Drug Poisonings Update for 2017

The CDC has released the mortality data for 2017. This post will be my annual update. In this post I’ll point out some of my pet peeves with respect to news stories on the “opioid epidemic.” I’ll try to give my readers the tools needed to interpret the numbers and warn you against sloppy reporting.

In 2017, there were 70,320 total drug poisoning deaths, up from 63,536 in 2016. So far, this has nothing to do with the opioid epidemic. There are many kinds of drugs and people are poisoned by many combinations of substances. If an “opioid epidemic” story anchors you to the 70,320 figure, be skeptical. They are starting with an inflated number. It’s true that deaths from heroin, synthetic narcotics, and prescription opioids constitute a large fraction of these overdoses, something like 2/3 of the total in recent years. But a large fraction of these deaths involving cocaine, methamphetamine and prescription ADHD medicine (collectively called “psychostimulants with abuse potential”), benzodiazepines, and antidepressants, often in combination with alcohol. (Pure alcohol poisoning deaths are not counted in the 70,320 figure above; alcohol has its own set of ICD10 cause of death codes. But many drug poisonings involve alcohol, as alcohol interacts with many kinds of drugs.)

I think the “two-thirds of seventy-thousand” framing is a bit of a trick. German Lopez at Vox has done this multiple times. I criticize his piece here and I see a more recent piece where he uses the same trick for the 2017 data. I don’t get it. 

“Dude, imagine there’s this guy. And he’s nine feet tall!”
“Whoa, that’s pretty tall.”
“Right? Okay, the dude I’m going to actually tell you about now. He’s 2/3 of that!”
“Um, okay. So he’s six feet tall?”
“Yeah. So this guy…”
“Wait. Why did you start me off at nine feet? You’re only talking about a six-foot guy, right? Why did you start me off by imagining this giant?”
“Dude, the three feet of space above his head exists. It’s just that I’m going to focus on the six feet of guy.”
 “O…kay. Fair enough.”

Before we start splitting these deaths up by substance, let’s separate out accidental overdoses and suicides. Of the 70,320, a majority (61,487) were accidental deaths. The rest were suicides (4,977), “undetermined intent” (3,714), and murder (142). When people talk about the crazy mortality of the “opioid epidemic”, I think they are really only talking about the accidental overdoses. If someone makes the decision to end their own life and they happen to use an overdose of opioids as the means, I don’t think we can blame that death on loose prescribing practices by pain doctors. We can’t really say, “But for the opioids, that person would still be with us.” On the contrary, but for the opioids, that person would likely have chosen another means of ending their life. Someone could make the case for adding in a fraction of the “undetermined intent” deaths, because some fraction of these are suicides and some fraction of these are accidental. But then again, the growth of the “undetermined intent” category is small (growing from 2,476 in 1999 to 3,714 in 2017) compared to the growth in accidental deaths (which grew from 11,006 to 61,487 over the same period). When people talk about the explosive growth in drug poisonings, they are mostly talking about the growth in accidental deaths. So my post will focus on those deaths.

(Including suicides and undetermined intent makes the overall numbers higher, but makes the slope of the trend line over the past 20 years less intense. I’m happy to discuss how any conclusions change if you include or exclude these in the count totals. Also, someone might make the argument for including opioid-related suicides in the counts. "Loose prescribing practices for painkillers have created a lot of miserable, hopeless addicts. They become depressed with their lives and end them." But this kind of story-telling speculates wildly, beyond what the raw numbers imply and contrary to other evidence about opioid addiction rates. Also, this could easily cut the other way; it's likely that the lack of readily available opioids leads to a lot of suicides from desperate chronic pain patients.)

Here is a table of accidental drug poisonings by substance since 1999. (Apologies if these tables and graphs look small on your screen or overlay the stuff in the right pane of this blog. You can click directly on these figures to enlarge them for a clearer look.)

Each of these columns corresponds to an ICD10 code. Heroin is T40.1. "Other Opioids" is code T40.2, and it corresponds to Oxycontin, Hydrocodone, and some other common prescription opioids. Methadone is T40.3. "Synthetic narcotics" is T40.4, which corresponds to fentanyl and other super-powerful synthetic opioids. Cocaine is T40.5, benzodiazepines is T42.4, "Psychostimulants with abuse potential" is T43.6, and antidepressants is T43.0 and T43.2.

If you're a more visual person, here are the same numbers as a chart.

Naively, everything is going up.

"What's the trend in accidental poisonings from 'other opioids'?"
"It's going up!"
"What about benzoidazepines?"
"That's the  orange line. It's going up, too!" 

You have to be really careful here. Most drug poisoning deaths involve multiple substances. I've written about that here and here, and in many other previous posts. What that means is that if overdose deaths from one substance are skyrocketing, it can drive up deaths from other substances. For example, even if deaths from prescription opioids excluding heroin are declining and deaths from benzodiazepines excluding heroin are flat, the overlap with heroin-related deaths can mean all deaths involving prescription opioids and benzodiazepines are increasing. Here is a table showing how different substances relate to each other in 2017. The top table shows raw numbers. (For example, 3,851 deaths involved both heroin and cocaine; 2,552 deaths involved both heroin and benzodiazepines.) The bottom table shows how many deaths involving the substance named in the row label also involved the substance in the column label. (For example, 26% of heroin deaths also involved cocaine; 46% of benzodiazepine deaths also involved synthetic narcotics such as fentanyl or carfentanyl. Basically, divide all the numbers in each row by the diagonal element for that row, and that gets you the bottom table.) This should give you a rough idea of which substances are likely to be found in combination when there is an overdose death.

Given this, let's take a second look at the trends by substance. I'm going to do some grouping. I'm defining "prescription opioids" to mean other opioids or methadone (so it involves ICD10 codes T40.2 or T40.3). In practice, heroin overdoses are sometimes labeled with T40.1 and (the ICD10 code for heroin) and sometimes with T40.4 (the ICD10 code for synthetic opioids). So much black market heroin is tainted with fentanyl and other super-potent synthetics, and toxicology screenings often do not distinguish between them. So it makes sense to count deaths involving heroin or synthetic opioids as being in the same category (generically, we could call these "heroin overdoses"). In the table below, "Inclusive" means it's counting all accidental poisonings involving that substance. "Exclusive" means it's excluding heroin and synthetic narcotics (shortening to just "heroin" for the discussion below).

This tells a slightly different story than you would see by looking at the raw figures. The "Prescription Opioids Inclusive" column shows a steady increase since 1999 (with a slight dip in 2012 and 2013). The "Prescription Opioids Exclusive" column, which once again is filtering out heroin deaths, has actually been declining since 2011. Benzodiazepine-related deaths look like they are steadily increasing, but excluding heroin, deaths involving benzos have been basically flat or even declining since 2010. Cocaine-related deaths hit a high in 2006, then started coming back down, then rose again. This is still true when filtering out those cocaine-related deaths also involving heroin, but the rise is not quite as dramatic. Psychostimulants are rising dramatically, but not quite as dramatically if you filter out the heroin-related deaths. The take-away here: heroin is so deadly and has become so prevalent it is driving up deaths in all the other major drug categories.

Keep this in mind when someone reports on the CDC numbers by saying "deaths from prescription opioids are at an all-time high." Such a report is really a commentary on the deadliness of heroin, not of prescription opioids, and the reporter is failing to separate these into different topics. You could just as truthfully say "Deaths from prescription opioids have been declining since 2011" or "Deaths from prescription opioids are at their lowest level since 2007." In accidental drug poisonings involving both street heroin and prescription pills, it seems likely that heroin would be the more dangerous substance and more likely to be responsible for the death. Given that, I think the "prescription painkiller deaths are declining" framing is more accurate.

If you're a visual person, here's the same information in graph form:

Some context for this. Some recreational opioid users use benzodiazepines to "potentiate" the effects of whatever else they're taking. This is incredibly dangerous and apparently accounts for a large number of accidental deaths, but some people apparently take this risk willingly.

Some context for the cocaine figures. Cocaine and heroin is a common combination, known as speed-ball. It was involved in the deaths of John Belushi, Chris Farley, Mitch Hedberg, and Philip Seymoure Hoffman. What's likely happening here is that the "heroin" used to create this concoction is increasingly tainted with fentanyl, sufentanyl, carfentanyl, etc.. What might look to a naive observer like an increasingly severe "cocaine problem" is really a "black-market heroin" problem. (Note the automatic axis scaling on this one; the bottom of the y-axis is not at 0. See the table above for the numbers.)

Policy Implications

Facts and number by themselves have no policy implications whatsoever. You need a theory or narrative to interpret them. This hasn't stopped some commentators from drawing unwarranted, even silly, conclusions from recent evens. Some have claimed that recent events have caused them to start "questioning my drug-libertarianism." (As if drug libertarianism has been tried as a matter of actual policy. It hasn't.) Supposedly the increase in prescription painkillers and the subsequent rise in opioid-related drug poisonings are a black eye for drug-libertarianism. That might have been a semi-defensible position up until 2010, when people started noticing the rise in opioid-related deaths. But the 2010 to present period represent a huge black eye for drug prohibition. In recent years, legal restrictions, informal guidelines, and attitudes about prescribing practices have made it harder to obtain opioid painkillers legally. This led to an increase in the use of street heroin by recreational drug users. Drug dealers have "innovated" by adding fentanyl and even more potent synthetic opioids to the "heroin" they sell (in some instances it seems to be a complete substitution of the synthetic for the heroin rather than simply adulterating).

The 2010 to present increase seen in the figures above is enormous, even overshadowing the 1999-2010 increase. (What was happening prior to 1999 is less clear, because the then-used ICD9 system did not separate out deaths by type of substance, at least not as well as ICD10. Still, it looks like overdoses were rising prior to 1999.) The libertarians had this one right all along: you can't control drugs, and you can't stop drug use. There are too many work-arounds. When dealers can collapse a one-ton shipment of heroin down to something that fits into a briefcase, it becomes impossible to stop drug trafficking. In his process, there is no guarantee that the final product has been properly mixed or diluted. The end users have no idea what dosage they are taking, which results in a large number of accidental fatal overdoses (as is evident in the dramatic rise in synthetic opioid deaths).

The 1999-present period did not actually see an increase in prescription painkiller abuse. Some commentators will attempt to blame today's heroin epidemic on prior loose prescribing practices, as in "Today's heroin users are yesterday's legitimate opioid patients; loose prescribing practices turned these people into addicts." But the flat-line in the "abuse" statistics is a huge break in that chain of causation. Apparently you can triple the amount of opioid prescriptions (roughly what happened in the 1999-2010 period) and not actually get any more addicts. Even putting that aside, it's not necessarily problematic if prescriptions for a drug triple and deaths associated with that drug also triple. If an enlightened decision is made to increase the use of a particular treatment tool, which I would argue is an accurate description of the decision to use opioids to treat more chronic pain, we should expect the side effects of that treatment to triple. It's not even clear that all of these deaths that are labeled "drug poisonings" are actually that. A standard textbook on drug-related pathology warns repeatedly against assuming something was a drug overdose based only on toxicology screenings, circumstantial evidence, or convenience. The repeated warnings, and statements made by the author elsewhere, imply that this mistake is common.

________________________________________________________

Here's a back-of-the-envelope calculation regarding mis-labeled causes of death. About 3% of the US population are on high-dose opioids (so about 9.7 million people). These people have a higher-than-average mortality, about 1.6 times the typical all-cause mortality (because they're older and sicker, not primarily because they are dying of overdoses). The population average mortality in the US is ~850 per 100k. So about 130,000 people on high-dose opioids die in a given year from all causes (9.7 million * 850/100k * 1.6). If 1% of these are mis-classified as drug poisonings (and that must be a tempting diagnosis if someone in this population dies suddenly and mysteriously, say from a heart arrhythmia), that's 1,300 deaths mis-labeled as drug overdoses. If 5% are mis-classified, there are 6,500 mis-classified drug poisonings related to prescription opioids (out of a total of 8,691 in 2017, again counting accidental only and excluding heroin-related deaths). Even if the mis-classification rates for cause of death are low, mis-labeled "opioid poisonings" could represent a large fraction of the total and could over-state the trend. Seen in this light, the rise in "opioid deaths" in the CDC's official numbers may not be alarming at all. (The discussion in this paragraph is strictly about prescription opioids, not heroin. I believe the recent rise in heroin-related deaths is very real. The "at-risk" population for heroin is a lot smaller, so it's not likely that a low mis-classification rate could be responsible for such a large increase in the official figures.)