Current state of health in the US

Dr. Christopher Murray shares findings from a series of papers published in The Lancet on US health.

Read the research

This transcript has been lightly edited for clarity

Rhonda Stewart: Hello, everyone. Welcome to IHME’s webinar on the State of Health in the United States. I’m Rhonda Stewart, Director of Global Impact. And today, Dr. Christopher Murray, IHME’s Director, will present new findings on challenges and progress in improving health in the US. Our work is part of a special series published by The Lancet.

If you do have questions for Dr. Murray, please use the Q&A to submit those, and we’ll get to as many as we can at the end of the discussion. Chris, over to you.

Chris Murray: Thank you, Rhonda, and welcome everybody in whatever time zone you are.

I am going to walk through our findings that were published in the special issue of The Lancet, but I want to start with framing the lens that we’re using to look at US health, and that is starting with the Global Burden of Disease. And for those of you not familiar with the GBD,

I’ll just give you a few slides as to how we go about thinking about the size of health problems. And then we’ve applied that to the United States.

So the GBD has been around for 30-plus years. If you’re interested in the history of it, there was a piece that I wrote in Nature Medicine a few years back on the GBD at 30. We like to think of it as a rules-based evidence synthesis for global health.

It is not necessarily right, but it reflects all the data that’s out there. And it follows some very strict rules for how we synthesize and incorporate data across the full spectrum of health outcomes.

The first GBD, begun in 1991, was only at the regional level, had a hundred-plus conditions and 10 risk factors and five broad age groups. And over time, the GBD has become more and more elaborate. So the last published GBD, which is 2021, was for a time series from 1990 to 2021 on an annual basis for 371 diseases and injuries, many clinical outcomes for those diseases and for nearly 100 risk factors, and for 204 countries, and for 21 countries at the subnational level, and that includes the United States. In fact, as you’ll see, funded by National Institute of Minority Health Disparities, down to the county level by race and ethnicity.

Now a core part of the GBD model for analyzing health data and coming up with observations about health over time and across place, is that it is a large collaborative. In fact, we have 13,600-plus members of the collaborative.

They come from 164 countries, 2,500 different institutions, and so the collaboration behind the GBD is sort of baked into how we go about analyzing data. And this map shows you that there are some countries, usually ones with subnational burden of disease estimation, where there are particularly large numbers of collaborators that are engaged, and those are shown in the dark blue.

Now, I’m going to show you results, some of which have just been published. But the core of the GBD results are available with each cycle of the GBD online and in a number of online data visualizations. Our flagship data visualization is called GBD Compare.

There are many views in GBD Compare ,and you can tap into the database of many billions of results by location, by year, and different views for that. So if it’s something you’re interested in to dig in more deeply, there are a lot of resources online for you.

So that’s the GBD, and then we have a special effort on the United States. The goal of that effort, funded by NIH, is to bring this machinery of measurement and representation that’s been built up over 30 years of the GBD to look at very fine-grained results for the US – and then to use those fine-grained results spatially by race and ethnicity, as well, to explore disparities and what are the key drivers of health in the United States.

Now, as in the basic GBD, we have lots and lots of sort of what I would think of as standard metrics of disease – disease incidence, disease prevalence, death rates, age-standardized death rates – but we also have rolled up summary measures.

So we have years of life lost, trying to capture that not all deaths represent the same loss to the individual of healthy life.

And then we take the prevalence of all the conditions measured in the GBD multiplied by the public’s view of how bad it is to have those conditions, collected through household surveys, 3 and that quantity is called years lived with disability. When we put the two together, that’s DALYs, one of our apex measures – disability-adjusted life years. And then the other apex measure in the GBD is healthy life expectancy – exact same information, but one’s a positive measure, how many years in full health you are expected to live, and the other is a measure of loss of health compared to some societal norm.

Now, on the risk factor part of the GBD, we look at both exposure to a risk, the relative risk of different outcomes from that, and then what is the minimum risk level of exposure. So think for tobacco, when I show you results for risk factors in the United States, it how much would the burden of disease be reduced if people didn’t smoke at all. But for some risk factors, the optimal level – think of body mass index – is not zero. Obviously, it is some other number. And so what we call the theoretical minimum risk exposure level is part and parcel of our quantification of risk factors.

Now, in this series of papers that are published in this issue of The Lancet this week, we have a number of papers. One of them is the burden of disease at the state level. What do we see in those trends for the last 30 years or 31 years? What are the leading risk factors? What are the most important diseases? How does it vary over states?

Secondly, we use this huge body of data from the county race and ethnicity level and try to aggregate the US into 10 Americas. And I’ll talk a little bit more about that in a moment. This builds up from a paper from 20 years ago called the Eight Americas, and I'll explain why we’ve expanded to 10.

But in a sense, think of it as a heuristic to take what’s perhaps too complicated, 15,000 units of analysis, and boil it down to the sort of key dimensions of disparity.

We forecast health out to 2050, using the forecasting framework that’s been now published several times – applied at the global level published last May. And now we dig into the US, what’s most likely to occur, as well as alternative scenarios.

And we have two other deeper dive papers that help buttress and support the analysis around disparity. The paper on the Human Development Index is an attempt to answer the question of who and where are the worst off in the US – very important for targeting of interventions for the future.

And then, because obesity is such an important risk factor, the number one cause of DALYs in the US, we look at overweight and obesity and what’s likely to come.

Okay, these papers, along with a viewpoint summarizing Ali Mokdad’s and my personal views about what are the main strategies to address the disparities that we see, and the problems in the US. Along with other pieces, not from IHME, very useful pieces, something from Victor Dzau at the National Academy of Medicine, something about Flint, Michigan, and a number of other thoughtful analyses are all in this issue of The Lancet, called the presidential briefing book on US health care.

Okay. When we look at US burden, or the state of health in the US, what we find – one way to think of the results, and I think one that’s super important for us to grapple with – is that the rank of the US in terms of the metric around mortality of life expectancy at birth across the 204 countries has been declining pretty steadily, as you can see for females, for essentially the last 40 years. And for males, there was a period of sort of stagnation where we ranked about 30th, and then it’s been declining for the last 25, 26 years.

So the US, comparatively, was not the best at any point in time in terms of life expectancy. But it was, you know, 20th for both sexes combined, among the other high-income countries back in 1980. Now we rank below essentially all high-income countries and some middle-income countries, and our rank in terms of life expectancy is below 40th across countries.

So a tale of steady decline, in relative terms. In terms of healthy life expectancy, because of the prevalence of a number of conditions – diabetes among them, depression, anxiety, musculoskeletal disorders – is quite high in the US compared to our peers, our rank for healthy life expectancy is lower and has also been steadily declining, such that by 2021, we’re ranked somewhere between 65th and 75th, depending on males or females in terms of healthy life expectancy.

So you know, really, really poor outcomes, given our wealth and given our health expenditure. And I think this is such an abject story of failure for public health and medicine in the US, that compared to all other countries, we just keep getting worse on that relative scale.

Now, it’s important to think about, what are the risk factors that explain that pattern of healthy life expectancy, and the best way to do that is to look at DALYs attributable to risks. On the x-axis, here is the percent of DALYs in the US in 2021 attributable to each of these risks, analyzed one by one.

So the largest risk factor in terms of DALYs is high body mass index. That means overweight and obesity, and that is atypical that it’s number one. If you were to look around the world or in high-income countries, high body mass index is certainly in the top five or 10, but it’s unusual to be the largest risk factor. That’s followed by high blood sugar. Those two are very intimately linked, because one of the main outcomes of high body mass index is to raise your blood sugar. So in some sense you can almost think of those two risks as reflecting a similar process in society.

Then tobacco is number three, despite progress, particularly in males and some in females, in reducing tobacco prevalence, it’s still the third largest cause of burden. Then diet in aggregate, and those are quite strongly linked to high body mass index. High blood pressure, drug use, particularly with the epidemic of fentanyl deaths, and then kidney dysfunction, high alcohol use, high cholesterol, occupational risks, and then on down. Notice that non-optimal temperature, both cold and heat, is on this list, as well as air pollution.

Now this is years of healthy life lost, or DALYs. By age-standardized, I’m removing the effect of aging that’s occurring in the US over this time period. In the left column is 1990, middle is 2010, and then 2021. And you can see quite a lot of change occurring over this period of time in the US. And you see that by the crossing lines, with the huge increase in drug use disorders, shooting up from quite a low level, but also seeing a rise of depression, anxiety going up on the rank list, chronic kidney disease going up, big drops for lung cancer, some drops for ischemic heart disease and low back pain, some drop for congenital disorders or defects.

And of course, in 2021, at least, COVID-19 shooting to the top as the cause of DALYs. That wouldn’t be the case in 2024. Okay, so that’s a very super high-level view of leading diseases, leading risk factors and really poor performance in the United States as a nation, compared to other nations.

We have a paper that focuses on obesity and overweight, because it’s the leading risk factor of DALYs. And I think it’s an important component, not the only one at all, to explain the difference between us and the high-income or upper-middle-income countries, but certainly an important component is our high levels of overweight and obesity.

So this is in adolescence, males on the left, females on the right – adolescence in this case being defined as 15 to 24. There’s different definitions of adolescence out there, but that’s the one we’re showing here, showing the prevalence of overweight and obesity on the same color scale. So you can see overweight and obesity is higher in adolescent females than in adolescent males.

But you get to a state like Mississippi, and now nearly 65% of adolescents are overweight or obese, and in males in Texas. It’s over 50% in all those states that show yellow, so already high at this young age.

When we go to adults over age 25, we get prevalence of overweight and obesity that are approaching or over 80% in places like North Dakota, Nebraska, Iowa, West Virginia. Colorado is the lowest on the scale. Well, actually, that’s not true. The District of Columbia is the lowest, and then Colorado.

For females, Colorado is the lowest, and you can see not quite as high for overweight and obesity for females, although when you look just at obesity, these maps flip with higher rates of obesity in females than males.

If you look by birth year cohort, each of these colored lines is tracing a birth year cohort through the period of observation where we have data, 1990 to 2021. So you get multiple observations, and you can see for all of the birth cohorts that are later in time at the same age, obesity is just going up.

And so this is a general phenomenon that’s affecting every age group where, in males and females, that obesity rises as we go to the next birth cohort, and so this phenomenon of rising overweight and obesity is affecting everybody. And if we look by state, even though there is a state pattern that I showed you on those maps, it is increasing everywhere at a pretty profound rate. So take, for example, age group 45.

The birth cohort of 1950 was about 20% obesity at 45. And now we’re at about 60% at that same age but at later birth cohorts. And then the paper also includes some forecasting models. These are fitting splines and it’s an ensemble model to use to forecast, which includes some nonlinear modeling. And this is the expected trend: overweight and obesity is expected to continue growing, but not particularly fast.

But we are seeing the whole distribution of body mass index shifting so that obesity goes up a lot in the future and overweight actually goes down as many people that are overweight move into the obese category. Remember, obese by WHO definitions is a body mass index over 30.

So no relief in sight. Now, there’s been some discussion of the last round of NHANES data. There was a writeup in the media where the argument was, well, maybe obesity has maxed out, and we’re seeing already the effects of GLP-1s.

The problem with that is that there are quite large uncertainty intervals around each round of NHANES, the National Health and Nutrition Examination Survey, and we have seen seemingly peaks of obesity in the past as well that didn’t turn out to be more than just the statistical noise in the measurement.

Self-reported obesity continues to march up. And so that makes us think that that we probably have not seen the peak, and that the forecasts are likely right.

Okay, moving along to the next paper, which is to take advantage of the work that’s being done, and some of it published at the county level, and to look at life expectancy, actually, age-specific mortality, using small area methods calculated for each county, 3,110 counties, and for five racial and ethnic groups, and then removing from the analysis populations of less than a thousand because of statistical noise.

We then have taken that information and try to create groupings to simplify the understanding of the patterns across different counties. This is building on a paper that we published, or I wrote with others some 20 years ago, called the Eight Americas, and the addition here, that is the new development from the Eight Americas is that now we have better data on Latino status.

And we were able to separate out from white Americans, Latino Americans, and divide them into two groups. So our 10 Americas here are, at the top, America 1, Asian Americans, who have the highest life expectancy.

And then one of the Latino groups is America 2, the second highest life expectancy. And that’s to distinguish Latinos in most parts of the US from America 5, which is Latinos in the Southwest, who have lower life expectancy.

America 3 is the sort of majority of White Americans.

America 4 is White Americans in rural and the sort of Midwest northern counties.

Then we see America 6, which is the first of three Black groupings, Black America, which has had quite substantial improvement in life expectancy from 2000 to 2010.  

Then America 7, which is the highly segregated Black populations living in highly segregated metropolitan areas – worse outcomes, but they did have similarly an improvement.

America 8, which is the sort of low-income Appalachia, Lower Mississippi Valley, White populations.

And then American 9, The rural, low-income South Black populations, and American Indians and Alaska Natives living in the West, the last of our Americas.

If you just look visually, what you see is that there has been improvement, particularly in Asians, Latinos, some improvement, as we mentioned, in the Black populations, not much improvement at all up until COVID, in Appalachia, Mississippi Valley Whites, and declines in American Indians, Alaska Natives [AIAN], as you can see. And then COVID hit, and you see the drop in all of the Americas. But the largest drop is in AIAN, and we’ve been slow to recover post-COVID.

Over this period, taking into account the stagnation and decline in AIAN populations, and then the huge hit from COVID, disparities, or the gap in life expectancy from Asians at the top to American Indian and Alaska Natives at the bottom, has widened considerably over the last 21 years. So, despite all the efforts to try to enhance access to health care through, for example, the ACA, policies around trying to address some of the drivers of disparities, some of the focus in states on

looking at social determinants and trying to intervene on them, in fact, over this period things have gotten worse in terms of disparities, and certainly not aided by the COVID epidemic.

So with that picture we wanted to explore more fully who and where are the worst-off in the US, because we think that part of the solution around these huge disparities is going to be targeting resources and programs that address things like obesity or smoking or high blood pressure to the people who are the worst off. And we wanted to not reinvent the wheel, so we adopted the framework from the United Nations Human Development Report, which has been around for 31 years, as long, or slightly longer than the GBD, where they sought and continue to measure human development using three components.

And we’ve adopted the same approach: lifespan, or life expectancy in their case, lifespan in our case, educational attainment, and income.

And so we looked at those three dimension in the US using the American Community Survey. And we had some interesting innovation in this. Normally, in the Human Development Index literature, it’s calculated at the population or community level. But we have brought this idea down to the individual level using the ACS data and our county race and ethnicity life tables. And then to calculate this for different age groups, we have used a construct called expected lifespan, because we wanted to look by age as well, not just life expectancy at birth. And so that’s your age, plus your expectation of life at your age, and that’s expected lifespan.

We’ve used years of education for the education measure. And then we’ve used household consumption, per consumption equivalent, using a standard technique from the economics literature to approximate consumption equivalence by using the square root of the number of the people in each household.

So here’s the distribution of HDI, this is Human Development Index within each race and ethnicity group by sex.

So this is for males, and so each column is for different race and ethnicity groups and the colors are what decile of HDI do you belong to?  So if you look on the far right, young Asians, men under 44, between ages 25 and 44, 39% of them are in the top decile and 2% of them are in the bottom decile.  

And if you go over on the far left, if you look at AIAN, you get 2% of young AIAN are in the top decile, and 66% of that group are in the bottom decile.

For young Black men, it’s 46% in the bottom decile, 1% in the top decile. Latinos are overrepresented in the bottom, but much more evenly distributed across deciles. And for White populations, for young White men, only 2% in the top decile but pretty even distribution across many of the other deciles, as shown.

For females, a similar story, but an even larger fraction of Asians and Asian women in the top decile. But if we focus on the younger or more recent cohorts, more than half are in the top decile for women.

For White women, 16% of younger White women are in the top decile, 3% in the bottom. For Latinos, it’s 12% in the top decile, 7% in the bottom. And then we look at the more disadvantaged groups, for Black women, it’s 2% in the top, 12% in the bottom with the younger ages and gets worse as you get into the older age groups. And for AIAN, it’s 28%. So another way to look at this is to look at the composition, which now reflects not only your likelihood of within a race and ethnicity group of being in different deciles, but the population composition. So there’s many more Black Americans than there are AIAN.  

So when we look at who makes up the bottom decile by age, this will reflect both the rates within a race and ethnicity group as well as the size of the population, but it does tell us who are the worst off that we might want to be targeting.

So in the bottom decile, in the younger age groups 25 to 44, the bottom decile is, 3% AIAN and 30% Black males,15% Latino, 27% White males, and then the remaining 25% are sort of the analogous groups that are female.

And if we then look at the top decile over on the far right, you find that 80% of the top decile in the younger age groups are females, and that changes to be about 50/50 in the oldest age groups.

And if you look at the race and ethnicity groups, 11% are Latino females, 47% of that younger age group in the top decile are White females, and 19% are Asian females. So dramatic change by cohort in the makeup of the top and the bottom, and very different patterns across top and bottom. I think, from a public health point of view, the makeup of the bottom decile, and the next decile as well, is probably the more actionable or useful thing from this analysis.

Where – that’s the sort of who by age and race and ethnicity,  and here is where. The part on the left is the bottom decile, and on the right is the top decile, and you can see that the fraction of the population in each county in the bottom decile is very high in Eastern Kentucky, parts of West Virginia, the Mississippi Valley, on the border in Texas, some of the Native American reservations like Pine Ridge and Rosebud in South Dakota. And then in terms of the best off, it’s major urban areas, and then a number of places, particularly in Colorado, where they’re high-income and people are moving there often for lifestyle or for outdoor activities. And you see a little bit of that in Jackson Hole in Wyoming as well.

The last of the analyses in this paper are trying to look into the future. And so what we’ve done is we’ve taken the burden of disease and used all that information by disease and cause and risk and used that to make forecasts in the future. But with the forecasts, you forecast each of the risk factors separately, then what’s not explained in the past by risk factors, and also builds into the modeling migration as well as fertility. And so it’s a quite complex forecasting framework.  

If you’re interested in the details published in May in The Lancet at the global level, with country-specific results. And then we use this framework to make forecasts for the United States and special scenarios for the US.

Now, the advantage of this framework is that by putting together these different components, we have the combination of both good out-of-sample predictive validity, that is, how good are we at forecasting what’s going to come? And we have all these drivers in it, all the risk factors. So we can answer “what if?” questions – What if we did a better job of dealing with smoking or obesity as an example?

Now, forecasting has been an integral part of the GBD for a long time, and this is just a reminder that it was actually part of the very first GBD set of papers. And then we’ve been building this more elaborate Bayesian forecasting framework actually over the last nine years. And we’re now starting to see a number of analyses get published from that. At the global level, what the forecasts tell us is there’s going to be this continued epidemiological transition, a continued shift from the communicable, maternal, neonatal causes in pink here to the non-communicable causes in blue.

And that general transition at the global level is the most important macro view. If you drill the next level down, you see big increases in cancer in black, cardiovascular disease in red, diabetes, chronic kidney disease in orange at the global level.

When we apply this to the United States, what we find out is that we don’t expect life expectancy to increase a lot. Some continued increases. The models, by the way, include the effects of climate change. So if people are interested in that, we can talk more about that. And when we look at forecasts of healthy life expectancy for women, we may see complete stagnation. That’s what the most probabilistic forecast says, and some very modest improvement for males.

What does this mean in terms of what we started with? Well, what it suggests is that the rank for life expectancy for the US on those current trajectories should continue dropping.

So we expect that we will make no more progress or less progress than our peer nations in the coming years.

Here’s a little more detail for deaths, years of life lost, years lived with disability, and DALYs. In these forecasts, the top row is all-age rates. The next row is age-standardized rates, and then counts which you know, capture aging of the population. And you know, while they’re sort of complicated, they speak to the importance of what we saw at the global level.  

Plus this green color here, which is the rise of dementia and other neurological disorders in terms of their contribution to DALYs in the future and the potential role of drug use disorders that might continue expanding.

We included a number of scenarios, and I’ll just show you the combined scenario for the forecast that says we can do something about this future trajectory. So the gray lines here are all the countries of the world in the past, very noisy in some cases, and then the future forecasts.

The red lines are the United States states, and the solid red is the US as a country. And you can see our forecast, which doesn’t look like we’ll see a lot of progress. But if we can address behavioral risks, diet, obesity, physical activity, you could be on the green trajectory, in which case we would actually rise in the rankings if we made a concerted effort on tackling risk.

So you know, it’s not a doom and gloom story here at all. It says we have the potential to actually make a difference in the United States if we can really be serious about tackling some of these major risk factors.

Now let me round out this presentation with some of the other resources we’ve built off these results. We have briefings at the state level if you’re interested in state-specific analyses that are sort of boiled down to be more useful in particular places. And we’re happy to make those available to you.

And let me end with what Ali Mokdad and I had in our viewpoint, which again, is a little bit more of a personal view on how we might go about really changing the trajectory for the US and addressing some of the disparities. We see in some of the analyses here, and in many of the other analyses that IHME has done, that there’s a super strong relationship between educational attainment and health outcomes. We think that’s causal. It’s its own separate discussion, but we believe very much that if we can raise educational attainment or reduce disparities, we should see that effect on health.

However, when you look at the data, what the studies on disparities in school readiness in educational performance suggest is that disparities have already emerged by age 4 or 5.

So there’s a real need to learn the lessons from some of the successful early child development programs that intervene very early, sometimes before birth, to actually change the trajectory for educational attainment, narrow the gaps and disparities.

Part of that is addressing this enormous gap opening up between males and females where boys are way behind girls in educational attainment, and that’s getting worse every year.

Second, we aren’t going to make a lot of progress if we can’t tackle obesity, diet, and physical activity, and that’s got to be through many routes because there aren’t any success stories at the community level anywhere in the world that we see in the data where obesity’s rise is being reversed.

And there is some important role to be played for GLP-1s in the concerted effort on obesity. We believe that for many reasons, both because it’s the right thing to do, it’ll reduce catastrophic spending, and it’ll make it easier to address prevention through primary health care is that universal health coverage is a key to future solutions, and then we also need to tackle, with known proven strategies, the big risks like tobacco and high blood pressure. So I’ll stop there and turn it over to Rhonda to moderate some questions. Thank you.

Rhonda Stewart: Great, thanks so much, Chris. If anyone does have a question, please do put those into the Q&A and I’ll begin to read those in just a moment. I also just wanted to go back to what Chris shared about finding this information. Our briefings and details on these papers, all of that information is available on our website, which is healthdata.org.

Okay, so let me go now to a few questions. So, Chris, this is one on the GLP-1s. And this participant asks, How do you see the uptick in weight loss drug utilization impacting obesity and overall health moving forward?

Chris Murray: You know, this is a super important question. We have modeled out some scenarios on this. And of course, the potential impact depends very much on what fraction of the population are willing to use GLP-1s, I mean, who need them, given the side effects from the current generation of GLP-1s.

And then who will pay? Because these are very costly drugs at this point, and so in our scenarios that we’ve run with what we think are reasonable assumptions, there’s some effect of GLP-1s. It is less dramatic than one would hope. But you know, probably you can address 20% of the burden related to overweight and obesity through GLP-1s. Maybe somewhat higher, but unlikely much past 30%. Given what we see now, there’s new generations of GLP-1s being researched. And so when we get purely oral and maybe better side effects profiles, maybe that story changes and we can see a bigger impact in the future.

Rhonda Stewart: Okay. And we have a few questions about the 10 Americas information. So one participant is asking if you can speak to disparities that may be missed when combining Asian and NHPI groups. And so why not analyze the five racial categories laid out by OMB standards, for example.

Chris Murray: So what we’ve shown here is using the five OMB categories. We are trying to move to the newer classification, and that is work underway. It’s part of our work with NIMHD. So we expect to see those available, and certainly splitting Pacific Islanders from Asians is super important because they have much worse outcomes than Asians. So that is an important part of the work.

We’re also trying, although the data is very sparse in many cases, to look into different heritages within Latinos. And that’s interesting work underway. And so it’s an ongoing part of our work with NIH on sort of how far can you push the fact that some of the newer race classifications and ethnicity classifications are much more recent. And how much can we map back in time? And then we also start to run into small number problems even more.

And remember, when we’re doing this work, you have to separately think about the denominator population, which is much easier to grapple with because people are alive, and they tell you what their race and ethnicity heritage is. But on the numerator, where it’s being filled out on a death certificate, sometimes the person filling that out is not so well informed. And so there’s this need to do these linkage studies, which are critical to understanding why, for example, we undercount deaths in Latinos, unless you correct for that on death certificates, as an example.

Rhonda Stewart: What about prevention regarding obesity? And are there ways to assess the effectiveness of those prevention efforts?

Chris Murray: Yeah, I mean, this is what everybody, every government is trying to grapple with. And of course there are ways to assess prevention. There are small-scale studies that show benefits from coaches for diet and physical activity.

Those aren’t so small but just very costly. There are small-scale studies on vending machines and on school lunches. There’s some evidence on taxation of sugar-sweetened beverages, although it’s a bit mixed, interestingly.

However, the real challenge there is that despite a lot of efforts to try some of these programs at the community level, we don’t have any community-level success stories. We have specific studies, usually very small, and so coming up with the menu of things that will work is really quite challenging on obesity given we’re not seeing much in the way of success.

Lots of good ideas. I think it’s such a challenge. I don’t think you can wait for the ideal evidence – we should be researching and, hopefully, doing more research on prevention strategies. But the urgency, the burden is such that some things that seem sensible probably should be tried in terms of getting rid of subsidies on certain foods, and programs that encourage physical activity.

Well, there’s a host of things to try. It’s just proving them is a different story.

Rhonda Stewart: And there are several questions about social determinants of health.

And one question one participant asks, considering the high health spending in the US, is it fair to say that the US has one of the least efficient health systems among high-income countries? Or is the declining rank due to social determinants of health? And another participant asked about why the GBD doesn’t factor in social determinants of health. So there are some questions about that aspect.

Chris Murray: I think it’s very fair to say we have a horribly inefficient health care system, because we spend the most or one of the most. Norway is certainly catching up, and Switzerland, or it may have exceeded us at this point. But we have demonstrably poor outcomes. It would be hard to point to the decline in the US being due to worsening of social determinants,

because, in fact, social determinants or disparities in them have been poor in the US for a very long time. Right? We’ve had income inequality. Education is actually going up.

Poverty, mean incomes are going up in most cases. So I think it’s difficult to ascribe the slide of the US to something like worsening social determinants. They are very important, and in fact, we will, working with NTNU in Norway, the CHAIN group, which is a European collaboration on social determinants. We have been now working a number of years with them, and we’ve reached the point now where we will be adding to the next cycle of GBD quantification of low education as a risk as the first social determinant, where we’re very happy with the evidence, and that’ll turn out to be the largest risk factor – larger than obesity or tobacco, for example, or high blood pressure.

So we are progressively trying to capture them. Now, the one thing that happens when we look at social determinants from the published studies

Chris Murray: and look at the relative risks by cause is, as you might expect, because they’re upstream, they’re much more heterogeneous, the effects. Right? Because some societies, despite having disparities in education, have many things, social policies and others that mitigate those disparities in terms of their impact on health. So when we look at the spread of the relative risks for education, for example, they’re much larger than for many other risk factors.

So there’s that challenge. And then when we started to try to dig into adding poverty, or low income, then we also need to tease apart the causality between education and income, or at least get  what we call the mediation of one risk through each other, even if it’s hard to tease the causes apart. So that’s ongoing work. First off, the first result there will be education, and that should be out next year.

Rhonda Stewart: And income inequality was a question that people have asked about. So one participant asks whether income inequality is what’s driving the disparity seen in the 10 America slide that you presented. For example, others are asking about what can be done in low-income communities where people maybe don’t have access to parks, or gyms, or healthy food, things of that nature.

Chris Murray: Yeah, both very important questions. So, when people talk about income inequality, often they’re referring to two things, and they’re worth distinguishing. One is the strong relationship between poverty and health outcomes. We know that whenever we look at this relationship, it’s present.

How much of that is mediated through education? People with lower education have lower incomes. We know education is a super powerful driver, but not all of it is through education. And that’s such a consistent finding across the world that it is definitely part of the 10 Americas, but not the only part, because, for example, one of the better-off groups for the US is actually comparatively lower-income White households in the north of the US that have pretty good outcomes.

So you know,  it’s not a simple translation of income and deterministically to outcomes. But it’s on average, super important. The other dimension is whether living in a place where you see a lot of income inequality also makes everybody, controlling for their income and education, worse off.

And there’s literature on this that goes back to Wilcox and others. And we’re trying to – with better datasets that we now have, and more detail – we’re trying to investigate that. That effect is certainly, if present – and I say, if, because some of the analyses we’ve done don’t really find that – but that effect is much smaller than the direct poverty, low-income effects that that we do see.

The second part of that question, Rhonda, can you remind me?

Rhonda Stewart: So there are communities, low-income communities in particular, where people have less access to parks, to healthy food, things of that nature. How does that factor into some of these results?

Chris Murray: So I think food deserts, or you know, safety – people can’t exercise because they can’t go walking in the street – are very important. There’s a lot of good studies in specific localities showing that. And I think if we want a concerted effort on the nexus of diet, physical activity, and obesity, those really need to be addressed, and think about things like subsidizing high-fructose corn syrup as opposed to subsidizing fruit and vegetables. There’s some macroeconomic policy interventions that will help.

But I’m sure it needs more than just that, because of the phenomenon where it’s unaffordable to purchase a healthy diet, and in many places it’s not safe to exercise.

Rhonda Stewart: And another participant raises a question about BMI as a health measure that’s been used, given that this person is saying that it’s been debunked in some literature sources. Can you talk a little bit about BMI and how that is factored into the research?

Chris Murray: Yeah. So you know, there’s a lot in the media about this, a lot of it not particularly soundly written, to be honest. And so let me just give you the facts, and then there’s interpretation that everybody can make on their own. So there are different measures of overweight and obesity. The one that is still the global standard from the World Health Organization is BMI.

There are something on the order, across big cohort studies, including the big UK cohort from the CPRD, close to 100 million person-years of observation on the linkage between BMI and many different health outcomes. So the data are unbelievably strong – there is no other risk that has ever been studied in this detail – and consistently show very strong relationships for a wide range of outcomes, cardiovascular disease, multiple cancers, diabetes. And so far, those risk curves look very similar for different race and ethnicity groups. There is a subset literature that suggests, maybe, but it’s sort of within the uncertainty intervals on those risk functions, that for Asian populations you start to be at risk sooner in terms of BMI. So instead of at risk from 25 on, it may be lower: 22, 23, 24. It’s contentious, because the uncertainty intervals on those different race and ethnicity–specific risk functions overlap. So it could just be statistical error.

Now, is there a better measure than BMI – waist to hip ratio? Many proposals out there, and there’s always papers coming up saying, this is a better predictor – and it may well be there are better measures.

But we don’t have 100 million person-years of observation on those better measures. We have very tiny numbers. So you will see the media or studies with 50,000 person-years, or even 5,000, and then they’ll say this measure was more predictive, and that’s certainly quite possible.

But when you go to ask the question, is there enough statistical evidence to tie that metric to different outcomes and say that it’s a better measure across multiple studies? No, not yet. It doesn’t mean we won’t get there.

But the evidence on BMI is really, really strong. We have an evidence scoring system in the GBD which published in Nature Medicine, the methodology, and then we star rate the strength of association. Many of the BMI associations are 3-, 4-star associations, some of them like BMI and Alzheimer’s, which goes the opposite direction, have a much lower star rating. But in general, there’s a lot of evidence on BMI that is quite compelling.  

Now, even within the family of measures of weight in the numerator and height raised to some power in the denominator, there’s no particular reason that squaring height is the right thing to do. And I think with the existing cohort data, if you had access to all the individual records as opposed to just what’s published, you could test whether some other exponent on height would be a better measure – or not better, because what was better means more predictive of outcome. And that’s a direction that I’ve been surprised that people haven’t pursued, given that there are some pooling studies out there. You know Oxford has a pooling study where they have access to the microdata for many measures, so you could imagine testing out is weight over height to the 2.3 power, a better measure?

So there may be better metrics coming. But at least for now, BMI is very predictive of outcome.

Rhonda Stewart: And last question is about policy. So one participant asks actually, what are your thoughts on conducting analyses of implementing policies hypothetically? So, for example, if we were to tax fast food and provide rebates for produce and fresh fruits and vegetables, what might that do? How could that connect, perhaps, to the work on forecasting? So what are your thoughts on that.

Chris Murray: Yeah, I mean, the reason we spent so many years building this forecasting framework is we can now try to answer that type of question. So if a state or somebody wants to answer the question, what might future health look like if we could shift diet towards higher-quality, lower-risk diet, or we could reduce sugar-sweetened beverages, or high-fructose corn syrup, or whatever it is, we should be able to answer those. We can run those scenarios and try to put in front of politicians and policymakers the idea of what a package of strategies might actually translate into in terms of health benefits. And you know, there’s also a component that’s a sort of quirk of the diet literature, where in in the diet world, and we reflect that in that community of that body of evidence, there is this separation between what’s called diet quality and the total caloric intake, which has got to be one of the drivers of obesity and overweight.

And so when people talk about diet and only look at diet quality, they are underestimating the full impact of changing diet, because that’s both going to be on the diet quality, but also potentially on obesity.

And in the forecast, we can capture both dimensions of that. If we intervene on diet, it’ll change both body mass index and diet quality, and we can try to capture the totality of those effects.

Rhonda Stewart: Great. Well, thanks so much, Chris, and thanks very much to everyone for joining us today. And please go to our website, healthdata.org. That’s where you will find all of the papers. You will also find the briefings that Chris mentioned. Thanks again, and this concludes the webinar.