Friday, March 13, 2020

Exponential Increase in Drug Poisoning Deaths

A recent paper published in Science examines trends in drug poisoning deaths over the past 40 years. It finds that drug poisoning mortality (and this is per-population, not simply the raw totals) has been increasing at an exponential rate since 1979. That was the year we switched over from ICD-8 to ICD-9, so presumably the prior data was not directly comparable. Otherwise I’m guessing they would have gone back further to see how far back the trend holds.



An unbroken exponential trend with an r-squared of 0.99 begs for some sort of explanation. Most opioid epidemic narratives describe discrete events or policy changes or changes in prevailing attitudes, like the release and aggressive marketing of OxyContin in 1996, the change in attitudes regarding pain management and opioids over the 1990s, the rise of the “Xalisco Boys” heroin cartel in the 90s and 00s, the reformulation of OxyContin to be abuse-deterrent in 2010, and the emergence of illicit fentanyl as a substitute for heroin in the 2014 to present period.  All of these things happened. All of these things were the mechanism by which drug poisonings increased at the time they were increasing. Each discrete item might explain, in a narrative sense, a few years of the time series. But it seems like there must be some kind of physical law compelling the trendline, and all the narrative is just detail. It’s as if someone wrote an economic history book describing a few specific inventions and innovations and said, “This explains the modern world,” but then failed to note that there was a nearly unbroken trend of exponentially increasing productivity and consumption. Yes, the assembly line, the shipping container,  electricity, and the transistor were the means by which we experienced exponential growth in productivity.  But there’s some kind of underlying law driving the emergence of these inventions. It's quite possible that even but for these particular developments, the exponential growth would have happened anyway. 

If something is growing exponentially, it’s worth looking for some driver that would explain an exponential pattern. Population growth comes to mind, but that’s obviously not right. It is the rate per population that is growing exponentially, not the raw number of poisoning deaths. My best guess is that the driver is economic growth. We are getting richer at an exponential rate. That allows us to buy more of all goods and services. When we get richer, we don’t just keep doing the same things and simply buy more stuff with our higher earnings. We don’t simply buy more houses and more automobiles and greater tonnage of food. To some degree we buy nicer versions of the things we already had. So bigger homes, safer (and more reliable and more fuel efficient and more comfortable) automobiles, tastier food of a much wider variety than was available decades ago. Even granting this, there’s only so much consumption we can do. When we get more productive, we “consume” some of that capacity by working less. There is, as economists say, a substitution effect and an income effect. Our incomes rise as we get more productive, but we trade work hours for labor hours as is becomes easier to satisfy our material needs. If your pay doubled, you could enjoy twice as much stuff with no extra leisure (by not working any fewer hours) or the same amount of stuff with a lot more leisure (by cutting your working hours in half). Most people tend to prefer something in between either of these extremes.

[A temporary bump in your hourly wage is likely to cause you to work more hours, assuming you know the change is temporary and your wage will predictably revert back to baseline. You will make hay while the sun shines. But if the sun promises to shine more brightly permanently, you'll eventually notice that you can meet your hay quota in fewer working hours and can thus afford more leisure.]

There is a lot of commentary about how “wages have stagnated” in the United States since the 1970s. Supposedly only the rich have enjoyed the rising incomes, and the middle-class and below have stagnated or even seen declining incomes. I think that narrative is basically wrong. I actually think that basically everybody has seen an enormous expansion of their overall option set, it's just that some have chosen to consume more leisure and others more work. (Survey data on how people spend their time backs up my story fairly well.) “Productivity” in general, the option of working for more pay and more command over material goods, has increased basically for everybody. But people are heterogeneous in their preferences for work vs. leisure. More and more prime-age males are leaving the workforce voluntarily. Many of them are taking prescription opioids, recreationally or medically. I think it’s wrong-headed to try to explain this as an “income inequality” story or to claim that these people “lack opportunity.” These people actually have a lot of command over material resources. It is not the case (as many commentators imply) that their zero market income implies zero access to resources. They generally have access to non-market income or other resources. Family members. Government safety net programs. The black market. Their labor market income vastly understates their actual command over material goods and services. In my view, these people are enjoying the same exponential growth in productivity as the rest of us. They are simply choosing to consume more leisure. The fact that everyone else is working productively increases their non-market remunerations. A higher-paid parent can afford to give a higher allowance to his/her man-child, and a higher-paid tax base can fund a more “generous” social safety net. Some people are pushing a universal basic income in anticipation that robots will make a lot of us obsolete. I think these people are implicitly acknowledging my point: the enormous productivity of the robot economy will presumably create the cornucopia from which we can afford the UBI. I do not support a UBI, I am just pointing out that I'm not the first to notice that non-workers enjoy the benefits of rising productivity. Some of these people are choosing to consume all of their share of enhanced productivity in the form of leisure. 

[Do not read any of this as scolding the life-choices of non-workers. This post is a judgment-free attempt to explain the trends in drug overdose deaths.]

I don't completely believe all of this. I don't think the 2010 to present trends were inevitable. I think this era is dominated by a surge of black market drug poisonings that never would have happened if people (including recreational users) had ready access to prescription opioids. I also think the rise in deaths from the late 1990s to 2010 are probably overstated. I think nobody has gotten a proper handle on how inaccurate cause of death assignments really are and how much this affects the national statistics. Also, an exponential rise in the death rate obviously can't go on forever. It has to hit some natural limit. At some point, all the hard core drug addicts die off, or the very high death rates scare off the next generation of would-be users, or the number of possible addicts hits a maximum beyond which no more normal people are tempted to join their ranks. Any exponential process that "explains" rising death rates is going to bump into a ceiling and stop working as an explanation. Something's gotta give. Mercifully, 2018 finally saw fewer drug poisonings than the prior year. Still, I do take seriously the idea that a richer society makes it easier to indulge vices. With exponential growth in productivity, with heterogeneous preferences for leisure and vice, and with transfers of income (government and private) between the working and the non-working, we should expect something like what we're currently seeing.

What's a better explanation? Or does this phenomenon even need one? Where else should we expect to see exponential trends if this basic story is right? Divorce rates? Alcohol-related deaths? What about the decline in smoking? Is that a fatal counter-example? What observable vices to we expect to indulge more as society gets richer?
____________________________________

This didn't fit into the flow of the post, but I wanted to draw a caricature of someone with an extreme preference for leisure trying to maximize his utility in today and compare that to thirty or forty years ago. Maybe thirty years ago, a marginally employed convenience store clerk (or perhaps a clerk at a movie rental store) would have earned enough for a small apartment. He'd have just enough income for the essentials and some disposable income for cable TV, maybe a nice stereo, and a couple of weekly video rental. If he is the intellectually curious type, he can get free books and magazines (and movie rentals, come to think of it) from the library. Today, an unemployed man-child would have much greater riches available to him. Maybe his parents get him a smart phone and put him on the family plan. Maybe he has his own laptop (a Christmas or birthday gift), or uses the family desktop computer (because Mom and Dad don't need it when they're at work anyway). He gets internet basically for free, and it's of no marginal cost to his parents for him to access it. He can watch tons of free videos all day long without subscribing to any service, and if he wants to watch movies he can use his parents' paid streaming service for no additional cost. Assuming he's acquired the smartphone or laptop described above, he can probably find a few public places with free WiFi. He might envy the freedom and independence of his 30-years-ago self, but he can get the basket of goods that his doppelganger has for free, and he has access to things that didn't exist 30 years ago (again, for free). Many forms of entertainment have gotten dramatically cheaper, and the people bankrolling his leisure are much richer. That lifestyle doesn't sound appealing to me, but I can understand why some people take it over the "bullshit job" that's their next best option. People get locked into bad patterns and it's hard to break out of them. I could also comprehend someone living this life for a few years and feeling guilty or regretful about their life choices and dulling their regret with drug use. Certainly there are people like this. I think we need to reframe our description of them from "They're completely missing out on economic growth" to "They're consuming their share of economic growth as excessive leisure and vice consumption."

Interesting paper here about how people who are outside the labor force are spending their time.

Sunday, March 8, 2020

Dreamland's Narrative Is Wrong

Sam Quinones' book Dreamland crafts a narrative of opioid use and addiction over the past 30 years. It uses a cast of compelling characters. Unfortunately it is far too heavy on anecdote and light on data. It is incredibly sloppy about its timelines. It could have used a few time series charts that marked key moments and pointed out the relevant changes in the resulting trend lines. I think this exercise would have alerted Quinones and his readers to basic weaknesses in his story.

Most opioid epidemic narratives start with OxyContin being released by Purdue Pharma in 1996. Some versions may start earlier, with attitudes in pain management slowly changing in the 80s or 90s. It is often claimed that OxyContin and other prescription opioids were over-prescribed, that this created a flood of new addicts, and that these addicts then switched to heroin. Quinones by contrast begins his story with heroin distribution in the early 1990s. A distribution network of Mexican farm boys from Xalisco (the "Xalisco Boys") pop up in major American cities. This expansion is happening several years prior to the release of Oxy in '96. The infrastructure for a heroin epidemic is slowly being built up. But it's not until large numbers of pain patients get addicted to prescription opioids that the heroin problem really begins to take off.

There is a timeline in the early pages of Dreamland. Some relevant events include the publication of papers regarding pain treatment in the 1980s, the expansion of the Xalisco Boys' heroin cartel in western states in the early 1990s then east of the Mississippi River in the late 90s, the release of OxyContin in 1996 (the same year the American Pain Society urged doctors to treat pain as a vital sign), and two major DEA stings against the Xalisco boys (one in 2000 and the other in 2006). Below is the second half of Quinone's timeline, where (I would argue) most of the narrative of his book takes place. You can see the entire timeline if you view his book on Amazon and click "Look Inside."

I think it's useful to look at some time series data and compare them to Quinones' timeline. Here are opioid-related deaths from 1979 to 1998, also shared in a previous post.


Opioid-related deaths are already rising before Purdue begins to market OxyContin and before the Xalisco Boys cells begin expanding across the country. I don't think anything in Quinones' story explains the doubling of deaths in the 1979 to 1990 period. (The population grew from 225 M to 250M in that time, not enough to explain a doubling. Here is US population by year.)  I'm also not sure if he's giving them credit for the more dramatic increases in the 1990s. Clearly the trend was in place prior to Oxy's release in 1996. (Drug-related death rates have been on an exponentially increasing trajectory since 1979. That's rates, as in per-population, not just the totals.) Unfortunately, pre-1998 deaths were coded according to ICD-9, which did not split out deaths by substance type, so it's hard to build an unbroken time series that goes all the way back to 1979. That is why you so often see 1999 as the first year in a time series of US drug poisonings. Below are deaths by type of opioid in the 1999 to present period.

OxyContin deaths would be counted in the blue line, "Other Opioids." Clearly these are rising over the period. What's interesting is that heroin deaths are pretty flat from 1999 until about 2007. They are then jump up to a slightly higher plateau from 2008 through 2010, and then increase dramatically after that. Synthetics (presumably this is mostly fentanyl) start low (~400 deaths in 1999) and march upward steadily (to ~2,500 deaths in 2013), then begin increasing dramatically. In a recent post I have tables showing the raw numbers. And in this post I give more details about exactly how I'm adding things up.

It's curious that heroin deaths are flat from 1999 to 2007. Weren't the Xalisco Boys expanding during this time? Weren't people getting addicted to Oxy and switching to heroin? Perhaps we should add the heroin and synthetic opioid deaths together? That would show a much clearer trend. But as far as I know nobody was talking about fentanyl being mixed with heroin at the time. That seems to be a more recent phenomenon, and Quinones doesn't really mention it at all. He explicitly describes the Mexican "black tar" heroin, its color and sticky texture being a function of the production of Mexican poppies into heroin. There is no hint at all that it's tainted with fentanyl. (More on this below.) It is worth keeping timelines and time series data in mind while reading Dreamland, because some sections can be misleading. He has a few paragraphs describing one Gary Oxman, chief health officer for Multnomah County, doing a statistical analysis of heroin deaths. He describes finding a "1,000% increase" from 1991 to 1999 (heroin overdoses having increased from 10 in 1991 to 111 in 1999). He might have been more straightforward with his readers that this was a very localized phenomenon and that a 10-fold increase over nine years isn't typical of what the nation as a whole experienced. (See chart above.) If he's telling a narrative of particularly hard-hit communities, that's fine, but then generalizing the relevance of that narrative to the rest of the nation becomes problematic. Anyone reading his book and thinking, "Yes, I totally understand the opioid epidemic and its roots over the past 30 years" is experiencing an illusion of explanatory depth.

These are the drug death statistics. What about trends in drug use and addiction? I've discussed this before. The big surprise is that prescription opioid abuse is flat from 2002 to about 2012, then appears to decline. This is according to the SAMHSA data (see here).


The chart above shows past month nonmedical use, and the "past year abuse disorder" numbers look similar. (See link in the above paragraph.) It's harder to say what's happening prior to 2002, but I've tried to do this before. Monitoring the Future is a survey of 8th, 10th, and 12th graders. See the numbers for narcotics other than heroin (in this case, only 12th graders). It looks like there's an increase from the 90s to the early 2000s, then it flattens and decreases. It's already on an upward trajectory from in 1995, the year before OxyContin was released. The break in the line around '01 is due to a change in methodology, basically meaning they changed how they asked the survey question. (Again, there is more in the link to my previous post above.)

The trends for the post-2000 period should be truly surprising for anyone who claims that Purdue turned us into a nation of addicts with its release of OxyContin. You don't see it in the numbers. Are they claiming that all the new addicts were created in the 1996-2002 period? I don't know any opioid alarmists who tell this version of the story. They all want to blame the increase in drug deaths over the full 1999 to present period on the dramatic increase in prescription opioids sold over the same period. The number of "milligrams of morphine equivalent" per adult increased roughly four-fold from 1999 to 2010, then started coming back down. (I'm looking at a chart in an unpublished paper that I can't share, but the four-fold increase is something I've seen in a number of places. If someone knows an authoritative ungated source for this information, please share.) Their story is: more opioids => more addicts => more overdose deaths. It's really awkward for their story if that middle step is broken.

It's less clear what was happening with opioid misuse prior to 2002. I tried to look into it in this post. Below is the chart I came up with (one of several; see the post for the full story). Unfortunately, this is just the raw "% of responses", which needs to be adjusted for the sampled population so it matches the US population. The reports I pulled these numbers from explicitly mention trying to sample high-use populations, which will cause the next year's "% of responses" to be higher even if drug use as a whole is unchanged. There is also some kind of methodology change in 1999, causing the percent of "yes" responses to double from what they are in 1998. This is clearly not real, and I think the reason the SAMHSA charts above start with 2002 because the SAMHSA people realize there is something screwy about those years. And once again, I don't think anyone's version of the story is that "drug abuse tripled from 1998 to 2002, then flatlined." It doesn't match the narrative that Quinones or anyone else in this space is telling.


Here is heroin use from MTF. Use goes up in the early 90s, flattens out, then comes back down. And perceived availability is trending down. Maybe all the new users are adults and the kids are alright?


Below are SAMHSA's data for heroin use disorders. Heroin use is so rare that it's hard to even measure. The report I pulled this from is rounding to the nearest 0.1%, so the "trends" look very step-like. There does seem to be an upward trend in the 18 to 25 cohort where heroin use is more common. It's not terribly obvious what the 26+ population is doing. Again, the rounding makes it a step-function going from 0.1% to 0.2%, switching to the higher plateau in 2013. Another issue here is that heroin addicts tend to be disproportionately homeless and won't necessarily be captured in a household survey. Presumably heroin is really is increasing over the time period, but it's hard to say exactly what's happening year-to-year. I doubt if we'll ever have a survey that accurately samples the homeless addict population.




Age of Typical Opioid User

Parts of the book focus on young people becoming addicted to opioids and their parents' grief over their children's drug abuse and (occasionally) death. One of my surprises when I first started looking into this was the fact that drug users tends to skew young while drug deaths tend to skew older. I think that Quinones does a disservice to his readers by fixating on young people when the vast majority of deaths are coming from people in the 30 - 60 demographic. Stories about dead teenagers do a good job of tugging our heart strings, but they just aren't representative of what's happening. Quinones has a long section about high school athletes, particularly football players, who supposedly go on prescription painkillers after suffering injuries. Some of them even use the painkillers to allow themselves to keep playing. The insinuation is that this leads to large numbers of addictions and overdoses. Quinones also quotes an addiction doctor named Richard Whitney, claiming his practice used to mostly be "middle age alcoholics" and that opioids brought in a cohort fifteen years younger than what he was used to. Once again, given the age demographics of the people who are dying it's just not plausible that this is a typical opioid story. Whitney may have been describing his own practice accurately, but as with many parts of the book, it's merely an anecdote and is not representative of nationwide trends.

Below I have plotted the average age at death for accidental overdoses involving opioids for years 1999 to 2018. There is a fairly clear upward trend in the age at death of people who die from prescription opioids (this is excluding heroin and fentanyl, though the trendline looks similar when I don't exclude them). There is no obvious, monotonic trend in heroin or fentanyl deaths. (It's notable that prescription opioid deaths are on average a few years older than heroin-related deaths.) There is no suggestion in the death-data that uses are getting younger.



Below are density plots showing the distribution of deaths by age for three years for prescription opioid deaths. (I tried a chart with more years, but the plot looked to busy.) You can see an aging in that the mode is shifting rightward and there is more probability mass at higher ages.


I have no data handy on how the average addict or user has been aging over time. Conceivably that trendline could be different from the trends in deaths, but it seems likely they'd be moving in the same direction. I have two earlier posts on the age-at-death distribution compared to the age-of-addicts distribution, one on opioids and one on cocaine. It's true that within a given year, the two distributions don't overlap well. Users skew young while deaths skew old. So either young people are less susceptible to drug overdoses because they are heartier, or old people who happen to use drugs are dying of other things and wrongly being labeled "drug overdoses." Plausibly both effects are real.

Very Recent Trends Are Not Really Part of His Narrative

As you read Dreamland, it's worth bearing in mind that Quinones' timeline does not really capture what's been happening in recent years. Dreamland was published in April of 2015. The CDC data on deaths for a given year come out late in the following year (November or December, although the 2018 numbers didn't come out until early 2020). So he would have had data through 2013 at the latest. Heroin deaths were increasing over the period in question, but the very recent explosion didn't start until 2011. See the chart below (same as the chart above but stopping at 2013); there is an obvious break in the trendline for heroin deaths between 2010 and 2011. It is clearly on a faster trajectory in the 2011 to 2013 period, and this upward trend continues through 2016. Assuming he was looking at this data at all, he would only have seen three years of this increase, and which is at least a decade removed from the Xalisco boys entering the market in the 1990s. Dreamland was out before the fentanyl crisis began, so that narrative's relevance to very recent trends is even more questionable. See the synthetic narcotics deaths. These were rising steadily between 1999 and 2013, but there was a sudden break in the trendline in 2014. They've been skyrocketing since then. The first data point on that new trajectory, 2014, wouldn't have been known until late 2015. (I may have been the first or one of the first to point out the break in the fentanyl trendline, because I started blogging about this as soon as that data came out in late 2015/early 2016.) Just keep in mind that all the discussion of Xalisco boys and OxyContin in the 1990s is separated by a decade and a half or more from the recent rise in heroin deaths. In my opinion, the stories he tells in his book are wholly divorced from the startling trends we've seen in the 2010 to 2020 period.


It would be rather Procrustian to use Quinones' narrative to describe any of these trends. The timing is just wrong.

The Jick Letter

Much ink has been spilt over the infamous Porter and Jick letter, which appeared in the New England Journal of Medicine in 1980. Here is the letter in its entirety:
Recently, we examined our current files to determine the incidence of narcotic addiction in 39,946 hospitalized medical patients who were monitored consecutively. Although there were 11,882 patients who received at least one narcotic preparation, there were only four cases of reasonably well documented addiction in patients who had no history of addiction. The addiction was considered major in only one instance. The drugs implicated were meperidine in two patients, Percodan in one, and hydromorphone in one. We conclude that despite widespread use of narcotic drugs in hospitals, the development of addiction is rare in medical patients with no history of addiction.
Supposedly opioid manufacturers and pain doctors latched onto this letter to support the expansion of access to opioids. The criticism is that relying on this letter was inappropriate. Inpatients getting clinical treatment under the supervision of a doctor is different from sending those same patients home with a bottle of opioid pills, it is claimed. The implied addiction rate of 0.03% (4/11,882) was (allegedly) too low an estimate. It's probably true that the total exposure to risk and opportunities for misuse are higher when patients are using opioids at home, out of sight of the doctors prescribing them. But I want to say two things in defense of the Jick letter. First of all, it really does demonstrate that popular notions about addiction are just wrong. A very large number of people are being exposed to opioids, and very few of them are feeling compelled to continue using. The voodoo pharamcology notion that these drugs dominate the will, that they cause addictive behavior by sinking their chemical hooks into people exposed to them, is plainly wrong. At some point presumably all of these 11,882 patients left the watchful eyes of their doctors. If pharmacology itself were so powerful, more of these people should have become addicts. At the very least, Jick's data should be a surprise to anyone who holds the voodoo pharmacology notion of addiction. (BTW, many commentators have pointed out that Jick himself has repudiated the use of his letter to justify the expansion of prescription opioids. I actually don't think it matters what Jick himself believes. If you publish some data or some piece of scholarship, you are free to qualify and hedge and add detail that's missing in the original, but you don't have a right to dictate all of the implications.)

Even setting that aside, subsequent information about addiction rates supports the claim that addiction is very rare. The 0.03% is probably low, especially if we're talking about the percent of people on very high dose opioids for long periods of time. There have been studies subsequent to the Jick letter, and Quinones makes almost no effort to scrutinize them. The only reference I could find in Dreamland to an update of the Jick letter was the following excerpt:
One 2007 survey of studies of back pain and opiates found that “use disorders” were common among patients, and “aberrant” use behavior occurred in up to 24 percent of the cases.
I don't know what meta-study he's referring to (there isn't a footnote or end note, at least in the Kindle version). But "aberrant behavior" can mean any kind of deviation whatsoever from use as directed. It could mean that the patient was called in for a pill count and the count was off by a few pills. As with most social problems (in fact, as with almost anything for which we can measure a distribution), minor problems are more common and severe problems are less common. He doesn't quantify what "common" means when claiming that use disorders were common. That's a shame, because this could have been an opportunity to quantify the actual claim and compare Purdue's marketing claims to Quinones' best estimate. It could have been an opportunity for Quinones to see a major weakness in his story, and qualify that for his readers.

Here is a meta-study on this question published in 2008, so it would have been available to Quinones before Dreamland was published. It finds that among chronic opioid patients who have no history of abuse, addiction arises in only 0.19% of cases. Aberrant behavior occurs in 11.5% of cases, but this drops to 0.59% when pre-selecting for people with no history of abuse (the same basis as the 0.19% figure). Here is another study finding that "abuse" only happens in about 0.6% of post-surgical opioid patients. Dreamland "calls out" Purdue Pharma over and over again for claiming in its marketing literature that addiction rates are "less than 1%", but Purdue was on absolutely solid ground here. There do seem to be a few studies finding much higher rates of abuse, sometimes in the double-digits. What should we do with these dueling studies?

My short answer is that those studies finding higher abuse rates are wrong and the lower abuse rates are correct. Let's look at some numbers at the population level. In 2015 (the first year, as far as I can tell, that SAMHSA asked about "any opioid use" in addition to "misuse"),  there were 97 million users of opioids (medical and non-medical, with the vast majority being medical). There were 2 million people who'd had a "past year painkiller misuse disorder", which I'll shorten by dubbing this "addiction". Naively dividing the 2 million addicts by 97 million total users yields just over 2%. A double-digit number for "percent of opioid patients becoming addicts" is already pretty unlikely. Now consider that most of those 2 million addicts don't have and in fact never had a prescription. Maybe 1/5 or 1/4 had a legitimate prescription at some point. Let's go high and say it's 1/4, and that for this 1/4 the prescription was indeed the nucleus of their addiction problem. Now we have 0.5 million addicts who started with a prescription (1/4 x 2 million); dividing by 97 million total past-year users yields 0.5% of opioid users eventually becoming addicts. This is the crudest back-of-the-envelope, and I'm using it only to be suggestive of the right ballpark. (I did this a slightly different way in this post.) Consider also the flat abuse rates in the charts above; there are just inconsistent with the higher values for addiction rates. With opioid prescriptions increasing four-fold or more in recent decades, it's just inconceivable that the addiction rate for these patients is in the double-digits.

How in the world are some studies getting double-digit answer while others are getting less than 1%? The meta-study described two paragraphs up is suggestive of what's going on here. Some studies are measuring "abuse" or "misuse", which will be much more common than full-blown addiction. (Very few drug users actually become addicts.) Also, qualifying for whether the patient has a history of abuse seems incredibly important and can dramatically change the answer. This shouldn't be too surprising, as many drug addicts probably try to convince their doctors that they suffer from chronic pain so they can get a legitimate supply. (Quinones gives a nod to this in Dreamland. There is a long section about people getting a cheap opioid prescription on their Medicaid card and then using the pills or selling them for a huge profit.) If we want to know what the odds are that a legitimate patient will be transformed against his will into an addict (whatever that might mean), then it's appropriate to remove people with a history of addiction from the analysis.  In addition to these considerations, some studies might have a too-broad definition of "abuse", perhaps counting a single aberrant pill-count as an instance of "misuse".

We have three effects that could be affecting the numbers here, some pushing in different directions. Patients are presumably trying to conceal their misuse from their doctors, which would push the observed addiction or abuse rates down compared to their true value. At the same time, addicts are trying to get legitimate prescriptions if they can, and some doctors might be overzealous about labeling someone an "addict" when in fact their behavior is consistent with someone who has under-treated pain. These latter considerations would tend to overstate the risk. We'd have to know more about the magnitudes of these effects to say for sure if these studies are over-stating or under-stating the risk.

[Edit: Quinones even mentions the number of people who have abused OxyContin in 2006, giving the number 6.1 million. This is where a time series would be handy. Is that number increasing? Descreasing? Flat? How do these changes correspond to events in his timeline? Presumably this number rises up from zero prior to 1996, so that would be a trend. But, as stated above, abuse and addiction rates were flat when the tonnage of opioids prescribed in the U.S. quadrupled. Why not disclose this to his audience? If he doesn't believe the trendlines, why give the 6.1 million figure, which presumably comes from the same source? Quinones isn't alone here. I've seen this omission elsewhere, in Vox for example.]

Death Certificates Can Be Misleading

Dreamland describes an epidemiologist for the state of Washington (Jennifer Sabel) giving a presentation to a room full of pain doctors. She is trying to convince them that opioid overdoses have increased.
She went on, detailing some of the cases from each year. Finally, Sabel turned to face her audience. The room was silent. This couldn’t be true, said one doctor, finally. There must have been some coding error, another said. Death certificates are notoriously unreliable, said a third. Others spoke skeptically of Sabel’s data. The message was clear: Several doctors in the room did not believe her. Sabel grew queasy, trying to defend the data but aware that she was new to the issue.
I think the intended take-away is that a flustered public servant is trying to point out a problem and the doctors are too dogmatic or self-interested to take it seriously. (That take would be consistent with the rest of the book.) I actually think that the unreliability of death certificates is a big deal and is massively understudied. I'm not the only one who thinks so. A standard medical textbook, Karch's Pathology of Drug Abuse, repeatedly warns its readers how hard it is to determine the cause of death. Specifically, Karch is warning the reader not to assume a death is a drug overdose just because drugs were present. He warns that posthumous testing of drugs and metabolites (a common "technique" for determining the cause of death) is fraught with error. See my posts about this here and here. I think the doctors in that room had good instincts and were right to grill Sabel. Modern vital statistics, even in an advanced industrial economy, are really questionable. It's very suspicious (to me anyway) that "drug poisoning" deaths tend to list a ton of other ailments, but it's rarely the case that non-drug-poisoning deaths mention drugs. You'd think there would be an equal smattering of "opioid overdose exacerbated by chronic lung disease" and "obesity-related sleep apnea exacerbated by an opioid habit", but in my reading the blame is always placed on the drugs. It's not a big stretch to say that anti-drug puritanism is reflected in our nations vital statistics. A cause-of-death determination is not "raw data," it's a collection of facts and evidence filtered through a person, whose judgments and biases are formed by living in our society. Summing these "causes of death" up across a population is just as fraught with error as the act of assigning a cause to any single event.

I do a back-of-the-envelope calculation at the bottom of this post. Even a very small mislabeling rate could explain a large fraction of the increase in prescription opioid related deaths. It is not at all clear that the trend in the CDC's official statistics reflect an actual rise in prescription-opioid related deaths. (I make it very clear in that post that I would only apply this argument to prescription opioids; the trends heroin and fentanyl deaths are almost certainly real.)

Failure to Distinguish Addiction from Physical Dependence

Quinones seems to have an incomplete understanding of addiction, and at times he does not clearly distinguish between addiction and physical dependence. Maybe he understands this distinction, and maybe I missed a passage or two where he hedged himself, but it just doesn't come through in his book. Addiction is the continued use of something despite negative consequences and (sometimes) despite a desire to stop. Physical dependence simply means that your body has become acclimated to a chemical. You require it to function, you have built up some kind of tolerance, and you will go through some form of withdrawal if you stop taking it.

Dreamland repeatedly refers to addicts being "dope sick", meaning they experience unpleasant withdrawal symptoms. Quinones describes how easy it is for cops to lean on desperate addicts, who are terrified of going to jail and experiencing withdrawal. He gives several anecdotes of addicts desperately trying to score to avoid "the dope sick." At one point, he describes the Xalisoco Boy's vacation-free, holiday-free work schedule. They don't even get Christmas off because "heroin addicts need their dope every day." Except they don't.

Jacob Sullum paints a fuller picture of the range of heroin habits in this classic piece. Many heroin users deliberately cycle off the drug and go through withdrawal so they can get high again (once their tolerance has come back down). Many do this out of economic necessity; getting dope every day is out of the question. I also recommend Maia Szalavitz's book Unbroken Brain. She at one point describes her own arrest and withdrawal. She describes her detention as the legal system is processing her, mentioning her craving for the dilaudid pills stashed in her apartment that she recons the cops didn't find. This is someone who has already gone through withdrawal and kicked her physical dependence and is still experiencing a craving. Addiction is a psychological phenomenon much more than it is a physiological one.

Quinones also mentions Vietnam veterans coming back "addicted" to heroin. This is an unfortunate failure of communication between opposing drug policy camps. The Vietnam experience is well known among drug reformers (reformers who favor liberalization anyway) as an example of addiction being a setting-dependent phenomenon. Very large numbers of American soldiers were using extremely pure heroin when in Vietnam, where it was readily available. No doubt, many of them continued to use when they got home, but the vast majority of them stopped. With a safe environment and a meaningful family and work life, heroin use was out of the question for most of these veterans. This is similar to the lessons of the Rat Park experiments by Bruce Alexander, where rats with an enriched environment (lots of toys and obstacle courses, other rats) used less drugs than rats with a more Spartan environment. When the opportunity cost of using heroin increases, people (and perhaps rats) use less. When you raise the cost of anything, you get less of it. 

The Complete Ineffectiveness of Drug Policing

It was jarring for me to hear Sam Quinones tell me over and over again how ineffective drug policing is at reducing the drug supply, given that his policy preference is to keep opioids under strict legal control. He at times has fawning, breathless praise for drug cops, which I found morally revolting. I assume he wants recreational heroin to remain illegal, and given the resolution he's defending in an upcoming Soho Forum debate with Jeff Singer he apparently approves of the recent crackdowns on opioids. Even his timeline (posted up top) hints at this. Operation Tar Pit happened in 2000, and then Operation Black Gold Rush happens six years later in 2006. Apparently the first attempt to uproot the Xalisco Boys didn't take, and I doubt if even Quinones would claim that the second one finished the job. In my Kindle I highlighted several excerpts admitting quite frankly that drug policing didn't work. Let me give a few examples. Emphasis mine in all cases:
As Dennis Chavez had seen in Denver, and as drug agents elsewhere would later discover, Ruplinger noticed that these Mexican drivers often had only a small amount of heroin and no weapons. So they never did much jail time. But it struck Ruplinger’s best police nerve that whenever he could determine where they were from, it turned out they were from a state in Mexico called Nayarit. What’s more, these guys were replaced within a couple of days. After a while, Ruplinger realized it didn’t matter how many Nayarit heroin drivers his task force arrested, more drivers filled the open slots.
Another excerpt from a different section:
At the Carolinas Medical Center, Bob Martin saw the change as well. Martin, a former New York cop, came to Charlotte in 1996 for a job as director of Substance Abuse Services at CMC. During his first years, when he heard of a big heroin bust, he threw up his hands. His beds were certain to fill with junkies checking in to ride out the drought. But by the early 2000s, whenever police corralled a dozen Xalisco heroin dealers, Martin saw no new flood of addicts rushing to his beds. There was no drought. The bust didn’t affect the area’s supply of heroin. “No matter how many million-dollar busts you guys do,” he told the officers, “it doesn’t show up on our radar.” 
And here is a description of a so-called "Len Bias" case, named for a young man who died of a cocaine overdose in 1986 leading to a full-blown moral panic. The idea here was to find a sympathetic-seeming dead junkie, find out who sold him the heroin, and lean on that person with the threat of a multi-decade prison sentence until he gave up his supplier. Rinse and repeat until you've found a high-level distributor. Again, I find this kind of behavior by law enforcement morally repulsive. But set aside any moral judgments, mine or yours, and just let Quinones tell you how effective this is in reducing the drug supply:
From Bickers’s office, I walked a few blocks to the office of a public defender to speak to an attorney who had agreed to talk to me as long as I left his name out of it. He had convinced many Xalisco Boys that their cooperation was the only way to avoid twenty years in prison under Len Bias. The attorney had a standing order to detectives to call him immediately when a Len Bias case began. He supported quick cooperation with investigators—a controversial idea among defense attorneys. “The value of your information is at its maximum the closer you are to the time of your arrest,” he said. “If you’re in really quick you can derive great benefit for your client.” Still, he didn’t see much effect from prosecutors’ new strategy. The pills were so widespread; new kids were getting addicted every day. They were switching to heroin all the time. Against that backdrop, he figured, prosecutors were only temporarily disrupting the market.
This next section admits clearly that shutting down the "pill mills" didn't really affect the total supply of pills. Note that the vast majority of pills were not coming from these clinics (the drop from 9.7 million to 7 million). This should have cued him in to the possibility that most of the prescriptions were legitimate and in some sense medically necessary. It would have been nice to hear him remark on that possibility, but in my reading he doesn't. Note that Quinones admits that heroin use and crime increased. He still calls closing the clinics a "necessary beginning", "an action townspeople took to determine their own future." But don't let the actual consequences of a bad policy get in the way of a good rhetorical flourish:  
Portsmouth did not avoid the new heroin scourge. Quite the contrary. Many in the town’s enormous population of opiate addicts switched to heroin. Crime went up. Detroit dealers of powder heroin began flowing down through Portsmouth. Addicts began going to Columbus for the cheap black tar that the Mexicans were selling like pizza. Before long, you could get either powder or tar heroin in the town where Dreamland once stood. The Scioto County pill mills illustrated how generalized opiate prescribing had become in America. In their last year of operation, 9.7 million pills were legally prescribed in the county of eighty thousand. But even two years after the pill mills were done, 7 million pills were still prescribed there. Nevertheless, closing the cynical clinics was a necessary beginning. Like the rescue of Mitchellace, it was an action townspeople took to determine their own future, instead of letting it happen to them. 
This is so frustrating. Passage after passage admitting that drug policy is ineffective or even counter-productive (to use a bloodless term for the human misery that's actually entailed), and not even a hint of soul-searching that the endeavor might be wrong-headed. This "...and damn the consequences" attitude of moral crusaders has saddled us with some terrible public policy, the war on drugs being the most extreme and (I would have thought) obvious example.

Here's an excerpt about drug law enforcement that doesn't necessarily contradict anything in Quinones' story, but is still revealing:
Other federal law enforcement agencies, the DEA man said, controlled their people from above to an almost stifling degree. Oversight limited the freedom an agent had in conducting an investigation. In part because of the kind of work the DEA did, individual agents had unmatched control over cases and could make them as big as their own abilities dictated. Unlike local policing, DEA agents had the chance to travel and live abroad, and the department would prize an agent like Kuykendall, fluent in Spanish. That sounded like an adventure. Not long after graduating in 1987, Jim Kuykendall applied to the agency where his father and his uncle were living legends. 
Maybe he's just passing on the words of other people, but it was hard to read this without detecting the author's approval. It seems like he's lauding the lack of oversight of DEA agents. And "That sounded like an adventure." seems like a flippant reaction to the destruction wrought by wrong-headed U.S drug policy. I think Quinones' head is in the wrong place here, and it's hard to give him credit for being "well-meaning" given that he seems to understand the futility and "unintended" consequences.

[Edit: I declined to mention in this section that the fentanyl phenomenon, discussed above, is caused by prohibition. The chart up top with the purple line showing synthetic narcotics deaths exploding? That's happening because Sam Quinones is basically getting his way in terms of drug policy. I wish we could go back to pre-2010 when recreational opioid users and even hardened addicts had relatively easy access to OxyContin (easy compared to today anyway). It's quite suspicious that heroin deaths jumped onto a new trendline in 2010 when OxyContin was reformulated to be "abuse deterrent", and I can recall seeing a "regression discontinuity" paper that establishes a connection. That year is also an inflection point for Hepatitis C, common among IV drug users. In other words, restrictions on the opioid supply have caused users to substitute to more dangerous drugs with large, unknown fluctuations in purity, and they are consuming them using less sanitary means (injection, as opposed to ingestion or snorting). The fentanyl phenomenon would never have happened if there were a legal market for recreational opioids. Overdose and blood-borne pathogens are the sequela of drug prohibition. They are not intrinsic hazards of the chemicals themselves. Chronic opioid use, unlike alcohol, cocaine, alcohol, or methamphetamine, doesn't do any kind of cumulative organ damage. It's actually safe to have someone using even heavy doses of opioids over a long period of time, as long as we can keep them from overdosing. Prohibition has made that nearly impossible.]

The Chronic Pain Patient Is Missing From the Narrative

I read Dreamland with the feeling that real chronic pain patients were getting short shrift. They are rarely mentioned, except as vectors to this disease of opioid addiction. I don't think that Quinones is giving adequate consideration to the pain patients who are being involuntarily tapered, thanks in part to sensationalist journalism such as his book. Some of these patients are committing suicide. I don't think he's given adequate thought to the notion that any kind of screening for "addicts" or "real pain patients" will be fraught with false positives and false negatives. His endorsement of the resolution for the upcoming Soho debate implies that he's okay with the recent brute-force reductions in medical opioids. I think someone needs to introduce him to some chronic pain patients who get real relief from opioids, because they are utterly missing from his narrative. 
_________________________

I couldn't fit these excerpts into the body of my post, but I wanted to share them anyway.

This part seemed like it was designed to annoy libertarians:
As I tried to chart the spread of the opiate epidemic, one thing dawned on me: Other than addicts and traffickers, most of the people I was speaking to were government workers. They were the only ones I saw fighting this scourge. We’ve seen a demonization of government and the exaltation of the free market in America over the previous thirty years. But here was a story where the battle against the free market’s worst effects was taken on mostly by anonymous public employees. These were local cops like Dennis Chavez and Jes Sandoval, prosecutors like Kathleen Bickers, federal agents like Jim Kuykendall and Rock Stone, coroners like Terry Johnson, public nurses like Lisa Roberts, scientists at the Centers for Disease Control, judges like Seth Norman, state pharmacists like Jaymie Mai, and epidemiologists like Jennifer Sabel and Ed Socie.
I'd love to see more of this "exaltation of the free market" and "demonization of government." Unfortunately, these are not exactly mainstream sentiments. Quinones says something similar in his interview with Russ Roberts, so this wasn't some one-off. It seems to reflect his true feelings about free markets and government.

Excerpt about age demographics, mentioned above in passing:
“That’s an interesting way to put it: Let’s feed this to everybody in the society and see what pops up,” he said. “Let’s, as a society, watch all of our potential alcoholics become opiate addicts instead. Had these opiates not appeared, I think we’d have seen a similar number of alcoholics, but later in life. My field used to be middle-aged alcoholics. It usually took twenty years of drinking to get people in enough trouble to need treatment. But with the potency of these drugs, the average age has dropped fifteen years and people get into trouble very quickly with oxycodone, hydrocodone, and heroin.”