Study of 18 Epigenetic Clocks Finds Most Give Unstable Readings Over Time

The study is a benchmarking analysis of existing data, not a trial of any treatment. The researchers, led by Sehgal and colleagues, pooled datasets of blood DNA methylation measurements using their TranslAGE platform. Sample sizes were small and varied by question: for example, 196 people with two replicates each on one array type, 128 on another, and 34 people in the meal and stress experiments.

Epigenetic clocks read chemical tags on DNA, called methylation, to estimate biological age. They are used to predict mortality and other health outcomes. They are often proposed as a way to track whether an intervention is working. For that, a clock must give consistent readings. The authors separate two kinds of consistency.

The first is technical reliability: do you get the same result when the same blood sample is run twice? Mostly, yes. Most clocks scored above 0.9 on a 0-to-1 reliability scale (the intraclass correlation coefficient, or ICC), which the authors call excellent. Older clocks such as Hannum, PhenoAge and PCPhenoAge were lower, around 0.7 to 0.8.

Some clocks were thrown off by lab details. When replicates sat in different positions on the testing slide, nine clocks fell below the "good" range. Three clocks lost reproducibility with different DNA extraction methods. Clocks built from principal components, notably PCGrimAge and SystemsAge, stayed robust across conditions.

The second kind is biological reliability: do repeated samples from the same person, taken hours or days apart, give similar results? Here the picture was much weaker. Across four studies, most clocks reached only "moderate" to "good" reliability (ICC about 0.4 to 0.7). None reached "excellent."

Everyday events mattered. After a single meal (34 people, four samples in a day), many clocks fell to moderate or poor reliability. Stress (also 34 people) had a similar effect, with DunedinPACE and GrimAge among the clocks rated poor. Pollution exposure (16 people, three samples over two days) also lowered stability. Altitude change (21 people, four samples over seven days) looked somewhat steadier. The authors report that only altitude produced a systematic shift in clock values, after seven days, so they removed later time points.

Two further results stand out. Adjusting for immune cell makeup, which researchers might expect to reduce noise, lowered biological reliability across nearly all clocks (p < 0.001). A few, such as DunedinPACE and PhenoAge in some conditions, improved. And technical reproducibility did not predict biological reliability. A clock that repeats well in the lab can still swing in the same person.

This matters for results. According to the authors, clocks with higher reliability produced more stable associations with cognitive outcomes and more consistent responses to interventions, such as a vegan diet. Less reliable clocks gave variable or misleading results. The paper's own example is that an intervention study that ignored fasting status could mistake meal effects for treatment effects.

For people weighing a consumer biological age test, the study supports caution about one reading, and about small changes between two readings. It does not say which commercial tests are better, and it did not test consumer products. The authors say improving biological reliability is essential before these clocks are used in clinical and interventional settings. Limits include the small samples and short time frames studied.

What to watch next

Whether newer clocks are designed and tested for biological reliability, not only lab reproducibility.

Home Light Therapy Beat Placebo on Sleep and Memory Scores in Mild Cognitive Impairment Exercising Outdoors in Nature Boosted Older Adults' Mental Health More Than Indoors AI Health Agent Reads Wearable and Medical Record Data to Coach Users HIIT Raised VO2 Max Most Consistently in Inactive People, Meta-Analysis Finds Too Much RNA Can Damage Mitochondria and Drain Cell Energy Ultra-Processed Calories Linked to Worse Daily Blood Sugar in Type 2 Diabetes Whole-Food Diet Changes Added About 12 Grams Daily Protein in Older Adults Breathing Practices May Calm Nerves and Brain, but Some People Need Caution Gut Microbe Transplants Eased Nicotine Withdrawal and Anxiety in Mice Vitamin B3 Linked to Younger Muscle Epigenetic Readings; Interval Training Mixed Sleeping an Hour Longer in Later Life Linked to Doubled Dementia Risk Cochlear Implant Users Had About 40% Lower Dementia Risk Over Time Only One in Five Women Start Hormone Therapy After Menopause Diagnosis Leafy Greens Linked to 11% Lower Heart Disease Risk, Meta-Analysis Finds Nudge-Based Activity Programs Linked to Better Frailty Scores and Mood in Older Adults New 'Pasta' Aging Clock Finds Drug Candidates That Rejuvenate Human Cells Recreational Runners Hit Higher VO2 Max With Less Work From Their Hearts Review Examines Gut Microbes in Preclinical Alzheimer's, With Timing Unresolved Not All Ultra-Processed Foods Are Equal, Study of 13,025 Products Finds Computer Model Finds No Single Cause of Aging Muscle's Protein Resistance Thinner Vagus Nerve Linked to Faster Heart Rate in Older Adults Five-Cohort Study Ties Consolidated Sleep and Brain Wave Coupling to Cognition Study of 18 Epigenetic Clocks Finds Most Give Unstable Readings Over Time Swinging Blood Pressure Readings Linked to Faster Cognitive Decline in Trials Aerobic Exercise Boosted Thinking Skills Most in Postmenopausal Women Protein MTR4 Falls in Aging Blood Vessel Lining Cells, Study Finds Gene Therapy Targeting a Mitochondrial Enzyme Slowed Ovarian Aging in Mice Review Proposes Gut-Brain-Fat Loop to Explain Ultra-Processed Food and Obesity Review Weighs Whether Years of Intense Endurance Training Harm Athletes' Arteries