I had an opportunity to compare the results of five epigenetic age tests in late 2021 - early 2022. My two blood samples were processed by TruDiagnostic, but I then obtained the .idat files (DNA microarray methylation values) from TruDiagnostic and sent these .idat files to the Horvath lab, where they were run against four other epigenetic clock algorithms. So this eliminated variance from blood samples or processing, and just showed the effects of weightings of CpGs in the epigenetic clock algorithms. My "biological age" results varied from 44.0 to 72.3 in late 2021 (chron age 70.7) and from 46.1 to 73.0 in early 2022 (chron age 71.0). The TruDiagnostic results were in the middle of the pack. So this is pretty consistent with Matt's report, but it shows a wide variation independent of lab practices and sample-to-sample variance.
Hey Matt - Thank you for pushing forward this conversation! This topic is absolutely needed and necessary, and we love your commitment to discussing the science behind biological age testing. That said, we’d like to clarify a few points to ensure viewers get the most accurate picture. 1.) Data Input and Algorithm Reliability: We noticed a critical issue wasn’t fully addressed in your analysis: bad input data = bad output data. As you said, discrepancies can arise when inputting incorrect chronological age data, so it's not surprising that the false age that you entered skewed your results. ** If we modeled your data with the correct chronological age, we would see a biological age discrepancy of less than 5%. ** Several published studies confirm that our algorithms, such as DunedinPACE and OMICmAge, maintain high reliability, with intraclass correlation coefficients (ICC) exceeding 0.96-a benchmark of consistency in the field. 2.) The Role of Chronological Age in the Calculations: Most epigenetic age algorithms, including GrimAge and OMICmAge, use chronological age as a covariate to sharpen the precision of their predictions. This design improves biological age accuracy and allows meaningful comparisons across data sets. Algorithms like DunedinPACE, which measures your speed of aging, do not include chronological age and thus demonstrate minimal variance across samples, as you mentioned. 3.) Published Clocks and Scientific Transparency: We pride ourselves on being the only epigenetic group that uses fully published and peer-reviewed algorithms. Transparency is foundational to our approach, so that our users can trust the science behind our tests. Many competing methods lack this level of validation, relying on proprietary or unpublished clocks, which compromises interpretability and scientific scrutiny. 4.) Variability across Testing Providers: You discussed the variability among different test companies. When doing so, it's crucial to call-out which companies use first-generation clocks, which are chronologically trained, and newer biological clocks that focus on health outcomes. Biological clocks, like the OMICmAge and SymphonyAge that we use, may show greater variability across samples due to their sensitivity to biological changes, but they significantly outperform chronological clocks in predicting health outcomes. It's unfortunate that you're comparing different generations of biological age clocks here. Platforms that use chronological trained clocks are certainly going to be much tighter grouped than 2nd generation biological clocks on the same data. This rewards chronologically trained clocks despite the fact that we know their hazard ratios are much worse to every disease outcome, including all-cause mortality. 5.) CLIA Certification & Our Commitment to Quality: Lastly, we want to emphasize that TruDiagnostic operates under CLIA certification, ensuring the highest laboratory standards. We collaborate with leading academic institutions like Harvard, Yale, and Duke, consistently contributing to peer-reviewed research and advancing the field of epigenetics. Thank you again for having this important discussion. We’ve sent you an email and would love to connect further on the nuances of these topics. Feel free to reach out if you'd like more data or references from our published studies to address these points further in detail! Warm regards, The TruDiagnostic Team
TruDiagnostics - 1) You can't be the ones to reveal the corrected biological age, you need to consider Matt's privacy. I agree the wrong age being declared isn't helpful to moving things forward, but there is a sequence of events that needs to happen for that number to be reported publicly, and you don't have that ability. The ICC is great, but you should also consider ISO 5725 on accuracy and precision, and state what those are, for the epigenetics and then after calculation, to encourage transparency. 2) Please encourage researchers to report model performance with and without age as a covariate. It would also be good to see some definitive studies on how different amounts of biomarker adjustments modify the epigenetics hazard ratios. 3) I agree and have already made similar comments. 4) These are the types of considerations that usually happen in "ensemble science". However I think you may have confused longitudinal variability with repeated sample precision here. If the repeated sample precision of OMICmAge and SymphonyAge is higher than would be expected, then those need to be declared and become a research project to understand why and be resolved. Also, its not rewarding clock generation, but exploring why these differences exist in the ensemble. I agree this could have been handled better. Ideally Matt would have also obtained the methylation file from all providers/tests, and conducted an analysis on how those differed. 5) Epigenetics is fundamentally a data science and although the CLIA cert. is great, transparency on accuracy and precision would be helpful to the consumers.
@@jamesgilmore8192 Hi James- Thanks for your notes and feedback! We've reached out to Matt to review all of these considerations in more depth but wanted to respond further to your comments here as well. First, currently all of the algorithms use age as a covariate. The models without covariate inclusion of age do not exist and therefore we cannot report both. As for your notes on the ISO guidance, we certainly calculate and follow these. However, the accuracy here is important to discuss. If we were just calculating a variable like chronological age, this would be very easy. However, what we are calculating is biological age and this definition widely varies. The definition of biological age differs among algorithms. That is why we measure accuracy by looking at Odds Ratios, Hazard Ratios, and C-index for disease. This is why we publish all of our new algorithms within biobanks as they would have this outcome data which shows our algorithms accuracy. Also, it is important to mention that the data used to create algorithms like OMICmAge and SymphonyAge had their raw data created within our lab at TruDiagnostic, which limits variation between the creation of the algorithm and the samples processed commercially. Last, we are also about to publish data with our collaborators from Yale on the longitudinal variability between a sample. We do know that immune cell subsets, time of day, stress, and feeding behavior can impact these markers and need to be incorporated into a model. Still, with all of this said, the biological age clocks show some of the most predictive hazard ratios for outcomes. This is the point - the reason why you would order it - to answer the question: How is biological aging affecting risk of death and disease? We hope this helps with clarity and really appreciate you extending the discussion! Warm regards, The TruDiagnostic Team
It sounds like chrono age dominates the predictions. It would be interesting to see a sensitivity analysis holding all other variables constant, and only changing the chrono age. Shapely values could then be calculated to identify potential biases or weaknesses, or to confirm the largest factors affecting bio age. This would help validate whether the model is accurately capturing physiological aging or just regressing to chronological norms.
@@Jrav27 Yes using an "explainable" framework for epigenetic clocks would be useful. TruDiagnostics has done some of that in the OmicmAge paper, so credit to them for publishing it. As for the other providers, who knows what they are doing. The field experts do have a lot more information, although it's trickling out slowly to others. The epigenetic clocks do seem to be capturing some aspects of physiological aging, although there are many questions, such as are the epigenetic clocks better when using measured biomarkers over epigenetic biomarker proxies? What exactly are the epigenetic clocks actually capturing, is it the underlying biomarkers, or is it control of epigenetic variance when everything is summed together for a final biological age. There are also other ideas out there like looking at features of the epigenome and how those change, rather than trying to regress to biomarkers.
Hi Matt, great explanation and pretty hilarious. These videos are important for everyone (including scientists like me) who are seriously interested in mechanisms of resilience and healthspan. Greetings from the aging research institute in the Netherlands.
Thanks Matt. Haven't the clocks backup science been already scrutinized by peers on scientific publications? Also, by the way, is Optispan company being regulated somehow? It would be nice if you can make a video to explicitly talk about this.
I think that your less than ideal biological test results and excessive melanin production are related: rapamycin -> mTOR down -> AMPK up -> PGC-1α up -> mitochondrial biogenesis up -> mitochondrial volume and activity up -> electron leakage up -> ROS generation up -> melanin production (as a visual indicator) up and DNA damage up. Chronically, not in pulses. As a way out of this, you may consider Nrf2/antioxidant activators, e.g. sulforaphane + GlyNAC.
Thank you, Matt and team, for all the valuable work you guys do. People need more experts like you on the internet. Thank you for your transparency. It is very honourable. Keep up this great work. ❤
Thank you Matt, I've been thinking about getting one of these but seeing the wide variation of results makes me think twice about the accuracy of any of them.
Dr. Kaeberlein. Several questions. Bryan Johnson stopped taking rapamycin after a study suggesting that it would accalerate aging based on several biological clocks. What do you think of that? What is your association with Novos? They seem to refer to you as a consultant. Your content seems to conflict with what they are doing (selling a combination of several supplements). It's a bit confusing. Thank you!
We can expect different tissues, such as saliva, blood etc to have different biological ages. It would be even better to have epigenetic ages of specific cells types, such as an epigenetic report on blood could have neutrophils, monocytes, lymphocytes classes etc. This would clear up the interpretation of these tests.
Matt -- The epigenetic clocks need advanced analysis to be reliable, see e.g. A computational solution for bolstering reliability of epigenetic clocks: Implications for clinical trials and longitudinal tracking, Higgins-Chen et al, Nature Aging 2022. In this paper they showed a principle component analysis increased reliability. The issue it seems is individual CpG site values are prone to noise, yet there is more robust information in the overall patterns.
@patruff You are welcome. Matt produces good content, but the epigenetic clocks are fundamentally a nexus of big data, complex systems and statistics and I'm not sure the nuances of this are a strength of this channel.
@jamesgilmore8192 yeah agreed. I do like the fact that the clocks exist but it's worth remembering that this is certainly new technology and should only improve over time, number of people using it, etc.
IMO the current most accurate biomarker of aging is how old you look to people that don't know you. For lolz put your blood test results into the Dunedin Pace calculator with different chronological ages.
There are AI picture analyzers which can guess age as well...then you don't have the issue with honesty, tact, and sample size. But if you want to use them more than once, don't give them your real age as feed back. It will just use that the next time, and it will just regurgitate what you told it. Of course, many cameras automatically clean up images, to make you happier. Gotta think that messes with the algorithm. Though, maybe it recognizes that, and adjusts accordingly. We rarely know what is happening under the hood.
@@ChessMasterNate IMO Their online dataset is too corrupted to ever be a accurate. Maybe if they could utilize hundreds of thousands of photos taken by the motor vehicle department under controlled conditions.
Its hard to compare epigenetic clocks from the four different providers because they all seem to be using different algorithms. The issue is none of the algorithms are transparent, so how do you interpret them?
MyDNAge require you to provide your chronological age before generating your result, saying they need it to to generate a results chart comparing your chronological and biological age. I was suspicious and refused. There was a bit of back and forth until they finally agreed to give the result without getting my age. I was happy with result but I don’t place any weight on it any more.
Great episode: very informative. I hear these tests trashed a lot, but no hard data that they are so unreliable. They're popular because people want a way to check if what they're doing is working. If they at least were consistent then they might be useful for that. It's a shame they aren't.
I did a sort of "biological age" test just by gathering data from ideal BMI, body fat (estimated ideal), and resting heart rate. For that I use the "damage" or deviation from ideals to reverse enginner the actual age of the subjects. Younger guys got older and older younger in my sample, but also older guys could change their age dramatically in mere days depending if they followed good habits for their profiles. I do think a big deal of it its maleable, as it makes sense if you predict age based on say BMI, if it is an obese person and then gets fit, then they would be translated as much "younger" based on that model variable.
Is this in reaction to Bryan Johnston’s decision to quit Rapamycin based on epigenetic tests that seemed to show it actually accelerates epigenetic aging?
Awesome review! I do think looking in the mirror candidly, and getting enough "unbiased stranger's opinion" could give one a clue as to the rate of aging they are experiencing. Also, bouts of stress or elation happenings can also greatly affect physical and emotional states (dahh, lol!) And then there's always that genetic component... like Matt said, do your best to eat, move, connect to maximize YOUR own potential 🤗
This was a fantastic overview and experiment. I had wondered the same thing as my epigenetic age showed a 10 year span (on one test I was supposedly 6 years younger, on the next test I was 4 years older and yet another test showed 9 years younger) over the course of a couple of years which seemed rather implausible. Furthermore, one of my recent results showed that I had worse VO2 max than 65% of the population, yet my actual VO2 max as measured (painfully on a treadmill 😂) in a proper metabolic lab showed my VO2 max in the “excellent” range (I.e. top 5% of the population). Regarding Rejuvenation Olympics, the only measure they use is Dunedin pace, not epigenetic/biological age. It would be interesting to do this same experiment with Dunedin pace and see how much variation there is in this measure.
I took the TruDiagnostic test few months ago when I was 59.5 yrs old. I took my blood sample at the end of a 3 day fast on purpose to see my stressed results. My pace of aging came out at .72, my Omic Age came out at 65 yrs old, my Symphony Age (organs) came out 47 yrs old. Seems the biological age is highly sensitive to current changes in health, like if you had the flu, a weekend partying, or under a lot of stress, etc. I think the pace is more of a stable reading like a trend, and the biological is more a daily noise. I guess we can look at it like the stock market. I put my body under stress last test with 3 days fasted. I'm taking another blood sample for a new test this weekend and will go in with just an overnight fast. Not really anything significant in any changes to my highly consistent health regimen. I know I'm really healthy for the most part just following a good consistent health maintenance routine. I like having the idea of being able to measure what we are doing. Over time I think we'll get better on how to do it more accurately.
This is a grey zone when looking to determine the bio age. We are at the beginning of defining what aging is. The same thing applies to heart rate, Vo2 max, etc.. These tests can only be used in a context as benchmark to define if your health is getting better or not. We will never define with precision what is our real bio age. It is just a nice to have tool.
Matt inadvertently presented on "model ensembles" in the post. That's when you form a collection of results from the same input through different models. In this case, a large variance is to be expected given the different algorithms and sample types used, especially with the clocks being relatively new scientifically speaking. I've worked with ensembles as a researcher and they are indeed tricky to interpret. IMHO, the critique of the differences could have been handled better by considering ensembles principles. When there is a large variance or e.g. one model in particular disagrees, this is usually an opportunity to investigate why. The issue, as rightly pointed out by Matt, is the for profit nature of this industry without transparency and evidence, which means this isn't possible.
I was considering saving up some money to get one of these tests, but realizing now I'd probably need eight of them to get any accuracy, I think I'll pass.
I consider these values as just some composite indexes aggregating my health data in some way. I can compare them to themselves over time, to track the effects of my life choices.
I think it is amazing that they can even estimate your calendar age just by analyzing some fluid samples. Obviously, those tests are not useful at this point in time.
I've done a few of these and was frustrated with the end results. Thank you matt. an example of more noise one test said my hormone status was 10 years older than my chronological age yet I have been normalizing my hormones with HRT for 10 years now and my testosterone and thyroid is that of 40 year old yet the age test said I had the hormones of an 80 year old. adding to the noise these tests can't track health practices that you're doing that they can't measure. I differ though in the case of hormonesthey could have looked at what was in my blood not just the epigenetics predisposition for hormone status. You're right Matt this is all embryonic it's nowhere near mature.
The difference in the Trudiagnostics result is not acceptable. It would not satisfy typical measurement standards, e.g. ISO 5725-1:2023 Accuracy (trueness and precision) of measurement methods and results. A 2 year difference (~4%) between the replicates would be considered acceptable for most biological measurements.
I’ve had my GlycanAge analysed as have over 8 people I know. Total waste of time. I’m apparently 21 years younger than my chronological age as were the vast majority of other people I know..
I never believed in the tests so kind of happy that you came to the conclusion that you did. I can well imagine they are not cheap. Thanks for the insight into a topic that has never really been investigated by comparison
Thank you very much for this particular podcast. I look at these tests as being like supplements and the hype that highly promotes some of them. They are not medically proven and I am highly skeptical about their usefulness. However, some of the epigenetic tests may be useful, just like some the supplements I use may be useful. But, I'm not totally sure!
Thank you for sharing such brilliant information! I have to agree-it doesn’t seem like these tests are particularly accurate. I tried the MUHDO test, and it showed my biological age to be close to my chronological age, but two years older. I’m currently waiting for new results after 3 cycles of fasting mimicking diet and taking supplements. Honestly, this is more of a curiosity-driven experiment for me, and it’s likely my last attempt-unless there’s new evidence proving the accuracy of these tests.
Matt, you said I don't know what to think about these epigenetic biological age assessments. Indeed, I know what to think about them ... at the moment they are at a level close to that of snake oil.
Hey Matt, thanks a lot! I've been looking for content like this since forever. Love to have you talk about these things on my Immortal Combat podcast, where I usually interview Rejuvenation Olympics athletes, but the insights in this video could serve as a foundational reminder for gaming with epigenetic age: th-cam.com/play/PL4nqc85w185sO4i7eR3oUO_lMmlJ2K1cL.html
Even children today have signs of artery disease. Fast foods and sugary drinks have long term consequences. There is an obesity epidemic today where over 60% of the population is overweight or obese. What is common today and "NORMAL" is not healthy. The average person today is not healthy and is taking medications. Heart attacks were once only common in people over 65. Today people in their 40 are having heart attacks. Add more plants and vegetables to your diet and less junk and overly processed foods Get off the couch. Get off the iphone or ipad. Get up and go for a walk, run, bike ride , garden, work outside. Move more, sit less.
I think your "high school level" N=1 experiment was a total waste of time. I guess you were able to "ridiculize" the providers, but it was done with such lack of rigour that is not helping anyone. With that lack of rigour and so biased approach is not surprising that the results were a complete mess. Of course the current biological clocks measure a mix of health and age. You did not need to make this tests to know that. And, of course they are using your chronological age as an input in their calculations, this because we all know these clocks are not a finalised science but a first generation of their kind which still need that input, so what? You knew that in advance and rather than taking the best and useful side of them you basically focused on highlighting their weaknesses of being the "dream biological clock". So yeah, they are not true biological clock measures (they are still a mix of health and age measures), we knew that already. Although I guess you would tell that with your "experiment" now we know. But now what? Why don't we focus on usefulness? Like: are any of the clocks available in the market telling some useful information to the user? Like weak points that could be improved or monitored? Or, for example, if you measure at baseline and then perform an intervention with Rapamycin, would you see a change in the clock marks? Would you see a change in any other biomarker or performance test? I think it is a better use of your time to invest in evaluating them to find which available clock would provide the best useful information and also which ones would show changes after the known positiive interventions that we know also show changes in known biomarkers and performance tests! I recommend you make a better use of your time! All these clocks have scientific papers backing them up. I would have expected from you at least to review the science in those papers and talking about that in your presentation rather than assuming them as just products that are out there. You should stop now the debate of the measure of biological age by this clocks. We all know they measure age and health. If you want don't call them biological clocks, just call them health-measures reporting time, or something similar and, as said, focus more on their usefulnes (or not) instead of on the definition of their name.
They have science papers backing them as laboratory tools, NOT as medical diagnostic tests. Their usefulness is within very limited contexts just for comparison in experiments, in an attempt to not have to follow study subjects with longer lifespans than studys lasting 1-5 years. It's not ideal, but it's what is needed to get beyond mice. Many are still skeptical that these tools even are useful for that. You seem really invested in these stastistical tools and want them to be more than what they actually are.
I had an opportunity to compare the results of five epigenetic age tests in late 2021 - early 2022. My two blood samples were processed by TruDiagnostic, but I then obtained the .idat files (DNA microarray methylation values) from TruDiagnostic and sent these .idat files to the Horvath lab, where they were run against four other epigenetic clock algorithms. So this eliminated variance from blood samples or processing, and just showed the effects of weightings of CpGs in the epigenetic clock algorithms. My "biological age" results varied from 44.0 to 72.3 in late 2021 (chron age 70.7) and from 46.1 to 73.0 in early 2022 (chron age 71.0). The TruDiagnostic results were in the middle of the pack. So this is pretty consistent with Matt's report, but it shows a wide variation independent of lab practices and sample-to-sample variance.
I love that Matt is so candid and his opinions cannot be bought by any company.
Love this style of video refuting the hype in the marketplace
Matt represents the
next generation of influences. Those with integrity and high levels of scientific literacy.
Hey Matt - Thank you for pushing forward this conversation! This topic is absolutely needed and necessary, and we love your commitment to discussing the science behind biological age testing. That said, we’d like to clarify a few points to ensure viewers get the most accurate picture.
1.) Data Input and Algorithm Reliability:
We noticed a critical issue wasn’t fully addressed in your analysis: bad input data = bad output data. As you said, discrepancies can arise when inputting incorrect chronological age data, so it's not surprising that the false age that you entered skewed your results.
** If we modeled your data with the correct chronological age, we would see a biological age discrepancy of less than 5%. **
Several published studies confirm that our algorithms, such as DunedinPACE and OMICmAge, maintain high reliability, with intraclass correlation coefficients (ICC) exceeding 0.96-a benchmark of consistency in the field.
2.) The Role of Chronological Age in the Calculations:
Most epigenetic age algorithms, including GrimAge and OMICmAge, use chronological age as a covariate to sharpen the precision of their predictions. This design improves biological age accuracy and allows meaningful comparisons across data sets. Algorithms like DunedinPACE, which measures your speed of aging, do not include chronological age and thus demonstrate minimal variance across samples, as you mentioned.
3.) Published Clocks and Scientific Transparency:
We pride ourselves on being the only epigenetic group that uses fully published and peer-reviewed algorithms. Transparency is foundational to our approach, so that our users can trust the science behind our tests. Many competing methods lack this level of validation, relying on proprietary or unpublished clocks, which compromises interpretability and scientific scrutiny.
4.) Variability across Testing Providers:
You discussed the variability among different test companies. When doing so, it's crucial to call-out which companies use first-generation clocks, which are chronologically trained, and newer biological clocks that focus on health outcomes. Biological clocks, like the OMICmAge and SymphonyAge that we use, may show greater variability across samples due to their sensitivity to biological changes, but they significantly outperform chronological clocks in predicting health outcomes.
It's unfortunate that you're comparing different generations of biological age clocks here. Platforms that use chronological trained clocks are certainly going to be much tighter grouped than 2nd generation biological clocks on the same data. This rewards chronologically trained clocks despite the fact that we know their hazard ratios are much worse to every disease outcome, including all-cause mortality.
5.) CLIA Certification & Our Commitment to Quality:
Lastly, we want to emphasize that TruDiagnostic operates under CLIA certification, ensuring the highest laboratory standards. We collaborate with leading academic institutions like Harvard, Yale, and Duke, consistently contributing to peer-reviewed research and advancing the field of epigenetics.
Thank you again for having this important discussion. We’ve sent you an email and would love to connect further on the nuances of these topics. Feel free to reach out if you'd like more data or references from our published studies to address these points further in detail!
Warm regards,
The TruDiagnostic Team
TruDiagnostics -
1) You can't be the ones to reveal the corrected biological age, you need to consider Matt's privacy. I agree the wrong age being declared isn't helpful to moving things forward, but there is a sequence of events that needs to happen for that number to be reported publicly, and you don't have that ability.
The ICC is great, but you should also consider ISO 5725 on accuracy and precision, and state what those are, for the epigenetics and then after calculation, to encourage transparency.
2) Please encourage researchers to report model performance with and without age as a covariate. It would also be good to see some definitive studies on how different amounts of biomarker adjustments modify the epigenetics hazard ratios.
3) I agree and have already made similar comments.
4) These are the types of considerations that usually happen in "ensemble science". However I think you may have confused longitudinal variability with repeated sample precision here. If the repeated sample precision of OMICmAge and SymphonyAge is higher than would be expected, then those need to be declared and become a research project to understand why and be resolved. Also, its not rewarding clock generation, but exploring why these differences exist in the ensemble. I agree this could have been handled better.
Ideally Matt would have also obtained the methylation file from all providers/tests, and conducted an analysis on how those differed.
5) Epigenetics is fundamentally a data science and although the CLIA cert. is great, transparency on accuracy and precision would be helpful to the consumers.
@@jamesgilmore8192 Hi James- Thanks for your notes and feedback! We've reached out to Matt to review all of these considerations in more depth but wanted to respond further to your comments here as well.
First, currently all of the algorithms use age as a covariate. The models without covariate inclusion of age do not exist and therefore we cannot report both.
As for your notes on the ISO guidance, we certainly calculate and follow these. However, the accuracy here is important to discuss. If we were just calculating a variable like chronological age, this would be very easy. However, what we are calculating is biological age and this definition widely varies. The definition of biological age differs among algorithms. That is why we measure accuracy by looking at Odds Ratios, Hazard Ratios, and C-index for disease. This is why we publish all of our new algorithms within biobanks as they would have this outcome data which shows our algorithms accuracy.
Also, it is important to mention that the data used to create algorithms like OMICmAge and SymphonyAge had their raw data created within our lab at TruDiagnostic, which limits variation between the creation of the algorithm and the samples processed commercially.
Last, we are also about to publish data with our collaborators from Yale on the longitudinal variability between a sample. We do know that immune cell subsets, time of day, stress, and feeding behavior can impact these markers and need to be incorporated into a model. Still, with all of this said, the biological age clocks show some of the most predictive hazard ratios for outcomes. This is the point - the reason why you would order it - to answer the question: How is biological aging affecting risk of death and disease?
We hope this helps with clarity and really appreciate you extending the discussion!
Warm regards,
The TruDiagnostic Team
It sounds like chrono age dominates the predictions. It would be interesting to see a sensitivity analysis holding all other variables constant, and only changing the chrono age. Shapely values could then be calculated to identify potential biases or weaknesses, or to confirm the largest factors affecting bio age.
This would help validate whether the model is accurately capturing physiological aging or just regressing to chronological norms.
@@Jrav27 Yes using an "explainable" framework for epigenetic clocks would be useful. TruDiagnostics has done some of that in the OmicmAge paper, so credit to them for publishing it. As for the other providers, who knows what they are doing. The field experts do have a lot more information, although it's trickling out slowly to others.
The epigenetic clocks do seem to be capturing some aspects of physiological aging, although there are many questions, such as are the epigenetic clocks better when using measured biomarkers over epigenetic biomarker proxies? What exactly are the epigenetic clocks actually capturing, is it the underlying biomarkers, or is it control of epigenetic variance when everything is summed together for a final biological age.
There are also other ideas out there like looking at features of the epigenome and how those change, rather than trying to regress to biomarkers.
Thank you for sharing this. Everyone considering epigenetic tests should see this video.
I love the minus 20 years reported age trick, as this exposes the shifty age calculation happening behind the scenes. Great work Matt!
Thank you for this video! I hope more people watch this
Great transparency! Thanks for doing this! 👍
So, are you dropping off the use of the True Diagnostics test in Optispan from now on?
Appreciate your disclosure of the data that are useful for people to understand aging at a deeper level.
Excellent episode. Thanks for doing this and helping us save time and money.
Hi Matt, great explanation and pretty hilarious. These videos are important for everyone (including scientists like me) who are seriously interested in mechanisms of resilience and healthspan. Greetings from the aging research institute in the Netherlands.
Thanks!
Appreciate your support!
- Nick
Thanks Matt. Haven't the clocks backup science been already scrutinized by peers on scientific publications? Also, by the way, is Optispan company being regulated somehow? It would be nice if you can make a video to explicitly talk about this.
I think that your less than ideal biological test results and excessive melanin production are related: rapamycin -> mTOR down -> AMPK up -> PGC-1α up -> mitochondrial biogenesis up -> mitochondrial volume and activity up -> electron leakage up -> ROS generation up -> melanin production (as a visual indicator) up and DNA damage up. Chronically, not in pulses. As a way out of this, you may consider Nrf2/antioxidant activators, e.g. sulforaphane + GlyNAC.
Thank you, Matt and team, for all the valuable work you guys do. People need more experts like you on the internet. Thank you for your transparency. It is very honourable. Keep up this great work. ❤
We really appreciate the feedback! - Tara
EXCELLENT INFORMATION MATT!
If they asked me my age, I'm out because aren't they're supposed to be telling me my age?
Great test ! I hope you do it again maybe next year to see if accuracy has improved.
Thank you Matt, I've been thinking about getting one of these but seeing the wide variation of results makes me think twice about the accuracy of any of them.
Dr. Kaeberlein. Several questions. Bryan Johnson stopped taking rapamycin after a study suggesting that it would accalerate aging based on several biological clocks. What do you think of that? What is your association with Novos? They seem to refer to you as a consultant. Your content seems to conflict with what they are doing (selling a combination of several supplements). It's a bit confusing. Thank you!
We can expect different tissues, such as saliva, blood etc to have different biological ages. It would be even better to have epigenetic ages of specific cells types, such as an epigenetic report on blood could have neutrophils, monocytes, lymphocytes classes etc. This would clear up the interpretation of these tests.
Matt -- The epigenetic clocks need advanced analysis to be reliable, see e.g. A computational solution for bolstering reliability of epigenetic clocks: Implications for clinical trials and longitudinal tracking, Higgins-Chen et al, Nature Aging 2022. In this paper they showed a principle component analysis increased reliability. The issue it seems is individual CpG site values are prone to noise, yet there is more robust information in the overall patterns.
Thanks for actual information! Not just random anecdotes.
@patruff You are welcome. Matt produces good content, but the epigenetic clocks are fundamentally a nexus of big data, complex systems and statistics and I'm not sure the nuances of this are a strength of this channel.
@jamesgilmore8192 yeah agreed. I do like the fact that the clocks exist but it's worth remembering that this is certainly new technology and should only improve over time, number of people using it, etc.
Thanks for Matt for taking this trouble and bring us this usefull information!
IMO the current most accurate biomarker of aging is how old you look to people that don't know you. For lolz put your blood test results into the Dunedin Pace calculator with different chronological ages.
There are AI picture analyzers which can guess age as well...then you don't have the issue with honesty, tact, and sample size. But if you want to use them more than once, don't give them your real age as feed back. It will just use that the next time, and it will just regurgitate what you told it. Of course, many cameras automatically clean up images, to make you happier. Gotta think that messes with the algorithm. Though, maybe it recognizes that, and adjusts accordingly. We rarely know what is happening under the hood.
@@ChessMasterNate IMO Their online dataset is too corrupted to ever be a accurate. Maybe if they could utilize hundreds of thousands of photos taken by the motor vehicle department under controlled conditions.
Thank you so much. Very enlightening
You mentioning that some of these influencers could just be showing the tests they want you to see was like a mind=blown moment for me....wow....
Its hard to compare epigenetic clocks from the four different providers because they all seem to be using different algorithms. The issue is none of the algorithms are transparent, so how do you interpret them?
MyDNAge require you to provide your chronological age before generating your result, saying they need it to to generate a results chart comparing your chronological and biological age. I was suspicious and refused. There was a bit of back and forth until they finally agreed to give the result without getting my age. I was happy with result but I don’t place any weight on it any more.
Great episode: very informative.
I hear these tests trashed a lot, but no hard data that they are so unreliable.
They're popular because people want a way to check if what they're doing is working. If they at least were consistent then they might be useful for that. It's a shame they aren't.
I did a sort of "biological age" test just by gathering data from ideal BMI, body fat (estimated ideal), and resting heart rate. For that I use the "damage" or deviation from ideals to reverse enginner the actual age of the subjects. Younger guys got older and older younger in my sample, but also older guys could change their age dramatically in mere days depending if they followed good habits for their profiles. I do think a big deal of it its maleable, as it makes sense if you predict age based on say BMI, if it is an obese person and then gets fit, then they would be translated as much "younger" based on that model variable.
These tests are expensive for the average person. Thanks for a realistic review.
Is this in reaction to Bryan Johnston’s decision to quit Rapamycin based on epigenetic tests that seemed to show it actually accelerates epigenetic aging?
This was filmed before that happened.
We are filming a video about Bryan Johnson quitting rapamycin tomorrow, will be out a week from today.
Great, thanks. Looking forward to it
Awesome review! I do think looking in the mirror candidly, and getting enough "unbiased stranger's opinion" could give one a clue as to the rate of aging they are experiencing. Also, bouts of stress or elation happenings can also greatly affect physical and emotional states (dahh, lol!) And then there's always that genetic component... like Matt said, do your best to eat, move, connect to maximize YOUR own potential 🤗
This was a fantastic overview and experiment. I had wondered the same thing as my epigenetic age showed a 10 year span (on one test I was supposedly 6 years younger, on the next test I was 4 years older and yet another test showed 9 years younger) over the course of a couple of years which seemed rather implausible.
Furthermore, one of my recent results showed that I had worse VO2 max than 65% of the population, yet my actual VO2 max as measured (painfully on a treadmill 😂) in a proper metabolic lab showed my VO2 max in the “excellent” range (I.e. top 5% of the population).
Regarding Rejuvenation Olympics, the only measure they use is Dunedin pace, not epigenetic/biological age.
It would be interesting to do this same experiment with Dunedin pace and see how much variation there is in this measure.
I took the TruDiagnostic test few months ago when I was 59.5 yrs old. I took my blood sample at the end of a 3 day fast on purpose to see my stressed results. My pace of aging came out at .72, my Omic Age came out at 65 yrs old, my Symphony Age (organs) came out 47 yrs old.
Seems the biological age is highly sensitive to current changes in health, like if you had the flu, a weekend partying, or under a lot of stress, etc. I think the pace is more of a stable reading like a trend, and the biological is more a daily noise. I guess we can look at it like the stock market.
I put my body under stress last test with 3 days fasted. I'm taking another blood sample for a new test this weekend and will go in with just an overnight fast. Not really anything significant in any changes to my highly consistent health regimen.
I know I'm really healthy for the most part just following a good consistent health maintenance routine. I like having the idea of being able to measure what we are doing. Over time I think we'll get better on how to do it more accurately.
Always love waking up to a cup of coffee alonside a Matt Kaeberlein video. Gets me motivated to do some anti ageing shit all through the day.
Aren't the "weird" aging speed, results from the sanity check, the product of noise around the mean, rather than a measure of true aging ?
This is a grey zone when looking to determine the bio age. We are at the beginning of defining what aging is. The same thing applies to heart rate, Vo2 max, etc.. These tests can only be used in a context as benchmark to define if your health is getting better or not. We will never define with precision what is our real bio age. It is just a nice to have tool.
Matt inadvertently presented on "model ensembles" in the post. That's when you form a collection of results from the same input through different models. In this case, a large variance is to be expected given the different algorithms and sample types used, especially with the clocks being relatively new scientifically speaking. I've worked with ensembles as a researcher and they are indeed tricky to interpret. IMHO, the critique of the differences could have been handled better by considering ensembles principles. When there is a large variance or e.g. one model in particular disagrees, this is usually an opportunity to investigate why. The issue, as rightly pointed out by Matt, is the for profit nature of this industry without transparency and evidence, which means this isn't possible.
I was considering saving up some money to get one of these tests, but realizing now I'd probably need eight of them to get any accuracy, I think I'll pass.
I consider these values as just some composite indexes aggregating my health data in some way. I can compare them to themselves over time, to track the effects of my life choices.
The same sample noise defeats any value!
thanks, Matt. You saved me a bunch of money.
I think it is amazing that they can even estimate your calendar age just by analyzing some fluid samples. Obviously, those tests are not useful at this point in time.
Your thoughts on PNOE? physical testing utilising oxygen/carbon dioxide output /heart rate etc? Thanks
I'd like to see a circadian series of age tests made over a 24-hr period.
I've done a few of these and was frustrated with the end results. Thank you matt. an example of more noise one test said my hormone status was 10 years older than my chronological age yet I have been normalizing my hormones with HRT for 10 years now and my testosterone and thyroid is that of 40 year old yet the age test said I had the hormones of an 80 year old. adding to the noise these tests can't track health practices that you're doing that they can't measure. I differ though in the case of hormonesthey could have looked at what was in my blood not just the epigenetics predisposition for hormone status. You're right Matt this is all embryonic it's nowhere near mature.
Good post-calls’em like I sees them!
The difference in the Trudiagnostics result is not acceptable. It would not satisfy typical measurement standards, e.g. ISO 5725-1:2023 Accuracy (trueness and precision) of measurement methods and results. A 2 year difference (~4%) between the replicates would be considered acceptable for most biological measurements.
What about Viome
Thank you and very useful
Why didn’t you include a test that measures glycans? Like GlycanAge for example? Isn’t that the most accurate test available at this time?
I’ve had my GlycanAge analysed as have over 8 people I know. Total waste of time. I’m apparently 21 years younger than my chronological age as were the vast majority of other people I know..
I never believed in the tests so kind of happy that you came to the conclusion that you did. I can well imagine they are not cheap. Thanks for the insight into a topic that has never really been investigated by comparison
Matt, this awesome.
Tell us your thoughts about Omega-3!
Thanks for saving us money, might as well throw a dice and subtract or add its value to our chronological age.
Thank you very much for this particular podcast. I look at these tests as being like supplements and the hype that highly promotes some of them. They are not medically proven and I am highly skeptical about their usefulness. However, some of the epigenetic tests may be useful, just like some the supplements I use may be useful. But, I'm not totally sure!
Thank you for sharing such brilliant information! I have to agree-it doesn’t seem like these tests are particularly accurate. I tried the MUHDO test, and it showed my biological age to be close to my chronological age, but two years older. I’m currently waiting for new results after 3 cycles of fasting mimicking diet and taking supplements. Honestly, this is more of a curiosity-driven experiment for me, and it’s likely my last attempt-unless there’s new evidence proving the accuracy of these tests.
Great reporting! It makes total sense.
I wasted $200!
It's too bad more people won't see this video...
Matt, you said I don't know what to think about these epigenetic biological age assessments. Indeed, I know what to think about them ... at the moment they are at a level close to that of snake oil.
sounds exciting
You forgot to tell us about your Dunedin Report
Hey Matt, thanks a lot! I've been looking for content like this since forever. Love to have you talk about these things on my Immortal Combat podcast, where I usually interview Rejuvenation Olympics athletes, but the insights in this video could serve as a foundational reminder for gaming with epigenetic age: th-cam.com/play/PL4nqc85w185sO4i7eR3oUO_lMmlJ2K1cL.html
You should be VERY supportive. These are indicators and are developing. Nothing else can match this potential.
I don't think one should be supportive of tests that are inaccurate just because they may be useful one day.
Even children today have signs of artery disease. Fast foods and sugary drinks have long term consequences.
There is an obesity epidemic today where over 60% of the population is overweight or obese.
What is common today and "NORMAL" is not healthy. The average person today is not healthy and is taking medications.
Heart attacks were once only common in people over 65. Today people in their 40 are having heart attacks.
Add more plants and vegetables to your diet and less junk and overly processed foods
Get off the couch. Get off the iphone or ipad. Get up and go for a walk, run, bike ride , garden, work outside. Move more, sit less.
Cholesterol and SFA, not sugar.
I think your "high school level" N=1 experiment was a total waste of time. I guess you were able to "ridiculize" the providers, but it was done with such lack of rigour that is not helping anyone. With that lack of rigour and so biased approach is not surprising that the results were a complete mess. Of course the current biological clocks measure a mix of health and age. You did not need to make this tests to know that. And, of course they are using your chronological age as an input in their calculations, this because we all know these clocks are not a finalised science but a first generation of their kind which still need that input, so what? You knew that in advance and rather than taking the best and useful side of them you basically focused on highlighting their weaknesses of being the "dream biological clock". So yeah, they are not true biological clock measures (they are still a mix of health and age measures), we knew that already. Although I guess you would tell that with your "experiment" now we know. But now what? Why don't we focus on usefulness? Like: are any of the clocks available in the market telling some useful information to the user? Like weak points that could be improved or monitored? Or, for example, if you measure at baseline and then perform an intervention with Rapamycin, would you see a change in the clock marks? Would you see a change in any other biomarker or performance test? I think it is a better use of your time to invest in evaluating them to find which available clock would provide the best useful information and also which ones would show changes after the known positiive interventions that we know also show changes in known biomarkers and performance tests! I recommend you make a better use of your time! All these clocks have scientific papers backing them up. I would have expected from you at least to review the science in those papers and talking about that in your presentation rather than assuming them as just products that are out there.
You should stop now the debate of the measure of biological age by this clocks. We all know they measure age and health. If you want don't call them biological clocks, just call them health-measures reporting time, or something similar and, as said, focus more on their usefulnes (or not) instead of on the definition of their name.
They have science papers backing them as laboratory tools, NOT as medical diagnostic tests. Their usefulness is within very limited contexts just for comparison in experiments, in an attempt to not have to follow study subjects with longer lifespans than studys lasting 1-5 years. It's not ideal, but it's what is needed to get beyond mice. Many are still skeptical that these tools even are useful for that. You seem really invested in these stastistical tools and want them to be more than what they actually are.
Great video advice, thanks, I think they're a money making scam.
Clean up your diet a bit and you can test lower.