The business card from the longevity clinic is made of wood.
Not wood-colored paper, but a thin, pale rectangle with the name of a private medical practice burned into the grain. It is an effective object. Most cards are designed to be discarded; this one seems to have already survived something.
My acquaintance David uses the practice advertised on this wood carving. To try to extend his life, every three months or so he visits the longevity clinic, which is located in Pacific Heights near Billionaires Row, a three-block stretch of San Francisco’s largest mansions.
David — himself a pathologist, trained to look at tissue and decide what is signal, what is artifact, and what cannot be known from the specimen in front of him — tells me he has had his biological age measured with GrimAge, an algorithm whose title is more candid than most medical branding.
GrimAge does not examine wrinkles or count birthdays. Instead, it serves as an epigenetic clock, a mechanism to tell how much damage has occurred to the body over time.
In 1992, Arline Geronimus called the cumulative process of health deterioration weathering. Epigenetic clocks seemed to make the weathering process visible.
Epigenetic clocks are valuable research instruments: They can track age-related risk across populations and may eventually help evaluate interventions. But their conversion into personal longevity products has moved ahead of evidence that an individual score should guide treatment. David can buy a precise-looking account of his future, but until the score reliably changes a clinical decision, the number is information, not care.
David tells me how he proactively addresses the results of his tests. He tells me he receives intravenous infusions and other new therapies that fall outside normal regulations. I do not know precisely what is in the infusion bags, where they are prepared or what diagnosis they are intended to treat. Without contents or provenance, there is no intervention I can evaluate.
The market surrounding such treatments is less reluctant than I am to treat without a diagnosis. A 2025 national study of intravenous-hydration spas found offerings that included magnesium, glutathione, nicotinamide adenine dinucleotide, anti-nausea drugs and pain medicines. Every website in the sample made benefit claims; 99% supplied no references. The median treatment cost was $179.
Those findings say nothing about David’s infusions or the clinical practice he uses to dispense them. They describe the market in which an unidentified bag now sits, ready to be dispensed.
David’s interest in GrimAge is not foolish. Population studies suggest the epigenetic clocks measure something consequential. But a strong population predictor of how long someone might live is not automatically a useful personal instrument.
Epigenetic clocks began not as tests of longevity but as attempts to recover calendar age from DNA methylation — chemical tags attached to DNA that are often associated with gene regulation. GrimAge, introduced in 2019, changed the target.
A 2025 review concluded that current epigenetic clocks do not meet ordinary standards for individual clinical decision-making: A single reading can shift with laboratory methods, blood-cell composition, recent illness, stress and time of day. Also, different clocks can disagree, and no accepted treatment tells a clinician what to do about someone with a short clock or with clock results in general.
Most importantly, there is no established evidence that making one person’s GrimAge number younger makes that person live longer.
All David can do is repeat the test. He can order another clock and compare the two. If the results disagree, he can purchase an interpretation.
Leon
Leon asks for less abstract measurements. He grew up on the other side of San Francisco, in the former Navy yard called Bayview–Hunters Point, where he learned welding and construction.
The work depended on knees, hips, shoulders and hands; it rewarded a body that could bear weight and hold position. He now has degenerative arthritis. When I see him in my clinic, we discuss pain, function, and what remains possible while he waits for an orthopedic appointment.
The referral covered by his insurance runs through Zuckerberg San Francisco General, the city’s public hospital. Months have passed. But that does not establish why his appointment has not occurred; a referral can stall for many reasons.
Long waits are not foreign to the institution. In a study of patients who eventually received hip or knee replacement, 98.4% of the San Francisco General group had public insurance. The mean interval from a surgeon’s recommendation to the operation was 292 days before the pandemic and 345 days in the pandemic comparison period. The study measured operations in 2019 and 2020, not Leon’s current wait for consultation. It is evidence of a narrow bottleneck, not an explanation of one man.
A 2025 review concluded that current epigenetic clocks do not meet ordinary standards for individual clinical decision-making: A single reading can shift with laboratory methods, blood-cell composition, recent illness, stress and time of day.
His mouth has its own referral history. Leon needs dentures and has repeatedly struggled to obtain dental care in time for the word “timely” to retain much meaning.
The Southeast Family Health Center, where he receives most of his care, lists both medical and basic dental services. Coverage exists as well: California restored full adult Medi-Cal dental benefits in 2018. Yet in 2024, only 24.6% of eligible adults used a Medi-Cal dental service.
The statewide figure does not tell me how long Leon waited or why. It tells me that his difficulty is not an exotic one.
A missing tooth is not a biomarker that indicates risk or predicts mortality. It changes what a person can chew, how he speaks, whether he smiles in a room of strangers. It does not require an algorithm to become legible. The treatment is also not speculative. Dentistry knows what dentures are for.
Weathering Health
Leon’s history in Bayview begins before either referral. During the Second Great Migration in the 1940s, wartime industry brought Black families west while housing discrimination restricted where many could live.
At Hunters Point, Black workers made up more than a third of the shipyard workforce by August 1945. After World War II, the Naval Radiological Defense Laboratory used the site to study and decontaminate ships exposed during nuclear tests. The shipyard became a federal Superfund site.
Leon was not a shipyard worker. Welding and construction provide enough ordinary explanations for aching joints; age provides another. Neither his arthritis nor his dental problems can be assigned to a contaminated parcel of land.
The clocks’ uneven performance across racial and ethnic groups does not make weathering disappear; it complicates any claim that one number measures its effects equally well in everyone.
The neighborhood history matters differently. It describes the set of exposures and institutions into which an individual medical history arrives.
Those exposures resist separation because through time they became so mutually embedded in one another. A study published in 2025, based on a 2020–21 survey of 125 residents of affordable housing in Hunters Point, found that 21% reported a chronic respiratory condition and 23% a cardiometabolic condition. Only 24% said their household income met monthly expenses. Food insecurity, discrimination and exposure to violence appeared in the same community health assessment.
The survey was small, self-reported and conducted during the pandemic. It cannot identify the cause of a resident’s disease. Yet its value is almost the opposite: It shows how rarely one cause arrives alone.
In 1992, Arline Geronimus called the cumulative process of health deterioration weathering. The hypothesis began as an explanation for the earlier deterioration of health among Black women exposed to repeated social and economic disadvantage. Fourteen years later, Geronimus’ group examined cardiovascular, metabolic and inflammatory markers, cumulative effects that are often called allostatic load. Black adults had higher allostatic load across age groups, especially from 35 to 64.
Income did not produce a clean health gradient: Nonpoor Black women had the second-highest probability of high allostatic load, after poor Black women.
Epigenetic Clocks Make Weathering Visible
Then epigenetic clocks seemed to make the weathering process visible.
Epigenetic clocks began not as tests of longevity but as attempts to recover calendar age from DNA methylation — chemical tags attached to DNA that are often associated with gene regulation. In 2013, Gregory Hannum’s group built a blood-based model from 71 methylation sites, while Steve Horvath reported a 353-site clock that worked across many tissues.
Both models were trained chiefly to predict how old an organ donor was. Their accuracy made the clocks attractive research tools: If a person’s predicted age ran ahead of the calendar, perhaps the difference recorded some aspect of aging.
An epigenetic clock is not a neutral gauge laid against the body. It is an algorithm taught by particular bodies and built from particular methylation sites, each chosen because it predicted age or mortality in the data used to construct the clock.
GrimAge, introduced in 2019, changed the target. Its developers first built methylation surrogates for smoking exposure and selected plasma proteins, then combined them with age and genetic sex in a model trained on time to death. The resulting score, expressed in years, predicted mortality and several age-related outcomes across cohorts.
But the same instrument David purchases could apparently detect the residue of everyday stresses from living in a neighborhood. A 2023 study examined residential segregation, neighborhood poverty, and four DNA methylation measures among 1,102 older adults. For non-Hispanic Black participants, a one-standard-deviation increase in segregation was associated with 0.42 years of GrimAge acceleration. The association appeared in GrimAge, not in the other three clocks, and was strongest in high-poverty tracts.
The newer studies complicate that symmetry. In March 2026, investigators reported repeated methylation measurements from 699 adults in the Italian InCHIANTI cohort followed for as long as 24 years. Baseline values and changes in several clocks predicted mortality. The clocks were detecting something that mattered. The evidence, however, came from a single Italian cohort.
In August, an eLife study evaluated several clocks in African American, Hispanic, and white cohorts. The clocks predicted age less accurately in admixed groups — groups with more homogenous genetics — and did not consistently identify accelerated aging in admixed patients with Alzheimer’s disease.
In one analysis, the Horvath clock’s correlation with chronological age was 0.72 in the white cohort, 0.51 in the African American cohort, and 0.45 in the Puerto Rican cohort. The result persisted in more than 2,500 additional participants. The paper’s title dispensed with euphemism: “Methylation clocks fail to generalize across genetically admixed individuals.”
The clocks’ uneven performance across racial and ethnic groups does not make weathering disappear; it complicates any claim that one number measures its effects equally well in everyone.
When the population changes, the accuracy — and perhaps the meaning — of the number the epigenetic clock produces can change as well.
An epigenetic clock is not a neutral gauge laid against the body. It is an algorithm taught by particular bodies and built from particular methylation sites, each chosen because it predicted age or mortality in the data used to construct the clock.
When the population changes, the accuracy — and perhaps the meaning — of the number the epigenetic clock produces can change as well. The clock therefore carries the history of who was measured when it learned to tell time. A test can be reproducible enough to package and sell before it has been shown to speak with equal accuracy, or equal clinical meaning, to everyone tested.
The latest intervention evidence is less orderly still. In the Baby’s First Years trial, mothers with low incomes were randomly assigned to receive either $333 or $20 each month after a child’s birth. At age four, the children in the higher-cash group showed a 0.17-standard-deviation slower DunedinPACE, another methylation measure. Researchers detected no effect on the children’s GrimAge or PhenoAge, on measured cognitive and health outcomes, or on the mothers’ epigenetic indices. Whether the difference will persist or matter clinically is unknown.
Money changed one clock in the children. It changed none of the clocks in their mothers, suggesting that children benefited the most from financial investments. Biology has timing. Repair may not be the inverse of injury.
Small, Practical Changes Can Have a Modest Effect on Longevity
My colleague Atheendar Venkataramani, a physician and economist, has studied a larger measure: economic opportunity, meaning the chance that a child born poor will rise above his parents’ position.
In 1,559 counties, his group found that greater mobility was associated with longer life among people in the lowest income quartile. A one-standard-deviation difference in mobility corresponded to 0.38 additional years of life expectancy for poor men and 0.29 for poor women. If every county had matched the highest observed mobility, the model projected that the longevity gap between the richest and poorest residents would narrow by about one-fifth.
The study was cross-sectional and ecological. It could not establish that mobility produced longevity, much less specify the policy that would do so. But it suggested a different unit of health — not income at one moment, but the distance a child can reasonably expect to travel.
Leon can keep calling about orthopedics and asking about dentures. I still cannot tell Leon when orthopedics will call. I can tell him, now, that the date is not written in his epigenetic clock.
One federal experiment changed that distance by lottery. From 1994 through 1998, the Moving to Opportunity program randomly offered families in very high-poverty public housing a voucher and counseling to move to a lower-poverty neighborhood. Ten to fifteen years later, the women offered that opportunity were 4.61 percentage points less likely to have a body-mass index of at least 35 and 4.31 points less likely to have a diabetic-range hemoglobin A1c than women in the control group.
Only 48% used the restricted voucher. The neighborhood differences diminished with time, the health effects were modest, and the mechanism remained unclear. Routine medical access did not improve. Still, the experiment made one fact difficult to dismiss: Bodies did not merely preserve the past. They responded to a changed environment.
That finding has an uncomfortable politics. A voucher can turn neighborhood repair into an invitation to leave. The choice need not be between remaining in a neglected place and escaping it. Opportunity can also mean making the institutions already present work better.
There are less dramatic experiments in this. When New York City’s public hospital system introduced electronic specialist consultations, 13% of referrals were resolved without an office visit. Among patients who still needed one, the proportion with an appointment scheduled rose from 66.5 to 82.3%, and the average wait fell by 8.2 days. The proportion who actually completed a specialist visit within 90 days did not improve. The queue moved; it did not disappear.
Dental coverage has proved similarly real and incomplete. In a study using state changes in adult Medicaid benefits, coverage was associated with a 12.9-percentage-point increase in annual dental visits and a 9.5-point reduction in untreated cavities. The gains were larger when Medicaid paid dentists more. An insurance card was not sufficient. It was not inert, either.
David can choose whether to test again, whether to continue an infusion, and whose interpretation to buy. Leon can keep calling about orthopedics and asking about dentures. I can renew medication, document the limitation, and send the referral again. But a referral queue is an arrangement, not a biomarker. It can be staffed, shortened, and measured.
I still cannot tell Leon when orthopedics will call. I can tell him, now, that the date is not written in his epigenetic clock.
David and Leon are pseudonyms. Both agreed to be portrayed anonymously; nonessential identifying details have been changed.
