by Laura Punnett, ScD
Perhaps you’ve heard that there are some unusual trends in U.S. death rates – specifically, that white people are experiencing an unexpected increase in mortality. Last year, the economists Anne Case and Angus Deaton published an innovative analysis calling attention to the alarming increase in mortality rates, specifically in white, non-Hispanic, middle-aged Americans. This trend contrasts dramatically with previous historical declines among whites, the continuing declines in U.S. Blacks and Latinos, and the declines in many other developed countries. The report has been cited at least 150 times in academic literature and elicited energetic reactions from lay and scientific audiences alike (e.g., here, here, here, here, here, here, here, here.)
Case and Deaton referred to these rising rates as indicators of distress, because they are almost entirely a result of drug overdose, liver disease, and suicide. They pointed to the ills stemming from globalization, international competition, economic insecurity, and income inequality, and the impact of these factors on individuals’ behavioral and mental health. These are potentially important causes for non-medical opioid use, high alcohol consumption, and suicide. The authors documented real increases in pain, functional limitations, and depression over the past 20 years, and they also noted the “increased availability of opioid prescriptions for pain” during the same time period. A further potential indication of distress is that an historically high proportion of working age men are not participating in the labor force, many of them due to pain severe enough to cause disability.
Although Case and Deaton made the connection between economic phenomena and outcomes affecting real people, they didn’t discuss the pathways that lead from one to the other. This led to the idea of engaging a group of work environment doctoral students at the University of Massachusetts Lowell to connect up what we already know about the impact of specific working conditions on a very large and urgent public health problem, one that is rarely discussed in relation to occupational health. This also provided a hands-on way to practice finding data available on-line and using them to calculate informative statistics such as attributable proportions, the fraction of injuries or diseases that are caused by a given exposure.
We decided to probe whether it was possible to estimate the contribution of work and working conditions to this increase in mortality among Whites. This is admittedly a complicated task, and we generated several ideas about how to dissect the question. One possible cause that we started with involves the effect of work overload and poor ergonomics on musculoskeletal pain and injury. These might initially be contributors to prescribed medications, eventually leading some people to self-medicate with alcohol or opioids. Another issue relates to increasing job insecurity, precarious work and economic instability. The impact of work stress and job strain on the mental health of workers is well-documented, although only recently has a study shown that they increase the risk of early death. We considered that part of that mortality might be mediated through substance abuse or even suicide.
We developed a set of possible causal pathways to each of the three “diseases of distress.” All three causal diagrams highlight the potential roles of workplace injury and of psychosocial stressors, such as discrimination, or jobs that are particularly demanding but in which workers have little control over how they do their job.
As we assessed the scientific literature and publicly available data, we found support for at least two of our proposed pathways to work-related suicide. One involves the impact of job loss and/or unemployment and the other is the long-term effects of psychosocial job strain. In the U.S., people in certain jobs, especially farming and law enforcement, are more prone to committing suicide due to factors like financial losses, work-home instability, and social isolation. Several international reports (e.g., here, here) have called attention to high suicide rates in the construction trades, attributed to long hours, high stress, and often transient lifestyle. Obviously there are many factors that affect the risk of suicide. We don’t claim that the work-stress/suicide association is the only possible pathway – just that it is one possible pathway.
One of the potential consequences of a stressful work environment is heavy and/or binge drinking, resulting in alcohol-related deaths. Daily work stress and job strain may interact with non-work situations and individual traits to influence individuals’ ability to cope with stress. Work stress and alcohol consumption may each cause the other: a study by Heikkilä and colleagues showed that the odds of reporting job strain were higher among chronic heavy drinkers, compared to moderate drinkers, but heavy drinking was also more common among people in high-stress job situations. We would need longitudinal studies to tease this apart, but the scientific literature on this relationship is fairly sparse. We also found very little evidence about the effects of work-family conflict, social support at work, or workplace culture on alcohol consumption.
Drug abuse is a worldwide public health problem, with markedly increasing deaths from opioid use and overuse in the U.S. in recent years. For our investigation we focused on injury-related pain treated with prescription opioids, which have some risk of causing addiction, leading to of-prescription use of the same pain killers or escalation to illegal drugs. One very recent study showed that even a single opioid prescription increases the risk of addiction and overdose. There are a number of tragic anecdotes showing that this has happened to individuals following work injuries, but again very few formal studies that help us to quantify this entire pathway and the number of people who might be affected in this way.
Example of how to estimate the contribution of work
Using published data on the prevalence of ergonomic hazards and job stress, and their relative risks for the diseases of distress, we can calculate attributable proportions and population attributable risk percentages. We selected to characterize one of these possible work-related pathways.
Low back pain is one of the most frequent work-related conditions in many countries. The fraction of back pain in the United States attributable to work was estimated in a WHO report to be 30%, and even higher (about 38%) for the age group of 45-59 year-olds. If we take a conservative estimate of the prevalence of primary chronic back pain in the U.S. as 8%, then – out of the total US population (318,900,000 in 2014) – almost 26 million people would have low back pain. That pain would be due to work for 30% of these people, or 7.7 million (representing 2.4% of the total U.S. population). Our next step will be estimating the number of the 7.7 million who would obtain opioid medications – either on or off prescription – and finally, how many go on to have substance abuse problems.
Recall that the original paper by Case and Deaton that got us started on this effort concerned the increasing rate of deaths due to drug overdoses and other ‘diseases of distress.’ Whether or not the number of people with work-related back pain has increased over time would depend on whether (a) the occupational risk factors had become more common or more severe; or (b) if the number of people exposed to those factors had changed. Both are possible.
Our goal is to define some potential pathways between occupational exposures and three causes of death. However, to avoid an excessive number of assumptions piled on top of each other, where possible we have skipped over some of the hypothesized intermediate routes. We decided to focus on simpler pathway explanations if they can be justified from the available evidence.
One big challenge is finding adequate data for some of the proposed associations in our models. Ideally, we would have data on the prevalences of different exposures (job stress, work overload, etc.) and their relative risks specific to the U.S., the years in question. We would also like to know if the risks are different for various age and racial subgroups. Unfortunately, epidemiologic surveillance on the diseases and causes of death is far more common than exposure surveillance, meaning the trends in job hazards.
Another challenge we face with this effort – a more empowering one – is having to decide our assumptions. The first and most important was about which evidence is relevant, especially concerning where the relative risks have been documented. When data are not available from studies conducted in the U.S., we have tried to find data from countries of similar levels of economic development or from pooled risk estimates derived from meta-analysis. However, differences in extent of good ergonomic practices, unionization of the workforce, and the social/economic safety net could have important effects on the risk.
The second assumption is that the relative risk was constant during the period of interest, although the exposure prevalence could vary. We made this supposition because it seems reasonable – even in the absence of statistical data – to assume that changes in ergonomic and psychosocial stressors may have resulted from the profound economic restructuring since the 1990s in the U.S., including reduction of the manufacturing sector, increase in the services sector, stagnation of wages, and loss of unionized jobs.
Unfortunately, it is not feasible to model all of the possible interactions among the causal factors, even though these seem very plausible and to some extent have been shown in the literature. Furthermore, we have not been able to factor in all of the later consequences of an original injury or stress-related condition. For example, there has been a lot of news about dramatic shrinkage of the workers’ compensation system in many states, shifting the burden of those costs onto the workers themselves and their family members, as well as tax-payer supported programs (see here, here and here.) Even before this, workers injured severely enough to lose time from work have lost a lot of income compared to what they were on track to earn over their lifetimes, despite compensation payments, and they are therefore unable to save as much as workers not injured, everything else being equal (see here and here.) They are also more likely to be injured again – which is especially predicted by early employment in hazardous jobs. All of this taken together means that many injured workers get caught in a vicious cycle of financial loss that could certainly impact their health over time.
We are continuing to review the literature to identify relevant estimates of the relative risks linking variables in our causal pathways, and to assess whether or not there have been recent increases in either exposure prevalences or the severity of their health consequences. For example, we are still seeking data sets and studies that have examined the relationship between job stress and substance abuse, recognizing that data are necessarily limited on the supply and characteristics of illegal drugs.
We will continue to quantify the attributable risks for the pathways from job-related injury or stress to suicide, alcohol and drug over-use. We will also try to examine differences in incidence among regions and subpopulations, as these differences might provide evidence of causal hypotheses.
Last, the assumptions involved in these computations are substantial. We plan to list them more fully than above and to examine the data that we can find which will help us to assess to what extent they can be justified.
We would welcome hearing from others who have undertaken similar computations, who have relevant studies (including grey literature) to share with us, or who would be interested in co-authoring a manuscript for peer review based on this work.
Laura Punnett, ScD is an occupational epidemiologist whose research interests include the contributions of working conditions to socio-economic disparities in health. This work was developed by a group of doctoral students at the University of Massachusetts Lowell who participated in an Advanced Research Methods seminar in the fall of 2016: Kelechi Adejumo, Scott Fulmer, Patricia Huerta, Gonzalo Lopez, Neha Sahasrabudhe, Kevin Schott, Noor Sheikh, and David Skinner. We thank David Kriebel for suggesting this investigation.