TL;DR
Public health can change people or change the conditions people choose inside. Frieden’s health impact pyramid ranks the second kind higher: interventions at the base reach the most people and ask the least of any individual, while counselling and education sit at the top — most labour-intensive, lowest population impact.1 The evidence backs the ordering. A meta-analysis of 18 studies and 44 effect estimates found smoke-free legislation associated with a pooled relative risk of 0.87 for acute myocardial infarction — roughly a 13% reduction — with no one needing to be persuaded of anything.2 And a review of systematic reviews found that mass media campaigns show evidence of increasing inequalities between socioeconomic groups, while fiscal measures like tobacco pricing and straightforward provision of resources show evidence of reducing them.3 Advice is not neutral. It is differentially takeable, and the people best able to take it are the people who needed it least.
A crossing, a signal, a marked lane. Every one of these is a structural health intervention, and nobody had to be convinced of anything to benefit from it. Photo: Florian Göpfert on Unsplash.4
Two Ways to Make a Population Healthier
There are broadly two places to intervene.
You can work on the person: awareness campaigns, dietary advice, exercise guidance, screening reminders, apps, incentives, warnings. Or you can work on the conditions: air quality, housing standards, what is in the food supply, whether the workplace is safe, what things cost, whether there is a footpath.
The first kind is enormously more popular, and it isn’t hard to see why. It is cheap. It is fast to deploy. It is easy to evaluate — you can run a trial on a leaflet. It requires no argument with any industry. And when it fails, the failure has somewhere to go that isn’t the institution that ran it.
The Ordering Almost Nobody Acts On
Thomas Frieden’s health impact pyramid, published in 2010, arranges public health action into five tiers and makes a claim about their relative power.
At the base sit socioeconomic factors — income, housing, education. Above that, changing the context so the default choice is the healthy one: fluoridation, smoke-free laws, taking trans fats out of the food supply. Then long-lasting protective interventions like immunisation, then clinical care, and at the top, health education — described as the most labour-intensive intervention with the lowest population impact.1
The relationship is inverse, and that’s the whole point. The further down you go, the more people are reached and the less any individual has to do. The further up, the more the intervention depends on a person receiving a message, believing it, and sustaining a change in behaviour against everything else in their life.
Public argument, public budgets and public attention concentrate at the top.
What Changing the Default Actually Does
The strongest natural experiment we have is smoking legislation, because it happened in many places at different times and the outcome is hard to fake.
A systematic review and meta-analysis pooled 18 eligible studies producing 44 effect estimates, drawn from nine US locations, three Italian regions, two Canadian cities, two Swiss areas, Great Britain and New Zealand. Smoke-free legislation was associated with a pooled relative risk for acute myocardial infarction of 0.87 (95% CI 0.84–0.91) — about a 13% reduction.2 There was a dose-response relationship: places where smoking prevalence fell more saw larger reductions in heart attacks.2
Consider what that intervention asked of the population. Nothing. No one had to attend anything, understand anything, or resolve to do better. The law changed where smoke was, and hospital admissions moved.
Compare the ask made by a campaign urging people to quit — sustained motivation, against addiction, for years, in whatever circumstances they happen to live in.
The Finding That Should Change the Conversation
Here is where it stops being a simple ranking and becomes something sharper.
Lorenc and colleagues conducted a rapid overview of systematic reviews looking specifically at whether interventions produce what they call intervention-generated inequalities — the phenomenon where an effective intervention increases inequality by disproportionately benefiting less disadvantaged groups.3
Their finding, in their own summary: mass media campaigns and workplace smoking bans show some evidence of increasing inequalities between socioeconomic groups. By contrast, structural workplace interventions, provision of resources, and fiscal interventions such as tobacco pricing show some evidence of reducing them. Their conclusion is that downstream preventive interventions are more likely to increase health inequalities than upstream ones.3
Read that carefully, because it is worse than “advice is weak.” Advice can be actively regressive. An intervention that works — that genuinely improves behaviour on average — can widen the gap, because uptake requires resources: time, money, stable housing, a job with predictable hours, the absence of more urgent problems. Give everyone the same message and you have not given everyone the same intervention.
This is the health version of an argument this site keeps running into. The same exam is not the same chance when preparation is purchasable. The same leaflet is not the same leaflet when acting on it costs some people nothing and others everything they have left over.
The Uncomfortable Case: When Both Are True
Notice that workplace smoking bans appear in both of the findings above. They are part of the legislative wave associated with a 13% fall in heart attacks, and they appear in the column of interventions with evidence of increasing inequality.
That combination is not a contradiction, and resolving it is the most useful thing in this argument. A policy can improve the population average substantially while widening the distance between the top and the bottom of it. A workplace ban protects the people who have a workplace with an enforceable policy — which is not everyone, and is least likely to be the people in insecure, informal or outdoor work.
So the correct conclusion is not “structural good, behavioural bad.” It is that average effect and distributional effect are two different measurements, and reporting only the first is the norm. An intervention with a good average and an unexamined distribution is not yet known to be a good intervention.
Three Questions, Not One
Which suggests replacing the usual single question with three.
Efficacy — can this work at all, under favourable conditions, with motivated participants? This is what most trials measure, and it is the easiest bar.
Effectiveness — does it work across an ordinary population, in ordinary conditions, at scale? Selection effects make this much harder, and they run in a predictable direction: the people who enrol in and complete a behavioural programme tend to be healthier, steadier and better resourced than the people who don’t, which flatters the result.
Equity — who can actually access and sustain it, and does the gap narrow or widen? This is the question the Lorenc review was built to ask and the one least often reported.
An intervention can pass the first, scrape the second and fail the third, and still be described in the press release as working.
Where I’d Hold This Loosely
Four limits, and the authors of the central paper name most of them first.
The IGI review is explicit that for many intervention types the data are simply lacking, and that most of its findings concern health behaviour outcomes rather than health status.3 The authors also make a careful methodological point: a differential intervention effect does not strictly establish an intervention-generated inequality, since you would need to show the intervention created a health difference rather than merely tracked one.3 I have used “evidence of” throughout deliberately.
Second, this is high-income-country evidence. Whether the same ordering holds where the binding constraint is the absence of a clinic rather than the presence of a bad default is a genuinely open question, and one that matters a great deal more for most of the world.
Third, none of this makes behavioural interventions worthless. Clinical advice works on the person in front of you, and the pyramid’s own logic is that a serious strategy operates on every tier at once. The critique is about proportion — about a field that spends most of its argument on its least powerful instrument.
Fourth, and cutting against my own framing: structural interventions are not automatically progressive either. A sugar tax is regressive in its incidence even where it is progressive in its health effect, and the smoke-free case shows a structural measure landing unevenly. “Upstream” is a useful heuristic, not a guarantee. The only reliable way to know whether an intervention narrowed the gap is to have measured the gap — which remains, on the evidence here, the thing least likely to have been done.
