Bad news for US life expectancy

The big health news this week was a report that, for the first time since 1993, life expectancy in the United States went down (incidentally this was also a story earlier this year when the preliminary numbers were published, coverage by the NYT here). Life expectancy decreased from 81.3 to 81.2 years for women and from 76.5 to 76.3 years for men. Those seem like small changes, right? So why is this such a big story? Because US life expectancy has been steadily increasing--with minor interruptions--for decades. When these numbers are reported each year (and usually don't make headlines) it is because the story is that life expectancy increased again, mortality is improving, and we are doing something right (or, at least, we aren't doing something wrong). The graph below shows trends in life expectancy at birth since 1929 for US men and women:

From published estimates by the US National Center for Health Statistics.

From published estimates by the US National Center for Health Statistics.

It's an impressive record, notwithstanding the fact that, despite this progress, the US still lags behind many other rich (and some poor) countries in terms of life expectancy.

This story has gotten a lot of coverage, and there are solid articles that have attempted to discuss the reasons for the decrease. Scientific American had an interview with Robert Anderson, chief of the Mortality Statistics Branch at the NCHS, who published the report, that talks about increases in cardiovascular disease, the spike in the 2015 flu epidemic that potentially spilled over to cause more cardiovascular deaths, and the continuing epidemic of overdoses from prescription opioids and heroin. 

The NY Times focused quite a bit on socioeconomic factors and the role they may have played in the decline. They quoted Columbia's Peter Meunning, who said that "popular theories for the cause of the decline, including an increase in obesity rates and an opioid epidemic, fail to explain a problem that feels broader."

Vox also had a nice write-up that discussed the wide range of diseases affected, potential mechanisms related to stress and poor health behaviors, but also (and positively, in my view) noted that, despite a recent focus on white mortality patterns in the press, this recent mortality increase did little to alter the gap in life expectancy between blacks and whites.

There are clearly important questions about what the underlying causes are of this decrease, especially if it is sustained and shows up in next year's statistics as well. Anderson from NCHS says that the preliminary data from 2016 suggest that 2015 may just be a "blip"): 

if we look at the age-adjusted rate for 2015, which is about 733 deaths per 100,000 population, and look at the age-adjusted rate for the 12 months ending in June 2016, it drops back to 720 per 100,000—and that’s actually below the 2014 level, so it may very well be a blip.
— Robert Anderson, NCHS, quoted in Scientific American

Most of these stories have centered on the fact that 8 out of the 10 leading causes of death showed increases, most distressingly for cardiovascular diseases (chiefly heart disease and stroke), which have been leading the cause of death tables for decades. The increase in cardiovascular diseases, because they represent the leading cause of death, is potentially what makes this a story about a decrease in life expectancy rather than a few causes of death that are higher from one year to the next. And there are, indeed, signs that the historic declines in deaths from cardiovascular diseases may be coming to an end. A recent paper in JAMA Cardiology (Sydney et al. 2016;1(5):594-599) shows that the declines have been stalling since around 2011 or 2012, which is of great concern since it continues to be the number one killer in the US.  

But none of these stories really tell us about which causes of death played the largest role in the decline of life expectancy. So, let's take a very early and simple look at this question.

Decomposing the decrease in life expectancy

We'll do this separately for men and women, since the mortality patterns and trends are often quite different by gender. The data on mortality by age and cause of death are publicly available and easily downloadable from CDC WONDER. I used the multiple cause of death database for 2014 and 2015, and asked for the "standard" list of 113 causes of death used by NCHS.

Because a decrease in life expectancy can arise due to changes in mortality at different ages, I used a technique developed by the demographer Eduardo Arriaga to decompose the change (there are other ways to do this, but this one is fairly straightforward and simple).

It's important to note as well that life expectancy is essentially a weighted average of age-specific death rates, and deaths at younger ages are given more weight (the steep increases in life expectancy in the early 20th century were largely driven by substantial reductions in infant and child mortality). Because it is a weighted average, life expectancy at birth can change in lots of different ways. It could decrease if mortality trends decrease in younger ages but increase at older ages, and vice versa. The decomposition helps to sort out which age groups made the most difference.


I calculated life expectancy for women as 81.36 in 2014 and 81.20 in 2015, a 0.16 year decrease, which is close to what was reported by NCHS. For men the values were 76.48 and 76.31, a decrease of 0.17 years. Below you can see the graph showing the proportion of the decline in life expectancy due to mortality changes at different ages, by gender:

Author's calculations. Source: CDC WONDER

Author's calculations. Source: CDC WONDER

A few things to notice here.

First, the contributions across age groups are quite different by gender. Nearly 60% of the (0.1 year) decrease in life expectancy among women came from unfavorable mortality changes at very old ages (85 and over). On the other hand, for men 70% of the increase was due to mortality increases among those aged 15 to 44 years. Although infant mortality can strongly influence life expectancy, it played no role in the worsening of life expectancy for either men or women (the fact that they did not help push toward increasing life expectancy probably is not a good sign).

Second, nearly all age groups (except the 45-54 year olds) helped to push life expectancy down for both men and women, which adds some weight to the story that this decrease may be associated with a broad range of factors affecting mortality.


With respect to causes of death, it obviously isn't feasible to try and look at all 113 causes, so I broke it down to some broad categories of disease (there are many other ways to do this). Here are the results:

Author's calculations. Source: CDC WONDER

Author's calculations. Source: CDC WONDER

Again, there are a lot of potentially interesting things going on here.

  • For women, most of the story appears to be worsening cardiovascular disease and Alzheimer's disease mortality. These two groups accounted for 94% of the net decrease (it is important to remember that the overall change is a net change that results from some causes that improve and some that worsen).
  • For men, most of the story is about deaths from injuries, both unintentional and intentional. The largest contributor (41%) was unintentional poisoning, but homicide also contributed 20%, motor vehicle crashes 13%, and suicide 7%. Altogether these causes accounted for 82% of the net decrease in life expectancy for men. 
  • Progress in cancer mortality for both men and women kept life expectancy from declining by more than it did (i.e., the decrease in life expectancy would have been greater had cancer mortality rates not declined from 2014 to 2015).

There is obviously quite a lot of additional work to be done exploring the decrease, but a few things to keep in mind.

One, reductions in life expectancy are obviously bad, but they are uncommon, and in this case they are also small in magnitude, so it is important to keep that in mind when interpreting the proportional contribution of different causes.

Two, there has been a lot of attention focused on the opioid epidemic, but it would seem that the decline among men is linked more broadly to a number of injury-related causes. I have yet to see many reports even mention homicide (perhaps because there has been so much focus on the middle-aged white population after the Case/Deaton study last year, where homicides do not figure prominently), yet homicide played a larger role among men than did Alzheimer's. My guess is that this played an even larger role among non-Hispanic black men.

Three, there are many potential hypotheses that may be consistent with some parts of the patterns above. Many of the researchers quoted talked about wide ranging stress, economic inequality, poor health behaviors, increases in obesity, and any number of additional "meta-theories". However, it seems unlikely (to me) that any one of these explanations will be sufficient to account for the diverse patterns by gender, age, and cause of death we see here, not to mention race, which may make things more complicated. It may be that a number of different channels are operating simultaneously rather than one "big thing".

Finally, and more controversially, there is a lot of literature that suggests that mortality rates worsen when the economy improves (and vice versa). Yes, this sounds counterintuitive, but there is a lot of work from Christopher Ruhm, Jose Tapia-Granados and others that makes a reasonably good case for this hypothesis. That being said, whatever increases in death rates that may come from an improving economy (one mechanism is usually increases in traffic accidents as more people are working and commuting, which we do see here and which has received some press attention), they are usually not large enough to overwhelm all of the other improvements in lifestyles, public health, and medical care that make death rates *usually* decline.

It will be very interesting to follow this story as it unfolds.

For those interested, the mortality data I used for the decomposition can be downloaded here, and the Stata do-file is here.