In 2002, the New England Journal of Medicine published one of the most influential results in preventive medicine. The Diabetes Prevention Program randomised 3,234 Americans with raised blood glucose to a placebo, the drug metformin, or an intensive lifestyle programme aiming for at least 7% weight loss and 150 minutes of activity a week. Lifestyle change cut the incidence of type 2 diabetes by 58%. Metformin cut it by 31%.
Twenty years later, health systems are still trying to turn that 58% into a public programme. Most of them are finding the number doesn’t survive the trip.
That is usually told as a failure story. It shouldn’t be. The better question is not “why don’t we get 58%?” but “what size of effect, delivered to how many people, at what cost, is worth having?” The programmes that scale are the ones that answered that question honestly.
The trial effect is not the programme effect
A trial and a national programme measure different things. The DPP trial compared people who were enrolled, supported intensively and followed closely. A national programme has to find people, persuade them to be referred, get them to turn up, and keep them coming for months.
England offers the clearest real-world data. The NHS Healthier You diabetes prevention programme, launched in 2016, offers adults with non-diabetic hyperglycaemia a nine-month behaviour change course. An independent evaluation funded by the National Institute for Health and Care Research compared 18,470 people who were referred with 51,331 similar people who were not, and found that referral was linked to a 20% lower risk of progressing to type 2 diabetes.
Twenty percent is not 58%. But it is measured on everyone referred, including those who never attended or dropped out. More than 1.2 million people have been offered a place, according to the NIHR. A modest effect across a very large population can prevent more cases than a dramatic effect in a small one.
This is the trade-off most commentary misses. Pushing for trial-level intensity often means fewer places, higher cost per person and stricter eligibility. Accepting a smaller effect can buy reach.
The American programme that almost nobody used
The United States took a different route, and it shows how scale can fail even when the evidence is accepted.
In 2016, Medicare’s actuaries certified that the National DPP lifestyle programme would reduce net Medicare spending, which allowed it to be expanded as a covered benefit. Suppliers began delivering services in 2018. By July 2021, according to CMS figures reported in Health Affairs, only about 3,600 Medicare beneficiaries nationally had used the benefit.
Evaluations point to familiar barriers: low referral rates, limited awareness, and a payment model that paid suppliers largely on performance milestones such as attendance and weight loss, at rates many providers found too low to cover their costs.
The lesson is uncomfortable. A benefit can be proven, certified as cost-saving and formally covered, and still reach almost no one if the organisations that must deliver it can’t make it work financially.
The long tail nobody budgets for
There is another reason to value prevention programmes that is easy to lose in short-term evaluations.
In 1986, clinics in Da Qing, China, began a six-year lifestyle intervention trial among 577 adults with impaired glucose tolerance. Thirty years later, results published in The Lancet Diabetes & Endocrinology in 2019 found that the combined intervention groups had 26% fewer deaths from any cause, 33% fewer cardiovascular deaths and 26% fewer cardiovascular events than the control group.
A six-year intervention was still paying off three decades later. Almost no health budget works on that time horizon. Programmes are judged on one- or two-year outcomes and funding cycles that rarely extend beyond five. The benefits that matter most arrive after the people who approved the budget have left.
What scalable programmes have in common
The programmes that reach large numbers tend to share a few features:
- They are built around referral pathways that already exist. England’s programme uses routine blood test results in primary care to identify eligible people. The work of finding participants is built into care that already happens.
- They accept multiple formats. Group sessions, remote sessions and digital delivery each reach different people. Insisting on one format limits reach.
- They pay providers enough to operate. A payment model that only rewards the best outcomes will attract few providers willing to serve harder-to-reach populations.
- They measure population effect, not just completer effect. The honest metric is what happens to everyone referred, not just the motivated minority who finish.
The metformin question
One more push against convention. In the original DPP, metformin achieved about half the effect of lifestyle change, from a cheap, widely available generic drug requiring far less infrastructure. Lifestyle programmes remain the right first offer. But for systems unable to build intensive behavioural programmes at scale, dismissing a medication option on principle may leave people with nothing at all.
The arithmetic of reach
It helps to put numbers on the trade-off, using simple, illustrative figures rather than real programme data.
Imagine a region with 100,000 adults with prediabetes. Suppose that without any intervention, a certain share of them will develop type 2 diabetes over the next few years.
Option one is an intensive programme modelled closely on the original trial. It is expensive per person, requires many in-person sessions, and has strict eligibility. With the available budget, it can enrol 5,000 people. Among those who complete it, the effect is large.
Option two is a lighter programme with fewer sessions, remote and group options, and broader eligibility. It can enrol 40,000 people for the same budget. Its effect per participant is much smaller.
Which prevents more cases? Unless the intensive programme’s effect is many times larger, the lighter one does, simply because it reaches eight times as many people. And it probably reaches a more representative group, including people who could never attend weekly in-person sessions.
This is the arithmetic behind England’s approach. A 20% reduction in risk among those referred looks modest next to the trial’s 58%. But when more than a million people have been offered the programme, the absolute number of cases prevented can be large.
The trap is to judge programmes by effect size alone. The number that matters for a population is effect multiplied by reach.
Why referral is the real bottleneck
In most prevention programmes, the step that limits impact is not the course itself. It is getting people into it.
People with prediabetes usually feel well. They have not asked for help. They must be identified, typically through a blood test, told about their risk in a way that motivates action, referred, and then persuaded to attend. Each of those steps loses people.
The US Medicare experience illustrates how fragile this chain is. The benefit existed, but beneficiaries had to find a supplier, and suppliers had to find beneficiaries, in a system where clinicians were not routinely referring and suppliers were struggling with low payment. The result was that very few people used it.
England built referral into routine primary care, where blood tests that identify non-diabetic hyperglycaemia are already done. That doesn’t solve every drop-off, but it removes the first and often largest barrier: finding the people who are eligible.
The strongest counterargument
The case against scaling lighter programmes is that they may be buying very little.
Observational evaluations can overstate effects, even with careful matching. People who are referred may differ from those who are not in ways the data don’t capture, such as motivation or contact with their GP. A lighter programme with modest attendance might produce much smaller real benefits than a 20% headline suggests. Meanwhile, funds spent on it could have gone to proven treatments for people who already have diabetes.
There is also the question of what happens after the programme ends. Weight regained over the following years could erode benefits, leaving a short-term delay in diagnosis rather than true prevention.
These are fair concerns. Observational designs carry uncertainty, and long-term benefits of lighter programmes are not yet established.
But the Da Qing results show why it is dangerous to dismiss prevention on short-term evidence. A six-year intervention produced measurable reductions in death and cardiovascular events thirty years later. Even if scaled programmes deliver a fraction of that, the long-term payoff could be substantial. And the choice is not between prevention and treatment. Health systems need both, and the cost of preventing diabetes is modest compared with a lifetime of treating it and its complications.
Second-order effects
New drugs are changing the landscape. GLP-1 receptor agonists and related medicines produce weight loss far beyond what most lifestyle programmes achieve. They are also expensive and currently prioritised for people with diabetes or obesity. As prices fall, pressure will grow to use them in prevention. That could shift diabetes prevention from a behavioural model to a pharmaceutical one, with very different cost and equity implications.
Digital delivery is widening access and raising questions. Remote and app-based versions can reach people who can’t attend groups, but they may also reach mainly the digitally confident, leaving others behind.
Prevention exposes inequity. People in poorer areas have higher diabetes risk and often lower programme attendance. A programme that reaches mainly better-off participants can widen health gaps even while improving the average.
What gets lost in the averages
There is a final trap worth naming. Average effects hide who benefits. A programme can show a respectable population-wide reduction in risk while doing almost nothing for the groups at highest risk, if those groups are less likely to be referred, less likely to attend and less likely to finish. Because type 2 diabetes is more common in poorer communities and in some ethnic groups, a programme that reaches mainly lower-risk, better-resourced participants will prevent fewer cases than its headline suggests.
The fix is not complicated, but it has to be deliberate: track referral, attendance and completion by deprivation and ethnicity, and adjust formats, locations and payment until the gaps close. A prevention programme that isn’t measuring who it misses will keep missing them.
What this means for decision-makers
For ministries and payers, it means evaluating prevention programmes on population impact, effect multiplied by reach, and funding them on time horizons long enough to capture benefits.
For primary care leaders, it means building identification and referral into routine care, so that finding eligible people doesn’t depend on individual clinicians remembering.
For programme providers, it means offering multiple formats and tracking who drops out, and why, especially among higher-risk groups.
For clinicians, it means treating a prediabetes result as a moment for action: a clear explanation of risk and a direct referral, not a note to “watch it.”
What to watch next
Watch for long-term follow-up from England’s programme showing whether reduced progression translates into fewer complications. Watch how health systems decide whether and when to use the new weight-loss drugs for prevention. And watch whether US policy fixes the payment and referral problems that have kept Medicare’s programme small.
The takeaway
The 58% figure was never a promise. It was a proof of concept. Diabetes prevention at scale looks like a smaller effect, delivered to many more people, through systems that are paid enough to keep doing it.
That is less exciting than the trial headline. It is also how prevention actually happens.
Would you trade a smaller effect for much wider reach in your prevention programmes? We’d like to hear where you draw that line.
