Drug Poisoning Update for 2021 Data

I haven't done one of these in almost three years, but I wanted to do an update. Here is my most recent one with data through 2018, also with links to prior years' analyses. I think not much has changed about the overall narrative, so I'll direct readers to the previous years' posts for those details.

The CDC's large mortality file gets posted here, in case anyone wants to download and do their own data mining. 

There were 107,027 drug-related deaths in 2021. This includes 98,722 accidents, 4,239 suicides, 3,852 incidents of "undetermined intent", and 214 murders. (Not included in the 107,027 figure is 7,017 "other" deaths, where drugs were mentioned on the death certificate but weren't listed as the immediate cause of death. Some of these really are caused by drug use, so presumably they should count for anyone who's concerned about drug-related public health issues.) In the past I've made a big deal about the difference between the total and accidental deaths. Sensationalist journalists reach for the largest number they can find, so they cite the total. This will often be in the context of an "opioid epidemic" story. But it doesn't make much sense to blame suicides on the opioid epidemic. A suicide is a death that would most likely have happened anyway, just by another means. At any rate, as the years go on the "accident" category is becoming such an overwhelming majority that it feels a little pedantic to make this distinction. Still, bonus points for honesty to any journalists who cite the 98k figure rather than the 107k figure. Here's a chart of the proportion of deaths by the various categories over time, where you can see "accidental" swallowing up a greater and greater share of the total:

I do suspect a lot of these "accidental" overdose deaths have a question mark over them. A lot of these people are taking such tremendous risks (like shooting up alone with a powder of unknown origin and potency) that it's worth asking whether it should count as an attempted suicide. But the best I can do is analyze the data we have and caveat with my personal suspicions. 

Above is a chart showing the deaths by major substance categories. The pink line that dominates in the most recent years is "synthetic narcotics", which mostly refers to fentanyl being sold as a street drug. Fentanyl has largely replaced heroin as the major opioid of abuse. This view can be slightly misleading in the following sense: most drug poisonings are multi-drug interactions. So if one category of drugs is skyrocketing, it can drive up the apparent deadliness of other drugs (even if the risk or abuse rates of, say, cocaine or benzodiazepines are basically unchanged). This is actually my value-added. Because this CDC file is at the individual record level, I can count, say, cocaine deaths excluding fentanyl and heroin. I think what looks like a general trend of increasing drug deadliness is in reality almost entirely a fentanyl story. 

Below I have created a similar time series, but this time I'm splitting each substance into two lines. The first line is unconditional count, and it should match the chart above. The second (necessarily lower) line excludes deaths involving heroin or fentanyl. The normal green line is cocaine deaths, while the pukish-green line is cocaine deaths excluding those that also involved heroin or fentanyl. The first looks like it's taking off, the second looks like it's relatively flat. This is clearly a case of fentanyl driving up cocaine-related deaths. Presumably in most of these deaths, the cocaine was either not a factor or not sufficient to kill without the added effect of a potent opioid. 

For clarity (the above chart is too busy to read clearly), here are prescription opioids inclusive and exclusive of heroin/fentanyl. The dishonest way to report on this is to look at the raw number and say "prescription opioid deaths are still rising." A more honest assessment is that prescription opioid deaths are falling, but the extremely high mortality of illicit fentanyl users is spuriously driving up the number of death certificates that mention prescription opioids. There are fentanyl users who happen to have prescription pills on them when they fatally overdose, and there are people who abstain from fentanyl but who fatally overdose on prescription pills. The non-overlap in these categories explains the huge gap between the two lines.

Here's benzodiazepines. Almost all benzodiazepine-related deaths involve some other substance, typically an opioid but sometimes alcohol or some other prescription medication. This is once again a story of fentanyl driving up the total.

Here are "psychostimulants with abuse potential", which mostly means methamphetamine. I think I have heard some commentators suggest that meth is on the rise again, which possibly exonerates opioids. The rising rates of meth use, this narrative goes, are simply a continuation of a longstanding trend of increasing overdose rates, which extends back to the 1970s (possibly earlier). Meth just happens to be in vogue again. This appears not to be the case. Undoubtedly the red line is increasing dramatically, but in the most recent year the blue line is almost three times as high as the red line. Pure meth-related mortality is several times higher than it was twenty years ago, but at the same time the vast majority of meth-related drug poisonings also involved fentanyl or heroin. 

By the way, heroin deaths seem to have reached a peak in 2017, at 14,870. By 2021 they were down to 9,064. So overwhelmingly the most fatal street drug is fentanyl, which appears to have almost totally taken over the heroin market. 

If you want more information about which drugs in combination are killing people, here is a cross table. (To read the lower table, take the second column of the first row. Read this as "22.48% of heroin deaths also involved cocaine.") 

The elephant in the room here is that I've added the Covid years to the analysis. My previous update went through 2018, where drug-related deaths appeared to be leveling off. Well, 2019 looks like a clone of 2017, if not slightly worse. 2020 and 2021 both saw large increases over the prior year. I don't know if the "deaths of despair" narrative makes sense here, or if we're just seeing a continuation of that longstanding trend reaching back to the 1970s. Shouldn't despair have been worse in 2020, when lockdowns were harsher, society was more closed, and drug users would have been more likely to shoot up alone (thus not having anyone who could call for help if they overdosed)? Or was there some kind of accumulation of despair from 2020 to 2021? That is, even though 2021 was milder in terms of social isolation, that isolation was still wearing people down and driving them to risky drug use. 

I have limited patience for the notion that drug-related mortality can exponentially increase for multiple decades. Obviously this can't go on forever. I haven't done a population dynamics calculations to check this, but at some point you'd start running out of drug users. A little Googling suggests there are around a million heroin users in the United States. (Maybe double or triple that if you think most heroin users are homeless and out of reach of household surveys of drug use). If these are the people being killed by fentanyl, then with the numbers we're seeing it would only take a few years for them to all die off. Even with new addicts refreshing their ranks, the mortality is high enough that some depletion should be happening. And these very high mortality rates should be scaring off a lot of potential new users. (Presumably that's happening, but some countervailing force is attracting them?) We should probably take seriously the notion that a large number of deaths are misclassified. Does anyone have a story for why methamphetamine-related deaths, excluding those involving heroin/fentanyl, have increased 30-fold since 1999? Were we perhaps undercounting back then and/or are we overcounting now? I don't want to selectively dismiss or down-weight government data, but at the same time this kind of dramatic increase makes me really suspicious. My attempt to find some pattern in all-cause mortality or a redistribution from one cause-of-death category to another was pretty inconclusive, but that's not to say a deeper dive wouldn't turn up something interesting.