RWE in the era of N-of-1 medicines & ultra targeted therapies

If drug development in the 20th century was a factory line for blockbusters, the 21st century is rapidly becoming a tailor’s atelier. A child with a lethal urea-cycle defect receives a base-editing construct custom-built for their mutation. Regulators review the file in days. There is no control arm, no forest plot, no Kaplan–Meier curve. Yet the therapy may still earn a place on the market with a substantial price tag; the story telling must be convincing though.

The question is no longer whether such bespoke and ultra-targeted therapies will arrive. They are here. The harder question is what sort of real-world evidence (RWE) will satisfy regulators and payers once the first patient has been treated, the headlines have faded and the invoices begin to land. “Trust us, it worked for K.J.” will not do.

From blockbuster trials to plausible mechanisms

A recent Sounding Board in The New England Journal of Medicine by Prasad and Makary sketches what it calls a “plausible mechanism” pathway: a regulatory route for highly personalised therapies where a classic randomized trial is infeasible. The basic deal is straightforward; sponsors may come to market on the back of a tight chain of biological plausibility and carefully documented early experience in a handful of patients – but they must then live with demanding post-marketing obligations built on RWE.

The article lays out the ingredients. First, a well-defined molecular abnormality, not a fuzzy syndrome. Second, a product that directly targets that abnormality. Third, a solid natural-history picture of what happens without treatment. Fourth, evidence that the target has actually been hit – in vitro, in animals or, where possible, in patient tissue. Finally, sustained improvements in clinical course sufficient to rule out regression to the mean or random fluctuation.

But even if all that holds, the authors envisage that sponsors will be “tasked, as a postmarketing commitment, with collecting real-world evidence.” This is critical, it will not be just on durability of benefit, but on off-target edits, growth and development, and unanticipated safety signals. The tolerance for risk will be explicitly tied to the severity of the underlying disease and the availability of alternatives.

The message is unmistakable. For N-of-1 and ultra-targeted therapies, the randomized trial is being partially replaced by a lifelong evidence-collection programme in which every treated patient is enrolled. Marketing authorisation becomes less of an endpoint, more of a checkpoint.

Payers: show me the counterfactual

If regulators are grappling with how little pre-market data they can live with, payers are wrestling with a different problem: how much post-market evidence they can demand without choking off innovation. Nowhere is that tension sharper than in the US Medicare system, which has acquired new powers to negotiate drug prices.

A recent JAMA viewpoint by Carey and Maciejewski on RWE for Medicare price negotiations spells out what America’s single largest payer thinks “good” evidence looks like for high-expenditure drugs. The Centers for Medicare & Medicaid Services (CMS) explicitly flag RWE as a crucial input to determining the health value of medicines whose pivotal trials are old, small or unrepresentative of today’s beneficiaries.

CMS’s guidance makes three notable moves. First, it encourages rigorous RWE studies built on claims or electronic-health-record data to show how a drug actually performs in the Medicare population; including off-label use. These studies are not just about the primary endpoint from the original trial; CMS is particularly interested in long-term outcomes and medical-cost offsets, such as reduced hospitalisations.

Second, the agency urges researchers to focus on direct measures of health or function – activities of daily living, life-years – rather than intermediate surrogates, and to do so using transparent methods drawn from target trial emulation and modern causal inference. The ideal RWE study, in CMS’s eyes, behaves like a randomized trial run in claims data: clear eligibility, explicit time zero, well-defined comparators and prespecified outcomes.

Third, CMS signals that RWE may matter even more for off-label indications than for labelled ones, because the latter often have some trial data and the former typically do not. For drugs whose use has drifted far from their pivotal studies – a likely fate for many ultra-targeted therapies as clinicians experiment – Medicare is effectively saying: if you want high prices across all these uses, bring us data.

So what does “good” RWE look like for N-of-1 therapies?

Put the regulatory and payer perspectives together, and a picture begins to emerge. RWE for N-of-1 and ultra-targeted therapies will need to do four things at once:

1. Anchor itself in biology and natural history. The bespoke era is not an excuse to abandon rigor. An N-of-1 product should sit on a platform with a shared mechanistic rationale and harmonised outcome measures across patients and mutations. Natural-history cohorts, which often must be created long before the first edit or infusion, provide the backbone for external controls.

2. Emulate a trial, even when you cannot randomise. Every treated patient can still enter a protocol-driven observational “trial” with explicit inclusion criteria, a defined index date and prespecified estimands. Comparators may be unmatched natural history, propensity-score-weighted controls, or synthetic cohorts built from registries and claims – but the logic of the target trial remains.

3. Capture outcomes that matter to both regulators and payers. For regulators, this means clinically meaningful improvements given the disease trajectory: fewer crises, delayed transplantation, preserved organ function. For payers, it means the same and downstream resource use, productivity, care-giver burden and longevity. If the data cannot speak to value, price negotiations will be brutal.

4. Be transparent, reproducible and (ideally) plural. One clever analysis is not enough. CMS hints that “triangulation” of results across several rigorously designed RWE studies will carry more weight than a single flagship paper. Regulators, likewise, are likely to be reassured when different sources of real world data point in the same direction.

   
   

From anecdotes to learning systems

For sponsors, the task is not to wrap a one-off story around each bespoke therapy; it is to construct a learning system in which every dose is both medicine and measurement.

Practically, that means designing platform registries that cut across individual products. In a world where 150 different mutations in the same gene each demand their own construct, it is madness to treat each as a separate post-marketing obligation. Regulators are already signalling that once a sponsor has “demonstrated success with several consecutive patients” on a platform, they may move towards marketing authorisation and allow subsequent variants to lean on those shared data. A single, well-governed registry can serve as the spine for that platform.

Second, sponsors will need to create the minimum dataset collected for every treated patient, regardless of geography or paying system. That dataset must cover not only safety and disease-specific outcomes but also function, quality of life and healthcare utilisation. Here, RWE for regulators and RWE for payers can and should be the same data, viewed through different lenses.

Third, companies should pre-negotiate evidence expectations with both regulators and major payers. For ultra-targeted therapies, the traditional sequence of approval first, coverage conversations later is rather risky. A more sensible model is coverage with evidence development: conditional reimbursement linked to participation in the platform registry and adherence to agreed analytic plans. That arrangement aligns incentives. Sponsors get early revenue; payers get a say in what is measured and how.

The methodological nuts and bolts

None of this works without credible methods. Fortunately, the toolbox is richer than it was even a decade ago. Target trial emulation, doubly robust estimation and data-fusion techniques are now part of the mainstream RWE vocabulary, not exotic footnotes.

In practice, three techniques are likely to dominate RWE for bespoke therapies:

• External controls using natural-history cohorts. For diseases with grim, well-characterised trajectories, untreated outcomes can provide a powerful counterfactual. The key is to ensure that treated and control patients were, in principle, eligible for the same imaginary trial: similar baseline severity, diagnostic timing, concomitant care.

• Within-patient designs. In episodic conditions, each patient can serve as their own control, with careful attention to regression to the mean and time trends. Prasad and Makary already hint that they will consider “patients as their own control” in conditions with waxing and waning courses, provided the data are strong enough.

• Bayesian borrowing across variants. If different constructs target the same pathway, hierarchical models can share information across them, shrinking wildly unstable estimates towards a common effect while still allowing for mutation-specific differences. This is conceptually similar to master protocols in oncology – but executed in RWE rather than randomized trials.

Methodology will not rescue poor data or lack of pre-specification, of course. But in the bespoke era, it can extract far more from the messy observational traces of practice than was previously possible.

Three (potentially) uncomfortable questions

A few questions still have to be asked.

First, who owns the learning system? Today’s natural-history studies and disease registries in rare conditions are often run by charities, academics or public–private partnerships – access becomes an issue.  Creating a bespoke disease registry may seem out of reach for a biotech innovating in ultra rare conditions due to cost and timelines.  Regulators and payers may need to demand – or even fund – shared infrastructures to avoid that fate.

Second, what about equity? Ultra-targeted therapies will initially reach those with access to sequencing, specialist centres and advocacy networks. If post-marketing evidence is drawn chiefly from this privileged subset, value estimates may be badly skewed. Designing RWE programmes that deliberately reach into under-served communities will be both ethically and politically essential and decentralised approaches to data generation may be particularly useful.

Third, how long is long enough? For one-shot gene or cell therapies, the most important questions, durability of effect, late toxicities, are determined over decades. Yet the commercial life of a product, and the attention span of investors, is shorter. Regulators can mandate long-term follow-up; payers can threaten price cuts if data are not produced. But someone must pay for the infrastructure to watch these patients grow up, age and, eventually, die.

A new social contract

The bespoke era invites a new social contract around evidence. Regulators will flex their rules to allow therapies for which randomized trials are impossible, leaning more heavily on mechanistic plausibility and early signals. Payers, armed with tools like Medicare price negotiations, will insist that high prices be justified by RWE that looks and behaves like a trial, even if no one was randomized.

For drug developers, this world is simultaneously more permissive and more demanding. It allows them to treat patients who would otherwise have no options, relying on modest pre-market datasets. But it also expects them to treat every subsequent patient as part of an ongoing experiment, with transparent methods, shared infrastructures and a willingness to have prices revisited in the light of new data.

What does real-world evidence look like in the era of N-of-1 medicines? Not a hastily assembled case series or a glossy slide deck of quotes about the patient experience, but a disciplined, quasi-experimental learning system that runs for as long as the therapy does. In the end, the atelier will need something the old factory line never quite mastered: the ability to learn from every garment it cuts.