The Translation Gap: Why Good Research Rarely Becomes a Tool Clinicians Use

A promising research finding fails to become a tool clinicians use for four reasons, and only one of them is the science. The finding has to be validated outside the lab that produced it, fit the way a real clinic runs, earn the trust of the people who would rely on it, and satisfy whatever regulator governs the claim. Most clear the first hurdle and stall on the other three. The distance between a true result and a used result is the translation gap, and closing it is a separate discipline from making the discovery.

I have spent my career on both sides of that gap, in diabetes genetics at the Lund University Diabetes Centre, in systems medicine at Stanford, and later in global development at Novo Nordisk. I have watched good findings sit unused because nobody did the unglamorous work of carrying them across. The crossing is harder than the lab work, and it gets far less credit.

What is the translation gap?

The translation gap is the space between a finding that is true and a finding that changes what happens to a patient. A result becomes a tool only when it survives validation, fit, trust, and regulation, and any one of those left undone is enough to strand it.

This is not the long drug-development pipeline, where molecules fail for biological reasons. The translation gap sits downstream of "does it work" and asks whether a working result ever reaches the hands that could use it. A risk model with an impressive accuracy figure can be correct and still die quietly because no one rebuilt it into something a busy clinician could use on a Tuesday.

Validation: does it hold outside the room it was born in?

The first thing that breaks is generalization. A finding learns the particular population, machines, and habits of the place that produced it. Move it elsewhere and performance often sags, not from fraud or sloppiness but because the world is more varied than any single dataset.

Validation in a new setting is therefore not a formality, it is where most of the truth gets decided. Diabetes shows why: it is a cluster of overlapping diseases whose relationships shift across groups. I co-authored a meta-analysis in Diabetes Care on ethnic differences in the relationship between insulin sensitivity and insulin response, cited widely since, and the lesson was humbling. A relationship that looks fixed in one population can change shape in another, and a tool built on the first will mislead the second with confidence.

So the honest standard is external validation, ideally prospective, in settings the developers did not control. A finding checked only against the data that birthed it has been admired, not tested.

Fit: can a clinician actually use it during a real visit?

The second failure is subtler, because the tool can be entirely correct and still useless. A recommendation in the wrong format, at the wrong moment, or buried two screens deep will be ignored, and an ignored tool is worse than none, because it trains people to click past it. Fit is about the short appointment, the incomplete chart, the patient with several other conditions, and the clinician who cannot stop to read a paper. I learned how much this matters when my colleagues and I built EASY Diabetes, an AI clinical decision-support system for type 2 diabetes, where I was Head of Medical and Science. The principle that did the most work was restraint: stay quiet when standard care was already right, and speak up only with something worth saying, inside the workflow the clinician already used. We then tested it the way you test a drug. The EASY-1 randomized controlled trial (NCT03258268) was a multi-clinic study that compared the tool against standard of care. That mattered more than any internal benchmark, because a tool that shines on past data can still fail in a waiting room.

Trust: will the people who must rely on it believe it?

The third failure is human, and it is the one builders most often underestimate. Clinicians carry the responsibility for what happens to the patient. A tool that cannot explain why it is suggesting something, or that has a reputation for crying wolf, will be quietly set aside no matter how good its numbers are. That caution is appropriate in people who answer for the result.

Trust grows from transparency about how a tool reaches its conclusion, honesty about where it is uncertain, and evidence gathered in conditions that resemble real practice rather than a curated demonstration. It is also earned by co-design. When the people who will use a tool help shape it, they trust it more, and it gets better. The fastest way to lose it is to hand clinicians a finished black box.

Regulation: does the claim match what the law allows you to say?

The fourth failure is regulatory, and many capable teams stumble here because they treat it as paperwork at the end rather than a design constraint from the start. The moment a tool makes a claim that influences diagnosis or treatment, it usually becomes a regulated medical device, with rules that differ by jurisdiction and by claim.

I took the Medical Device Regulations training at KTH Royal Institute of Technology because this knowledge is not optional for anyone crossing the gap. Software that supports a medical decision sits under frameworks such as the EU's device and in-vitro diagnostic regulations and the FDA's approach to software as a medical device, with adaptive AI raising its own questions about accountability over time. The practical point is simple: what you are allowed to claim shapes the evidence you must gather, so it should inform the first study you design, not the last. A common trap is building something impressive, then finding the claim needed evidence no one collected.

How do you actually cross it?

You plan the crossing before the discovery is even finished. Decide early what claim you intend to make, which sets the regulatory path, the evidence you need, and where you validate. Build with the clinicians who will use the tool, not for them, and measure success in the field.

None of this is as exciting as the original insight, which is the quiet injustice of the field. The discovery gets the headline. The crossing gets the patient. Both deserve the same rigor.

This article is educational and is not medical advice. For decisions about your own care, talk to a clinician who knows you.