We Keep Discovering Biomarkers. We Keep Failing to Build Them.

This is my personal mission and process of learning in public to solve chronic pain. Always open to feedback, discussion, collabs, etc. Please, never hesitate to reach out.

In my previous essay, I surveyed the landscape of objective pain measurement—the approaches that have been tried, the signals that have been found, and the frustrating gap between promising science and clinical tools that actually exist. Despite decades of work, no validated composite biomarker for chronic pain is routinely used in clinical practice.

There are still many technical limitations to objective measures of chronic pain, but I argue here that the real bottlenecks are no longer scientific, but structural.

The way biomarkers get built, who pays for them, and the incentives that drive most biomedical innovation are insufficient. This piece breaks down why chronic pain biomarkers, and biomarkers more broadly, are still not in the hands of clinicians, scientists, and, most importantly, patients, despite years of promising science.

A biomarker is most useful when everyone uses it.

LDL and HDL are biomarkers and a medical standard. Nobody really “owns LDL” since it’s just a molecule you’re measuring in your body. But everyone uses it, and acknowledges it as a tool to understand our health. It produces immense societal and economic value, but very little of it is captured by a private company because they don’t “sell” it to you directly. A drug, by contrast, is proprietary. If you develop a successful painkiller, you own it, you sell it, and you profit from it. The incentive to invest is therefore straightforward.

Okay, so if biomarkers don’t produce much profit, what if we made it proprietary and just offer it in a private clinic as a “boutique” option?

You can, but if only one company uses it, it by definition is not a standard. With time, it creates mistrust by the broader medical and scientific community, because if it really was so profoundly helpful, why wouldn’t it be available to everyone? Without mass adoption, it can easily fall into the category of “premium Beverly Hills doctor service” and raise eyebrows.

But making it broadly available means it has to be accepted by clinicians who diagnose patients, by drug developers who design clinical trials, by regulators who approve new therapeutics, and by payers who decide what gets reimbursed. For trial stratifiers, adoption is narrower but still requires cross-sponsor trust, cross-site reproducibility, and standardisation. This is a high bar, and an expensive one at that. This takes around 3-5 years of consistent funding at the $20M+ range, excluding discovery R&D.

This creates a paradox: by the time a biomarker is universally accepted as useful, it becomes harder to own, so private value capture collapses even as the societal value rises. Yet the societal value is enormous. Economist Heidi Williams has estimated the net present value of better cancer biomarkers alone to be over $2 trillion. But that value mostly accrues to patients, to healthcare systems, to drug developers who can now run better trials, not solely those that develop it.

This is the single largest reason why most biomarker development does not happen. In startup speak, the valley of death is incredibly, incredibly, long, with very little payout at the end for private investors. The traditional private business model simply doesn’t work.

A biomarker is ultimately a statistical test that helps make a specific decision. The problem with most statistical tests is not that they don’t work at all—it’s that they don’t work consistently enough.

To prove that a test works consistently across different contexts, you need to prove it over large datasets. A machine learning classifier that distinguishes chronic pain patients from healthy controls with 85% accuracy in one lab, using one scanner, with one patient population, is promising, but it’s not a biomarker.

To be a biomarker, that classifier needs to maintain its performance across different labs, different scanners, different patient populations, different sites, and different clinical contexts. This requires large sample sizes to train the model and even larger out-of-sample datasets to validate it.

The problem is that those “samples” are patients. Every additional patient enrolled means IRB and ethics approval, clinical overhead, and regulatory compliance, all adding up with hundreds to thousands of dollars per patient. Each site added comes with protocol harmonisation, data standardisation, quality control, and coordination. Broad, coordinated infrastructure at this level rarely exists.

After all, this would look like an open medical centre, with patients waiting on a list, with ethics approval that took a whole year to obtain, and independent data contractors paid and ready to start. There is very little economic incentive for this to come together in a “business as usual” world.

For early-stage biotech companies or academic labs, building this all from scratch for a single biomarker is incredibly difficult. This general lack of iterative clinical infrastructure (aka, a standing, patient-first “clinical testbed”) where new biomarkers can be quickly and safely tested and refined, is one of the central bottlenecks in translational medicine. This of course is not just for pain biomarkers, but for all biomarkers.

To date, we’ve discovered many promising biomarkers in the lab, but most have been left unvalidated due to these structural, infrastructure-level bottlenecks.

Suppose, against the odds, someone does validate a composite pain biomarker. They publish it in Nature. It shows robust performance across multiple sites and populations. Does this spark mass clinical adoption?

Usually, no.

Biomarkers suffer from a failure of adoption problem. Clinical and research behaviour does not change because of a high-impact publication. For a biomarker to be adopted by clinicians, drug developers, and regulators, it needs more than scientific evidence, but clinical-grade statistical evidence, generated within a specific “context of use,” demonstrating that this tool is predictive and useful for a very specific clinical decision. This is what biomarker qualification pathways in the FDA and MHRA, for example, require.

The “final exam” of biomarkers is likely not a paper, but regulatory submission to a federal institution with in-person evidence that the biomarker meaningfully improves trial success for a drug trial in a specific disease.

But designing such clear “final exams” from the beginning and building an entire program backward from that endpoint is hard. For an early-stage biotech company or an academic lab, this kind of end-to-end program design is not in their realm to scope, pitch for, and execute.

Successful biomarker development requires:

• Pre-competitive coordination across academic labs, clinical sites, device companies, drug developers, and regulators.

• Long time horizons—four to six years minimum from discovery through clinical validation, mostly because clinical trial data takes 8-16 months to get back.

• Large patient cohorts enrolled across multiple sites under the same protocols, for powered statistics.

• Regulator engagement from day one—because the data regulators want to see may not be the data you generate.

• Multiple shots-on-goal—because there may be several ways to measure the same health outcome.

This combination of tools are mostly unfundable by venture capital. Traditional NIH grants can fund the early work like academic discovery studies, but they do not fund the kind of integrated, milestone-driven, multi-institutional program required to take a biomarker from “promising in-lab” to clinical validation and regulatory qualification. The NIH HEAL Initiative has made important investments in pain biomarker discovery and validation around opioid addiction, but HEAL’s grants are understandably not concentrated in a single coordinated program.

What is missing is therefore the coordination layer—the entity that creates pre-competitive partnerships bringing clinics and academic medical centres together with biomarker and device companies, drug developers, and regulatory agencies under unified governance. A “general manager,” if you will.

The role of philanthropy is to fund the pre-competitive R&D that unlocks downstream funding, often by private markets.

An effective program must fund across the technology readiness spectrum, mostly staying in the Pasteur’s Quadrant, using stage-gated milestone structures (like those used in ARPA-style programmes) that down-select from several approaches to a select few in a “competition” style, bringing the final one or two through to the final exam stage. All this while engaging regulators from day one, to make sure the data you generate is what they actually want to see.

This is what impact philanthropy is capable of. It can fund shared infrastructure and pre-competitive collaboration, take coordinated, long-horizon bets that neither government grants nor venture capital are structured to make at first. Later, once the TRL of the product (biomarker) goes up, so does its fundability and likelihood of follow-on private funding.

The question that remains is where to start. Not all chronic pain conditions are equally suited to be the proving ground for a composite biomarker programme. The ideal first target would be a condition where the clinical need is acute, where the structure-pain discordance is severe enough that objective measurement would be transformative, and where the patient population is large and motivated to support the clinical recruitment required for validation.

Ultimately, I believe identifying a chronic pain biomarker in even just one disease would begin the transformation of multiple chronic pain conditions making the end of chronic pain inevitable.