Can an Algorithm Predict Your Biological Age?
“We must make haste then, not only because we are daily nearer to death, but also because the conception of things and the understanding of them cease first.”
― Marcus Aurelius, Meditations
If someone gave you a pill to live forever — would you take it?
Perhaps not. Many people scoff at the concept of physical (or even metaphysical) immorality; citing either biological implausibility or a fear of “growing bored with living” as reasons they’d like to someday shed their mortal coil.
Regardless of one’s stance on eternal existence, most people, if given the option, would at least take the option of extending their years of healthy living. Increasing the “years of life” and the “life in those years” is a concept that almost everyone can get behind.
Longevity is a hot research topic lately. Scientists are calorie-restricting mice, monkeys, and humans to see if we can eke out a few more trips around the sun. Drugs like Metformin and Rapamycin are being looked at for their lifespan-enhancing effects — and some show promise. Notable researchers like David Sinclair are conducting cutting-edge studies in the lab, mostly on rodents, to determine why we age — and how we can stop it.
Unfortunately there is no “single cause” of aging. Rather, aging is the result of an accumulation of biological changes in our body — DNA damage, “death” of cells, a wearing out of arteries and bones, and degeneration of muscles.
Our internal biological environment plays a part in a lot of these changes. Blood glucose, insulin, cholesterol, antioxidant systems, inflammation, blood pressure, etc. — all of these so called “biomarkers” change with age and in response to our lifestyle habits including exercise, diet, and exposure to toxins, pollution, and radiation.
Having high levels of some biomarkers is bad, and high markers of other, good. For instance, the “rate” at which you age might be influenced by how much insulin you are exposed to throughout life — with more insulin exposure associated with a faster rate of aging and an earlier death (in animals, at least).
We can easily access our own levels of biomarkers using a simple blood test. Since we have loads of data on what “reference” ranges are associated with positive or negative health outcomes, a blood test can at least give us a “snapshot” of overall health, at least in comparison to the dataset where the reference ranges come from — usually the U.S. (if you live there) or another population-wide database.
But what if these numbers could tell us more than just our “health status” — but our overall life status? How near are you do death, based on your most recent blood test? Is this a question you’d want answered?
What would happen if we could utilize these biomarkers to help us guide life decisions in order to bring ourselves “less near” to death?
The concept that you might be younger or older than your “chronological” age (the actual time, in years, that you’ve been alive) is known as your “biological age.”
Your “biological age” reflects your physical attributes rather than simply how long you’ve been on earth. Since the process of aging occurs at a different rate in everyone, biological age can give a more accurate picture of “aging” than the number of birthday candles on your cake.
Take two individuals who both have a chronological age of 50.
The first individual is a life-long athlete, consumes little to no alcohol, and lives “off the grid” without much exposure to pollution or city life.
Our second individual is a heavy drinker, lives in a busy downtown area, and rarely leaves the couch.
Case #1 might be 50, chronologically speaking, but biologically much younger. — 42 or 43. On the other hand, individual #2 might have a biological age around 65 or 70 — much higher than is chronological age (50).
But how can biological age be determined?
A few options exist, all have their flaws. One way is to measure something called telomeres — these are the “caps” on the end of DNA that grow shorter each time a cell replicates. In addition, telomeres can shorten due to DNA damage. Longer telomeres, on average, are associated with increased longevity, and vice-versa (though there are some exceptions).
Several companies offer laboratory and at-home tests of biological age using saliva or blood samples; measuring epigenetic modifications in the blood or levels of other proteins. At present, no “gold standard” really exists.
Perhaps a more interesting possibility involves taking a handful (or a few handfuls) of biological markers (biomarkers) that are known to be associated with health and using them to “predict” one’s biological age. This would have the advantage of taking a variety of biochemical factors vs. a single measurement to determine biological age.
An algorithm such as this has been developed and tested in a recent paper under peer-review titled: “An interpretable machine learning model of biological age.”
Researchers developed a machine-learning algorithm based on the premise that easily accessible biomarkers could potentially “assist in the rapid assessment of promising interventions aimed at increasing longevity.”
Basically: if an algorithm can be used to accurately predict one’s biological age using an individual's biomarkers, then we can find ways to reduce biological age through certain interventions that increase or decrease these biomarkers. It’s “simple” math.
In addition to the algorithm’s ability to spit out biological age, it can also “provide individual weighting for how each biomarker affected the final output (age).” This could answer the valuable question of what are the strongest contributors to biological age.
How does the algorithm work?
Using a database of 46,739 participants and a total of 39 biomarkers, the predictive model was “trained” on the dataset. Then, the “predicted age” of each participant was compared to their actual chronological age — assessing the “accuracy” of this model.
The biomarkers used in this dataset included things like red blood cells, cholesterol, iron, blood glucose, sodium, calcium, and uric acid; things you can have measured using a routine, comprehensive blood test and metabolic panel.
The accuracy of the algorithm proved to be quite strong — biological age correlated significantly with chronological age.
For each biomarker, the algorithm predicted a range (i.e. level in the body) where it was associated with the lowest predicted age. This would be assumed the “most optimal” level, in theory (in reality…physiology is much more complex. For instance, the authors discuss how while high cholesterol is associated with poor health outcomes…so is cholesterol that’s too low).
In addition, we are given an “inflection point” for each biomarker — the level above or below which a change in any biomarker causes an increase in biological age. For instance, potassium has an inflection point of 4.1 millimoles per liter. Blood potassium levels above 4.1 would cause a net increase in one’s biological age.
Which markers had the greatest influence on biological age?
For females: blood urea nitrogen (BUN) topped the list. Urea is a waste product produced by the liver during the digestion of protein.
For men, albumin was the most influential biomarker. Albumin is a protein produced by our liver that’s needed to transport hormones, proteins, and drugs/ligands throughout the body.
For both men and women, fasting plasma glucose (blood sugar) was the second most influential biomarker.
What proves to be most interesting about this algorithm is that it’s fully personalized to a given individual. The study actually took laboratory data from one of the lead authors (C.K.), input his data, and observed how each individual biomarker contributed uniquely and with varying magnitudes to his biological age. In this individual, the following markers ADDED to his age: BUN contributed an addition 3.5 years to his biological age, cholesterol 2.8 years, potassium 1.7, phosphorus 1.2, and lactate dehydrogenase (LDH) 0.9.
On the other hand, these factors subtracted to his biological age: lymphocytes took off 1.2 years, red blood cells 2.3, albumin 2.7, fasting glucose 3.1, and triglycerides 3.9.
Taken together, his predicted biological age was 43 years — while his actual age was also 42.
This is all interesting…but is it useful? Definitely, at least in theory.
One can imagine a scenario where this algorithm can weigh which biomarkers are most heavily influencing your biological age, and in what direction. Taking this data, one could then implement lifestyle and/or pharmacological interventions targeted at increasing or decreasing these influential biomarkers. If reducing your fasting blood glucose from 95 to 85 could cause a net reduction in biological age by 4 years, then perhaps a low-carb diet might be beneficial. The possibilities are endless.
Furthermore, this algorithm could provide feedback on whether something you’re doing is actually working in the direction you’d like it to. We rarely get quantitative feedback on our dietary or supplementation strategies, and a fancy, personalized algorithm such as the one developed here might be a biohacker’s dream.
The ability to personalize this equation is perhaps the most alluring. The age of personalized medicine is upon us, and we’re realizing that cookie-cutter approaches to exercise, nutrition, and medicine are no longer sufficient. What works on a population level, in theory, rarely plays out perfectly on an individual basis.
As we have increasing access to our own biodata, there will be an elevated desire to put this data to use. While it will surely take place on a large scale, with countries and nations using population data to design policy and do research, self-interested people will also want to use their own data to make intuitive decisions about their life choices.
Your personal biochemistry is important — perhaps the most important thing when speaking of health and longevity. Fortunately, we no longer have to let our biology control us, but rather, we can take control of our biology. Along with a bit of complex mathematical modeling, we may perhaps someday hack ourselves a few more years on this planet.
Study cited
Wood T, Kelly C, Roberts M and Walsh B. An interpretable machine learning model of biological age [version 1]. F1000Research 2019, 8:17 (doi: 10.12688/f1000research.17555.1)
