One of my favorite questions for scientists is: say you were the benevolent dictator of all science...what experiment would you run? The idea is to explore potentially disruptive projects that, for whatever reason, have not been pursued by the current research ecosystem.

Here’s my answer. I’d collect a large cohort of healthy people and use Theranos to measure blood-based “lifestyle” biomarkers longitudinally. The goal would be to better understand natural variation in blood biomarkers, so we can establish a quantitative basis of how our bodies progress over time. This would lead to better tools for health assessment and targets for wellness interventions. Read on for the details.

Blood biomarkers as a lever for wellness

Blood tests are a core tool of modern medicine. Blood can provide indicators of disease state, both near-term (ie. blood glucose) and longer term (ie. hemoglobin A1c).

However, my hypothesis is that blood biomarkers have been underexplored as a tool for wellness. Consider the following:

  • When one starts a diet or exercise plan, the most common target is weight loss: I want to lose X pounds. Yet we know that weight is an extremely poor proxy for overall health and fitness.

  • When your doctor runs blood tests at a physical exam, her only normative feedback is if a test falls outside a usually very permissive range: your cholesterol levels are too high. You're conditioned to ignore any difference between the 40th and 60th percentiles; both are in the healthy or green range.

  • Many blood biomarkers are known to track with exercise, diet, and age. DHEA is one example - there is evidence that DHEA levels increase the more you exercise, the healthier you eat, and the younger you are.

Taken together, it’s conceivable that one could instead start a fitness program with I want to increase my DHEA by 50 µg/dL. To my knowledge, we don't know the answer right now.

What data to collect

How could we figure that out whether that is possible? The experiment I’m imagining would do the following: 

  • Track healthy individuals ages ~20-50, collecting ~100 blood biomarkers every three months.

  • Collect basic measurements of weight and body fat at time of testing

  • Choose biomarkers that are known to correlate with fitness and/or age, and span a large range of biological function. Examples are DHEA, HbA1C, Free Testosterone, etc.

  • Target an effective cohort size of ~10,000 individuals tracked for 10 years (though similar statistical power could be achieved in a number of ways.)

This experiment is made feasible by Theranos, an emerging technology that could potentially run all 100 tests with a single drop of blood, and with very high precision.

Using the data

Say we collect those data as a public resource - what could we learn? A number of things:

Understanding biomarkers

The primary goal is to establish optimal targets for each of the biomarkers - the value that your ‘perfect’ body would expect. This is a very hard problem and one that will require extreme care - obviously in practice there is no ‘perfect’ body. However, we can take clues from other correlations. If a marker consistently increases with lean body mass and decreases with age, you probably want to see it increased yourself, regardless of your age.

Note that this does not imply anything about direct effects - we are looking for indicators of health, not causal inference. DHEA supplements almost certainly don’t work.

Advanced wellness metrics

No single biomarker will be a perfect proxy for health. To this end, statistics could be derived from sets of biomarkers that aim to estimate fitness and/or wellness on one or a small number of dimensions.

An example is expected age. It's possible that a single test panel contains enough information for one to reliably say I'm 42 and I have the body of a 39 year old.

Tools for longevity research

A key component of this experiment is that it measures changes in biomarkers as one ages, not just natural variation in the healthy population. Studying these changes could lead to a more nuanced understanding of aging. There is much excitement about telomeres lately - I suspect that many of the claims about telomere length and epigenetic signatures as proxies for aging also apply to some blood biomarkers. Rather than simply testing whether an intervention increases telomere length, we could test how it impacts these blood biomarkers and derived statistics.


Nutrition science is science's biggest fail. If blood can provide a more precise measure of one’s recent health, it could provide better ways to test and measure dietary interventions. Imagine A/B testing two similar diets (low glycemic v. raw foods) and measuring effectiveness by testing subtle changes in blood tests.

Short term wellness monitoring

I'll leave my personal favorite application for last. Have you ever gone through a particularly unhealthy period (perhaps last holiday party season?) and felt like you're not as fit as before?

Confirmation would be motivating. You could step on the scale - but again, that's a poor indicator, particularly if you have high muscle mass. This type of experiment could lead to another option: you could go to Wallgreens, get your 100 blood tests, and have a precise measure of exactly how much you've fallen off - and what you need to do to make up for it.

Making it happen

Encouragingly, there has been progress in this space in recent years. Most notably, the 100K Wellness Project is collecting similar data, though in a systems biology context. This is also closely related to much of the life sciences work at Google X, which is perhaps uniquely qualified to execute this type of experiemnt. There are also academic efforts to study natural variation in blood biomarkers, particularly in cancer.

However, there are reasons why this experiment hasn't been performed yet - a large longitudinal study with touch points every three months is extremely hard. I'll end with some thoughts on why this is a different class of experiment than most research studies:

  • Study participants will likely need to be seen in person every three months. This makes me think that workplaces, rather than clinical or research sites, are the best venues for these tests.

  • Success will hinge on minimizing participant dropoff, so a critical component of any study design will be the interface (app or website) that participants use to interact with the study.