Reading the evidence: surrogate markers and what they do not tell you

A supplement company publishes a clinical trial. Participants who took the product showed a statistically significant improvement in a specific blood marker compared to those who took a placebo. The press release describes this as proof that the supplement supports cardiovascular health. The trial is real. The biomarker change is real. The conclusion is not justified.

This gap between what a trial measured and what it claims to show is one of the most pervasive problems in supplement research. It operates through the use of surrogate markers: measurable variables that are thought to be related to health outcomes but that do not directly measure them. Understanding what surrogate markers can and cannot tell us is essential to reading any supplement trial critically.

What a surrogate marker is

A surrogate marker, sometimes called a surrogate endpoint or intermediate endpoint, is a measurement used in a clinical trial as a substitute for a clinical outcome. Clinical outcomes are things that matter directly to the person: reduced symptoms, avoided disease, improved function, longer survival. Surrogate markers are measurements thought to be related to those outcomes: a blood test result, a physiological variable, an imaging measurement, a score on a questionnaire.

The rationale for using surrogate markers in research is practical and legitimate. The outcomes we most care about often take years or decades to develop. Following a large cohort long enough to measure cardiovascular events, cancer incidence, or cognitive decline is expensive, slow, and logistically demanding. A surrogate that changes in weeks or months and that reliably tracks the clinical outcome offers a faster and cheaper way to gather evidence. When the relationship between a surrogate and its clinical outcome has been rigorously established, using the surrogate is a reasonable scientific choice.

The problem is that the relationship between a surrogate and a clinical outcome is often assumed rather than established. A biomarker may be associated with an outcome in observational studies without the intervention effect on that biomarker reliably predicting the intervention effect on the outcome. These are different things, and the distinction is not a technical nuance. It is the difference between a measurement that tells us something useful and one that tells us almost nothing about what matters.

Why correlation is not surrogacy

The most common misunderstanding about surrogate markers is the assumption that if a biomarker is associated with a clinical outcome in the population, then changing that biomarker through an intervention will change the clinical outcome correspondingly. This does not follow, and the history of medicine is marked by cases where it failed badly.

The formal criteria for surrogate validity, set out by Prentice in 1989 and developed by subsequent methodologists, require more than an association between marker and outcome. They require that the treatment effect on the surrogate reliably predicts the treatment effect on the clinical outcome. This is a substantially stronger condition. A biomarker can be a genuine risk factor, robustly associated with disease in observational data, and yet prove to be a poor surrogate when tested in intervention trials.

Homocysteine is perhaps the most instructive example available from cardiovascular research, and it is relevant directly to supplement marketing. Elevated homocysteine levels are consistently associated with increased cardiovascular risk in observational studies. Folic acid and B vitamins reliably lower homocysteine levels. The inference that B vitamin supplementation should therefore reduce cardiovascular events is superficially logical and was widely promoted in supplement marketing for years. The HOPE-2 trial, a large randomised trial published in the New England Journal of Medicine in 2006, enrolled 5,522 patients with established vascular disease. Treatment reduced homocysteine levels substantially. It had no beneficial effect on the primary outcome of cardiovascular death, myocardial infarction, and stroke. A subsequent meta-analysis of eight randomised trials involving over 37,000 individuals reached the same conclusion. The surrogate changed. The outcome did not. Lowering homocysteine through this intervention did not improve clinical outcomes, indicating that the epidemiological association did not translate into an effective intervention in this context. Whether homocysteine is causal, contributory, or a marker of other processes remains unresolved — what is clear is that changing it via B vitamins did not help.

This is not an isolated failure. A meta-epidemiological analysis reported that trials using surrogate primary endpoints overpredict treatment effects by more than 40 percent on average compared with trials measuring patient-relevant outcomes directly, though the magnitude varies by field and methodology. The gap is not occasional. It is systematic.

The spectrum from biomarker to validated surrogate

Not all biomarkers are equally unreliable as surrogate endpoints. The relevant question is always the same: has the relationship between this biomarker and this clinical outcome been rigorously validated in large intervention trials, across multiple populations and treatment contexts?

Where that validation exists, a biomarker can function as a legitimate surrogate. Systolic blood pressure is the most studied example. Decades of randomised trials across multiple drug classes have shown that reducing systolic blood pressure reduces the risk of stroke, with a reasonably consistent relationship between the magnitude of blood pressure reduction and the magnitude of stroke reduction. The relationship is not perfect, and it is more predictive for stroke than for myocardial infarction or overall mortality, but the evidence base supporting blood pressure as a surrogate for stroke outcomes is substantial.

LDL cholesterol sits in a similar position in cardiovascular medicine, though the picture is more nuanced than it is sometimes presented. Multiple statin trials have established a consistent relationship between LDL reduction and cardiovascular event reduction, supporting LDL as a reasonable surrogate for cardiovascular outcomes in trials of that drug class. Whether the same relationship holds for other mechanisms of LDL reduction has been more contested.

Most biomarkers used in supplement research do not come close to this standard. Inflammatory markers, antioxidant measures, hormonal variables, cognitive test scores, and various metabolic parameters are commonly used as primary endpoints in supplement trials without any established relationship to the clinical outcomes the marketing copy implies. A compound that raises a marker of antioxidant activity has not been shown to prevent oxidative stress-related disease. A compound that modestly reduces a measure of inflammation in a twelve-week trial has not been shown to reduce the incidence of inflammatory disease. The word "shown" carries specific meaning here: it requires trials measuring the actual outcome, not the proxy.

How supplement marketing exploits this gap

The gap between biomarker change and clinical outcome is where most supplement overclaiming lives, and it is exploited in a consistent pattern.

A mechanism is identified through preclinical research. A biomarker is found to be associated with the pathway of interest. A short trial shows that a compound changes the biomarker in the hypothesised direction. The marketing copy presents this sequence as evidence that the compound produces the clinical benefit at the end of the chain.

Each step in this sequence may be internally defensible. The mechanism may be real. The biomarker association may be genuine. The trial may be well-conducted. The conclusion is still not supported, because the critical link, whether the intervention effect on the biomarker actually predicts the intervention effect on the clinical outcome, has not been tested.

The interactive diagram below traces this pathway and shows where it can break down. It also illustrates several examples of where the assumption that changing a biomarker translates to clinical benefit has been tested and failed.

What this means for reading a supplement trial

When reading a trial that reports a positive finding, the first question to ask is what was actually measured. If the primary outcome is a biomarker, the next question is whether that biomarker has been validated as a surrogate for the clinical outcome being implied. In most supplement research, the honest answer is that it has not.

This does not mean the trial is worthless. A well-conducted trial showing that a compound changes a biomarker in a consistent direction contributes to the mechanistic picture. It may identify a signal worth following up with larger trials measuring clinical outcomes. What it does not do is establish that the compound produces the health benefit implied by the marketing.

Evidentia applies this distinction throughout the evidence library. A rating of Moderate for a biomarker change is not equivalent to a Moderate rating for a clinical outcome, and those are stated separately wherever both types of evidence exist. The distinction is not pedantic. It reflects a genuine difference in what the evidence base can support.

A note on self-reported outcomes

A specific category of surrogate deserves attention in the context of supplements targeting energy, mood, cognition, sleep, and stress: self-reported outcome measures. These are questionnaires and scales on which participants rate how they feel across various dimensions.

Self-reported outcomes are not inherently unreliable, and in some contexts, such as pain or fatigue, they are the most direct available measure of patient experience. The problem in supplement research is that they are particularly susceptible to placebo effects, expectation effects, and demand characteristics, especially in unblinded or inadequately blinded trials. A participant who knows they are taking a product marketed as an energy supplement and who pays conscious attention to their energy levels over twelve weeks is likely to report improvement regardless of whether the compound has any physiological effect.

This does not mean positive findings on self-reported outcomes should be dismissed. It means they should be interpreted in the context of blinding quality, effect size, consistency across studies, and whether the finding has been replicated in well-blinded trials. A small, unblinded trial showing improved self-reported energy scores is very different from a well-blinded, adequately powered trial showing the same.

The broader implication

The surrogate marker problem is not unique to supplements. It is a recognised challenge across clinical research, and it is the reason that regulatory agencies require evidence of clinical benefit for drug approval in most circumstances, and why validated surrogate endpoints require extensive testing before they are accepted as substitutes for clinical outcomes.

In supplement research, the regulatory bar is lower, and the pressure to demonstrate efficacy quickly with small trials measuring easy-to-change proxies is considerable. The result is a literature dominated by biomarker studies whose clinical relevance is largely assumed rather than established. Reading this literature well requires holding that assumption to account consistently, asking not just whether a marker changed but whether changing that marker has been shown to produce the clinical benefit being claimed.

The final article in this series addresses a related problem: why studies on the same supplement often appear to conflict with each other, and how to think about a body of evidence that points in multiple directions simultaneously.

Where evidence is limited or outcomes are uncertain, conclusions should be treated as provisional and subject to revision as the evidence base develops.

Key references

Prentice RL. (1989). Surrogate endpoints in clinical trials: definition and operational criteria. Statistics in Medicine, 8(4), 431–440.

Lonn E et al. for the HOPE-2 Investigators. (2006). Homocysteine lowering with folic acid and B vitamins in vascular disease. New England Journal of Medicine, 354(15), 1567–1577.

Clarke R et al. (2010). Effects of lowering homocysteine levels with B vitamins on cardiovascular disease, cancer, and cause-specific mortality: meta-analysis of 8 randomized trials involving 37,485 individuals. Archives of Internal Medicine, 170(18), 1622–1631.

Ciani O et al. (2016). Time to review the role of surrogate end points in health policy. Value in Health, 19(6), 726–735.

Prasad V, Kim C, Burotto M, Vandross A. (2015). The strength of association between surrogate end points and survival in oncology: a systematic review of trial-level meta-analyses. JAMA Internal Medicine, 175(8), 1389–1398.

For individual supplement evidence reviews, see the Evidence library on Evidentia Nutrition.