I noticed something interesting in the CDC’s mortality data. I’ve recently started looking into the age distribution of drug poisoning deaths, as in what’s the average age of someone who dies of a drug poisoning, how sharply peaked is the distribution, has this distribution changed much over time, etc. I also compared this with the age distribution for drug use. (See here. The age distribution figures start around page 150 or so. This is a big 10 MB pdf, so it may take a few moments to load.) Drug users skew young, but overdose deaths skew old.
Here is what got me thinking about the age distribution of drug overdose deaths. I knew that age is an overwhelmingly powerful driver of mortality rates. Duh!, you say. Obvious as it is, I was surprised to see the magnitudes (see this post on my blog on mortality by age). My thinking was that if the population of drug users has been aging, there are many more chances for people to die from other causes, and many more chances for a death to be labeled an overdose (even if it isn’t). An aging population of drug users combined with a tendency to misclassify deaths as “drug related” will introduce a spurious increasing trend in drug overdose deaths. (That question requires looking at how the distribution of drug users changes over time, and I haven’t looked into that yet. Perhaps in a future post…) I also wanted to see if the average drug OD decedent has been aging over time. I was expecting to see an aging of the average decedent, considering that the “opioid epidemic” is supposedly exploding into the general population. Drug use is mostly a youth phenomenon, so I was expecting to see that 15 years ago, before the drug death rates started rising, the average decedent was young, and this age should have crept up over time. So these were my “priors,” and what I actually saw did surprise me a bit.
The average overdose death didn’t age by much. The average decedent in a drug poisoning death was 40.9 in 2000 and 42.8 in 2014. A two-year aging doesn’t describe the massive demographic shift I was expecting, of opioids saturating the whole of society rather than just being a youth phenomenon. Poisoning deaths *did* get a little less sharply peaked.
This graph shows you how the drug poisoning deaths in each year are distributed across ages, compared to the distribution of illicit drug use for opioids and then for cocaine based on the SAMHSA drug use survey. (So if you sum up the % for each age, you get 100% for each of these curves.)
Notice the illicit opioid use rates and cocaine use rates (dotted lines) and contrast them with the death rates. This confirms the general notion that illicit drug *use* is a youth phenomenon that most people grow out of in their 20s. The surprise here was how poorly the death curves and the use curves overlapped. Notice also how the peak of the drug OD line has been moving; while the *average* decedent has only aged about 2 years, the modal (peak of the distribution) decedent has aged about 10 years.
(Note: Use rates look like a step function for part of the graph because the use rates by age are in categories that span 5 years, as in “40-44”, “45-49”, etc. Use rates are given for each individual age up to 25, but after that they are in age groupings. I assumed that the number of total users from the SAMHSA data was evenly distributed over the years within these groupings.)
The above graph is plotting the distribution of *all* drug related deaths, which potentially muddies the waters. Different drugs have different age of use distributions, so let’s fixate on cocaine. It’s a good choice for analysis. It kills a large number of people each year, about 5,000 in 2014. While that makes it a tragic social problem, it’s also makes it a statistically well-behaved social problem because the data volume is sufficiently large for analysis. If you were looking at, say, hallucinogens, there are so few deaths that you couldn’t even analyze the issue, other than doing very crude sums and saying, “It killed six guys last year.” Big social problems are tragic, but at least we can study and understand them.
There are also a fairly large number of users, so cocaine use shows up in a statistically significant way in drug use surveys. These surveys are done on a random sampling of about 30 or 40 thousand people. Cocaine use is common enough that these surveys can sample its use effectively. Cocaine use is in the 0.5% of total population range for past month use, and in the 1.5% range for past year use, so the magnitudes found in these surveys are significant. Heroin, which kills more people than cocaine (as of very recently anyway), has so few users that these surveys don’t register it very well (past month use of heroin is in the 0.1-0.2% range). Cocaine should be statistically “well-behaved” for the sake of mortality and usage analysis. Also, I’m not muddying the analysis by comparing drug deaths (most heavily driven by legal prescription painkillers) to usage rates (most heavily driven by marijuana use, which doesn’t actually kill anyone). I’m comparing cocaine use rates to cocaine deaths rates.
Here is what I get for the usage distribution and the mortality distribution. Notice again that usage is a youth phenomenon, but it’s mostly older people who are actually dying of cocaine-related drug poisonings. In 2014, the average *user* is only 33, but the average decedent is 43, ten years older. People 30 and under make up about 50% of users but only 18% of overdose deaths. For people 40 and under, these figures are 70% and 41%. For people under 20 these figures are 12% and 2%. If you like, you can look at this in the other direction. People 50 and over make up 17% of users but 36% of overdoses. People 40 and over make up 31% of users and 61% of deaths. Older people die of cocaine use *way* out of proportion to their usage rates.
I also want to show you a mortality picture (as in “deaths per 100k”), but there are important caveats to this. It’s hard to know what the appropriate denominator is for calculating cocaine-related mortality. Do I assume *all* past year users are at risk? In that case, I’d divide the number of cocaine deaths in 2014 by the number of past-year users in 2014. But wait, the number of past month users is something like 1/3 the number of past-year users (giving the lie, btw, to the notion that this drug is hopelessly addictive). Maybe it’s just this population of continuing users who are at risk. Surely the hard-core addicts are in this group, not the “past year use” group. Naturally, this will give me a higher mortality, because I’m dividing the same 2014 cocaine death count by a smaller denominator. Anyway, I’ll just plot them both so you can see what this looks like.
Again, you can see clearly that cocaine-related mortality is heavily age-driven. I’m plotting the overall (all-cause) mortality at each age for comparison. So what’s going on here? Are older people just more susceptible to drug overdoses? Or are older people just more likely to die *for any reason*, so if cocaine is present in the toxicology screening it will be blamed for the death? In this latter explanation, there are just more older corpses that happen to have cocaine in them, even though it had nothing to do with the cause of death and even though living bodies with cocaine in them tend to skew younger.
If you like looking at the underlying numbers, the following data table is the basis for the previous two charts. Deaths are from CDC data files, use numbers are from the SAMHSA survey linked to above:
This next paragraph is a bit of an aside, but it relates to my quandary about calculating mortality rates. “What is the risk of death from cocaine use?” becomes an almost philosophical question, because it’s impossible to know what the relevant denominator is. This is a general problem for quantifying the risk of drug use. One could simply divide the number of death by the number of users in order to get a very rough-and-ready estimate of risk, but that’s probably not right. There are people who dabble in drug use for very short periods and then never touch them again (as you can see if you compare “past month”, “past year”, and “lifetime” usage). Surely it’s the continuing users who are really at risk, but then maybe it’s only a subset of *these*. Perhaps even addiction and dependence per se are not dangerous, it’s just that some individuals are *very* poorly informed and engage in unnecessarily risky behavior. The risk profile of cocaine users may be very heterogeneous; they don't all face an identical risk of death. It’s nice to have a solid easy-to-calculate figure in mind when you’re thinking about these issues, but you always have to remind yourself that this figure doesn’t mean what it appears to mean.
I worked up the above charts for prescription opioids too, and they look very similar. Once again, there is a 10-year difference between the average illicit user (about 33) and the average decedent (about 43). But I’m cautious about drawing any conclusions from these charts and mortality figures for opioids. I suspect that it’s wrong to divide the opioid related deaths by the number of illicit users, because the at-risk population is the full population of people who use opioids, not just those who use them illicitly. Remember that many people have legal prescriptions and legitimate medical need for painkillers; the usage rates from the SAMHSA survey only measure illicit use. My preferred explanation for the recent increase in opioid deaths is that normal users are accidentally mixing them with other drugs, not that we’ve created a growing population of high-risk drug addicts. We don’t see any increase in illicit use of opioids over time in the drug use surveys, so I think my preferred explanation makes more sense. So anyway, I *could* calculate an “opioid mortality per illicit user”, but it would be irrelevant because I’d be using the wrong denominator.
I looked at the average age of a drug overdose for the dozen most lethal drug categories, and I saw roughly similar patterns for all. The average age of decedent was in the 40-42 range for all categories, and there was a slight upward trend in age for 2000 to 2014. Heroin was the closest thing to an outlier, with a moderately young average age of death of 38. And “psychostimulants with potential for abuse” (meth and ADHD medicine) was the closes thing to an outlier in terms of the trend, with the average decedent aging from 36 to 43 over the 2000-2014 period. I didn’t do a thorough distributional analysis for each one of these, mostly because I don’t have the relevant corresponding drug use data (for all I know it doesn’t exist) at that level of detail. But considering the average age of decedent is similar for all these drug categories, I'd be very surprised if the result were any different. I suspect this general patter of younger average users and older average decedents holds up pretty well.
"So what?" you may be saying. Is this just a pointless exercise in data manipulation? No, I believe there are relevant lessons for drug policy here. For one, I think this supports my view that a lot of “drug overdoses” just aren’t. That is, many of these people are dying for other reasons related to age and health and they are wrongly being labeled as drug poisoning deaths. (See my previous post here on illnesses listed on the death records of drug poisonings.) It’s not smoking gun, but it pours a little cold water on the notion that there is a runaway opioid epidemic. The “spurious trend” story looks more credible in this light. But put aside the spurious trend story for a moment. Even supposing that *all* these drug poisoning deaths are correctly labeled and there’s no spurious trend or upward bias in the totals, there are still some policy implications. This is another case where a targeted deterrence policy makes a lot more sense than a general deterrence policy. If people in their 30s, 40s, and 50s are at a massively higher risk than people in their teens and 20s, we should warn them. “Just because your heart could take it at 20, doesn’t mean your heart can take it at 40,” might be a good slogan for a targeted anti-cocaine public service announcement. Or for that matter, and anti-meth or anti-opioid or anti-heroine public service announcement. It's trivially true that anyone can die of a drug overdose, but the magnitude of the risk is *very* different for different ages.
It would be interesting to see what the use vs mortality figures look like for legal opioid use, or for benzodiazepine use (another leading killer drug). If I manage to pull that together somehow, I’ll turn it into another post.