Death seems like it would be a pretty discrete data point with little ambiguity. After all, you either are or are not dead. But as it does so often, the meaning of death in health research is more complex than that simple binary choice. A new study of death and dying in Denmark provides food for thought about how best to use that discrete endpoint to better understand the care we provide.
I have written previously about what mortality is not always the best metric when considering the care we provide. The most significant advantage of death as a metric is that it is definitive. Of course, COVID did remind us that death certificates often reported what we thought happened, not what actually transpired.
Another popular way to consider death is years of life lost (YLL). Economists love that measure of the difference between when you died and the expectation of an average lifespan because it provides a way to quantify the $ value of disease or treatment. But those calculations rely on several “simplifying” assumptions, including estimating an individual's normal lifespan.
The researchers considered the mortality risk ratio (MRR), "the ratio of mortality rates between persons with and without a diagnosis of the specific disorder adjusted for age, sex, and birth date.” The MRR changes when we are diagnosed with a disease, especially those we would all consider life-altering, like cancer or diabetes. We can also calculate the MRR associated with less lethal diseases, like allergies. Often these risks are expressed as survival, a variation in life expectancy. We tend to think of this as all-or-none, too; think of all the popular press about cancer survivors; is there some lower limit on how long you must survive to be considered a survivor?
The researchers used a relatively complete dataset of Danes’ diagnosis and subsequent survival, uncovering some fascinating insights. They considered the fate of 7.3 million Danes over 18 years, 2000-2018. In addition to categorizing patients by their diagnosis, they classified deaths as "external" when they involved trauma, suicide, and homicides, and “natural,” for any other cause. They also correlated residential air pollution with fatalities, “as air pollution is a prominent environmental health threat.”
- 7.38 million individuals, slightly more females, with an average age at entry of 31, were followed for a mean of 13 years – resulting in 101.7 million person-years (Don’t you love these mashups?)
- 14% of the cohort died during the 19 years of follow-up; slightly more females than males.
- 5% died from external causes – accidents, suicides, and homicides.
- Based on ICD-10 classifications, the most prevalent disorders were circulatory, e.g., cardiovascular, followed by neurologic, and mental health.
As expected, MRRs and YLL ranged dramatically across the 39 categories. For nearly every case, MRRs increased from as little as 1.09 (a 9% increase) for vision problems to 7.85 (a 785% increase) for chronic liver disease. Same for YLL, with a reduction of 0.31 years (4 months) for allergies and 17.05 years for chronic liver disease. Two categories, hearing loss, and migraines were associated with a slightly decreased MRR and a somewhat longer life expectancy. (40 days for hearing loss, 180 days for migraines)
Rather than get lost among the trees, it would be better to consider the view of the forest.
Developing a chronic illness was associated with increased MRRs and YLL – duh. But it is when you look at the pairing that the “important nuances were revealed.” The researchers cite the example of cardiovascular and pulmonary disease. Both have very similar MRRs, 2.90 and 2.94, respectively. But YLL is a different story; cardiovascular disease shortens your life by 3.8 years, pulmonary disease by 8.27 years. The underlying reason for this discrepancy is the earlier onset of pulmonary disease (55) rather than that of cardiovascular disease (66).
Reflection on the underlying mathematics would tell you that the earlier the onset of a disease, the greater the years lost for any given mortality rate ratio. That is why conditions that are particularly lethal around the time of childbirth and childhood have very large YLLs
COVID taught us that MRRs, which in that context are mortality rates per infected individual, varied by age but also changed over time. As it turns out, this temporal variation in mortality, what the researchers called a “temporal signature,” occurs for most chronic diseases. During the first six months after a diagnosis, MRRs are elevated – perhaps reflecting the presence of more advanced disease at discovery, as would be the case in ovarian and pancreatic cancer.
After that initial six-month period, MRRs followed three main trajectories, although remaining elevated compared to the population without the chronic illness in question. For some chronic diseases, the MRR stabilized, slowly chipping away at the cohort. In other conditions, the MRR lessened, reflecting perhaps more efficacious treatment or remission. Finally, for a vanishing few, like dementia, the MRR increased over time.
Mental illness, chronic liver disease, and epilepsy were the three conditions where years of life lost were greater than one year because of “external causes.” Remember, external causes include suicides and accidents, both readily understandable reasons for the additional years lost.
“Although air pollution has been found to be associated both with mortality and a wide range of health conditions, we found that the relationship between the disorders and mortality rates was not substantially altered in models adjusting for air pollution exposure.”
In this study air pollution did not affect mortality. That certainly goes against both intuition and the mainstream view; the researchers were quick to point out that “air pollutants were modeled at the residential address during only 1 year, and they might be poor indicators of the real individual accumulated exposure.” Of course, we have argued the same point for the studies showing an adverse impact of air pollution on mortality. Could this suggest, despite the drumbeats of the media, social and otherwise, that this is not “settled science?”
The study has other limitations. It did not collect data from general practices, only patients seen in a hospital setting. It does not tell us anything about the quality of their lives or their socio-economic circumstances. Most importantly, it looks at solitary chronic illness, which is rarely the case – 27 % of US residents have more than one chronic illness.
If we want to gauge the success of healthcare in the US or elsewhere, deaths while easily measured and ambiguously interpreted. Perhaps we should consider to what degree we “bend the curve” and alter the mortality risk ratios’ trajectories for chronic illness. Or quantify a reduction in years of life lost as a measure of how many more we keep alive.
For those of you with an interest, the researchers have gathered their data into an online interactive dataset that you can use to see how these variables change with your chronic illness(s). It can be found here.
Source: Analysis of mortality metrics associated with a comprehensive range of disorders in Denmark, 2000 to 2018: A population-based cohort study PLOS Medicine DOI: 10.1371/journal.pmed.1004023