post by Paul Kelleher
The British epidemiologist Geoffrey Rose made the following simple but powerful observation:
a large number of people at a small risk may give rise to more cases of disease than the small number who are at a high risk.
The point can be usefully illustrated with a toy example. Suppose the population of your country has a ritual of tossing coins at noon, and suppose you are a policy maker wanting to maximize the number of coinflips in the population that turn up heads. The vast majority of the population usually flips fair coins (50% chance of landing on heads, 50% chance of landing on tails), while a small segment of the population has coins with tails on both sides (giving them a 100% chance of getting tails). Assume that you can choose one of two options in your quest to maximize the number of coinflips turning up heads: (1) You can replace the coins held by the majority with coins that come up heads 51% of the time (rather than 50%); or, (2) You can replace the tails/tails coins held by the minority with heads/heads coins, changing their probability of flipping a heads from 0% to 100%. Now ask: Which strategy should you choose if your goal is to maximize the number of coinflips that come up heads? The answer, Rose tells us, is that in a very large population, you should pick option (1), since the 1% change in probability for the vast majority will yield more additional cases of heads than a 100% change for a much smaller group. Similarly, reducing the majority's average cardiovascular risk by a small amount may prevent more deaths than eliminating altogether the high risk faced by a disadvantaged minority.
This is a powerful insight. It leads to the conclusion that if we wish to maximize population health, the best strategy may be to bring about small but favorable shifts in the entire curve (i.e. the population's distribution of risk), rather than focus all our efforts on those folks with the highest risks who reside in the "tail" of the curve. Here's how that might look (figure taken from Frohlich & Potvin). Note, the x-axis measures the relevant health risk factor, with "bad" to the right and "good" to the left; and the y-axis measures the percentage of the population that carries that level of risk:
The before-and-after depicted above shows the effects of a population health intervention that reduces everyone's risk by the same degree. That is what is meant by saying that the entire curve has been shifted in a favorable direction. But sometimes our best bet for boosting population health comes from interventions that do more for the advantaged majority than they do for disadvantaged minorities. Here's what that would look like.
The above before-and-after reflects an intervention that actually widens inequality at the same time it boosts population health. Indeed, population health is boosted even as those with the highest risk (i.e. those residing in the rightmost tail of the distribution) are not helped much at all. This raises the question of whether and how the goal of boosting population health should be balanced against the goal of constraining health inequality.
In a recent paper that will become a staple in my course syllabi, Joan Benach et al. provide a useful typology of different ways in which changes in population health can be paired with changes in health inequality. In the following scenarios (click to enlarge), the solid lines represent the starting distribution and the dotted lines represent the distribution after the intervention. In general, distributions that move to the left signify an improvement in population health, while a flatter resultant distrubtion signifies increasing inequalities.
The authors discuss each scenario with reference to real-world interventions that have or might have affected popluation health in the manner depicted. For example, in the context of scenario (e), where population health improves while inequality increases, Benach et al. note that it mirrors "the historical effects of anti-smoking campaigns, in which there is an overall improvement in the population health, but mostly for the better-off."
Consider now what Benach et al. say about scenario (g):
Scenario (g) portrays a complex picture in which health inequalities are reduced but overall population health remains the same in such a way that for social groups with better health outcomes, the situation worsens. This scenario contrasts with the so-called “leveling up” approach, which argues that inequalities must be tackled by improving health levels for less privileged groups without damaging the situation of the better-off, and therefore that a “leveling down” approach is not equitable under any circumstances.
I don't agree with the authors' claim that (g) is an example of "leveling down." To see why, consider a passage from Whitehead and Dahlgren (pdf), which is the source Benach et. al. cite as making the case that "a 'leveling down' approach is not equitable under any circumstances" (emphases added by me):
Policies Should Strive to Level Up, Not Level Down
Nobody would seriously suggest trying to close the health gap by bringing healthier people down to the level of the least healthy. A worsening in the infant mortality rate of the babies of rich parents, for example, with no change in the mortality rate of poor babies, would not be seen as a success, but would rather be seen as a tragedy – even if it led to a narrowing of the differences between the two groups, purely in terms of measurement. Yet, opponents of an equity policy have warned of this danger. Therefore, to make it absolutely clear, the principle set out in this paper emphasizes that the only way to narrow the health gap in an equitable way is to bring up the level of health of the groups of people who are worse off to that of the groups who are better off. Levelling-down is not an option.
The example of leveling down given by Whitehead and Dahlgren is one in which inequality is lessened by worsening the health of rich persons' babies while holding the health of poor babies fixed. Here's what that would look like:
In (i), inequality is reduced solely by increasing the health risks of the better off. In my view, the term "leveling down" should be applied only to (i), not to (g). So Benach et al. can rest assured that in supporting some interventions that look like (g), they are not supporting leveling down. The example they give is resdistributive policies designed to improve the health of the worst off, but which carry the side effect of reducing the health of the better off (mediated through a relationship between income and health). If there is something morally problematic with this sort of policy, it cannot be because it amounts to leveling down.