Sunday Science Lesson: “X people die from smoking” – what does that even mean?

by Carl V Phillips

In the comments, I was recently asked a version of the question in the title. I started an answer and was going to add more, but decided it would make a good post. So here it is. The original question (slightly edited) was:

How long must a person be a nonsmoker before his eventual, inevitable, death is regarded as unrelated to smoking? Or is it that if you ever smoked, no matter how long ago, and eventually die of something that might be smoking related, then it is counted as a smoking caused death? For example if somebody smoked, socially, for 10 years between 1955 and 1965, and then ceased smoking. If he happened to die today at age of 75 of heart disease or bladder cancer, is that death listed as smoking related?

This question captures numerous points of confusion that thinking people run up against when they try scientifically interpret popular epidemiologic claims. When a casual reader reads epidemiologic claims at the level he reads about astronomy, he is unbothered. (“They just observed a new planet circling a distant star? Cool.”)  But when they try to really make sense of the simplistic statements in the news, they start to realize that what it seems to mean cannot be quite right. (“But, wait, what does it even mean to observe something at interstellar distances? They say something about about perturbation of motion, but what does that mean? And how much of that you need before you claim there is a planet? Also why do they say it is like Jupiter even though they cannot really see it? How precise are these claims that are phrased with the same precision of someone discovering a new species of newt that they actually picked up and looked at?”) Of course, with astronomy, the next thought probably is, “I have no idea what this really means, but I do not really care because I trust the people who are making the claims, and even if I did not, who cares whether it is even true?”

But for epidemiology, we often do care if it is true and (quite reasonably) do not trust the claims. At that point, the realization that one does not even really understand what is being claimed becomes rather frustrating.

I will unpack a few of the complications embedded in the above question. It is going to take a few posts, and you will probably wonder if I am wandering into the weeds and not addressing the question. But the proper answer is really buried that deep.

What does it even mean to say a death was caused by smoking (or whatever)?

First off, “caused” basically means “if this one factor had been absent and everything else the same, the event would not have happened.” You can read more about this in my recent paper.

But as noted in the question, it cannot possibly mean “caused a death that never would have otherwise occurred”. Such phrasing mostly makes sense for something like causing a case of lung cancer, which indeed probably would not have occurred otherwise. But not for deaths. It turns out that in epidemiology, the textbook definition of a cause is that it caused something to occur that otherwise never would have occurred or it caused it to occur sooner than it otherwise would have. So that explains how something can be a cause of a particular person’s death.

But, wait: Sooner? How much sooner? Well it turns out that the textbook definition is quite bad, in that it does not clarify this, and thus technically says that if the event occurred even one second sooner, it counts. That is obviously not what anyone cares about or thinks they are being told when causes of death are discussed. The practical importance of this disconnect is dramatically reduced by the fact that we can never have enough data to detect outcomes occurring one second sooner. On the other hand, it allows those who want to exaggerate claims about deaths from smoking to say most any death of a smoker or ex-smoker belongs in their tally. If a smoker dies from anything other than being instantly killed by trauma, chances are that the smoking hastened it by at least a few seconds. Smoking permanently damages to heart and lung capacity, and almost every time someone dies, he would have hung on at least few more seconds if his heart and lungs were stronger.

It turns out that those trying to exaggerate smoking deaths do not go quite so far as to just assume almost all deaths of smokers and ex-smokers were caused by smoking for this reason. They use other methods to inflate their estimates. But because of this technicality, they are technically right when they pump up the numbers. Indeed, by that technical definition, they are actually understating their claims.

But, of course, that is not what anyone thinks they are being told when they read death statistics, nor what anyone cares about. Someone reading a claimed death risk or tally assumes it refers to deaths that were substantially hastened. The readers probably do not think it through (and if they do, they start running into those “hey, wait a minute…” questions), but probably implicitly assume it must refer to deaths that occurred years earlier than they otherwise would have. The primary upshot of all this is that just referring to deaths caused by something like smoking or eating badly — anything that  slowly erodes someone’s vitality — is largely meaningless. To be meaningful, such claims must be specified in terms of the minimum amount by which a death is hastened before being counted, or need to switch to a better metric like years of potential life lost (YPLL).

It turns out that attributing a case (incidence) of bladder cancer to smoking is rather cleaner than attributing a death to smoking. Few people (though not none) are on a downhill slide toward bladder cancer, which the smoking causes to happen just a bit earlier. However, even then, almost all the deaths from bladder cancer among smokers, even for cases where the cancer incidence was not caused by smoking, are bladder cancer deaths caused (i.e., caused to occur at least a second earlier) by smoking.

Did I mention this is complicated? It is far more complicated than most people spewing out epidemiology results are even aware of, let alone understand.

Circling back, when “deaths caused by smoking” are estimated from data (usually badly — read on) we can look at what was really done to see what it really means. In most cases what it really means is “caused someone to not reach her next birthday”. The good news is that this very rarely counts someone who died one second earlier, and averages out to a reasonable six months. But this is obviously not what you would ever choose to measure. The reason this is inadvertently measured is because any remotely valid method for estimating health risks separates people by age, for the obvious reasons that this is a big influence on their chance of dying during a given period of calendar time. Age is pretty much always measured by the step function {how many birthdays someone has had}. So if someone would have died at 75 years + 10 months had he never smoked, and smoking causes him to die the day after he turns 75 rather than ten months later, he would not be counted as dying earlier from smoking.

[An even more technical aside for those finding this too easy to follow: Correctly separating people out by age really calls for stratifying by age: comparing 75-year-olds only to other 75-year-olds. Age is just too big a confounder and its effects are too complicated to do anything else. Most studies, however, just put in a single “control” variable for age, a linear count of how many birthdays someone has had. This does not really work because it assumes (i.e., forces the statistics to conform to the assumption) that the effect on risk (usually measured as a ratio, but this depends on what statistical model is being used) of aging one year is the same for each year: aging from 54 to 55 changes risk of dying (or whatever) by the same amount as aging from 74 to 75. It also assumes that increase in risk from having smoked is the same for the 55- and 75-year-olds. These are obviously not going to be true.]

I am sorry to say that this is only the first step on the way to understanding the answer to our question.

What are we trying to measure in order to estimate this?

Any good scientist thinks through what causal contrast she is trying to measure. In epidemiology — where this is much more complicated than almost any lab or clinical science, and where data that may superficially seem to answer a question really does not do so very well — this is especially important. Yet it would probably be optimistic to say that 1% of the research projects in epidemiology involve any such thinking.

For the question of totaling up deaths from smoking (say, last year, for concreteness), the causal contrast of interest is “which of those deaths would have occurred later had none of those people smoked, but everything else in the world was held the same?”

[With apologies: Actually, that is not quite right. As already noted, “later” is not what we really want to know and that should be translated into “more than some specified amount of time later”. Also, not quite everything else in the world could be the same, since if someone did not smoke, this would have ripple effects that created other changes; these could not be held the same in the hypothetical experiment.]

If we were given that information by an omniscient force that can observe the alternate reality (call it divine data intervention, or DDI) then we would have what we needed. Not very practical, but if you do not think through what you would ask of DDI when starting your research, you are almost certainly doing crap work. The non-divine, though still annoyingly impossible, version of this would be to conduct an experiment about the causal contrast: re-run the history of the world, but with those smokers who died last year never smoking, and see if they also died last year in that history.

Oh, but wait. That is not quite right either unless you have very narrowly defined the rather strange question, “how many of the ever-smokers who died this year would have died (much) later had they never smoked?” You can ask whatever question you want, of course. And that is roughly what most of the reported statistics are attempts to answer. But that answer does not really means what people think it means. It is not a general measure of the real effect of smoking. For one thing, it does not subtract out the smokers who did not die last year, but would have died last year had they never smoked.

Wait, what? Smoking does not keep people from dying! Well, actually it inevitably does sometimes. Some people’s lives are extended by their smoking. The reported statistics are designed to ignore that, even though it should be subtracted out. (Though in fairness, it is not such a big number that it matters much, and it is basically an accident of the methods that are unthinkingly employed, rather than being an effort to intentionally bias the results.)

But that is not the major problem I was referring to. The bigger issue is that smoking causes quite a few people to not die in a given year because it caused them to die earlier. This is obviously in no way a consolation from a practical perspective, but it does make some interpretations of the statistics badly wrong. (E.g., Did 400K more Americans die last year, than otherwise would have, because of smoking? Nope. Were it not for smoking, basically the same number of deaths would have occurred, it just would have been a different list of people. This is critical for the bullshit claims about smoking increasing healthcare expenses, which skyrocket in the year someone dies.)

This means that if you really wanted a serious understanding of what seems like the right question, the hypothetical experiment described above is not really the one you want to run. Instead you would want to run a history of the world with no one smoking and compare it to reality. This would show what portion of smokers have their lives substantially shortened by smoking, and let you drill down the such questions as that risk for someone who smoked for ten years and quit in 1975. It would also let you compare numbers of deaths at the population level, and sort that out by age to report the age-specific effects of having smoked.

Oh but just one more complication. (Sorry.) If you are making a claim about the effect of smoking for the average person, as is common in anti-smoking propaganda, even this is not the right experiment. The people who chose to smoke are different from those who chose not to, probably in ways that change how much smoking affects their health. So really you need to run an alternative history where everyone smokes to get at the risk for the average person. Moreover, if the propaganda claim is the more typical claim about how much starting smoking now will affect someone’s future risk, you need to run the two alternative future histories of a world, since the next 50 years will be a lot different from the last 50 years in many ways (most notably, medical technology) that affect how big a risk smoking poses.

Fortunately the original question excuses us from having to deal with that complication, which is good because no real data can possible provide a good substitute for those future-history experiments. We can now move on to looking at how real data can be best used to substitute for that missing DDI or historical experiment. Or rather, we can do it next week, because my head is about to explode and I am guessing yours is too. As I said, epidemiology is complicated to a degree that it is far beyond the abilities of 99% of “public health” people.

43 responses to “Sunday Science Lesson: “X people die from smoking” – what does that even mean?

  1. All I did was try an ecig, I never guessed it’d mean I’d be re-awakening my Philosophy of Science brain, long dormant from disuse. Unintended consequences indeed, loving it.

    • Carl V Phillips

      I have no substantive response to that. I just had to say I appreciated it. However, I would say this is far from my most HPS post, so look through the archives if you haven’t.

      Interestingly, most of what appears in this one I credit to the (tiny) cadre of deep thinking epidemiology people (George Maldonado in particular) and not to my HPS influences. I learned a lot from them too, but it was a bit… well, less practical.

  2. “ head is about to explode and I am guessing yours is too.”

    LOL! Well said…..this post make Javascript seem simple:-)

  3. Carl
    I once would have assumed that, “smoking related”, was related to people who had smoked for a fairly longtime, or were smokers at or about the time of diagnosis. And that it was mostly involved diseases of lungs or heart disease- Thanks.
    Could I ask another question? If the hypothetical person in my earlier question was, by 2001 very obese and had type2 diabetes and eventually died of heart disease and other type2 related health issues. I presume that his death would be listed as a type2/ obesity related death,yes?
    Does this mean that there is also double counting going on, i.e that the combined number of smoking related deaths, and all other deaths attributed to other causes, could be more than the total number of deaths of ,individuals?

    • Carl V Phillips

      I ought to cover these points in the main thread as it continues, but just in case I miss them in all the chaos:

      It is not a matter of double counting. They both count legitimately. That is, every event has multiple causes. If an obese smoker would not have gotten a particular disease had he not smoked or had he not been obese than both the obesity and the smoking are causes of the disease. If he dies a few months earlier than he would have due to smoking and a few months earlier still due to obesity, then his death was caused by both. It was also caused by him being born in the first place, surviving the 1980s (it was a dangerous time!), and an infinite list of other causes.

      It is the case — and should be the case — that if you add up, across diseases or across causes of disease, all deaths caused by each entry on the list, you will get a sum that is far larger than the total number of deaths (or whatever endpoint you are using). It is really just a matter of what someone thought to put on the list. To take a very simple example for illustration, one of the causes of prostate cancer is being male (if that variable is different, the disease does not occur, so it is a cause) and another is reaching age 50 (which is similarly a cause of most cases); add up the cases of PC that were caused by each of those, and you get very close to double the number of total actual cases.

      One immediate implication of this is that you cannot legitimately calculate the total number of deaths due to smoking by adding up, across diseases, the number of deaths from that disease that are attributed to smoking. However, this is exactly what some of those junk calculations do. If each entry in that sum is estimated correctly, then the total will be far higher than the real number.

      The other point I want to riff off of this comment is that “smoking related” is not a phrase that should be used in science or other serious analysis. (As I am guessing you realize, thus the scare quotes. I am just running with it.) It is a sloppy journalism-quality term that is not well defined. A disease is related to smoking if smoking causes it, or if smoking protects against it, or if they happen to occur in the same populations by coincidence (all of those can properly be referred to as associated with smoking, which is a meaningful term, but it does not mean what people who use it think it means). It is also related to smoking if it starts with the same letter or is studied by some of the same people — not a very interesting relationship, but it is a relationship.

      These terms get used as weasel words in popular discourse, intentionally communicating them message “this is caused by smoking” without actually saying it.

      • Many thanks
        The scare quotes were deliberate; Ceci n’est pas une pipe.
        A representation is never what it represents, that’s the power, usefulness and risk inherent in ; the map is not the country.

  4. Carl, thank you for a very interesting and well-thought-out analysis. I’ve played with this kind of thinking myself a few times but the combination of not really knowing enough while simultaneously facing just too many variables usually left be scurrying back to my bag of chocolates in the corner. :>

    I’m unsure about one aspect of your analysis. Where you say:

    “So if someone would have died at 75 years + 10 months had he never smoked, and smoking causes him to die the day after he turns 75 rather than ten months later, he would not be counted as dying earlier from smoking.”

    It makes it sound as though the deaths are actually COUNTED… and I’m pretty sure this isn’t the case. SAMMEC, so far as I know, simply takes the various variables and formulae that are fed into it by (quite likely antismoking) researchers as regards different diseases and the theorized effect of smoking distilled from the studies on those diseases and comes up with a number/ratio that can then be applied to the total number of deaths from a particular disease in a particular age range. Change some bits ‘n pieces ‘n assumptions in the input and formulae and your final numbers will come out very differently.

    Do you remember the royal stink Glantz et al made about ten years ago when some foolhardy nutrition researchers tried to make the claim that obesity killed more people (in the US I think?) than smoking? I don’t remember all the details, but I *do* remember a LOT of hell being raised until the claim was “corrected” to satisfy the Antismokers so that they could continue parading smoking out there as the “most deadly scourge known to humanity!”

    In general, the way I think the figures are worked out (or close to it) is like this:

    Say the average age for everyone who dies from a heart attack is 75. Obviously half the people who die from heart attacks are then dying “prematurely.” In the ideal, non-political world of medical science (Yes, let’s pretend it exists for the moment.) We’d take the data on smoking, obesity, physical activity, drinking, drug use, marriage/divorce status & timing, socio-economic position, job & job loss timing and status, psychological health and treatments, geographic habitat, pet ownership, neighborhood noise levels, and a few dozen other “heart health factors” into account, examine and synthesize all the studies done on each factor, further examine and synthesize all the studies done on each combination of each factor doubled/tripled/quad’d/etc’d with each other for all of them…

    and eventually come out with an answer that would say, “This person died of his heart attack at 73 instead of 75, 1% because he ate too many deserts, 3% because he didn’t have enough sex, 2% because he spent too many hours reading on his computer, etc etc etc

    We’d do this for EVERYONE who died before 75, and then multiply all the factors out to be able to say that ON AVERAGE, of the 13,617 folks who died that way last April, 273 of them died because of their smoking, 14 of them died because their wife left them the month before, 225 died because they ate too many rich deserts, 114 of them died because they drank too much while 113 of them died because they didn’t drink enough, 42 died just because it’s the right answer, 3 died because of the kid next door playing his electric guitar too loudly every night… and so on.

    NONE of the individual deaths could be “blamed” on any one of those factors, and indeed, it’s quite possible that someone who had ALL of those bad factors might have died at exactly the same age from the same heart attack anyway even if he did NONE of those naughty things.

    Contrariwise, there would be a few folks over 75 who WOULD have died in their 20s or 30s if they hadn’t stopped at the street corner for a moment to light up a very lucky Lucky.

    IDEALLY all that stuff could be figured in properly and accurately so the blame for at least the gross number of deaths could be properly apportioned in case St. Pete was interested. But in reality the “properly and accurately” falls apart as soon as you have researchers WANTING some factors to be more “guilty” than others. The “fraud” rate for research in general may be only 15% or so, the fraud rate among researchers who are jiggling their results not just for money or jobs or publication…. but because the jiggled results will be “for the greater good” in modifying society and behavior … well at that point the 15% might rise to 20, 25, or even 30%. Throw in the fact that there’s no “double-blind” generally in this sort of research, and even HONEST researchers are likely to often be at least somewhat guilty of unconsciously biasing their design, gathering, reading, and interpretation of research. Remember the researcher I cited in TobakkoNacht who took a 1 point lowering of girls’ teenage blood-pressure as being “a possible cause of concern” if her parents smoked? Do you think that same researcher would have been “concerned” if an apple a day correlated with the same 1 point decrease?

    We do NOT live in anything even remotely resembling an ideal world when it comes to antismoking research, and because of that I have no real faith in there being any sort of accuracy at all in the SAMMEC type analyses that are made. They may correlate loosely with reality, but “loosely” is definitely the operative word.

    Carl, thanks again for your careful thinking and analysis. I hope my own blitherings here don’t detract from it — they’re certainly not as carefully thought out and stated — and I look forward to seeing more of the discussion your piece will generate!

    – MJM

    • Carl V Phillips

      There are many ways to come to a tally that are complete garbage. I am addressing (so far anyway) what might be done correctly, according to the standards of good epidemiology (which is far from perfect, but far better than tobacco controllers do). By those standards, the very simplified version is: look at what portion of 75-year-old smokers have heart attacks; compare to otherwise similar nonsmokers of the same age; attribute the difference to smoking. To the extent that smoking moves a heart attack earlier, but it is still at age 75, that would not be part of the difference.

  5. “compare to otherwise similar nonsmokers”

    Ay, there’s the rub (so to speak). “OTHERWISE SIMILAR.” Yes, I’m sure some researchers have tried to be diligent in that regard, but they certainly aren’t omniscient. In the secondhand smoke and children area, not even Judge Osteen tried to dispute the EPA’s finding that children exposed to smoke at home got more respiratory infections. But WERE they in conditions that were “otherwise similar”? Very few would dispute the statistic that smokers themselves are more prone to such infections, and if indeed that’s true, then their children would certainly be in a heightened position in which to catch such infections — not because of smoke exposure but because they hang around parents who get infections more often due to their own smoking. But I don’t recall that rather glaringly obvious confounder being corrected for very often… do you?

    And in terms of our topic at hand, there could be all sorts of correlated confounding factors that we simply aren’t aware of or haven’t admitted for various reasons. Did smoking cause that particular premature heart attack? Or could it have been an HPV infection? Did smoking pure tobacco cause it? Or was it only smoking commercial cigarettes with their various additives? Did nuclear fallout play a big role in making tobacco smoking far more dangerous due to the plants’ uptake of radiation from the soil and fertilizer? Or, a favorite speculation of mine for lung cancer, how much has the inhalation of the highly concentrated alcohol fumes from their own highly correlated drinking habits contributed to their lung cancers and where, if ever, have you seen this quite possible factor corrected for?

    And finally, the *REAL* “rub” of it all… I left out one of the most important categories in my original criticisms of these statistical analyses: what about the 517 (or 637, or 992 etc) smokers who died prematurely from heart attacks NOT because of any of those other factors… but just because… because…

    Well, just because. That’s life. Feces happens. Two identical twins raised in identical cubicals who spend their lives watching identical reruns of Oprah while dancing to identical oldies and eating an diet of identical wheat grass with soy curds sided by sprigs of parsley and twists of lemon are STILL unlikely to both die of a heart attack at the same stroke of midnight on New Year’s Eve of 2099. One will almost certainly die “prematurely” for absolutely no good reason at all except that St. Peter happened to ring his bell a bit early.

    One good thing about being into the second half of a century of living is that I’ve experienced first hand just how much everyone has seemed to think we “know it all” at all those times when we didn’t.

    And we still don’t.

    – MJM

  6. Can anyone tell me what is a premature death? I can never understand the term.

    • Carl V Phillips

      It is another one of those vague journalism-style terms that should not be used in scientific communication unless specifically defined. If it were defined for use in this context, it would presumably mean “earlier than would have occurred in the absence of the particular causal factor we are referring to.”

  7. Thank you Carl for a very complex but intensely interesting post, (thanks to the commentors as well, some interesting thoughts as well).

    I’d like to give an example of how a non-scientific person thinks about the death tallies produced by the tobacco control industry, from my elderly neighbour, after a recent discussion after a funeral of her friend , a lifelong heavy smoker, who died aged 92 from a ladder fall when he was cleaning out his gutters before the winter rains.

    This man would be considered a “smoking related death”.

    My neighbour asks, “why, if so few people in the population smoke now, (in Australia it is claimed by the tobacco control industry, that fewer than 13% of the population now smoke tobacco), that smoking is still reported to be the most common cause of death”? “Why are so many people dying from cancers that are caused by smoking when so few people now smoke”? (This question is directly related to the latest ad campaign plastered all over our screens in Australia, claiming that smoking “causes” 16 types of cancer).

    Tobacco control lies about smoking are becoming so obvious now, as was always going to be the case, yet they still keep on lying. When people like my neighbour start to question, you can guarantee this is the beginning of the end for tobacco control, and for public health in general, because they have so long supported the lies. My neighbour doesn’t have a clue about how the death stats are arrived at, she doesn’t know how the numbers are manipulated to come up with “X amount of deaths are caused by smoking”, but she is no idiot, and knows bullshit when she smells it.

    • Carl V Phillips

      Yes, it was a “smoking related death”. He smoked. He died. That makes those related. It turns out that this may be one of the rare cases where you could not say that smoking caused the death of this smoker, as I explained (unless he died slowly from his injuries, in which case that “one second earlier due to a bit of damage to his heart” thing kicks in).

      And yes, it is good to see that the average person is numerate enough that they are smelling the tobacco control bullshit.

      One more technical aside: Those claims about “causing the most deaths” are a different form of bullshit that closely relates to a point covered earlier in this discussion. Consider the question, “Which causes the most deaths? Is it smoking, cancer, or aging?” If your answer is “wait, what?”, then you got it right. It is meaningless to declare something “most” without specifying what list is being compared, and they never do. Then, if you offer a list of cross-cutting possible causes, it shows just how stupid the claim is.

      The non-technical explanation for this is that what the tobacco controllers are really trying to claim, and this is pretty clearly actually true, is that smoking is the lifestyle choice or consumption choice that causes (i.e., substantially hastens) more deaths than any other such choice. Presumably that was the slogan back when their forbears actually told the truth. But they morphed it into the current (false) claim without that caveat because they need to pretend that smoking is not a lifestyle or a choice. But they still want to make a claim about how it compares to other choices. Rather than give up either of the contradictory claims they want to make, they come up with some bullshit version that tricks the rubes.

      • Thanks for the reply Carl, your last point is nicely illustrated by this article:

        There is a danger in attributing smoking as the “cause” of many cancers, and subsequent deaths. The danger is that people will believe that if they don’t smoke tobacco, they will not get cancer. To illustrate this point I will tell you the story of a young woman I know personally, she is 32 years old, never smoked, and does not have any friends or family that smoke, or smoked in her presence. She was diagnosed 6 months ago with stage 4 lung cancer, and although she is undergoing treatment her prognosis is very poor. She ignored the early signs of this cancer, a persistent cough, pain in her upper back, weight loss, shortness of breath, etc, because she thought that there was no chance she could have anything serious because she did not smoke, and “everybody” knows that only smokers get lung cancer. There is also the bias by her medical doctor as well, as he did not even test her from the possibility that her symptoms were caused by lung cancer, and put her symptoms down to a chest infection lingering after she suffered a bout of flu last year.

        She is just one case, but it leads me to wonder how many others there are? This is a direct result of blatant lies and misinformation given out by public health, and the tobacco control industry.

        • Carl V Phillips

          There is definitely some risk of that. And you presumably saw this, where I point out that the reason her prognosis is so poor, and the diagnostic methods were so poor, is that public health people do not want lung cancer to be curable:

        • On my lung cancer site there are many cases of never smokers and those who gave up decades ago. As you said their symptoms are often ignored until the Cancer is advanced as neither they nor the doctors suspect it. Some are really young but most seem to be women in the 40 – 50 age range. I tell friends if they are at all worried to say they used to smoke, at least that way they will get checked and at least have ain x ray.

        • Carl V Phillips

          Unfortunately regular chest xrays rarely detect lung cancer cases early enough to be of much help. More thorough scans have shown more promise, though there is a lot of debate about that too.

        • That is true but it is better than not investigating at all because the patient is a non smoker. Mine was found at a very early stage when I had a CT for an unrelated problem, I had no symptoms at all. However it did not show up on subsequent x rays although I know many who were dx via x ray.

  8. Jonathan Bagley

    Food for thought. I won’t need to buy food for several days.

  9. Not exactly on topic, but I wonder if Carl has seen this:

    “Global burden of disease due to smokeless tobacco consumption in adults:…..”

    Click to access s12916-015-0424-2.pdf

    It seems to be fairly recent since it is dated 2015.

    • Carl V Phillips

      That one is complete garbage for a simpler reason: garbage in, garbage out. It is not clear that smokeless tobacco use increases the risk of any life-threatening disease, and anything times zero is still zero. Maybe some of the dirtier products, like the Sudanese-style are measurably harmful, but there is simply not enough data. The authors just used what are basically made-up numbers and applied it to all products. They also threw in a bunch of products that are not smokeless tobacco.

      Beyond that, the analysis had all the problems I will try to address in the continuation about how similar studies are done for smoking.

      • It’s the same deal as with the rest of their “research”: you have certain subgroups of Antismokers that are just in it for the money (grants/prestige), the control, or the “revenge” against anyone profiting from tobacco in any way, shape, or form because they believe someone close to them died or was harmed by it, or simply because they are crazy up the yin-yang and the focus of their psychopathy happens to be anything related to or looking like smoking.

        We’ve seen a lot more of it focused on vaping lately, but Carl, I’m sure you’re more than familiar with how it was focused on smokeless snus/chew/snuff for years before that. I think I mentioned running across some sitcom where some 8th-grader introduced a 7th grader to Copenhagen or somesuch and then the 8th grader is rapidly diagnosed and (and clearly expected to die by the friend and his dad) from mouth cancer.

        They lie, in every second of every hour of every day in every way — they lie.

        – MJM

        • Carl V Phillips

          Pretty much. Still, I think there is value in understanding what the truth would look like, and thus the exact nature of the lies. Thus this blog.

          That bit of fiction is a great example of innumeracy. I suspect whoever wrote it believed it was plausible. There are
          (I happened to already have a SEER window open to check my recall of when lung cancer peaked.) There are estimated to be less than a thousand cases a year of oral cancer among all Americans under age 35. Even if you had absolutely absurd beliefs about etiology — that smokeless tobacco caused all of them (rather than the reality of it apparently causing none of them), and the chance of that causation was the same for a 14-year-old who tried it a few times as for a 34 year old who had been using it regularly for 15 years — the chances of that would be something down in the range of 1/2000. That is just a failure to understand numbers, similar to worrying about dying in a plane crash or a terrorist attack. And oral cancer is not too likely to be fatal, unlike lung cancer.

  10. I too saw the one Junican quoted, and it made me laugh.

    The problem is that statistical analysis can never show causality, although we pretend it can – at least we do when results show what we want them to show. That’s (IMO) the short answer to your long analysis of the calculation of “deaths caused by smoking”. (Although I realize you are simply trying to explain what happens.)

    My new favorite is “deaths caused by poverty”.

    “…including 176,000 due to racial segregation, 162,000 to low social support, 133,000 to individual-level poverty, 119,000 to income inequality, and 39,000 to area-level poverty. By comparison, 119,000 people in the United States die from accidents each year, and 156,000 from lung cancer.”

    “Dr. Galea noted that “any time you try to say that death is attributable to a single cause, there’s a problem―all deaths are attributable to many causes. But what we did is just as valid as what was done to establish smoking as a cause of death.””

    Yes, I agree – just as valid. *eyeroll* I’m eager to see the death certificate that lists “area-level poverty” as a cause of death.

    • Carl V Phillips

      Nothing can ever prove show (taking the word to mean “make directly observable”) causation. Causation is an unobservable construct because it inherently depends on drawing a conclusion about what would have changed in the world if one particular occurrence in the past were changed. We can’t show that. We can only infer it. Particular statistical analyses is how we often do that, and it works quite nicely when done right (see: the entire history of science).

      That particular piece attributing deaths to poverty is silly, but that is mostly because it is a synthetic meta-analysis of incommensurate studies — that is always junk. Also it claims an absurd level of precision. However, it appears at a glance to be at least as good as the similar tobacco control research. The exercise suffers from worse confounding problems than measures of smoking, which does not mean it cannot be done right, but does mean that most of the studies they summarized were probably bad even if they were trying to be good. I can pretty easily guess at the statistical model that they used to make all those separate claims, and I can tell you that is bad methodology. But it is no worse than is typical for epidemiology.

      That “death certificate” bit is a common complaint in discussions like this, but it is misguided. As I assume we all know, a death certificate is a very rough measure of the most prominent physical condition that was the proximate cause of death. It does not list the causes of that cause (e.g., smoking or poverty), and obviously does not list all of the infinite factors that caused that death then. For everyone who dies at 75, one of the causes of that death was “not dying at 74”, but you will never see that listed either.

  11. Carl Thanks again, its fascinating (sort of).
    I had not realised that smoking “related” deaths did not necessarily mean : deaths mostly , or primarily, related to smoking.
    Presumably 10 yearly trend figures for lung cancer deaths and at what age ( at least within one country) would be a better indicator of the overall trend direction and scale, for all smoking caused deaths, and also those that occur prior to average life expectancy . Would that be so?
    I would hate to think that planning for future hospital bed numbers and the like was based on nonsense like “smoking related”.
    What are the trend figures for lung cancer in the US like?

    • Carl V Phillips

      Lung cancer is the sentinel disease for smoking, which basically means that you can estimate how much smoking there is and how much total disease it is causing by following the LC trend. It has been going down in the USA since a peak c.1990. I am not sure offhand when other rich countries peaked, but I would expect they are all going down. It will be increasing in many poorer countries.

      Re planning, smoking has such a small effect on healthcare demands that it does not matter much. It shifts some demand earlier in time, but that is a pretty minor net effect if you think about it. Remarkably close to zero. The claims about increased demand (the standard anti-tobacco lie that smokers cost the system money) are based on intentionally ignoring the fact that a disease that occurs earlier results in both an increase in cost earlier and a decrease later — they only count the first side of the ledger.

      • Carl
        One of Australia’s escalating costs is catering for the needs of people who need long term, sometimes 10 years or more, High care nursing homes and other forms of intensive but non hospital care -mainly but not exclusively Alzheimer’s . Therefore estimates of how many are likely to get sick go to hospital get hospital treatments and within a few months die, as opposed to the numbers who will have long slow declines in a ‘home’ are important to decisions as to what kind of facilities will need to be built and in what proportion.

        • Carl V Phillips

          It turns out that a couple of weeks in intensive care, with some surgeries and whatnot, balances out a lot of months of extended care costs. But it is true if smoking shifted substantially toward or away from one of these patterns, it could have net effects, but mostly it doesn’t. It usually kills a bit cheaper than average (heart attacks and lung cancer) though sometimes produces lingering treatment (COPD). It pretty much evens out.

    • Anyone over the age of fifty or thereabouts can testify as to the ENORMOUS decrease in nonsmokers’ exposure to secondary smoke throughout the 1980s, 90s, and 2000s. This anecdotal perception would be backed up by studies done on salivary cotinine etc (Sorry… don’t actually have that research at hand… Anyone else out there have it?) I believe a reasonable estimate for exposure reduction would likely be on the order of 80%.

      Back in 1986 to 1988 the Antismokers first came up with and approved of the “fact” that 3,000 US nonsmokers a year were getting Lung Cancer from secondary smoke exposure. An 80% reduction would mean that today there would be only 600 nonsmoking deaths per year caused by smokers.

      Hmmmm…. Oddly enough though… If we go to the Holy Bible Of The Surgeon Generals’ Reports and read the chapter and verse of the Infallible Words of The Year of Our Lord 2014, we find that instead of the number decreasing from 3,000 down to 600, it INCREASED from 3,000 to 7,330! We’ve gotten rid of three-quarters of the supposed causative factor, and the Highest Holy Authority In The Land has declared that the “effect” of that decrease has been more than a DOUBLING of the cases of lung cancer!

      This would seem to indicate that if we continue banning and reducing smoking everywhere we will begin to see more and more and more lung cancers springing up in nonsmokers everywhere. By the time we get down to there just being a half dozen or so smokers hiding out in various caves and such, the trend would indicate that there might be literally HUNDREDS OF MILLIONS of lung cancer deaths from those diluted wisps of secondary smoke EVERY SINGLE DAY!

      It will be a Public Health Disaster of Unprecedented Proportions! Truly an ETTWAWKI!

      – MJM, busily outfitting a basement bunker that will be supplied with a constant stream of secondary smoke to ward off the Lung Cancer Coprophiliacs! (It is rumored that Coprophilia runs rampant among a good number of the prominent Antismokers that Carl has featured on this blog over the years, explaining their pot-bellies, haggard visages, and breath that defies adjectival limnation.)

      • Carl V Phillips

        Well without getting quite so, um…animated on the point :-)….
        There is probably no better simple evidence they are just making up their numbers (all of them) than the fact that they have indeed forced a dramatic reduction in ETS exposure over the last 25 years, but they claim that the harms from ETS are basically unchanged over that period (and, as you say, are increased compared to the initial estimates from before that period).

      • Well exactly. The fact that this is now being observed by those people who have no dog in this fight is testament to the thought that not all people have lost their ability to think critically in matters of “public health”.

        I have noticed the same garbage coming from the tobacco control industry, and fortunately even non-smokers are noticing, but to counter this increase in critical thinking, the tobacco control industry is providing a deluge of hate campaigns and mountains of junk science and outright lies, as well as using social media to encourage the most paranoid and ignorant anti-smokers/vapers, to fill the comments sections with vitriol and made up stories about “evil” smokers/vapers. When in reality the vast majority of non-smokers rarely are exposed to tobacco smoke.

        I wonder at the increasing pressure from the ANTZ to bring in outdoor smoking bans, and vaping bans, is this done to get the nutters worked up, (probably), and to ensure that the issue is kept in the public eye, so that they also ensure they keep their jobs.

        The exaggeration of the issue has become laughable, and again, non-smokers/vapers, (not “anti-smokers/vapers), are starting to notice the exaggerations. This to me, signals the beginning of the end for the zealots, the tipping point has been reached and breached, its all downhill from here.

        • Jude, you wrote, “as well as using social media to encourage the most paranoid and ignorant anti-smokers/vapers, to fill the comments sections with vitriol and made up stories about “evil” smokers/vapers.”

          About five years ago one of the Aussie “Chapman Minions” by the name of Marita Hefler authored a study whose purpose seemed to me to be about how professional Tobacco Control could use social media to infiltrate the minds of children to better control their thinking and behavior so it moved along desired channels. I emailed her, asking for a copy of the study, and she responded that she’d be happy to send it to me but was curious as to who I was and what I’d be doing with it. I responded honestly and she responded by cutting off all communication with me until finally, after repeated emails to her, I got a response saying she had no intention of helping me with my own research by sharing hers.

          So much for the integrity of science, eh?

          – MJM

  12. WRT the death certificate remark – Of course the poverty study is junk. But that’s not why it’s silly. My point was that when people hear “deaths caused by smoking”, or “deaths caused by obesity”, they imagine that what is being represented is a some known physiological cause-effect relationship, as smoking and obesity are at least physical actions/attributes. Deaths caused by poverty, having no particular identified physiological mechanism in this study *appear* to jump to another level of abstraction. (Only in practice they do not, as Dr Galea feels the need to point out). The death certificate remark was meant to illustrate the absurdity of accepting this type of conclusion, based on non-randomized data (and skip the hard work of determining physiological processes, if any, and how they relate to individuals), wrt deaths caused by area-level-poverty (anyone can see that is by itself medically meaningless), and thus the related absurdity of accepting the similar “deaths caused by smoking/obesity”. I thought the appearance of a higher level of abstraction would illustrate that – but perhaps not.

    Not quite the same as your not dying at 74 example.

    WRT your reply “Particular statistical analyses is how we often do that, and it works quite nicely when done right (see: the entire history of science).” Well, of course I mis-spoke about “statistical analysis” when I should have simply said (non-randomized) simple association or simple correlation (with or without the pretend precision of statistical control). However, I would suggest that, rather than working quite nicely, public health has largely departed from “the entire history of science” with its peculiar attachment to these methods as representing meaningful conclusions.

    • Carl V Phillips

      Actually, I suspect that most of the studies that were inappropriately mashed together in that metaanalysis are, like most attempts to tally the deaths from smoking, based on adding up estimates by disease. So they do identify a disease pathway downstream of poverty or smoking. It turns out that this is not really the most robust way to do it. The better way to do it is to just look at mortality (from all causes). The disease-by-disease approach is a legitimate choice as a tradeoff of using cleaner data that is a rougher estimate of what you really want. Looking at impoverished places and trying to figure out a comparison that lets you estimate how much of the higher death rate is caused by living in an impoverished place, per se, is pretty hopeless, even though it is a very legitimate question that theoretically could be answered that way.

      All claims of causation are inferred (never observed or proven) based on observations. Sometimes we have an opportunity to improve the usefulness of our observation (e.g., by turning it into a randomized experiment) though sometimes that is not and option and sometimes it does not improve the observation even if it is possible. The vast majority of science conducted is, and was always, based primarily on studying things where simple lab experiments are not possible, and a large portion of sciences that are often experimental still rely on non-intervention methods. It is always just a matter of how good the observation is and whether it is properly used.

  13. Pingback: Ecig proponents need to learn lessons from other activists | Anti-THR Lies and related topics

  14. Sorry, but this is just too many words to be useful for explaining things to people. Also, it doesn’t address that person’s first point of confusion, which is believing that peoples’ individual deaths are “listed as smoking related” or not, and then someone comes along and adds them all up. WRONG! All the supposed smoking related deaths are completely theoretical. To describe it simply, they compare death rates of smokers versus death rates of non-smokers to come up with a risk ratio (or odds ratio or relative risks, the differences between which get into the weeds). Then, they gather data on the total number of deaths and on the proportion of smokers in the population of interest, and calculate the supposed number of smoking related deaths from that. It makes no difference even if no individual person’s death is listed as smoking related because they don’t use it at all anyway. This is how the CDC’s SAMMEC operates, and it employs data from the American Cancer Society’s CPS studies to come up with the RR. Those studies have no data on the role of infection in the diseases they blame on smoking, and smokers are more likely to have been infected for socioeconomic reasons, so the studies are rigged to falsely blame smoking itself rather than infection. AND, their little ritual of “adjusting for socioeconomic class” does NOT eliminate the problem, nor can it address the issue of the age at which people were infected.

    • Carl V Phillips

      I’m afraid I cannot simplify how epidemiology of this type is done down to a fortune cookie. I would probably spend a couple of weeks of class time teaching this, and have a few hours of required reading to go with it. So I kinda think I am doing a pretty good job of simplifying it.

      As for the rest of what you say, it is about right, though there are some bits that are either not quite right or are right in spirit but don’t quite hit the most important point, and some that seem to confuse a legitimate methodology that is applied in faulty ways with illegitimate methodology. I am going to explain all of this. I did point out that this is to be continued.

    • Carol, you’re right about the explanation being too involved for most people reading our stuff casually out on the net, but Carl’s readership has a bit more background and I think that’s more the audience he intended to hit.

      When I’m out in the wild and post about this I usually boil it down to something like:

      “These numbers all come, not from death certificates and real bodies, but out of a computer program called SAMMEC (Smoking Attributable Mortality, Morbidity, and Economic Costs) and the number can be any numbers at all that the programmer wants them to be: they simply depend on what formulae and population figures are put in right at the beginning.”

      For the general Netizen, that’s really all that’s needed. Sometimes I might refer them to the Levy/Marimont piece at: if I think people want and can handle more information — the article is almost 20 years old, but I think it’s still about the best explanation out there for popular consumption!

      – MJM

      • Carl V Phillips

        I would definitely not call that link recommended reading for understanding this stuff. It is not worthless — the authors demonstrate a pretty good non-scientist investigative reporter approach to the issues, and identify some problems. This includes some key bits of insight, like the low magnitude of one study’s estimate of ETS effects on lung cancer compared to the estimated effect for a presumably unrelated covariate. But they then get themselves into trouble with their limited understanding, trying to make a big deal about statistical significance and the point estimate RR being less than 2. They are basically repeating simplifications used to dumb down complicated analysis into soundbites for medics as if it were the right way to do a critical analysis. Even back in 1998 when this was written, anyone who knew what they were talking about knew better than this.

        It goes downhill from there. I am not even going to try to critique it. They actually do have some useful bits of insight and explanation peppered throughout. But it is so mired in error and naivety that it is definitely not recommended reading.

        • The actual SAMMEC process of calculation is like a story problem most people would have encountered in high school (at least they used to). Most people figured it out, but then forgot how to do it because they didn’t have to use it in regular life. Its big problem is that it gets its RRs from American Cancer Society data. And those RRs are not modifiable by the user of the software.

  15. Pingback: Sunday Science Lesson: So how would you estimate how many deaths are caused by smoking? | Anti-THR Lies and related topics

  16. Pingback: Sunday Science Lesson: How they estimate deaths from smoking etc. | Anti-THR Lies and related topics

Leave a Reply

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

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

Facebook photo

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

Connecting to %s