Evolution of the PhenoAge Algorithm

Imagine if you could know your biological age — the age your body seems to be — rather than just your birthday. That’s exactly what the PhenoAge algorithm tries to do. Over time, scientists have improved this algorithm. And today we're going to walk through its evolution. We’ll break it down step by step and even look at the math behind it!

A Quick Timeline

1. PhenoAge1 (April 18, 2018): The original method using clinical biomarkers from NHANES III data.
2. Adjusted PhenoAge1 (November 1, 2018): A tweak to convert the clinical age into a DNA methylation (DNAm) based age estimate.
3. Uncorrected PhenoAge2 (December 31, 2018): A new version using NHANES IV data, but with a small mistake.
4. Corrected PhenoAge2 (February 25, 2019): The mistake is fixed by adding a missing conversion step.
5. Adjusted Corrected PhenoAge2: Finally, the corrected version is adjusted the same way as PhenoAge1 to yield a refined algorithm.

Step 1: Calculating Clinical PhenoAge

Both PhenoAge1 and PhenoAge2 start by using clinical tests (blood tests) that measure nine different things in your blood (like albumin, creatinine, glucose, and others) along with your actual age. They combine these values into one number using a formula. We call this combination "xb." The formula looks like this:

xb = –19.907 – 0.0336×(albumin) + 0.0095×(creatinine) + 0.0195×(glucose) + 0.0954×ln(C‑reactive protein) – 0.0120×(lymphocyte percent) + 0.0268×(mean cell volume) + 0.3356×(red cell distribution width) + 0.00188×(alkaline phosphatase) + 0.0554×(white blood cell count) + 0.0804×(chronological age)

Next, using a model that describes how likely a person is to die in 10 years (called the Gompertz model), the algorithm converts xb into a 10‑year mortality risk, "M." The formula for M is:

M = 1 – exp{ – exp(xb) × [exp(120×γ) – 1] / γ }

Here, γ (gamma) is a small number (approximately 0.0076927), and 120 represents 120 months, or 10 years.

Once we have M, the algorithm converts this risk into a clinical Phenotypic Age, "P." For PhenoAge1, the math is:

P = 141.50225 + [ ln( –0.00553 × ln(1 – M) ) ] / 0.090165

For the PhenoAge2 method, the process is nearly the same, but in its uncorrected version (released on December 31, 2018) they skipped the step of converting xb into M explicitly. In the corrected PhenoAge2 (released on February 25, 2019), they fixed that by using the same conversion as above — except they use very slightly adjusted constants:

P = 141.50 + [ ln( –0.00553 × ln(1 – M) ) ] / 0.09165

Notice the tiny differences. PhenoAge1 uses 141.50225 and 0.090165, while corrected PhenoAge2 uses 141.50 and 0.09165.

Step 2: Adjusting the Clinical PhenoAge

Scientists noticed that the clinical PhenoAge (whether from PhenoAge1 or PhenoAge2) could be fine-tuned further to account for sparse representation of older people in the original data. They developed an adjustment formula. This adjustment can be applied to the clinical PhenoAge (P) to give the final DNAm PhenoAge, which we can call "D(P)." The formula is:

D(P) = P / [1 + 1.28047 × exp(0.0344329 × (P – 182.344))]

This equation tweaks the clinical PhenoAge so it more accurately reflects biological aging for more people.

Comparing the Two Sequences

Now let’s compare the two combined sequences:

• Adjusted PhenoAge1
  • Step 1 (Clinical PhenoAge1): Compute xb using the nine biomarkers and age, then calculate M with the Gompertz model, and finally compute P = 141.50225 + [ ln( –0.00553 × ln(1 – M) ) ] / 0.090165
  • Step 2 (Adjustment): Apply D(P) = P / [1 + 1.28047 × exp(0.0344329 × (P – 182.344))]
• Adjusted Corrected PhenoAge2
  • Step 1 (Clinical Corrected PhenoAge2): Again, compute xb with the same formula, then get M using M = 1 – exp{ – exp(xb) × [exp(120×0.0076927) – 1] / 0.0076927 }, then convert M into age with P = 141.50 + [ ln( –0.00553 × ln(1 – M) ) ] / 0.09165
  • Step 2 (Adjustment): Apply the identical adjustment formula, D(P) = P / [1 + 1.28047 × exp(0.0344329 × (P – 182.344))]

The main differences are in the constants used during the conversion from mortality risk to age. PhenoAge1’s numbers are 141.50225 and 0.090165, whereas corrected PhenoAge2 uses 141.50 and 0.09165. Also, the corrected PhenoAge2 properly includes the explicit step of calculating M from xb, which was missing in the uncorrected version.

Even though the differences in numbers may seem tiny, they can change the final biological age estimate. Scientists work hard to refine these models so that they can more accurately predict how well a person is aging — information that is becoming increasingly useful for health and medicine.

Conclusion

To sum up, evolution of the PhenoAge algorithm went like this:

1. PhenoAge1 (clinical age calculation from NHANES III) was created first.
2. Then an adjustment was added to better match DNA methylation data, leading to adjusted PhenoAge1.
3. Next came uncorrected PhenoAge2 (using NHANES IV data), which originally skipped a step (the uncorrected version).
4. This was later corrected by explicitly calculating the 10‑year mortality risk, producing corrected PhenoAge2.
5. Finally, the same adjustment was applied to get the adjusted corrected PhenoAge2 — the refined algorithm.

Each step in the timeline has brought scientists closer to an accurate and useful biological clock that can measure aging at the molecular level. Hopefully, this simple breakdown helps you understand how math and science work together to refine tools that can help us all live healthier lives!

Thrivous develops Thrivous Clock and a free calculator to measure biological aging, help people make better lifestyle decisions, and enhance healthy longevity. When first launched, Thrivous Clock was based on the adjusted PhenoAge1 algorithm. We have since updated it to use the adjusted corrected PhenoAge2 algorithm. Thrivous Clock is available to purchase online now in the Thrivous store.