Does anybody want to calculate their LinAge2 (better than Levine), using blood test biomarkers?

#1

Hi everybody,

https://www.nature.com/articles/s41514-025-00221-4

This paper basically describes a better version of the Levine Phenotypic Age. This one uses 60 parameters and can also provide some useful insights.

For example, based on your results you will get a biological age and it tells you which of the subgroups in the dataset that you are most like, and what their causes of death were.

It’s more complex than Levine, and cannot run in a simple Excel. However, I’ve made a simple template and if any of you want to fill it in and send to me, I can run the R. script and share the results with you.

In the Excel template, the parameters are to be filled in each column, from B to CI. There are three example datasets provided. Two are patient datasets from subjects with a real (chronological) age of 72. The first patient (8881) had diabetes, had a LinAge2 of 88.4 (i.e. +16 years) and he actually died at 77. The second patient (9106) had a LinAge2 of 64.4 (-7.6 years) and he actually died at 91. (The calculator shows that patient 1 had strong likelihood of dying from cardiometabolic syndrome, which indeed came true.)

If you want to fill the template, simply fill in the column highlighted in yellow. Be careful with units, because the calculator expects a lot of mmol/L. I’ve given you conversion factors in the sheet. The third example dataset is the NHANES dataset median. I put it there so that you should know whether your own test results are approximately correct. For the first few columns you need to use 1 or 2 to designate yes/no or male/female. Again, instructions are in the sheet.

userData for Rapa.news.xlsx (14.8 KB)

I ran my own, and it gave me a biological age of -15.7 years, which is similar to Levine at -17. However, my wife scores -14 years on Levine but “only” scored -7 years on LinAge2. Her biomarkers are almost perfect on paper (hsCRP 0.011, LDL-C of 70, fasting glucose of 75, HBA1C of 4.5% etc), but LinAge2 strongly associated her results with immunity and vascular inflammation-related issues, and she does indeed have some autoimmune disease along those lines. So there is some value IMO of having the 60 parameters and the relationships between those numbers are important, not just the actual raw number.

If anybody wants a go, simply fill the template and post here or PM me if you want to keep it private.

3 Likes
#2

Related to this thread: LinAge2: providing actionable insights and benchmarking with epigenetic clocks

#3

Yes, but I think my offer to run the calculator for people may have been buried after the discussion about running R, python etc :slight_smile:

3 Likes
#4

Converting the R code and spreadsheets to a more user friendly format feels like an easy task for GPT-5.1-Codex-Max set to high reasoning, if you haven’t tried it yet. It’s included in plus subscriptions.

1 Like
#5

I was missing maybe a half dozen of the answers. Hard test.

2 Likes
#6

Thank you for your generosity and I’m definitely going to take you up on your offer… stay tuned for my data dump :slight_smile:

#7

Which lab is this sensitive?

I was assuming mg/l, but having looked at your spreadsheet I see it is stil mg/dl

However, even 0.11mg/l is more sensitive than any lab I am aware of in the UK.

2 Likes
#8

@relaxedmeatball I finished!! Well, sort of!

I filled out the spreadsheet with the units of measurement I have in my labs.

Before I take the time to covert them (and hurt my brain), I’m making sure it’s not a waste of time because I don’t have all of the results needed.

Specifically, I don’t have: Natriuretic, bicarbonate, or lactate levels. I also don’t have urine results for albumin or creatinine, but I have the serum results.

Do you think it’s worth just inserting an average result for those? Or do I just not bother.

@relaxedmeatball never mind, I just saw your notes at the top of the screen on what to do if we don’t have all the results. They didn’t show up on my ipad until I turned it.

#9

Try running them through an AI system… perhaps easier and faster.

1 Like
#10

@RapAdmin I can’t believe how long this took me, and that is even with your brilliant tip to use AI!

@relaxedmeatball, I’ll link it here. This should still be in excel for you, but if it’s not, I’ll have excel tech support (aka husband) convert it for me
https://docs.google.com/spreadsheets/d/1ieHM7Ezk0T51RUe1-h-KVGgLUkb_RL84/edit?usp=drivesdk&ouid=105774201770561638601&rtpof=true&sd=true

1 Like
#11

Very difficult, I think. There’s a whole bunch of log transformations and other things to do. Plus, I don’t see how you could replicate all the other functions like generating principal component plots.

If they are relatively unimportant ones you could use the NHANES median.

Good question. I’m not in the UK any more, and my local lab reported back the 0.011mg/dl this time. In previous times they have simply stated “<0.3 mg/dl”. Sorry I’m not sure about the assay.

For me, ChatGPT did a very good job of converting units. I double checked them, and it made 2 errors out of around 20 parameter conversions. Luckily, they were quite obvious because if the calculator is expecting something in the range of 8,000 to 12,000 and your number is 27, something is wrong!

Hey @Beth , the sheet is not public. Can you please change the permission settings so that I can access it?

1 Like