Surrogate vs Clinical Endpoints

The number that looks like a result but isn't

Forty-three percent of drug approvals by the US FDA between 2008 and 2012 were based on surrogate endpoints alone. That figure came out of a 2019 analysis in JAMA Internal Medicine, and it stuck with me. Not because surrogate endpoints are inherently bad science , they're not , but because that gap between "we measured a proxy" and "we improved something real in patients' lives" is where a lot of confident claims quietly fall apart.

If you read research into cannabinoids, phytomedicines, or peptides, you will hit this problem constantly. A study finds that a compound shifts a biomarker in a particular direction. The press release says the result is "promising." And technically, that's true. But promising compared to what, and measured how? That's the question worth asking every single time.

What surrogate and clinical endpoints actually mean

A clinical endpoint is a direct measure of how a patient feels, functions, or survives. Did the person live longer? Walk further? Report less disruption to their daily life? These are the outcomes that matter to the person enrolled in the trial.

A surrogate endpoint is a stand-in , a biological or physiological marker assumed to correlate with those real-world outcomes. Blood pressure instead of stroke incidence. Tumour shrinkage instead of survival. A receptor-binding affinity score instead of any observable patient change. The logic is: if we know X correlates with Y, and we move X, we've probably moved Y.

Sometimes that logic holds up perfectly well. Sometimes it doesn't hold at all.

The classic cautionary tale in clinical pharmacology is the CAST trial (Cardiac Arrhythmia Suppression Trial), published in the New England Journal of Medicine in 1989. Researchers knew that post-heart-attack arrhythmias correlated with sudden death. They found drugs that suppressed the arrhythmias beautifully; the surrogate looked great. Mortality went up. The drugs were killing people. The surrogate endpoint had moved in the right direction while the clinical outcome moved in the wrong one.

That's an extreme example, but the underlying problem, that a surrogate can be statistically associated with an outcome without being causally on that pathway, is genuinely common.

Why researchers use surrogate endpoints anyway

Honestly, the reasons are mostly practical and sometimes defensible. Clinical endpoints take time. If you want to know whether a compound extends survival in a slowly progressing condition, you might need a decade-long trial. Surrogate endpoints can be measured in weeks or months, which makes early-phase and proof-of-concept research feasible.

They're also cheaper, easier to standardise, and less dependent on subjective patient reporting. A blood level is a blood level. Whether someone's quality of life improved is harder to quantify and varies across populations and measurement tools.

In exploratory research; the kind that asks "is there something here worth investigating further?", surrogates are often the only reasonable option. The problem isn't using them. The problem is treating them as if they've answered the clinical question when they've only answered a related, smaller question.

Where this matters in cannabinoid and phytomedicine research

A lot of early cannabinoid research is mechanistic: a compound interacts with receptors in the endocannabinoid system, or modulates a signalling pathway in a cell line, or shifts a cytokine level in an animal model. These are all, in endpoint terminology, surrogates, or pre-surrogate, really. They're useful for building a mechanistic picture. They don't tell us what happens in humans at the clinical level.

For something like cannabidiol, the research landscape has matured enough that some genuine clinical-endpoint data now exists in specific, narrow contexts. But for many minor cannabinoids; cannabigerol, THCV, and others, we're largely still in the mechanistic and animal-model phase. That's not a scandal; it's where early research lives. The issue is when that early data gets translated into strong efficacy claims before the clinical work has been done.

The same applies to bioavailability studies. Showing that a formulation achieves higher plasma concentrations than a previous one is a surrogate finding. It may be relevant. But higher plasma concentration is not, by itself, a clinical outcome. The gap between pharmacokinetic data and patient benefit is a real one, and it's often glossed over in product-adjacent content.

For peptide research, and I've spent a fair bit of time going through the published literature here; the surrogate-to-clinical gap is particularly wide in some areas. Preclinical data showing a peptide affects a particular pathway in rodents is interesting. It's not evidence of clinical benefit in humans. Worth knowing the difference before you read anything claiming otherwise.

Validated vs non-validated surrogate endpoints

Not all surrogates are equal. Regulatory bodies like the TGA, and international equivalents, distinguish between validated and non-validated surrogate endpoints.

A validated surrogate has strong epidemiological and biological evidence that changes in the marker reliably predict changes in the clinical outcome. HbA1c as a surrogate for long-term glycaemic control is one of the better-established examples. It's not perfect, but decades of data support the link.

A non-validated surrogate is one where that causal relationship hasn't been robustly established, it might be plausible, biologically interesting, even widely used in early research, but it hasn't been proven to predict real-world clinical change. A lot of the biomarker endpoints you'll see in phytomedicine and cannabinoid trials fall into this category. The TGA's framework for evaluating clinical evidence, referenced in the Australian Regulatory Guidelines for Prescription Medicines, explicitly considers the validation status of surrogate endpoints when assessing the strength of an application.

I'd argue this distinction is underexplained in popular science writing, people see "statistically significant result" and assume the hard work is done, when really the question of whether the endpoint itself is validated is equally important.

How to spot the gap when you're reading a study

A few things to look for. First, check what the primary outcome actually was. It should be declared upfront in the methods section, not buried in the results. If the headline result is a secondary or exploratory endpoint; common in smaller trials, treat it accordingly.

Second, ask whether the endpoint is patient-reported, observer-reported, or a laboratory measure. All three have legitimate uses, but they answer different questions. A receptor-binding study answers a biochemistry question. A validated quality-of-life instrument answers a patient-experience question. These aren't interchangeable.

Third, look at the trial phase. Phase I is primarily safety. Phase II is often dose-finding and early efficacy signals, frequently using surrogates. Phase III is where clinical endpoints and statistical power to detect meaningful differences are expected. A Phase II result reported as though it were Phase III evidence is a common inflation in press coverage.

Finally, and this is worth doing for any study that interests you; look up whether the authors themselves acknowledge the surrogate limitation. Good researchers do. They write something like "these findings warrant clinical trials with patient-centred endpoints." That sentence is them being honest about where the evidence currently sits. Take it seriously.

Between sea swims last week I went back through a few recent minor-cannabinoid papers with this lens, and honestly, the ones that acknowledged their surrogate limitations clearly were also the ones with the tightest methodology overall. That's not a coincidence. Epistemic honesty tends to cluster.

The hierarchy isn't about dismissing early research

None of this means surrogate-endpoint research is worthless. It means it occupies a specific, early place in the evidence hierarchy, and reading it well means understanding that place. Mechanistic data is genuinely useful for understanding how compounds interact with pathways like the endocannabinoid system. Biomarker shifts are useful for screening candidate compounds. Animal models generate hypotheses worth testing in humans.

The hard yards are the long clinical trials that measure what actually happens to people. Those take time and resources, and for many compounds in this space, that work is still underway. The research literacy skill here is holding both things at once: this early finding is interesting, and it hasn't yet answered the clinical question.

That's not scepticism for its own sake. That's just how evidence accumulates.

Sources

• Surrogate Endpoints in Clinical Trials: Are They Valid?, NCBI / JAMA Internal Medicine
• Australian Regulatory Guidelines for Prescription Medicines; TGA
• Evaluation of Biomarkers and Surrogate Endpoints in Chronic Disease, NCBI / National Academies Press
• Preliminary Report: Effect of Encainide and Flecainide on Mortality (CAST Trial), New England Journal of Medicine

, Hannah Bui, Evidence & Research-Literacy Writer