JGC 

Wonderful information, thank you. I am traveling but will get back asap on this. I recollect CRP was given as mg/dL and log in Levine's paper. I have it in mg/L so how do I need to input it in the calculator? log10, log2 or?

albedo 

Thanks for asking this question, which caused me to realize I had made an error.  The entry in Table 1 and Table S1 of the paper for C-reactive protein wants the value on mg/dL and has a parenthetic log in the description line.   However, I had just put in the value in the value in mg/L as given on my blood analysis.   When I convert to mg/dL and take the natural logarithm, my 10 year mortality probability drops to 0.891 and my wife's to 0.527.  My DNAm PhenoAge goes to 92.7 and hers to 80.6.  Those values are are still higher than I would like, but more in the ballpark.  (I think "log" means natural log, as used elsewhere in the paper, but that's a guess.)  I'll replace the Mathematica notebook in DropBox with the corrected one.

albedo 

OK I made an Excel spreadsheet that does the DNAm PhenoAge calculation from blood test inputs.  The link to the spreadsheet is HERE.  On the blue INPUT line, you simply replace my data from blood tests with your own, and on the red output line the spreadsheet will calculate your probability of mortality in the next 10 years and your DNAm PhenoAge.

In my previous posts, I had the Albumin variable in the wrong units (gm/dL instead of gm/L) and the calculation was giving suspiciously large numbers.  Now is says that my Mortatlity Score is 0.497 and my wife's is 0.207, both reasonable.  My DNAm PhenoAge is 79.7 years and my wife's is 67.7 years, both lower than our calendar ages.

JGC 

Thank you ! Wonderful piece of work and so telling about you that you have openly shared it! Independently and a bit in the hurry, I went through the same formidable exp's and ln's effort directly in Excel (but in a rather awful way when comparing to you elegant spreadsheet!) and can confirm all your calculations. It was good also my lab, in EU, gave the biomarkers values directly in SI units ;-) which made the task of comparing with your work easier.

If I may, I suggest a coupe of minor improvements, maybe the last subject to further discussion:

spelling: albumin

spelling: creatinine (vs. creatine which is a different thing)

IMHO, what we calculated here is what they call "phenotypic age" in their paper (Step 1). I would rather reserve the term of PhenoAge to what they perform in Step 2 when regressing the cohort DNA methylation data on the phenotypic age. I need to better understand this step though and wonder if you have additional insight. OTOS, you use PhenoAge vs DNAn PhenoAge so it looks right. You might also refer to this presentation by Dr Morgan Levine:

http://gero.usc.edu/CBPH/files/4_25_2018_PAA/1.Levine.pdf

JGC This is fantastic: thank you so much for doing this and sharing!

albedo 

Thanks for the corrections and suggestions, which I implemented.  I replaced the old spreadsheet with the updated one, which is HERE.

In the "Terms" row of the spreadsheet, in which one can compare relative contributions to the linear combination variable, I'm still bothered by the fact that the dominant terms in the calculation are Age, Red Cell Distribution Width, Mean Cell Volume, and Glucose, while the C-reactive Protein (indicating inflammation) is in next-to-last place.  It seems counter-intuitive to me that inflammation would have such a small (and logarithmic) effect on predicted mortality and that two essentially "geometric" blood factors would be so important.  I sent the corresponding author of the Levine paper an e-mail asking about this, but he has not answered.

albedo 

On the Phenotypic Age graph from the Levin talk that you show, it is interesting that in the 80+ region (where I am)  the data leaves the red line and flattens as the Phenotype Age begins to over-predict the DNAm PhenoAge.  It appears that  there needs to be a fit with a polynomial instead of a straight line,  leading to a correction.  That would give a better prediction of  DNAm PhenoAge in the 80+ region.

JGC 

I did also notice this and intrigued by the weight of CRP inflammation, e.g. when compared to RDW. I do not have an easy answer lacking the required expertise. However I try to give some food for thought.

First, to me this is another proof point you need to integrate a larger fraction of biomarkers of aging. One critic a bit along these lines was done to a previous paper by Levine's by Mitnitski and Rockwood on frailty indexes in particular.

I feel more progress will happen when a system biology approach will be taken to the biological age determination, integrating several approaches such as the epigenetic (as DNA methylation), clinically biomarkers (as for the Phenotypic Age), anthropometric (as in Mitnitski/Rockwood) and AI/Machine Learning (as in Aging.Ai). IMHO this is also to be expected as, similarly, a system biology and multi-level approach to aging is important for a better understanding of what aging really is (which BTW does not mean to seat and wait for doing something against it as I wrote in a previous post in this Forum ;)). Levine replies in this sense to the Mitnitski and Rockwood critic to a previous paper of her:

"... I agree with Mitnitski and Rockwood that a systems biology approach is important, and in moving forward, algorithms need to incorporate interactions between various systems/levels, which may rely on more advanced computational techniques—such as machine learning...."

Levine ME. Response to Dr. Mitnitski's and Dr. Rockwood's letter to the editor: Biological age revisited. J Gerontol A Biol Sci Med Sci. 2014;69(3):297-8.

https://academic.oup.com/biomedgerontology/article/69A/3/297/707010

Second, it is not a matter of one biomarker – one disease but it is rather a matter of, again, integration and network of biomarkers.

For example CRP is a marker of chronic inflammation which might turn to increase mortality risk maybe later in the course of life while CVD or other might take their toll earlier. And sure enough, RDW, which turned out to have the highest weight in the phenotypic age determination, has an higher impact for determining heart coronary disease and mortality. E.g. see:

Pilling LC, Atkins JL, Kuchel GA, Ferrucci L, Melzer D. Red cell distribution width and common disease onsets in 240,477 healthy volunteers followed for up to 9 years. PLoS One. 2018;13(9):e0203504. Published 2018 Sep 13. doi:10.1371/journal.pone.0203504

 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6136726/

Perlstein TS, Weuve J, Pfeffer MA, Beckman JA. Red blood cell distribution width and mortality risk in a community-based prospective cohort. Arch Intern Med. 2009;169(6):588-94.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3387573/

Third, one biomarkers might measure different things and I think that is always the challenge of determining its clinical relevance.  E.g. I recently discovered, for different reasons, that CA 19-9, a marker typically used in pancreatic cancer, is very useful to assess the therapeutic effectiveness of lung infection to NTM (non-tuberculosis mycobacterium) possibly mediated by the bronchial inflammation level. So maybe, to ease your concern, it is quite possible RDW is very much related to inflammation and is maybe “telling” similar things and while you pointed (righty) to CRP for inflammation, you might be better off with RDW for the same reason, e.g. see:

Lippi G, Targher G, Montagnana M, Salvagno GL, Zoppini G, Guidi GC. Relation between red blood cell distribution width and inflammatory biomarkers in a large cohort of unselected outpatients. Arch Pathol Lab Med. 2009;133(4):628-32.

https://www.ncbi.nlm.nih.gov/pubmed/19391664

JGC 

I also noticed that flattening, but it is real?

Yes indeed another fit might be better but I do not know. Did Levine tried it out? I do not think she explains this in detail in her paper.

I think there two problems, one which is statistical in nature and the other one related to the concept of biological age itself.

Statistically, the few data in that region of the model makes me wonder about the value of any extrapolation. Would you agree?

Second, it is quite known that the predictive value of a biomarker of aging, maybe independently from its nature, decreases due to the increased heterogeneity of the sample population at old age.

albedo 

The Lervine data for very old individuals is sparse, but it's consistently below the linear dependence line, so I take it seriously.  I read some points off the Levine graph and then used the fitting capabilities of Mathematica 11.3  to get a rough correction procedure from those points.  Here, to be taken with a grain of salt, are the results.

If the Phenotypic Age, as calculated from my Excel spreadsheet, is P, then the corresponding DNAm PhenoAge estimate D is:
D(P) = P/(1 + 1.28047*Exp(0.0344329 (P-182.344))).
The Mortality Score MS corresponding to this value of D is:
MS(D) = 1 - Exp[-0.0005203635*Exp[0.090165*x]]
Below is a plot showing my fit to the points that I extracted.

For me, my calendar age is 84, and the spreadsheet says that my Phenotypic Age is 79.73 and my Mortality Score is 0.498.  These fits give my corrected Phenotypic Age, i.e., my estimated DNAm PhenoAge, as 76.86 and my corrected Mortality Score as 0.413. 

JGC But, hang on: the DNAm PhenoAge is built up from the Phenotypic age, not the other way around — and "the phenotypic age measure used to select CpGs is a better predictor of morbidity and mortality outcomes than DNAm PhenoAge." "Correcting" the Phenotypic Age to match up more closely with the DNAm PhenoAge would be putting the horse before the cart.

Iðunn 

In the Levine paper they report two approaches, which they connect.  Their Step 1 is based on blood work and Step 2 on methylation of DNA.  In the latter, they focus on evidence that a subject's DNA has become methylated as a marker of aging by analyzing some 500+ regions of a DNA analysis for evidence of methylation.  They correlate this with Phenotypic Age from Step 1 to get weighting factors for producing what they call the DNAm PhenoAge.  Since the latter is based on a direct analysis of DNA damage, I take it to be a better indication of aging effects than the blood work.  However, I suppose that is a matter of choice and opinion.

Of course, there are large uncertainties in either of these aging measures, but at least my fitting allows one to work both sides of the street.  Assuming that the main effects of aging are from the buildup of senescent cells, such buildup will show up both in average DNA damage and in the increasing presence of senescent cells in the bloodstream, which the Levine Step 1 weighting suggests shows up in average blood-cell geometry.

My present goal is to self-experiment with the plausible and currently accessible senolytic treatments (e.g., Fisetin and D+Q) on my wife and me, and to look at both of these markers of aging after each treatment sequence to see if either or both of the senolytic treatments have observable effects. 

JGC 

Thank you for sharing. I also applied the correction to my data and got similar results (proportionally less than yours due to my younger calendar age). Yet, I still wonder about the effect of heterogeneity of population at old ages and predictability (the second point in my post above) but I also start to believe the effect. Flattening of DNAm age clock has been shown, barring confounders and issue of repeatability with other cohorts, in centenarians by the team of Claudio Franceschi in Bologna (btw, he is the father of "inflammaging" which I am sure resonates with your thinking).

albedo 

I updated the Excl spreadsheet HERE to include the correction for the DNAm age and the corresponding mortality score.