A birthday is wonderfully unambiguous. Once a person turns 60, no recovery period can make them chronologically 59 again. A biological-age estimate is a different kind of number. It is calculated from molecular patterns that tend to vary with ageing, health and exposure, and those patterns may be far less steady than the calendar.

That is the genuinely counterintuitive result of a 2023 Cell Metabolism study. Across experiments involving mice and analyses of human blood, several biological-age markers rose under severe physiological stress and later fell as the stress resolved.

The timescale could be startlingly short. In the clearest human example, some estimates rose by the morning after emergency surgery and returned toward their earlier level four to seven days later.

That does not mean nine patients became old overnight and young again before leaving hospital. It means something subtler: part of what researchers call biological age appears to include a reversible state of physiological strain.

Biological age is an estimate, not a hidden birthday

The term “biological age” sounds as though every body contains one true number waiting to be read. It does not. Researchers have developed many clocks, trained on different biological measurements and designed for different purposes.

The best known use DNA methylation, chemical tags attached to DNA that can affect how genes are used without changing the genetic sequence. An algorithm combines methylation at selected sites into a score. Other clocks draw on gene activity, metabolites, proteins, organ function or clinical measurements.

First-generation methylation clocks were built largely to predict chronological age. Later clocks such as PhenoAge and GrimAge incorporated patterns associated with health and mortality. DunedinPACE was designed to estimate the pace of ageing rather than an age in years. As the US National Institute on Aging explains, the generations are related but not interchangeable.

A number produced by one clock therefore need not match a number from another. Nor does movement in a blood-based clock prove that every tissue, organ and form of molecular damage moved in concert.

Nine emergency operations revealed the fastest swing

The human surgery analysis included nine older adults who required emergency repair of a fractured hip. Blood was collected before surgery, the next morning, and again four to seven days later. Their mean age was 77.9 years.

PhenoAge, GrimAge and DunedinPACE rose sharply after the operation, then returned toward their preoperative readings at the third sample. The NIA’s account of the research describes the pattern as an increase the morning after surgery followed by a return to pre-surgery levels within four to seven days.

That compact timeline is the strongest support for saying these biomarkers can move in days. The result is still based on nine patients, without an unoperated control group experiencing comparable trauma. It needs replication in larger, deliberately designed studies.

The pattern also did not appear after every kind of surgery. The researchers examined elective colorectal operations and did not see the same consistent rise. A planned procedure in a screened and prepared patient is not biologically equivalent to a fracture, emergency admission and urgent operation arriving together.

Pain, tissue injury, inflammation, anaesthesia, disrupted sleep and the stress response could all contribute. The study could not separate their individual effects. What it captured was the combined physiological upheaval and the early recovery from it.

Pregnancy traced a slower rise and recovery

Pregnancy is not an illness, but it places sustained demands on cardiovascular, metabolic, endocrine and immune systems. The 2023 team analysed several existing methylation datasets collected across human pregnancy.

The precise pattern depended on the cohort and clock. In 54 US participants sampled in one trimester, GrimAge rose between the first and third trimesters and DunedinPACE rose across successive trimesters, while PhenoAge did not change significantly. Other longitudinal datasets also showed increases on some second-generation measures as delivery approached.

One dataset followed 14 women through six weeks after birth. PhenoAge showed a trend upward around delivery followed by a significant decline postpartum. That was suggestive, but 14 people are too few to treat as a final description of pregnancy.

A larger follow-up published in 2024 strengthened the basic observation. It used blood from 119 women in early, middle and late pregnancy; 68 also provided a sample about three months after delivery. Biological-age estimates rose during pregnancy, then declined after birth across all five epigenetic biomarkers studied.

The size of that apparent reversal varied by clock and maternal circumstances. The study also associated postpartum readings with pre-pregnancy body mass index and breastfeeding, but it was observational. It cannot turn either factor into a prescription or prove that one caused the clock change.

It is safer to say that pregnancy and postpartum recovery alter age-related methylation patterns than to say childbirth makes someone younger. Pregnancy can carry real and sometimes lasting health consequences even when a selected biomarker later falls.

Severe COVID provided a more complicated case

The researchers next followed 29 people who had COVID-19, entered intensive care and survived. The group comprised 10 women and 19 men, each providing as many as four blood samples from near ICU admission through at least seven days after ICU discharge.

The first sample was usually taken after critical illness had already begun. There was no healthy pre-infection measurement, so the analysis was better suited to seeing recovery than to measuring how far each person had moved from their own normal baseline.

DunedinPACE was already about 25 per cent above its reference pace near ICU admission in both sexes. It declined after discharge, though not completely to baseline. Other clocks produced results that differed by sex and sampling point, and the older first-generation clocks found no significant change.

This is an important restraint on the headline. Severe COVID did not make every biological clock rise and fall neatly in every patient. The result was clock-specific and sex-specific, and it came only from people who survived intensive care.

The mouse work showed the pattern beyond methylation

The study did not rest entirely on repurposed human datasets. In one experiment, researchers surgically joined young and old mice so that they shared circulation. The young mice acquired older-looking molecular profiles while connected to the older animals. After separation and recovery, those profiles moved back toward their prior state.

The shifts appeared not only in DNA methylation but also in patterns of gene activity and metabolites. That agreement across different molecular layers makes a purely technical quirk less likely, although a mouse experiment cannot establish what the same process means for human health or lifespan.

Pregnancy in mice produced another rise-and-recovery pattern. Together, the experiments support a broad biological idea: acute conditions can push age-associated molecular systems away from their usual state, and some of those systems can reset when the condition ends.

Why a biological clock might run backwards

Acute stress changes almost everything that a blood sample can reveal. Hormones surge. Inflammatory signals rise. Energy use shifts. Gene expression changes, and the proportions of immune-cell types circulating in blood can change as well.

A clock trained partly on health-related or mortality-related patterns may register this disturbed state. When inflammation subsides, metabolism settles and blood-cell populations normalise, the same calculation may move in the other direction.

Changing cell composition is a particularly important concern because methylation is cell-specific. The researchers estimated blood-cell proportions and concluded that composition alone was unlikely to explain the pregnancy and COVID findings. Even so, indirect adjustment is not the same as measuring each purified cell type, and short-term immune changes can still complicate interpretation.

This helps explain why the newer clocks moved while first-generation clocks generally did not. A tool trained mainly to estimate years since birth should be relatively insensitive to a four-day recovery. A tool incorporating mortality-linked proteins, smoking exposure or physiological decline may respond more readily to acute illness.

Previous clock stories need the same caution

ScienceBlog recently examined an eight-week exercise study in which a 1.4-year improvement appeared only in the eight women with the highest initial age acceleration. The other ten participants did not show that change, and the study had no non-exercising control group.

A separate ScienceBlog report found that purpose in life tracked some health-oriented clocks but not nine clocks trained on chronological age. That cross-sectional association could not establish that purpose itself slowed ageing.

Those results are not contradictions. They show why “the biological clock” is misleading shorthand. Each model selects a particular set of signals, and each study design supports a limited kind of conclusion.

The current stress-and-recovery paper offers a related warning. A measurement made immediately after surgery may capture an acute peak. A sample taken a week later may capture recovery. Neither necessarily describes the person’s long-term trajectory.

What this research does not establish

No participant in these human datasets was randomly assigned to surgery, pregnancy or severe infection. The cohorts were small or assembled for other research questions. They did not test whether a temporary clock rise caused later disease, shortened life, or disappeared from every tissue.

The study also did not show that ordinary psychological stress produces the same movement. Its human examples involved major physiological events: trauma surgery, pregnancy and critical illness. Stretching the findings to a difficult work week would go beyond the evidence.

Nor does a falling score validate a treatment as an anti-ageing therapy. In the COVID cohort, patients who had received the anti-inflammatory drug tocilizumab showed greater recovery on several clocks, but treatment was not randomised and the groups were small. The result was exploratory, not clinical guidance.

Most importantly, reversible biomarkers do not make accumulated damage disappear. A person may recover from an acute molecular disturbance while retaining scar tissue, lost function or other lasting consequences. A temporary state and a long-term ageing process can coexist.

A more realistic picture of ageing

Ageing can still be understood as a long-term rise in damage and vulnerability. The new element is that a biomarker tracing that rise may also contain short-lived excursions caused by the body’s present condition.

That makes recovery scientifically interesting in its own right. The ability to return toward a previous molecular state may be one component of resilience, and repeated failure to recover could eventually matter more than any single peak. Those possibilities remain hypotheses to test, not conclusions already established.

For researchers, the immediate lesson is practical. Timing, recent illness, surgery, pregnancy and recovery status can all affect how an ageing clock should be interpreted. Studies of supposed rejuvenation need repeated measurements, suitable controls and evidence that a lower score corresponds to meaningful health outcomes.

The deepest change is conceptual. Biological age may not be a hidden odometer counting irreversible wear. It may look more like a long upward trend overlaid with peaks and recoveries, observed through several imperfect instruments.

The calendar still moves one way. Some of the physiology measured beneath it does not.