Surrogate endpoints in longevity trials, and why a moved marker is not a longer life

Why surrogates exist at all

A trial that waits for deaths or for disease onset in a healthy population must run for a long time and enrol a very large number of participants. Almost nobody can fund that design routinely, so trials substitute a measurement expected to sit on the causal path between treatment and outcome. That substitute is the surrogate endpoint, and the entire logic of the trial rests on the assumption that moving it moves the thing that matters.

The assumption is sometimes correct, which is why the practice persists in regulated medicine under carefully specified conditions. It is also sometimes wrong in ways that become visible only when somebody finally runs the long and expensive trial.

The conditions a surrogate has to meet

It is not enough for a surrogate to correlate with the outcome, because correlation is entirely compatible with the surrogate being a bystander. The surrogate has to lie on the causal path, and the treatment effect on the outcome must run through it rather than around it. Establishing that requires evidence from trials where both the surrogate and the outcome were measured, which is precisely the expensive design being avoided.

As a result, many surrogates in wide use were adopted on plausibility and observational association rather than on formal validation. The strength of that underlying evidence varies enormously between fields and is almost never stated when a result is summarised.

How surrogates fail in practice

The classic failure is a treatment that moves a marker in the expected direction while harming patients through a mechanism unrelated to the marker. A second failure occurs when the marker sits downstream of the disease rather than upstream, so changing it alters the reading without altering the process. A third arises when a drug affects the measurement itself, through assay interference or through a physiological change that biases the test.

Each of these has occurred in cardiovascular and cancer medicine, which is why regulators now demand outcome data for many approvals. The failures were not obvious in advance to the people running those trials, and that is the part genuinely worth remembering.

Why longevity work is particularly exposed

The outcome of interest here sits decades away, so reliance on surrogates is not a shortcut but a structural necessity of the subject. The available surrogates, including inflammatory markers, methylation clocks and functional tests, have been validated far less thoroughly than those used in cardiology. Commercial pressure compounds the problem, because a product can be marketed on a moved marker long before anyone could demonstrate an outcome.

The result is a literature in which biomarker improvements are abundant while hard outcome evidence is close to absent. That imbalance is a feature of the timescale rather than a failing of the researchers, and it deserves stating plainly rather than papering over.

Reading a trial that used one

The first thing to establish is whether the reported endpoint was an outcome that matters to a person or a measurement standing in for one. The second is what evidence exists that the substitute predicts the outcome under this specific treatment rather than in some general sense. A marker that responds to many unrelated interventions is often a broad indicator of state rather than a specific link in a causal chain.

Trials also differ in whether the surrogate was specified in advance or chosen after the data were seen, and that distinction is decisive. Where a decision about treatment turns on any of this, the interpretation belongs in a conversation with a clinician who can see the whole picture.