The US Life Expectancy Mess, Part 4: So What About The Original United States Question?

Introduction:

This series is an examination of the life expectancy of the United States compared to “other rich countries”, inspired by graphs like this:

In the first article, I noted that United States life expectancy has fallen behind the life expectancy of a fixed set of “Other Rich Countries” pretty steadily, accelerating more recently:

In the second article, I wrote about how “life expectancy” is essentially a statistical artifact aggregating death rates by age into a single number, and two countries with similar “life expectancies” might have death rate differences at different ages “cancelling out”:

In the third article, I discussed how this means, in my opinion, “breaking down life expectancy” is a bit of a waste of time — the death rates by age are more like the “real information”; life expectancy is just a summary of them.

In this fourth article, I will try to present a way to compare death rates by age for different countries over time in a way that addressed the original question. I will then use this to ignore my own third article and crudely try to estimate the contributions of different causes on “the change in the life expectancy gap”.

Focusing On The Change In The Difference:

Recall that our original question became: “Why did United States life expectancy improve by 2–3 years less than the life expectancy of Other Rich Countries from 1990 to 2017?”. That is, it was a second-order or third-order question, “how do you explain the change in the difference”.

Therefore, to answer this question, we need to illustrate not just United States deaths in 1990 and 2018, not just Other Rich Countries deaths in 1990 and 2017, and not just the differences between them, but the change in the distance.

The following graphic shows death rates by age and cause for the United States and “Other Rich Countries” for 1990 and 2017, along with the differences, the change, and — again, what’s most important to our original question — the change in the difference, in the bottom-right. Here are the actual death rates by age that go into all the other metrics (at least theoretically).

(Following what I believe is a common practice and for visualization purposes, I have collapsed the 222 available causes of death into 11 categories. Note that these categories were chosen with this analysis in mind and may not fit other countries or comparisons very well. Note also that combining causes into categories does risk “hiding” relevant information. All data is again from the Global Health Data Exchange or GHDx.)

Looking at the bottom-right graphic, we can see that “deaths of despair” are not necessarily a majority of the difference in death rates between the United States and “other rich countries”, but they are a pretty big share of the change in the difference, and especially for younger age groups.

To clarify that distinction with an example: It’s not new that the United States has unusually many deaths from interpersonal violence compared to “Other Rich Countries”, but that gap has shrunk over time, not widened:

So interpersonal violence can’t explain why the gap has increased (unless you have some counterfactual where interpersonal violence “should have dropped by more” in the United States). It’s not new that more people in the United States are murdered. What is new, at least for relatively younger people, is the United States having many more deaths from substance abuse, self-harm, and cirrhosis (actually, mostly the former):

But these “deaths of despair” are not the only causes that leap out in the bottom right “change in the difference” graph. While deaths by cardiovascular disease at older ages have shrunk dramatically in both the United States and in “other rich countries”, they have shrunk more slowly in the United States than in most:

Cardiovascular disease is such a major killer that even “improving somewhat more slowly” can add up to hundreds of thousands of potentially “excess” deaths. Chronic respiratory disease and diabetes also stand out for older ages. And while these older deaths are weighted somewhat less under either “years of life lost” or “life expectancy”, their weight is not zero even there. You could come up with a plausible public health metric that completely ignores people after they turn 75, but “life expectancy” isn’t it.

(Actually, a lot of the “cardiovascular disease effect” here is a little subtle. As far as I can tell, basically, in 1990 the United States had more people die of heart attack than “other rich countries” but this was more-or-less cancelled out by fewer older people dying of strokes. Over time, the other countries caught up to the US in strokes, but the US didn’t catch up to the other countries in heart attacks. The net effect is “cardiovascular disease” “explaining” more of the gap over time.)

How Much Do These Different Causes “Explain”?:

The graphic suggests that “deaths of despair”, cardiovascular disease, respiratory disease, and diabetes all might contribute to the “change in the difference” of “life expectancy”. Can we quantify how much each one contributed?

For “life expectancy” this is a difficult question as I’ve discussed. However, “age-adjusted years of life lost per capita” and the “age-adjusted death rate” really are (more or less) just weighted sums, so the question can be answered more exactly for those metrics. For the “age-adjusted death rate”, “deaths of despair” explain about 23% of the “change in the difference”, while cardiovascular disease deaths explain about 36%:

For “age-adjusted years of life lost”, “deaths of despair” explain about 50% of the “change in the difference”, while cardiovascular disease deaths only explain about 24%:

(To be clear, this is just an exercise in arithmetic, there’s no attempt here to consider survivor bias or other such effects.)

Again, this isn’t a paradox, they’re just two different weighted sums, applied to causes that are pretty sharply split by age. Recall the Denmark/United States example. Just as Denmark “did better” than the United States in 1990 on measures that placed a higher weight on younger deaths, so too here. “Years of life lost” weights younger deaths higher, so it will place more importance on drug overdoses. Neither is the “right answer”, unless either one happens to match your beliefs about how important deaths at different ages are.

Finally A Kind Of Answer To The Original Question:

So, what about “life expectancy”? That was the original motivating question for this series, after all. Honestly, I wish I could answer “who cares” at this point and let that be the takeaway, but I probably shouldn’t.

Just for illustrative purposes, let’s run some crude counterfactuals. For these counterfactuals, I’m simply going to arithmetically replace the United States death rate for with the “Other Rich” death rate for a set of causes of death within each age group and year.

Again, this shouldn’t be confused with a “real counterfactual” or “real simulation”, but hopefully it will convey some sense of the arithmetic influences of different causes of death on the complex “life expectancy formula”. (And yes, this is basically treating “life expectancy” like a weighted sum even after I explained it really isn’t, but oh well.)

Based on the previous graph, I’m going to successively “fix” United States deaths from “deaths of despair”, cardiovascular disease, chronic respiratory diseases, and diabetes and kidney diseases. All four “hypotheticals” start out with similar life expectancy in 1990 as the United States really had, so that’s good:

Following this as a VERY crude exercise, you might say “deaths of despair” and cardiovascular disease each contributed ~30% of the “change in the life expectancy” gap from 1990 to 2017, with chronic respiratory diseases and diabetes contributing another ~10% each, and other causes contributing the remaining ~20%. If you use 1994 instead of 1990 as your base year, then you might move “deaths of despair” up a bit and cardiovascular disease down a bit —you might notice there’s a funny bump in the data around there, seemingly related to heart disease.

(Which “other causes”? “Falls” and Alzheimer’s disease — which is a little hard to interpret without getting into survivor effects — along with stomach cancer, pedestrian and motor vehicle deaths, etc.)

Note also that, without its initial high level and subsequent dramatic decline in AIDS-related deaths, the United States would have dropped even more compared to “Other Rich Countries” over this period. But now we’re getting into a few too many hypotheticals.

Conclusion:

It is hard to deny that the United States currently does worse on a number of “public health metrics” than most or all “other rich countries”, that this gap has grown over time, or that recent increases in what are commonly (if not necessarily rigorously) called “deaths of despair” are a significant contributor to that. Depending on exactly how you want to weight “extra” deaths at younger ages, they might even explain a majority of that change or close to it.

However, it is also hard to deny that other causes have played a role as well. The United States had already drifted near the bottom of the pile of “other rich countries” by “life expectancy” even in the mid-2000s, before the surge in opioid deaths and so on. Deaths by cardiovascular disease, chronic respiratory conditions, diabetes and kidney issues, transit injuries, interpersonal violence, these all arguably contribute as well, depending on your country/year of comparison, your counterfactual, and the exact question you’re asking.

More generally: I would suggest that getting caught up in the details of “life expectancy” in particular can be a waste of time. (Look how much time it took me to write these articles!) “Life expectancy” is a statistical artifact that surely does a pretty good job tracking general well-being in most cases. But in some sense, it’s just one particular choice of age weights and interactions of different deaths, out of many possible defensible choices.

In other words: Whether opioids contribute 20% and cardiovascular disease contributes 40%, or whether opioids contribute 40% and cardiovascular disease contributes 20%, to the change in a particular metric is probably not really knowable and probably not that important — at least, not nearly as important a question as how to go about reducing those deaths.

Notes and Sources:
Public health data from:

Global Burden of Disease Collaborative Network.
Global Burden of Disease Study 2017 (GBD 2017) Results.
Seattle, United States: Institute for Health Metrics and Evaluation (IHME), 2018.
Available from http://ghdx.healthdata.org/gbd-results-tool.