3 replies to this topic

[Michael Lustgarten]

Attached Files

•  Screen Shot 2021-07-26 at 6.16.23 PM.png   542.71KB   0 downloads

[albedo]

Thank you Michael for all the pain you take to post this information. Great work!

 

I will retrieve a pretty complete HVR analysis I have got several years ago to compare with your data. I wonder if you also tracked your HVR in function of daily status (e.g. sleep, stress, working, exercise, etc...). I will let you know...

 

I am always very pleased reading your work using both Levine's Phenotypic Age and Insilico's Aging.Ai as for the moment those are the BA scores I use. I know there also other blood based similar papers on biological age not using the chronological age.

 

(BTW, I am happy I was the one pointing to the Levine's team the correction in their formula in the paper you cite (Liu et al. 2018, NHANES IV) as shown then into https://doi.org/10.1...al.pmed.1002718)

[albedo]

See also this from Michael on HRV and related:

[Michael Lustgarten]

Yes, I remember, and thanks for that, albedo!

 

Thank you Michael for all the pain you take to post this information. Great work!

 

I will retrieve a pretty complete HVR analysis I have got several years ago to compare with your data. I wonder if you also tracked your HVR in function of daily status (e.g. sleep, stress, working, exercise, etc...). I will let you know...

 

I am always very pleased reading your work using both Levine's Phenotypic Age and Insilico's Aging.Ai as for the moment those are the BA scores I use. I know there also other blood based similar papers on biological age not using the chronological age.

 

(BTW, I am happy I was the one pointing to the Levine's team the correction in their formula in the paper you cite (Liu et al. 2018, NHANES IV) as shown then into https://doi.org/10.1...al.pmed.1002718)

 

[albedo]

Yes, I remember, and thanks for that, albedo!

 

Old thread, sorry Michael.

 

As (with Levine's) we both also looked at AI (e.g. aging.ai), you might like this contribution:

 

"...We asked whether AI-predicted biochemical age predicts death. On 45,033 NHANES adults with
5,213 deaths, the answer is: barely, and much less than the same data can deliver. Age acceleration
from chronological-age models adds at most +0.0124 to the C-index over chronological age and sex,
and the more accurately a model predicts age the smaller that gain becomes — the correlation
between R2 and C-index gain across model families is −0.94. The same analytes fitted against
mortality add +0.0367 and keep 85% of that after age acceleration is accounted for. The bottleneck
is the training target. A model that hits chronological age accurately is, for the purpose of predicting
death, systematically the wrong model, and the metric conventionally used to judge such models —
R2 against chronological age — is largely a description of the sample rather than of the model...." (in the conclusions)

 

Feng J. Does AI-predicted biochemical age predict death? A cross-validated re-examination in NHANES 1999–2018 with linked mortality follow-up. Preprint posted online August 6, 2026. doi:10.21203/rs.3.rs-10599297/v1

 

https://www.research.../rs-10599297/v1