Tuesday, April 12, 2016

Prescription Painkiller Drug Poisonings: Four Competing Explanations

The number of deaths involving prescription painkillers has increased in the past 15 years. Consider a few different explanations.

1)      Doctors have loosened the constraints on prescribing painkillers. With more people taking habit-forming narcotics, more people are getting addicted, developing harmful drug habits, and overdosing.
2)      There aren’t really any more addicts or users with unhealthy drug habits. The increase in overdose deaths is entirely from within the population of normal people with legal, medically legitimate need for painkillers.
3)      There aren’t really any more addicts or users with unhealthy drug habits (same as in 2). But the drug habits of the (roughly unchanged) population of addicts have become more severe. Addiction has become more deadly. The increase in painkiller-related poisonings is happening within this population of addicts.
4)      There is a reporting bias that is grossly exaggerating the true trend, or even creating a trend out of nothing. Medical examiners and coroners are doing more thorough toxicology screenings than they used to, so naturally they are finding more evidence of drug use in the recently deceased. Deaths are inappropriately labeled “overdoses” when there is evidence of drug use, even if there isn’t good evidence that the drug contributed to the person’s death.

So what’s really going on here? We need to pick the narrative that best fits the available data. These  narrative explanations aren’t all mutually exclusive, as we’ll see. But we need to pick the one or two that best fit with the available information. What follows is a discussion of the evidence that should be brought to bear on any narrative that attempts to explain the recent rise in prescription opioid painkiller deaths. Below I list the pieces of evidence I found most salient, which must be reconciled with any narrative.

CDC Drug Poisoning Data

The claim that opioid-related drug overdoses are increasing is based mostly on CDC cause of death data. The CDC keeps a record of every single death that happens in the United States, including information about the decedent and the causes of death. Up to 20 causes of death are listed in the CDC’s data file. “Other Opioids”, marked by cause of death code T40.2, is the cause of death most often listed on accidental drug poisoning deaths (cause of death codes X40-X44). It is the category that most prescription painkiller overdoses fit into, although there is also a Methadone, “Other Synthetic Narcotics”, and an “Other and Unspecified Narcotics” category. In 2014, there were 12,657 deaths in this category, followed by 10,754 “Heroin” deaths, 8,295 “Benzodiazepines” deaths, 5,992 “Cocaine” deaths, and 5,695 “Other synthetic narcotics” deaths. Were it not for a recent increase in heroin deaths, “Other Opioids” would be the leader by a wide margin. (Then again, were it not for a recent decrease in cocaine deaths, which started declining in 2006, it might not be the top category.) Our best available data is showing an increase in “Other Opioid” deaths. So at the very least some kind of explanation is needed.

Information About Prescription Drug Abuse Rates

Are there any more addicts than there were 15 years ago? Are more people abusing prescription painkillers? Apparently not. See the National Survey on Drug Use and Health, page 7 (page 11 of the pdf here). The number of people using painkillers illicitly is flat or even decreasing for some age groups. You will find a similar story in the Monitoring the Future survey here (page 31 of this report). MTF focuses on high school students, but the question about “Availability” is telling. Apparently our society is not awash in these drugs if high school students are finding them *less* available. If you look at this document (here, page 32), you can see where people are getting their illicit supply of painkillers from. The largest single source, representing 53% of responses, is “Free from a friend.” This is hardly a reliable source of drugs for a hardcore drug user. Most of the people who self-report illicit use of painkillers are getting the occasional pill from a friend or relative, not buying in bulk from a “pill mill.”

Maybe the survey results aren’t as telling as they seem. Perhaps there is one really hard-core, self-harming painkiller abuser for every 5 recreational drug users (or 1 for every 10, or for every 20, or whatever ratio seems plausible). If illicit use of painkillers is in the 2% of total population range, and if problematic illicit use is in the few-tenths-of-a-percent range, then this sub-population of abusers could grow substantially without changing the overall “Past month use” numbers. Well, see page 84 of this document, the 2013 SAMHSA drug abuse survey (once again). In the paragraph directly under the chart (which shows number of people in millions, *not* % of population), it says that about 0.7 percent of the adult population are dependent on pain relievers, while (recall from above) about 2% of the adult population are past-month users. So about a third of people who have recently used painkillers are experiencing some sort of problem with them, and this ratio has been very stable (and the numerator and denominator of that ratio have *both* been very stable). We might still posit that there are some *really* hardcore painkiller addicts who represent 0.1% of the population, and that the growth in problem usage and overdose deaths are all attributed to this subpopulation. But if we go down that path we are increasingly dismissing what the available data are telling us and grasping at straws to explain it away.

Evaluating the “Loose Prescription Pad” Hypothesis

There is conflicting information about the availability of painkillers. Supposedly it’s much easier to get a prescription, and indeed the number of legal prescriptions has been rising. It’s a few years old, but see the graph at the bottom of this page showing the tonnage of opioid pain relievers sold. It has since leveled off at about 7kg/10,000 population (I have data e-mailed to me by a CDC employee that shows the “OPR Sales KG per 10k” for the next several years), but clearly it has increased over the past 15 years. Both opioid prescriptions and opioid deaths per population have increased by about a factor of 4. This seems reasonable, and it’s a piece of information worth committing to memory: a given prescription is no more deadly that it was 15 years ago. It’s just that there are more prescriptions.

It’s odd that this increase happened at the time that it did. The DEA launched a massive crackdown on prescription painkillers in the late 90’s and early 2000’s. See Ronald Libby’s paper on the topic here. If you don’t want to take a CATO scholar’s word for it, you don’t have to. The National Association of Attorneys General was alarmed enough about the DEA’s crackdown that they sent a letter to the DEA’s director, registering concern and asking for an explanation (here ). Perhaps the crackdown succeeded in putting a few high-volume pain doctors out of business; there were certainly some high-profile prosecutions. But the tonnage of opioids prescribed quadrupled anyway. What’s happening here? Did the remaining pain doctors pick up the slack, and then some? Was the medical community becoming more enlightened about the need for adequate pain management, and the DEA’s efforts just weren’t sufficient to stem the rising tide? Did naïve primary care doctors pick up the slack left by pain doctors and prescribe promiscuously? Whatever is going on here, it happened *in spite of* a government crackdown. Some people have claimed that the past 15 years demonstrates a need for prohibition and a refutation of drug-libertarianism. That explanation is *a bit* silly considering that the past 15 years have seen an increasingly severe regime of drug prohibition.  

Evaluating Drug Poisoning Trends

It’s impossible to know how accurate the CDC’s drug overdose figures are, but a spurious trend of rising overdoses shouldn’t be ruled out of the question. What’s interesting about the rising rates of overdose deaths is that we’re seeing them in almost every single drug category. See the lines in the above chart. Everything appears to be going up. Even if we focus on drugs that don’t kill very many people (Antiparkinsonism drugs, Cannabis, antitussives, beta-anrednoreceptor antagonists not elsewhere specified), death rates are rising 2- or 3- or 8- or 9-fold. (Less-lethal categories of drug plotted here, bottom figure.) There are for sure a few categories that are fairly flat or declining. Cocaine is only slightly up over 1999 and has in fact been declining since 2006; it used to be the biggest single killer and now it is 4th. Barbiturates, which only kill about 350 people a year, are also pretty flat over the period. A plausible explanation is that, with so many additional drug overdoses, medical examiners are checking those additional bodies and jotting down whatever other substances they find. In this view, the increase in opioid painkiller overdoses is real but the trends in the other drug categories are spurious. Another plausible explanation is that the medical examiner are simply doing more checking (as an exogenous change) and finding more of every drug category. In this (perhaps less plausible) view, the recent increases in opioid painkiller deaths are also spurious. It is notoriously difficult to diagnose an opioid overdose. A person who has been using opioid painkillers for a long time has built up a tolerance and may have blood levels of the drug that would be toxic to a new user. I don’t really think the *entire* increase in opioid poisonings is spurious, but it’s clear that death certificates are including more information about the drugs in the decedent’s system than they used to. It’s implausible that some two dozen drug categories have increased in lethality by a factor of 3 or 4, all at the same time. Clearly some of the substances that are being written down on death certificates aren’t actually related to the cause of death. If we’re willing to apply this observation to Cannabis and antiparkinsonism drugs (for which there is no obvious narrative to explain the increase), we should go a step or two further and apply it to opioid painkillers, too.

In support of the “spurious trend” hypothesis, cocaine use is correlated with death rates after about 2006 (when both are falling) but uncorrelated with death rates before 2006 (when use rates are roughly flat but overdose deaths are increasing). Heroin deaths are hard to measure, because there are so few users, and so many heroin addicts probably aren’t part of a household where a drug use survey would reach them. Survey numbers tend to be noisy, and even a large apparent year-to-year difference in the number of users can be “statistically insignificant” (indistinguishable from noise). Still, use rates on the SAMHSA survey are correlated with heroin overdose deaths from the CDC data. Someone relying on the SAMHSA use rates would have predicted the recent surge in heroin-related deaths. So clearly the self-reported drug use in these surveys is *somewhat* accurate. If the CDC’s numbers for cocaine-related drug poisonings was rising between 1999 and 2006, when the survey data said that cocaine use was flat, this is evidence that there was some kind of reporting bias, at least for part of the period we are talking about.

In evaluating narrative 4), the “spurious trend hypothesis”, I would fixate on the fact that opioid poisonings, expressed as a rate per legal prescriptions or per kg of opioids sold, are roughly constant. It makes sense that this ratio wouldn’t move much. Whatever risk prescription painkillers pose to their users, that risk doesn’t change when you add more users. In this view, the trend is real and is roughly accurate (a 4-fold increase over 15 years), but the total number of deaths might be higher or lower than the CDC data shows.    

Multi-drug Interactions

Most prescription opioid “overdoses” aren’t true overdoses. They are multi-drug interactions, usually with benzodiazepines or alcohol. In combination these drugs decrease respiration, often with fatal results. Just over 30% of prescription opioid poisonings also involve benzodiazepines, and about 15% involve alcohol. Only about 15-20% of opioid poisoning deaths involve only one substance. (By that I mean the CDC death record has only one drug-related “multiple cause of death” code in about 15-20% of cases.) Overwhelmingly these are multi-drug interactions, which bolsters hypothesis 2) above: the additional overdose deaths are coming primarily from normal people with legal prescriptions who unknowingly mix dangerous combinations of drugs. If this were specifically a “prescription opioid epidemic”, I would have expected to see more drug poisonings that are exclusively caused by the “Other Opioid”, “Methadone”, or “Other Synthetic Narcotics” categories. The pattern of multi-drug deaths I’m describing here is also pretty stable across the past 15 years. If I go back to 2000, I see the same pattern: most poisonings involve multiple substances, the distribution of number of substances is about the same across all years, and certain substances tend to be found in combination at the same rates, regardless of which year you’re looking at.

A Reiteration: Competing Narratives and Evidence

I found it impossible to write this post without interjecting my own discussion and opinions, but let me try to reiterate the competing hypotheses and the data that need to be squared with them.

Hypothesis 1: Looser prescription practices led to more opioid addicts, which led to more overdose deaths.
Hypothesis 2: There aren’t any more addicts. The increase in overdoses came from an increase in the total number of people with legal prescriptions, mostly *not* using the drugs recreationally.
Hypothesis 3: The population of addicts and recreational users of painkillers has been constant, but the intensity of drug use within this population has increased, leading to more overdoses.
Hypothesis 4: The increase in drug poisonings isn’t real. It’s a spurious trend created by a reporting bias.

Whichever hypothesis you prefer, you need to square it with the following information:
  1.       CDC data shows a dramatic increase in opioid painkiller deaths.
  2.      CDC data shows a dramatic increase in *all* categories of drug related deaths, even those for which there is little evidence of increased use.
  3.       Most drug-related poisoning deaths involve multiple substances.
  4.       Many death records include mentions of drugs that aren’t plausibly related to the cause of death.
  5.       It is difficult to determine a cause of death.
  6.       A toxicology screening can be misleading, because a chronic drug user can tolerate a dose that would be lethal to a naïve user.
  7.       Drug use surveys show that the proportion of the population using prescription painkillers illegally is unchanging.
  8.       Drug use surveys show that the proportion of the population with painkiller-related drug abuse disorders is also unchanging.
  9.       Most illicit users of prescription painkillers received their drugs for free from a friend (not from a doctor or drug dealer).
  10.    CDC drug poisonings are correlated with survey-reported drug use rates for cocaine and for heroin, but *not* for opioids (or other prescription drugs).
  11.    The government crackdown on prescription painkillers began before painkiller-related deaths started to rise.

Happy hunting.

No comments:

Post a Comment