I’ve been looking over the CDC’s mortality data in great detail to see what insights I can gleam from it. One of the open questions is whether most of the drug overdose deaths (which have supposedly increased dramatically over the past 15 years) are from hard-core drug users or from normal users who don’t have a serious drug problem but just get careless one day and poison themselves. This distinction between hard-core and normal users might make less sense for heroin or cocaine, where an overdose probably indicates a serious drug problem. But the difference is very salient for things like prescription opioids and benzodiazepines, for which most users are normal people with legal prescriptions. This is a big deal, because different answers to this question can have different policy implications. Most of the media reports of the rise in prescription painkiller overdose deaths assume that loose prescribing habits have created a lot of addicts, who then turn into problematic drug abusers. An alternative narrative is that loose prescriptions per se don’t turn people into addicts, but suddenly cutting people off from their doctors when they “look like drug addicts” might induce these people toward risky behaviors, like seeking drugs on the black market. (Doctors can get into serious trouble for prescribing opioids to people who look like they’re abusing them, so doctors are extremely cautious and often do cut people off from their supply of painkillers if they see any suspicious behavior.) Another explanation is that the increase in deaths isn’t driven by an increase in addiction or risky recreational drug use, but is rather driven by people carelessly mixing drugs that they have a legal prescription for, as well as having a legitimate medical use for.
I had an idea that I thought would allow me to determine how many of these deaths were normal/legal users and how many were addicts with a serious drug problem. In the CDC death records data, there is room for up to 20 causes of death to be listed, and drug abuse disorders are often listed as contributing causes of death. These are cause of death codes F10-F19.9, for anyone who’s curious. They all start with “Mental and behavioral disorders due to use of…” and then list one or more drug types and “withdrawal”, “harmful use”, etc. At first blush, this settles it. There’s a code for whether the decedent had a serious drug habit, so just calculate the percent of drug poisonings that have such a code listed on the record. It’s a simple counting problem that Excel’s sumifs() or countifs() formula could handle. This should be able to tell me if the recent rise in opioid deaths is mostly attributable to people with a serious drug problem or not.
Unfortunately, it looks like these F-codes have not been used consistently over time. In 2000, only 29% of heroin overdoses deaths and 32% of cocaine overdose deaths had such a code; in 2014 it was 58% and 59%. I don’t know what the appropriate level is for these numbers, but my guess would be that someone who overdoses from one of these drugs is probably someone with a serious drug problem. I’m thinking both the 29% and the 58% numbers for heroin are low. The proportion of these overdoses that are due to serious drug users has supposedly increased by a factor of roughly two (1.99 for heroin and 1.83 for cocaine). I see a similar, if slightly larger, magnitude for other opioids, benzodiazepines, methadone, and “psychostimulants with potential for abuse” (meth and ADHD medication). If this increase was very different for heroin/cocaine than it is for benzodiazepines and prescription opioids, I might have believed these figures. But it looks like there is a general upward trend in the use of these “f-code” causes of death that probably has nothing to do with an actual increase in addiction rates. My conclusion is not that the proportion of drug poisonings due to serious drug addicts has increased; rather I conclude from this that the death record data isn’t very good and contains many spurious trends. I’m chalking this one up as another of those “data quality issues” I mentioned in this earlier post. It’s too bad, really, because the coding system actually has the potential to answer the questions I’m asking of the data, if only it were used properly. But to get accurate answers, the coding system would have to be used in a consistent way over time and across the country (even across different countries). This necessary consistency is sorely missing.
Why does this matter? If a life is tragically cut short by a drug poisoning, who cares whether it was a helpless addict or a careless normal user with a legal prescription? A corpse is a corpse, right? It actually matters quite a bit for drug policy. Popular media reports tell us that the increase in opioid prescriptions is creating an addiction epidemic, and these helpless addicts, their will having been overtaken by drugs, poison themselves. If this story is right, then some kind of massive drug abuse treatment initiative might be in order. (I don’t think this story is correct, because we don’t see any increase in illicit use of prescription painkillers in drug use surveys, and we don’t see a substantial increase in the number of people with substance abuse disorders.) Another explanation is that drug use, even drug dependence, is mostly safe, but addicts turn to risky behaviors when they are suddenly cut off from their legal supply. In this story, we should make opioids *more* accessible, not less, because it is the *lack* of access that is driving people to risky behaviors. My preferred explanation is that most of the drug overdoses are coming from normal users with legal prescriptions, who accidentally take too many pills (single-drug overdoses are rare) or mix them with benzodiazepines, alcohol, or some other drug that causes a fatal interaction (multi-drug poisonings are much more common). In this story, these additional overdoses actually represent a very small hazard. Considering that there are roughly as many legal prescriptions each year as there are adults in the US, 10 to 20 thousand deaths represents a risk of only about 1 death for every 15,000 prescriptions. (20,000 deaths /260,000,000 prescriptions gets you 1/13,000; in the previous sentence I’m being imprecise on purpose because I think the underlying figures are imprecise.) This is probably an acceptable risk for someone who is suffering severe chronic pain, especially if it is a risk that the user can actively mitigate (say, by avoiding dangerous drug combinations and taking only the recommended dosage). In this view, the policy implication is a stern warning from the doctor (perhaps reinforced by the pharmacist, and then again by the drug’s packaging), but blanket restrictions on opioid prescriptions are unnecessary. If these deaths were accurately coded, we would have been able to see whether the increase in deaths is driven by hard-core addicts or normal non-addicted users, and we’d have a better idea of the policy implications.