.

Other reports I looked at:

Fish (+Fish oil)

Fish is pretty good for you! Fish oil might contribute to fish “consumption”.

“Risks and benefits of omega 3 fats for mortality, cardiovascular disease, and cancer: systematic review” (2006, N=unknown subset of 36,913) looked at both fish consumption and fish oil, finding that fish/fish oil weren’t significantly different:

» High omega-3 (both advice to eat more fish, and supplementation) vs low, RR = 0.87[0.73, 1.03]

Note this analysis only included RCTs.

“Association Between Omega-3 Fatty Acid Supplementation and Risk of Major Cardiovascular Disease Events: A Systematic Review and Meta-analysis” (2012, N=68,680) looked only at fish oil supplementation:

» Omega-3 supplementation vs none, RR = 0.96[0.91, 1.02]

Note that both of these results have relatively wide CI covering 1.0. Additionally, the two studies seem to differ on the relative effectiveness of fish oil.

There’s plenty of exposition on mechanisms for why fish oil (omega-3 oil) might help in the AHA scientific statement “Fish Consumption, Fish Oil, Omega-3 Fatty Acids, and Cardiovascular Disease”.

Also make sure that you’re not eating mercury laden fish while you’re at it; just because Newton did it doesn’t mean you should.

Other studies I looked at:

Nuts

This study of 7th Day Adventists by “Nut consumption, vegetarian diets, ischemic heart disease risk, and all-cause mortality: evidence from epidemiologic studies” points in the right direction (1999, N=34,198):

» Eating nuts <1 time/week vs >=5 times/week, fatal heart attack RR ~ 0.5[0.32, 0.75] (estimated from a graph)

However, I don’t trust it. Look at how implausibly low that RR is: eating nuts is better than getting the maximum benefit from exercise? How in the world would that work? Unfortunately, I wasn’t able to find any studies that weren’t confounded by religion, so I just have to stay uncertain for now.

Sleep

We spend a third of our lives asleep, of course it matters. The easiest thing to measure about sleep is the length, so plenty of studies have been done on that. You want to hit a Goldilocks zone of sleep length, not too short or not too long. The literature calls this the aptly named U-shape response.

What’s too short, or too long? It’s frustrating, because one study’s “too long” can be another study’s “too short”, and vice versa.

However, from “Sleep Duration and All-Cause Mortality: A Systematic Review and Meta-Analysis of Prospective Studies” (2010, N=1,382,999):

» Too short (<4-7h), RR = 1.12[1.06, 1.18]

» Too long (>8-12), RR = 1.30[1.22, 1.38]

And from “Sleep duration and mortality: a systematic review and meta-analysis” (2009, N=unknown):

» Too short (<7h), RR = 1.10[1.06, 1.15]

» Too long (>9h), RR = 1.23[1.17, 1.30]

So there’s range right around 8 hours that most studies can agree is good.

You might be fine outside of the Goldilocks zone, but if you haven’t made special efforts to get into the zone, you might want to try and get into that 7-9h zone the studies can generally agree on.

Again, most of these studies are survey based. I can’t find the source, but a possible unique confounder is that sleeping unusually long might be a dependent, not independent variable: if you’re sick but don’t know it, one symptom could manifest as sleeping more.

And, if you get enough sleep but feel groggy, you might want to get checked out for sleep apnea.

Other studies I looked at:

Less Effective

Flossing

The original longevity guide was enthusiastic about flossing. Looking at “Dental Health Behaviors, Dentition, and Mortality in the Elderly: The Leisure World Cohort Study” (2011, N=690), it’s hard not to be:

» Among daily brushers, never vs everyday flossers, HR = 1.25[1.06, 1.48]

Even more exciting is the dental visit results (N=861):

» Dental exam twice/year vs none, HR = 1.48[1.23, 1.79]

However, the study primarily covers the elderly with an average age of 81yo. Sure, one hopes that the effects are universal, but the non-representative population makes it hard to do so. So while flossing looks good, I’m not ready to trust one study, especially when I can’t find a reasonable meta-analysis covering more than a few hundred people.

As a counterpoint, Cochrane looked at flossing specifically in “Flossing to reduce gum disease and tooth decay” (2011, N=1083), finding that there’s weak evidence for reduction in plaque, but basically nothing else.

I’ll keep flossing, but I’m not confident about the impact of doing so.

Other studies I looked at:

Sitting

Sitting down all day might-maybe-possibly be bad for health outcomes.

There are some studies trying to measure the impact of sitting length. From “Daily Sitting Time and All-Cause Mortality: A Meta-Analysis” (2013, N=595,086):

» +1 hour sitting with >7 hours/day sitting, HR = 1.05[1.02, 1.08]

However, the aptly named “Does physical activity attenuate, or even eliminate, the detrimental association of sitting time with mortality? A harmonised meta-analysis of data from more than 1 million men and women” (2016, N=1,005,791, no full text) claims the correlation only holds at low levels of activity: once people start getting close to the exercise limit, this study found the correlation between sitting and all-cause mortality disappeared.

From “Leisure Time Spent Sitting in Relation to Total Mortality in a Prospective Cohort of US Adults” (2010, N=53,440):

» Sitting >6 hours vs <3 hours/day (leisure time), RR 1.17[1.11, 1.24]

Note that this is the effect for men: the effect for women is larger. Also, this study directly contradicts the other study, claiming that sitting time has an effect on mortality regardless of activity level. And who in the world sits for less than 3 hours/day during their leisure time? Do they just not have leisure time?

Again, these studies were survey based.

The big unanswered question in my mind is whether exercising vigorously will just wipe the need to not sit. So, I’m not super confident you should get a fancy sit-stand desk.

(However, I do know that writing this post meant so much sitting that my butt started to hurt, so even if it’s not for longevity reasons I’m seriously considering it.)

Other reports I looked at:

Air quality

Air quality has a surprisingly small impact on all-cause mortality.

From “Meta-Analysis of Time-Series Studies of Air Pollution and Mortality: Effects of Gases and Particles and the Influence of Cause of Death, Age, and Season” (2011, N=unknown (but aggregated from 109 studies(?!))):

+31.3 μg/m3 PM10RR = 1.02[1.015, 1.024]

+1.1 ppm CO, RR = 1.017[1.012, 1.022]

+24.0 ppb NO2RR = 1.028[1.021, 1.035]

+31.2 ppb O3 daily max, RR = 1.016[1.011, 1.020]

+9.4 ppb SO2RR = 1.009[1.007, 1.012]

(I’m deriving the RR from percentage change in mortality.)

By itself the RR increments aren’t overwhelming. But since it’s expressed as an increment, if there are 50 increments present in a normal day that we can filter out ourselves, then that adds up to some real impact. The increments aren’t tiny compared to absolute values, though. For example, maximum values in NYC during the 2016 summer:

PM10 ~ 58 μg/m3

CO ~ 1.86 ppm

NO2 ~ 60.1 ppb

O3 ~ 86 ppb

SO2 ~ 7.3 ppb

So the difference between a heavily trafficked metro area and a clean room is maybe twice the percentage impacts we’ve seen, which just doesn’t add up to very much. Beijing is another story, but even then I (baselessly) question the ability of household filtration systems to make a sizable dent in interior air quality.

There are plenty of possible confounders: it seems the way these sorts of studies are run is by looking at city-level pollution and mortality data, and running the regressions on those data points.

Other studies I looked at:

Hospital Choice

Going to the hospital isn’t great: medical professionals do the best they can, but they’re still human and can still screw up. It’s just that the stakes are really high. Like, people recommend marking on yourself which side a pending operation should be done on, to reduce chances of catastrophic error.

Quantitatively, “A New, Evidence-based Estimate of Patient Harms Associated with Hospital Care” (2013) says that 1% of deaths in the hospital are adverse deaths deaths. However, note that many adverse deaths weren’t plausibly preventable by anyone other than Omega.

If you’re having a high stakes operation done, “Operative Mortality and Procedure Volume as Predictors of Subsequent Hospital Performance” (2006) recommends taking into account a hospital’s historical morbidity rate and volume for that procedure: if you’re getting heart surgery, you want to go to the hospital that handles lots of heart surgeries, and has done so successfully in the past.

Other studies I looked at:

Green tea

Unfortunately, there’s no all-cause mortality data on the impact of tea in general, green tea in particular. We might expect it to have an effect through flavonoids.

As a proxy, though, we can look at blood pressure, where lower blood pressure is better. From “Green and black tea for the primary prevention of cardiovascular disease” (2013, N=821):

» Systolic blood pressure, -3.18[-5.25, -1.11] mmHg

» Diastolic blood pressure, -3.42[-4.54, -2.30] mmHg

There’s a smaller effect from black tea, around half the size.

Cochrane also looked at green tea prevention rates for different cancers. From “Green tea for the prevention of cancer” (2009, N=1.6 million), it’s unclear whether there’s any strong evidence of effect for any cancer, in addition to there being a possible garden of forking paths.

If you’re already drinking tea, like me, then switching to green tea is low cost despite any questions about efficacy.

Borderline efficacy

Baby Aspirin

The practice of taking tiny daily doses of aspirin, mainly to combat cardiovascular disease. From “Low-dose aspirin for primary prevention of cardiovascular events in Japanese patients 60 years or older with atherosclerotic risk factors: a randomized clinical trial.” (2014, N=14,464):

» Aspirin vs none, aggregate cardiovascular mortality HR = 0.94[0.77, 1.15]

That CI width is very concerning; you can cut the data so you get subsets of cardiovascular mortality to become significant, like looking at only non-fatal heart attacks, but it’s not like there’s a breath of correcting for multiple comparisons anywhere, and the study was stopped early due to “likely futility”.

The side effects of baby aspirin are also concerning. Internal bleeding is possible (Mayo clinic article), since aspirin is acting as a blood thinner; however, it isn’t too terrible, since it’s only a 0.13% increase in “serious bleeding” that resulted in hospitalization (from “Systematic Review and Meta-analysis of Adverse Events of Low-dose Aspirin and Clopidogrel in Randomized Controlled Trials” (2006)).

More concerning is the stopping effect. “Low-dose aspirin for secondary cardiovascular prevention – cardiovascular risks after its perioperative withdrawal versus bleeding risks with its continuation – review and meta-analysis” looked into cardiovascular risks when stopping a baby aspirin regime before surgery (because of increased internal bleeding risks), and found that a low single-digit percentage of heart attacks happened shortly after aspirin discontinuation. (I’m having trouble interpreting this report.)

I imagine this is why professionals start recommending baby aspirin to folks above 50yo, since the risks of heart attack start to obviously outweigh the costs of taking aspirin constantly. And speaking of cost: baby aspirin is monetarily inexpensive.

Other studies I looked at:

Meal Frequency

Some people recommend eating smaller meals more frequently, particularly to lose weight, which is tied to health outcomes.

From “Effects of meal frequency on weight loss and body composition: a meta-analysis” (2015, N=unknown):

» +1 meal/day, -0.27 ± 0.11 kg of fat mass

It’s not really an overwhelming result; taking into account the logistical overhead of planning out extra meals in a society based on 3 square meals a day, is it really worth it to lose maybe half a kilogram of fat?

Other studies I looked at:

Caloric Restriction

Most longevity folks are really on board the caloric restriction (CR) train. There’s an appealing mechanism where lower metabolic rates produce fewer free radicals to damage cellular machinery, and it’s the exact amount of effort that one might expect from a longevity intervention that actually works.

A common example of CR is the Japanese Ryukyu islands, where there are a surprising number of really old people, who eat a surprisingly low number of calories. However, say it with me: con-found-ed to he-ll! The fact that a single isolated subsection of a single ethnic group have a correlation between CR and longevity doesn’t make me confident that I too can practice CR and tell death to fuck off for a few more years.

So we want studies. Unfortunately, most humans fall into the state of starving and lacking essential nutrients, or having enough calories and nutrients, but almost never the middle ground of having too few calories but all the essential nutrients (2003, literature review). Then there’s the ethics of getting humans to agree to a really long study that controls their diet, so let’s look at animal studies first.

However, different rhesus monkey studies give different answers.

» From “Impact of caloric restriction on health and survival in rhesus monkeys from the NIA study” (2012, N=unknown, no full text), there was no longevity increase from young or old rhesus monkeys.

» However, from “Caloric restriction delays disease onset and mortality in rhesus monkeys” (2009, N=76), there was a 30% reduction in death over 20 years.

Thankfully they’re both randomized, but it doesn’t really help when they end up with conflicting conclusions. You’d hope there would be better support even in animal models for something that should have huge impacts.

What else could we look at? We’re not going to wait for an 80-year human study to finish (the ongoing CALERIE study comes close), so maybe we could look at intermediate markers that are known to have an impact on longevity and go from there.

A CALERIE checkpoint study, “A 2-Year Randomized Controlled Trial of Human Caloric Restriction: Feasibility and Effects on Predictors of Health Span and Longevity” (2015, N=218), looks at the impact of 25% CR on blood pressure:

» Mean blood pressure change, around -3 mmHg (read from a chart)

Pretty good, but that’s also around the impact of green tea. Then, there’s the implied garden of forking paths bringing in multiple comparisons, since the study in the same cluster looks at multiple types of cholesterol and insulin resistance markers.

Finally, there’s the costs: you have to exert plenty of willpower to actually accomplish CR. For something with such large costs, the evidence base just isn’t there.

Chocolate

Chocolate has some impact on blood pressure. “Effect of cocoa on blood pressure” (2017, N=1804, Cochrane) finds that eating chocolate lowers your blood pressure:

Systolic blood pressure, -1.76[-3.09, -0.43] mmHg

Diastolic blood pressure, -1.76[-2.57,-0.94] mmHg

However, if you’re normotensive then there’s no impact on blood pressure, and only taking into account hypertensives the effect jumps to -4 mmHg. Feel free to keep eating your chocolate, but don’t expect miracles.

Social Interaction

Having a social life looks like a really great intervention.

From “Social Relationships and Mortality Risk: A Meta-analytic Review” (2010, N=308,849):

» Weaker vs stronger relationships, OR = 1.50[1.42, 1.59]

And from “Social isolation, loneliness, and all-cause mortality in older men and women” (2013, N=6500):

» Highest vs other quintiles of social isolation, HR = 1.26[1.08, 1.48]

And from “Marital status and mortality in the elderly: A systematic review and meta-analysis” (2007, N>250,000, no full text):

» Married vs all currently non-married, RR = 0.88[0.85, 0.91]

You can propose a causal mechanism off the top of your head: people with more friends are less depressed which just has good health outcomes.

However, the alarm bells should be ringing: is the causal relationship backwards? Are healthier people more prone to socializing? Do the confounders never end? The kicker is that all these studies are looking at the elderly (above 50yo at least), which reduces their general applicability even more.

Other studies I looked at:

Cellphone Usage

Remember when everyone was worried that chronic cellphone usage was going to give us all cancer?

Well “Mobile Phone Use and Risk of Tumors: A Meta-Analysis” (2008, N=37,916) says it actually does:

» Overall tumor, OR = 1.18[1.04, 1.34]

» Malignant tumor, OR = 1.00[0.89, 1.13]

Since we’re worried about malignant tumors, it’s hard to say we should be worried by cellphones.

Other studies I looked at:

Unproven

Confusing thirst with hunger

Some people recommend taking a drink when you feel hungry, the idea being that thirst sometimes manifests as hunger, and you can end up eating fewer calories.

Unfortunately, I couldn’t find any studies that tried to look into this specifically: the closest thing I found was “Hunger and Thirst: Issues in measurement and prediction of eating and drinking” (2010) which reads like a freshman philosophy paper, and “Thirst-drinking, hunger-eating; tight coupling?” (2009, N=50?) which fails to persuade me about… anything, really.

Stress Reduction in a Pill

There are some “natural” plants rumored to have stress reduction effects, Rhodiola rosea and Ashwagandha root.

Meta-analysis on Rhodiola, “The effectiveness and efficacy of Rhodiola rosea L.: A systematic review of randomized clinical trials” (2011, N=unknown) found that Rhodiola had effects on something, but the study was basically a fishing expedition. Even the study name betrays that it doesn’t matter what it’s effective at, just that it’s effective.

Another meta-analysis, “Rhodiola rosea for physical and mental fatigue: a systematic review” (2012, N>176) looked specifically at fatigue and found mixed results.

Meta-analysis on Ashwagandha, “Prospective, Randomized Double-Blind, Placebo-Controlled Study of Safety and Efficacy of a High-Concentration Full-Spectrum Extract of Ashwagandha Root in Reducing Stress and Anxiety in Adults” (2012, N=64) found reductions in self-reported stress scales and cortisol levels (and with RCTs!).

Look, the Ns are tiny, and the studies the meta-analyses are based on are old, and who knows if the Russians were conducting their side of the studies right (Rhodiola originated in Russia, so many of the studies are Russian).

I’m including this because I got excited when I saw it in the original longevity post: stress reduction in a pill! Why do the hard work of meditation when I could just pop some pills (a very American approach, I know)? It just doesn’t look like the evidence base is trustworthy, and my personal experiences confirm that if there’s an effect it’s subtle (Whole Foods carries both Rhodiola and Ashwagandha, so you can try them out for yourself for like $20).

Other studies I looked at:

Water Filters

Unfortunately, there’s basically no research on health effects from water filtration in 1st world countries above and beyond municipal water treatment. Most filtration research is either about how adding any filtration to 3rd world countries has massive benefits, or how bacteria can grow on activated carbon granules. Good to know, but on reflection did we expect bacteria to stop growing wherever it damn well pleases?

So keep your Brita filter, but it’s not like we know for sure whether it’s doing anything either. Probably not worth it to go out of your way to get one.

Hand sanitizer

So I keep hand sanitizer in multiple places in my apartment, but does it do anything?

I only found “Effectiveness of a hospital-wide programme to improve compliance with hand hygiene” (2000, N=unknown), which focused on hospital health outcomes impacted by hand washing adherence. First, not all doctors wash their hands regularly (40% compliance rates in 2011) (scholarly overview), which is worrying. Second, there’s a positive trend between hand washing (including hand sanitizers) and outcomes:

» From moving 48% hand washing adherence to 66%, the hospital-wide infection rate decreased from 16.9% to 9.9%.

However, keep in mind that home and work are usually less adverse environments than a hospital; there are fewer people with compromised immune systems, there are fewer gaping wounds (hopefully). The cited result is probably an upper bound for us non-hospital folk.

(There’s also this cute study: hand sanitizer contains chemicals that make it easier for other chemicals to penetrate the skin, and freshly printed receipts have plenty of BPA on the paper. This means that sanitizing and then handling a receipt will lead to a spike of BPA in your bloodstream. I presume that relative to eating with filthy hands the BPA impact is negligible, but damn it, researchers are doing these cute small scale studies instead of the huge randomized trials I want.)

Other studies I looked at:

Doctor visits

Should you visit your doctor for a annual checkup? My conscientious side says “of course”, but my contrarian side says “of course not”.

Well, “General health checks in adults for reducing morbidity and mortality from disease” (2012, N=182,880, Cochrane) says:

» Annual checkup vs no exam, RR = 0.99[0.95, 1.03]

So basically no impact! Ha, take that, couple hour appointment!

However, The Chicago Tribune notes some mitigating factors, like the main studies the meta-analysis is based on are old, like 1960s old.

Metformin

I didn’t look at metformin in my main study period: I knew it had some interesting results, but it also caused gastrointestinal distress, better known as diarrhea. It brings to mind the old quip: metformin doesn’t make you live longer, it just feels like it[9].

However, while I was reading Tools of TitansDominic D’Agostino floated an intriguing idea: he would titrate the metformin dose from some tiny amount until he started exhibiting GI symptoms, and then dialed it back a touch. I don’t think people have started even doing small scale studies around this, but it might be worth looking into.

Other

There’s some stuff that doesn’t have a cost-benefit calculation attached, but I’m including anyways. Or, there are things that won’t help you, but might help the people around you.

CPR

From “Effectiveness of Bystander-Initiated Cardiac-Only Resuscitation for Patients With Out-of-Hospital Cardiac Arrest” (2007, N=4902 heart attacks):

» Cardiac-only CPR vs no CPR, OR 1.72[1.01, 2.95]

So the odds ratio looks pretty good, except that CI is really wide, and the in absolute terms most people still die from heart attacks: administering CPR raises the chances of survival from 2.5% to 4.3%. So, spending more than a few hours practicing CPR is chasing some really tail risks[10].

However, have two people in your friend group that know CPR, and they can provide a potential buff to everyone around them (two, because you can’t give CPR to yourself). In a similar vein, the Heimlich maneuver might be good to know.

Other studies I looked at:

Smoke Alarm testing

Death by fire is not super common. That said, these days it’s cheap to set up a reminder to check your alarm on some long interval, like 6 months.

Quikclot

It’s unlikely you’ll need to do trauma medicine in the field, but if you’re paranoid about tail risk then quikclot (and competitors) can serve as a buttress against bleeding out. Some folks claim that tourniquets are better, but the trauma bandages are a bit more versatile, since you can’t tourniquet your chest.

It’s not magical: since the entire thing becomes a clot, it’s basically just moving a life threatening wound from the field into a hospital. Also make sure to get the bandage form, not the powder; some people have been blinded when the wind blew the clot precursor into their eyes.

Cryonics

Of course, this post wouldn’t be complete without a nod to cryonics. It’s the ultimate backstop. If there all else fails, there’s one last option to make a Hail Mary throw into the future.

Obviously there are no empirical RR values I can give you: you’ll have to estimate your own probabilities and weigh your own values.

WTF, Science?

The overarching story is that we cannot trust anything, because almost all the studies are observational and everything could be confounded to hell by things outside the short list that every study incants they controlled for and we would have no idea.

Like Gwern says, even the easiest things to randomize, like giving people free beer, aren’t being done, much less on a scale that could give us some real confidence.

There is too little disregard for the garden of forking paths in this post-replication crisis world, and many studies are focused on subgroups that plausibly won’t generalize (ex. the elderly).

And what’s up with the heterogeneity in meta-analyses? If every single analysis results in “these results displayed significant heterogeneity”, then what’s the point? What are we doing wrong?

What am I doing?

Maybe you want to know what me myself am doing; I suspect people would be interested for the same reason journalists intersperse a perfectly good technical thriller with human interest vignettes, so here:

  • Continuing vitamin D supplementation, and getting a couple minutes of sun when I can.
  • Making an effort to eat more vegetables, less bacon/potatoes (to be honest, I’m more optimistic about cutting out the bacon than potatoes), more fish, and replacing more of my snacking with walnuts.
  • Keep taking fish oil.
  • Exercise better: I haven’t upped the intensity of my routine in a while. I probably need some more aerobic work, too.
  • Tell myself I should iron out my sleep schedule.
  • Get myself a standing desk for home: I have a standing desk at work, so I’m already halfway there.
  • Buy an air filter: low impact, but whatever, gimmie my percentage points of RR.
  • Switch from drinking black tea to green tea.
  • Cut back on donating blood. I’ll keep doing it because it’s also wrapped up in “doing good things”, but I was doing it partly selfishly based on the non-quasi-randomized studies. Besides, I have shitty blood.

TLDR

Effective and certain:

  • Supplement vitamin D.

Effective, possibly confounded:

  • Exercise vigorously 5 hours/week.
  • Eat more fruits and vegetables, more fish, less red meat, cut out the bacon.
  • Get 7-9 hours of sleep.

Less effective, less certain:

  • Brush your teeth and floss daily.
  • Try to not sit all day.
  • Regarding air quality, don’t live in Beijing.

There is also a presentation.


[1]  If you need me to go through the science of smoking, then let me know and I can do so: I mostly skipped it because I’m already not smoking, and the direction of my study was partly determined by what could be applicable to me. As a non-smoker, I didn’t even notice it was missing until a late editing pass.

[2]  The abstract reports results in terms of percentage mortality decrease, which I believe maps to the same RR I gave.

[3]  If I remember correctly, Due Diligence talks about this.

[4]  The Cochrane Group does good, rigorous analysis work. Gwern is an independent researcher in my in group, and he seems to be better at this sort of thing than I am.

[5]  Annoyingly, some meta-analyses don’t report the aggregate sample sizes for analyses that only use a subset of the analyzed reports.

[6]  For example, Scott’s review of The Hungry Brain points out that some people think potatoes are great at satiating appetites, so it might in fact work out in favor of being okay.

[7]  These category comparisons are loose, since some studies will report quartiles and others will use tertiles, so the analysis simply goes with the largest effect possible across all studies.

[8]  Yes, it’s fucking stupid I have to stoop to this.

[9]  Originally “marriage doesn’t make you live longer, it just feels like it.”

[10]  I know, it’s ironic that I’m calling this a tail risk, when we’re pushing something as stubborn as the Gompertz curve.

Filed under: Uncategorized

Leave a Reply

All content is licensed under CC-by-nc-sa
Creative Commons License