What does it mean for a disease or a medical condition to cause a death? In the easiest case, there is a straightforward story that one can tell about why someone died. “He was in good health, he got Covid, it caused pneumonia, and he died.” The implication, presumably, is that if he had not caught Covid, he would have lived a completely “normal” life for several years thereafter.
Subtleties arise when we begin relaxing some of these conditions. Suppose the man was not in good health to begin with. Or suppose that Covid weakened his heart and he died of a heart-related condition months later, well after the virus could be measured at detectable levels in his body. (It’s worth noting that this latter type of death would not be counted in our official statistics.) In the straightforward story, health is an almost binary condition. It is either good or bad, and the disease caused it to switch from one state to another. The chain of causal links is thus very taut. In the more complicated stories, that chain slackens. Health is less of a switch and more of a spectrum, and death becomes an accumulation of cosmic insults. For something to “cause” death, in the most elastic interpretation, it need not be the sole, first, last, or even primary contributor. It simply needs to be the case that, in the counterfactual scenario, where it did not exist, the person would not have died, at least right then and there.
Naturally this type of causation is more difficult to reason about. For one, it manifests itself more readily in aggregates than in individual cases. One can imagine a disease that killed hardly anyone, at least directly, but weakened the body considerably. (Arguably, HIV and cancer-causing viruses work like this, although I’m a bit out of my depth here.) No one’s death certificate would have “HPV” written on it, but, if HPV did not exist, thousands of people would otherwise have lived many months or years longer. It’s a bit like the relationship between global warming, weather, and climate. It is difficult to pinpoint a particular weather event that is “caused” by global warming. But it is quite easy to show that, in aggregate, the climate is getting hotter, more extreme, etc.
Many of the confusions about Covid mortality, whether innocent or not, stem from this fact. Right-wingers, including Donald Trump, have claimed that Covid deaths are grossly overreported, because many of those who died had “preexisting conditions”, such as obesity, diabetes, and asthma. It may be true that, without these preexisting conditions, these individuals would not have died from Covid. Yet it is equally true that, without Covid, these individuals would have lived years or even decades longer. Most of these conditions are perfectly manageable with medication and lifestyle changes, and, as an example, my parents have now lived with type II diabetes for 25 years. To suggest that they deserve to die (which is the implication of the argument) is both embarrassing and gross. There are related confusions: the difference between being hospitalized “with Covid” and “because of Covid”; the matter of how to mark death certificates; and so on. These difficulties are largely bound up with the multi-causal nature of ill health and death, and the trouble we have slotting that reality into our mono-causal way of thinking.
Some commentators have suggested circumventing this problem by considering “excess mortality”. This is a statistical measure based on subtracting the number of deaths that actually occurred, irrespective of cause, from the number of deaths that would have occurred in a “normal” year (based on “normal” death rates). Using excess mortality heads off some difficulties but creates yet others. It measures what would have happened in absence of the pandemic, not in absence of Covid. As such, it includes effects like increases in homicides, overdoses, and traffic deaths that arise from the societal disruption caused by the pandemic, and it also includes indirect effects like overburdening of hospital systems. To categorize all of these “Covid deaths” seems like a stretch. (Of course, we can try to adjust the figures to remove these “extra” deaths, but then we lose the appeal of a “simple” statistical measure.)
I don’t mean to spend this entire essay talking about Covid: it is simply the easiest example to grasp. Instead, I want to talk about other influences on ill health, disease, and death, ones that are even more difficult to grasp in a mono-causal framework. One example is environmental toxins. Here’s David Roberts at The Grist, nearly 10 years ago now, talking about the EPA’s new mercury regulations:
Wednesday, at long last, the EPA unveiled its new rule covering mercury and other toxic emissions from coal- and oil-fired power plants.
Anyone who pays attention to green news will have spent the last two years hearing a torrent of stories about EPA rules and the political fights over them. It can get tedious. After a certain point even my eyes glaze over, and I’m paid to follow this stuff.
But this one is a Big Deal. It’s worth lifting our heads out of the news cycle and taking a moment to appreciate that history is being made. Finally controlling mercury and toxics will be an advance on par with getting lead out of gasoline. It will save save tens of thousands of lives every year and prevent birth defects, learning disabilities, and respiratory diseases. It will make America a more decent, just, and humane place to live.
(One might wonder — how did these plants escape previous environmental regulation, like the Clean Air Act? Roberts explains that they were grandfathered in, “on the assumption that they were nearing the end of their lives and would be shut soon anyway. Well, funny story … they never shut down!” So, newer plants (post-Clean Air Act) emitted far less mercury than did older ones, but the older ones were still present in large enough numbers, and were inefficient enough, that they constituted a large fraction of all power plant mercury emissions.)
I found a funny article by the “Institute For Energy Research” (a Koch-backed industry think tank) about a Paul Krugman blogpost praising these regulations. The article is written by Robert P. Murphy, a “Senior Economist”. (Mr. Murphy, ten years later, is banging the same drum, and has a 3 part series called “Should We Trust the Climate Models?”). The article intends to be a comprehensive rebuttal, and it accuses Krugman of “botching” his commentary. “The deeper one digs into the actual science backing up the wild claims of Krugman & Co. on mercury regulation, the weaker their rhetoric becomes. The implausible assertions in Krugman’s blog post were generated by a comedy of transgressions, ranging from (perhaps honest) poor paraphrasing, to reporting the top number of a wide range, to justifying regulations on air toxics emissions based on dubious models of the health effects of particulate matter.”
A few of the claims warrant closer scrutiny. The author argues that the vast majority of the benefits from these regulations are not from regulating mercury at all. Instead, they are from regulating PM2.5, the micron-sized particles that are easily inhalable and get trapped deep in one’s lungs, exacerbating conditions like asthma. This claim actually appears to be correct (see picture above), and I was surprised that both Krugman and Roberts omitted this important detail.
The author, as I noted above, further complains about “dubious models of the health effects of particulate matter”. In other words, he argues that, even if we accept the claim that PM2.5, not mercury, is the primary toxin causing harm, the causal link between levels of PM2.5 in the atmosphere and ill health or death simply cannot be substantiated. He cites a “Dr. Anne E. Smith” (who, once again, has a background in economics, not health or medicine, and works for an economics consulting firm, NERA), who wrote
And yet, EPA identifies not a single death during 2005 that was attributed, even in part, to exposure to ambient PM2.5. If PM2.5 is indeed having this estimated effect on the public health, there is no evidence indicating when or where these events occurred, or who was affected. Rather, these mortality estimates are merely inferences drawn after making a host of assumptions about how to convert a statistical association into a concentration-response function. No one really even knows what types of deaths might be implicated.
Let me try to provide some context. Dr. Smith is correct that it is difficult to identify anyone who died of PM2.5. No one’s death certificate is marked as such. Yet it is likely true that PM2.5 is killing (tens of thousands of) people. Of course, proving this relationship requires using statistical models that connect levels of ambient PM2.5 to deaths. This is the “concentration-response function” alluded to above, where “concentration” refers to the concentration of PM2.5 in the atmosphere, and “response” refers to the outcome: in this case, the number of deaths occurring. These models make a host of assumptions to try to elucidate a causal relationship from non-randomized, observational data, as I discussed in another essay. I have certainly not read the papers that develop these models, and I cannot comment on their veracity. But it’s worth considering what Dr. Smith’s argument implies. If we cannot develop a strong, straightforward narrative that connects a small increase in a toxin to a particular person’s ill health or death, then can we really say that these things are “causing” ill health or death at all? If taken seriously, we are left with the repugnant conclusion that companies can dump whatever they want into the environment, so long as these toxins are not present at levels that will kill us outright (in a mono-causal, as opposed to multi-causal, way).
Murphy is correct to note that the concentration-response function is rather uncertain. We don’t know precisely how many lives will be saved by reducing PM2.5 levels in the atmosphere through regulation of these old power plants. But, even taking the low end of the estimate (i.e., being conservative), the benefits massively outweigh the costs. (Murphy also argues that “saving lives” and “preventing premature deaths” are not the same thing, although he appears to be regurgitating the exact same fallacy as Dr. Smith.)
Although this essay is already getting long, it might be worth explaining how these estimates of “benefits” are constructed. There are monetary values assigned to various health conditions (and death itself): a hospital admission for pneumonia is $23k, one for asthma is $10k, having to miss a day of school is $89, and dying prematurely is ~$7-8 million. (Some of these estimates strike me as being absurd, but no matter. They are based on a very economics-brain notion of “willingness to pay” (WTP). The EPA writes, “For example, the value of an avoided respiratory symptom would be a person’s WTP to avoid that symptom”. I can’t even begin to contemplate that question.)
This means, in practice, that if the EPA lacks a concentration-response function for a certain condition, or, instead, if it cannot assign a monetary value to a certain outcome, then it simply omits these benefits from its estimate. There is no monetary value assigned for better air quality/visibility from reduced levels of PM2.5. The same is true for effects of mercury on the “ecosystem”, and for “cardiovascular effects of Hg [mercury] exposure” and “other health effects of Hg exposure”. The EPA also admits, separately, that it cannot quantify the effects of PM2.5 on rates of “subchronic bronchitis cases”, “low birth weight” babies, “pulmonary function”, and other types of ill health.
In other words, if PM2.5 harms your health but that does not rise to the level of a medical condition in the EPA’s list, it simply does not care. So, if anything, the benefits of environmental regulation are likely underestimated.
PM2.5 is the clearest example of a toxin that is underregulated. The benefits of regulating it are in the (literal) trillions of dollars, and the Clean Air Act is one of the most successful pieces of legislation ever passed, from a cost-benefit perspective. Once we accept that death is a multifarious thing, and therefore that even small reductions in levels of toxins can markedly improve public health, the logical implication is that we should do even more. We should eliminate gas powered leaf blowers. We should regulate particulate emissions from brake and tire wear. We should regulate power plants, outlaw coal, and reduce traffic, particularly near heavily populated areas. And we should expand our scope well beyond PM2.5, to reproductive toxins, carcinogens, and many other chemicals for which there is a cost but, whether because of industry pressure or mere negligence, we have not yet quantified it.