Drug Poisoning Update for 2018 Data

This is my annual update based on the CDC's numbers. I will mostly be updating the data and not trying to introduce any new points or topics. This will mostly be last year's post with the data updated through 2018. See also my updates through 2014, 2015, and 2016.

The CDC has finally released the mortality data for 2018, which can be found here. They were quite slow about it this year. I'm not sure what the problem was this time. The data for given year is usually released late the next year. The data for 2017 was released in late November 2018. This year it didn't get updated until late January or early February. I kept checking that page to see if the new data was out, and at one point (spanning a week or two) the page was formatted weirdly with random links strewn across the page, overlapping  other links, which is pretty unuseable. Anyway, it seems like the CDC has its stuff together now. The data is out and I've done my analysis with the latest year added. Below I will list some notable observations.

Total drug poisonings are down for the first time in decades. What's not fully appreciated by most observers is the fact that drug poisoning rates have been increasing exponentially since before the so-called "opioid epidemic" started. (Seriously, click the link to the Science paper and at least read the abstract and look at some of the charts. Very interesting stuff. Note that this is a statement about rates per population. Obviously in a population experiencing exponential growth, and exponential growth in the raw count of something isn't surprising. The rate per population is experiencing an exponential increase.) This pattern goes back to 1979. Drug poisoning deaths tripled from 1979 to 1996, the year that Oxycontin was released. Clearly the mix of drugs involved in these poisonings is changing over time, and we could define periods of relatively high rates of cocaine, heroin, or prescription opioid deaths. But these different regimes collapse into a single pattern of exponentially increasing death rates. An exponential fit to this data yields an R-squared of 0.99, an extremely tight fit. It's possible that 2018 is just a blip, but hopefully drug-related deaths have hit some kind of "maximum" and are either flattening or coming back down.

In 2018, there were 67,464 total drug poisoning deaths, down from 70,320 in 2017. As I point out every year when I do this update, this number by itself is not very useful for characterizing the opioid epidemic. If anyone leads with this number, be very skeptical of anything else they say. They are anchoring to the highest possible number they can find. They sometimes start with this high figure and then, if the piece is about the opioid epidemic, they might inform the reader that "about 2/3 of these are opioid-related." (German Lopez at Vox sometimes does this, I have noticed.) Even though an alert reader could take in this information and compute the approximate number of opioid deaths (the number that's actually relevant to the story), I still think this is misleading. If someone is writing a missive against pharmaceutical companies or criticizing doctors with loose prescription pads, giving their reader a figure that includes cocaine overdoses and intentional suicides is incredibly misleading. There is a tendency among journalists to sensationalize a story by inflating magnitudes, and that certainly happens a lot with stories about drug overdoses. Don't count on your audience to pull out a calculator and multiply the number on the page by 2/3. The number you printed on the page is what's going to stick in the reader's head. If that number includes things that have nothing to do with the thing you're actually writing about, you going to mislead all but the most alert news consumers.

Of the 67,464 drug poisonings, 59,078 were accidental, 4,734 were suicides, 3,536 were "undetermined intent", and 116 were "murder". (Keep in mind these figures include all drug poisonings, not just the ones that are opioid-related.) So a large majority of these deaths were accidental overdoses. I think when people talk about the "opioid epidemic" as a problem, they are talking about accidental overdoses and not suicides. It just seem ridiculous to blame a suicide, the intentional ending of one's own life, on a doctor's loose prescription pad or a pharmaceutical company's marketing practices. You can't claim that but for the opioids, that person wouldn't have killed himself. If pills hadn't been so readily available, that person would simply have chosen another means of committing suicide. On the other hand, it plausibly makes sense to worry about a growing population of drug users leading to a larger number of accidental overdoses. Someone who is trying to quantify the drug overdose problem should be using the 59,078 figure (plausibly with some fraction of the "undetermined intent" added back in), not the 67,464 figure. Like I've said before, there is a similar problem with people who try to inflate the amount of "gun violence" by counting suicides. It just doesn't make sense to blame guns or drugs or any other inanimate object on a suicide when, in a but for sense, the person would very likely have died anyway. Note also that some researchers think the number of drug-related suicides is under-counted. That is, the true number of accidental fatal poisonings may be lower than the 59,078 figure. For whatever reasons (family shame, a policy of defaulting to "accident" when intent is ambiguous, other explanations?), medical examiners don't like to label a death a suicide. The possibility of mislabeling taints the raw numbers, and any conclusions we'd like to draw from them should keep this reservation in mind.

Here are the accidental deaths by substance for the past 20 years. Click directly on the table to enlarge if you have trouble seeing it:

Here are the same numbers as a (very busy) chart:

One has to be very careful with these numbers. They are not additive, not mutually exclusive. Someone could have had each of these substances in their system when they died and thus be counted in each row of the table. Multi-drug poisonings are far more common than a single-substance overdose. Here is a table showing how substances tend to be found together in an overdose death:

You can see that there is a strong tendency for certain substances to go together. A death invovling benzodiazepines is very likely to also involve either prescription opioids or illicit heroin/fentanyl. Heroin/fentanyl deaths are so pervasive and users of these drugs are such prolific users of other substances that illicit heroin/fentanyl is driving up death totals of other substances. One can look at deaths that do or do not include heroin/fentanyl and see if the trends are different. And this can end up telling a different story than if you weren't cautious about this point. Here is a table of several substances (or categories of substances) inclusive and exclusive of heroin/fentanyl:

Let's make the trends more obvious by plotting them. Here are prescription opioid deaths (involving either methodone or "other opioids"). The blue line is trending upward (with a reversal in the latest year). But the red line has been trending down since 2011. One might incautiously state that "the prescription opioid problem has been getting worse", but it seems a lot more likely that this is the heroin problem driving up the numbers of prescription opioid deaths. Prescription opioid deaths excluding heroin/fentanly are trending downward. The timing is interesting, too. Abuse-resistant Oxycontin was introduced in 2010, and many have blamed this for the surge in heroin overdoses. It certainly seems likely that making Oxy harder to abuse will cause some users to substitute other opioids, like heroin and fentanyl. Clearly Oxycontin and other opioids are still around, but they are in some sense being supplanted by more dangerous substances.

Now take a look at benzodiazepines. Again, someone naive of the multilple-drug overdose issue would naively say that benzo-related deaths have been soaring (with a reversal in the latest year). But clearly this is another case of heroin/fentanyl driving up the totals. Benzodiazepine-related deaths flatlined for several years and then began coming down (if we incautiously take the last two years as a "trend").

Below is cocaine. Cocaine seems to be rising on its own, but then again a large majority of cocaine-related deaths also involve heroin or fentanyl.

Below is "psychostimulants with abuse potential", which includes methamphetamine and ADHD medications. This trendline does seem to have a life of its own even excluding heroin/fentanyl. Then again, the last few years have seen a very large increase in deaths involving both substances. One can say of "psychostimulants" (and probably to a lesser degree cocaine) that there is an upward-trending problem that is independent of the heroin phenomenon. I would have guessed there wasn't much overlap between the users of meth and heroin, and evidently that was true until quite recently:

That's all for now. I'll share more if I notice anything. I think my past years' updates have given a pretty good summary of the issues with this data (particularly last year's), so feel free to go there for any grand conclusions or policy implications. See here or here for a contrarian take on the "opioid epidemic," which, if you're a new reader, should explain why this kind of analysis is important and what kinds of claims other people are making about these trends in drug poisonings.