I’ve written several posts on the recent dramatic increase in drug poisoning deaths. I think it’s still an open question whether there are *really* more drug overdoses or if there are just more people who happen to die with high opiate levels in their blood because there are so many more people using prescription painkillers. It’s possible that someone with a high-dosage prescription for painkillers will drop dead for some unrelated reason, and a lazy/incurious/ignorant medical examiner will mark it as a drug overdose because that’s the most handy explanation available. With any death, a human being using their fallible judgment has to ultimately decide on the cause of death, which ultimately gets marked down as *the* underlying cause of death in the Center for Disease Control’s mortality database. It’s always possible that one of many contributing causes gets singled out as the one underlying cause. Or perhaps a person is stricken down by an invisible cause (say a heart arrhythmia, which won’t leave any physical sign), but something visible, like a toxicology screening showing elevated levels of opioids, conveniently explains away the death.
The CDC maintains an excellent database, which lists every single death in the US for every year going back to 1968. (See the Mortality Multiple Cause Files here.) Each row in these files is a single death record, listing the age, gender, and other demographic variables, along with the causes of death. A single “underlying” cause of death is singled out, and then there are 20 spaces to list contributing causes of death. For example, the underlying cause might be “drug poisoning” and in the 20 spaces listing the contributing causes of death, it might say, “heroin”, “alcohol”, and “benzodiazepines” (plus a bunch of blanks). (The CDC files actually list codes, not the named descriptions. You need a file that lists the codes and what they mean to actually decode the CDC’s file.) Knowing that this database even exists, we’re off to a good start. Superficially it looks like we have all the information we need to determine how many drug overdoses are happening each year, and from which classes of drugs they are coming from. If one naively counts the drug poisoning deaths, there does indeed seem to be a recent rapid increase in drug-related deaths, particularly those related to opioids.
However, many of these “drug overdoses” also list various kinds of organ failures, illnesses, and vague infirmities among the up to 20 causes of death listed on the death record. That raises a serious question about what killed many of these individuals. Take an individual with “cardiomegaly.” Perhaps this really was a drug overdose and an irrelevant medical condition was simply listed on the death certificate. But it could be the other way around. Maybe this person had a heart condition that killed them, but an irrelevant drug habit was listed on the death record. Or possibly an underlying medical condition *did* contribute to the person’s death, as in the drugs would not have been lethal to a healthy individual. (Cue philosophy lecture on the nature of causation in a world that is dense with causal factors.) Or perhaps someone’s drug habit damaged his health and organs, ultimately contributing to his death. Some of these cases are true overdoses, some are not overdoses per se even though they are related to societies drug problem, and some have nothing whatsoever to do with drugs. I think we need to be very cautious about how we interpret this data, especially since there is a general bias toward exaggerating the harmfulness of drugs and blaming them for things that they didn’t do.
Now, some of these have an obvious connection to heroin use or an opioid overdose, and I’ve marked them in red. “Respiratory arrest” and “anoxic brain damage” sound like the effects of an opioid overdose. Usually these people die because respiration is suppressed so much that they asphyxiate. “Chronic viral hepatitis C” sounds like infection from intravenous drug use. I’m willing to chalk such a death down as indicative of society’s drug problem, but I think it’s fair to say this casts a lot of doubt on what killed these people. Were the hepatitis C deaths *really* drug overdoses, or were they cases of gradual organ failure?
It’s very hard to quantify this uncertainty, but here’s how I tried. When I counted the number of “drug poisoning” records that have one of the causes listed in this table, excluding the ones marked in red, I get that about 20-25% of records have at least one of these causes. (Don’t sum up the numbers in this table to get a total. That will overcount because a death might have, say, three of these causes listed on the death record and thus will be counted three times in such a total.) A more precise estimate for this figure would require someone to go through the list of contributing causes of death and saying, “This is a common drug-related illness, this is not, this one is, this one isn’t…” The full list of contributing infirmities is much longer than the table above; there are about 1,000 different causes of death to sift through that are related to organ failures and other infirmities. Someone with far more expertise on drug pathology would need to do this sifting.
I looked at another item that’s relevant to this discussion: How many deaths had a drug listed as one of the contributing causes of death, but weren’t counted as drug overdoses? (In other words, these didn’t have X40-X44, X60-X65, X85, or Y10-Y14 listed as the underlying cause of death code. For example, some of these had “other opioids” listed as one of the 20 causes of death, but the underlying cause was cancer or a heart attack or an automobile accident.) It was very rare to have any drug poisoning mentioned as any one of the contributing cause of loss codes, but *not* have the underlying cause of death be a drug poisoning. For example, only 1.7% of deaths that mentioned heroin (cause of death code T40.1 ) were *not* counted as drug overdoses. This number was 4% for methadone, 2.6% for “other synthetic narcotics”, 3.9% for “other opioids”, 4.3% for “other antidepressants.” When I averaged this figure for all the most lethal drugs (which captures an overwhelming majority of drug-related deaths), I get about 5% for this figure. So in other words, 95% of the time when there’s a drug on the death record, they count it as an overdose. But a significant proportion of drug poisonings list another type of illness as a contributing cause of death (25% by my reckoning, and the real number is probably quite a lot higher). I take these together to mean that there is a significant bias in favor of labeling deaths “drug overdoses” if there’s any kind of evidence of drug use.
If there really is a tendency to misattribute the cause of death to a drug poisoning, it could go even deeper than it appears. I’ve described above how many deaths are labeled in a way that leaves the underlying cause of death ambiguous. But many of the deaths that aren’t ambiguously coded could still be in error. If a sudden heart attack or pulmonary embolism kills someone, and there is outward evidence of opioid use, a lazy or incurious medical examiner might simply say, “Code it as X42, drug overdose. Done! I’m going to lunch.” From a previous post, autopsies are missing in about 20% or more of these drug poisoning cases, and some causes of death aren’t obvious even *with* an autopsy. Pathology of Drug Abuse, a popular textbook on the topic by Steven B. Karch, goes into great detail on this point: it’s easy to misclassify a death as a drug overdose. If a large proportion of these deaths are misattributed to drugs when they are really some other cause of death entirely, then the recent moral panic over opioids is overblown. It could be an exaggerated problem, or not a problem at all (at least not an *increasingly severe* problem, as we’ve been led to believe.) In short, if we are mistaken about the magnitude and causes of a problem, we will implement the wrong set of policies to address that problem. Based on what I’ve seen, this misattribution problem is a big deal.
I want to make another slightly different point about the policy implications of all this. It’s possible that drug use is relatively safe for most people, but risky for people with certain medical conditions. It is worth considering the implications of this for drug policy. If the risk of a fatal poisoning is different for different people, then there is a lot of low-hanging fruit to pluck. Instead of deterring drug use per se, a policy of deterring drug use for high-risk individuals could be more effective and ultimately save more lives. Public service announcements that warn against the dangers of drugs in general are bound to backfire, because people who encounter drugs and come out unscathed will learn that the government is just lying to them. But a public service announcement warning that people who experience sleep apnea are especially prone to drug overdoses might be taken more seriously. “Sleep Apnoea” and “obesity” appear frequently on the death records for drug poisonings, and it’s not hard to see why this might be the case. Opioid overdoses suppress respiration, and the decedent asphyxiates. Someone with sleep apnea or obesity, who has trouble breathing during sleep anyway, will be especially susceptible to suppressed respiration.
There’s a limit to what I can do with the CDC’s death record data. If these data are miscoded, as I’m arguing in this post, then the information I need to answer the relevant questions doesn’t exist. Someone would have to do a thorough investigation of a large sample of drug poisonings to see if a potential cause of death was missed. I’d like to know just how many of these deaths are *actually* drug poisonings, and I’d also like to know what medical conditions are risk factors for drug poisonings. Both would help in crafting good drug policy.