Health Behaviors Do Not Explain the Growing Education-Mortality Gradient

The gap in premature mortality between high and low educated people in the United States has grown considerably over the last few decades, even as life expectancy has increased overall. A common explanation is the changing distribution of risk factors: if the less educated are relatively slow to experience declines in mortality, it must be because they are also slower to manage hypertension and give up smoking, drinking, and fatty foods. (Paul Kelleher has an interesting recent discussion on whether the behavioral contributions to life expectancy gradients are different in the UK, U.S., and France).

David Cutler and four coauthors explore the role of behavioral risk factors in the education-health gradient in the latest JHE and find that: “despite the importance of smoking, obesity, hypertension, and cholesterol as determinants of population health, differential changes in these risk factors do not explain the widening educational gap in death rates since the 1970s… even if less educated populations were able to achieve risk factor profiles mirroring those with more education, widening mortality differentials would likely persist.” In other words, if we want to address mortality differentials we may need to look at factors beyond health behaviors.

Estimating the Education-Mortality Gradient

Before discussing what other factors might be at stake, I want to dig deeper into the key findings of the paper. The authors estimated survival rates by linking person-level health data collected in the 1970s to 1990s with five-year follow-up mortality records. They focused on two of the most important population health studies in the United States: The National Health and Nutrition Examination Survey (NHANES) and the National Health Interview Survey (NHIS). NHANES provides directly measured data on obesity, cholesterol, and blood pressure, while the NHIS relies on self reports with larger samples. The main comparison was between individuals with a high school diploma or less with those with at least some college. Because of dramatic growth in the college-going rate, they also test the sensitivity of their estimates by reclassifying some individuals on the borderline of the groups in earlier years (people who would have likely gone to college if they had been in the later samples).

Consistent with previous studies, the authors find that the decline in five-year mortality between the 1970s and 1990s is much higher among higher educated individuals (mortality actually increased slightly for lower educated women). There were considerable declines in current smoking among college-educated men (15 points) and women (12 points), but smoking rates did not decrease for non-college women. Gains in obesity were substantial and roughly equivalent across educational groups. Despite the increase in obesity, hypertension and cholesterol rates decreased substantially across groups. The table below illustrates the changes for the males.

The authors then go on to test how much mortality rates among lower-educated groups would have changed if they exhibited the same distribution of risk factors as better-educated groups. (They fit conditional proportional hazard models for survival in each period and then assess how much of the secular change is due to changes in the model coefficients). Although smoking and obesity significantly contribute to the education-mortality gradient at a point in time, the table below illustrates that counterfactually changing risk factors has a negligible effect on estimated mortality ratios (those are the small differences displayed in the third column). These results are robust to additional controls for smoking intensity and the timing of quitting smoking.

Why no Change?

The technical way of describing the impact of risk factors on the education mortality gradient is that the changing return on education and the return on risk factors (i.e. the extent to which having an educational profile or a risk factor influences mortality), and not the distribution of risk factors and education, is driving the differential gradient over time. Why would that be? There are a few obvious possibilities: the authors have not adequately controlled for the true risk factors (omitted variable bias), the results reflect compositional changes in the groups over time, and there are some other exposures that are changing the interaction between risk factors and mortality.

The authors rule out the first two explanations. They conduct a simple bounding exercise to assess how large an omitted behavioral risk factor would have to be in order to drive the differential change, and find that it would have to be (implausibly) large. They also re-estimate the models to account for the compositional changes in educational groups, and find that these changes only moderately change the observed differences.

They speculate that two exposures that might be influencing the gradient: differences in access to health services by educational status and differential environmental exposures. For health care they state that, “Adherence to prescribed regimes may also have become both more important and more difficult over time, yielding larger gains in life expectancy for highly educated individuals who have better adherence rates.” My own view is that the evidence is mixed about whether adherence to treatment has in fact become more difficult over time (see the comparison between AIDS and hypertension treatment in this paper). Nevertheless, there is little debate that the technological inputs that make a disease condition more likely to kill you have changed over time, and access to these technologies varies by education (at least for the non-elderly covered by Medicare).

Second, geographic variation in the exposure to harmful environment has changed over time. In particular, areas where the better educated live have done more to reduce the exposure to harmful pollutants, while less has been done to reduce exposures for the less educated. They draw on evidence from Janet Currie, for example, that shows that differential location near sources of pollution (such as plants), can explain about 6% of existing gaps in birth weight between the most advantaged (white college educated) and least advantaged (black high school dropouts) mothers.

Finally, it is worth noting that because this analysis focuses only on five-year mortality it may not account fully for the long-term effects of smoking cessation on mortality. The mortality gains of quitting smoking in the 1990s are likely to have extended beyond the early 2000s, and so we may need to revisit the mortality data in several years to fully account for the long-term gains of smoking cessation.

About Brendan Saloner

I am a postdoctoral fellow at the University of Pennsylvania in the Robert Wood Johnson Health and Society Scholars Program. I completed a PhD in health policy at Harvard in 2012. My current research focuses on children's health, public programs, racial/ethnic disparities, and mental health. I am also interested in justice and health care.
This entry was posted in Blog posts and tagged , , , . Bookmark the permalink.

4 Responses to Health Behaviors Do Not Explain the Growing Education-Mortality Gradient

  1. anon says:

    interesting paper but surprising that they don’t reference (or very little) the vast social epidemiology literature on this issue – see this Lancet paper for example and follow references and links from there
    http://www.thelancet.com/journals/lancet/article/PIIS0140-6736(08)61688-8/abstract

  2. Pingback: Health Promotion Headlines from Robyn & Penny December 13, 2011 | Health Nexus Today / Nexus Santé aujourd'hui

  3. Pingback: Life Expectancy in the U.S. is Getting Shorter for the Least Educated | Inequalities

  4. Pingback: Conditionally Accepted | Reflections On Pedagogy And Self-Censorship As The Semester Ends

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s