There are recurring howls in my work. One of them is this: in general, if you don’t know which intervention works best, then you should randomise everyone, everywhere. This is for good reason: uncertainty costs lives, through sub-optimal treatment. Wherever randomised trials are the right approach, you should embed them in routine clinical care.

This is an argument I’ve made, with colleagues, in endless different places. New diabetes drugs are approved with woeful data, small numbers of patients in trials that only measure blood tests, rather than real-world outcomes such as heart attack, renal failure, or death: so let’s roll out new diabetes treatments in the NHS through randomised trials. We rely on observational studies to establish whether Tamiflu reduces complications of pneumonia: that’s silly, we can do trials, and we should. Statin treatment regimes in widespread use have never been compared head-to-head, using real-world outcomes such as heart attack, stroke, and death: so let’s embed randomised trials as cheaply as possible in routine clinical care (we’ve done two pilots, to document the barriers).

This week a dozen colleagues and I published yet another application of this basic, simple principle, as an editorial in the BMJ. The Cancer Drugs Fund is being marketed as a way to generate new knowledge: but in reality, the data that will be collected is weak, observational evidence, riven through with confounders. There’s no need for us to squander patient experience like this. When we spent vast sums of money on new treatments with uncertain benefits (and hazards, and costs), then we should do so through randomised trials wherever possible. That’s how we can find out whether these expensive new treatments are effective, how effective they are, and whether they’re cost effective.

That’s the argument today. It was the argument last year. And it will be the argument next year. But it is part of a growing, furious thread: we should not be tolerating bad quality evidence. We should be randomising, as a matter of routine, throughout the health service, whenever we lack good evidence on what works best. We should be turning the NHS, and it’s sisters around the world, into learning health systems that test, learn, and adapt with every move they make. “Big Data” is a tedious buzzword. We need better data, and where that we requires randomisation then we should be identifying the ethical and practical barriers, and smashing those down. For that battle, we should reach outside the ivory tower and use every tool available to us, including harrassing policymakers, as a matter of routine.

There are lives at stake. Anything short of perfect data exposes patients to avoidable harm. Onward!