A recurrent topic on this blog is drug overdoses due to street drugs and the epidemiology of these overdoses. As an addiction psychiatrist it is obvious that access to drugs and alcohol is a critical feature of epidemics and the broad exposure has resulted in increased accidental overdoses. The clearest example I can think of is the spread of heroin overdoses to rural America. That phenomenon did not exist 20 years ago. The implicit aspect of the access argument is that other commonly held reasons for addiction and overdose deaths like poverty, race, culture, etc are really not reasons that people get addicted to drugs and alcohol. That happens because they have a biological predisposition and they have access. There is a lot of resistance to this basic idea because it runs counter to the idea of broad legalization of cannabis and other drugs like psychedelics. It runs counter to the idea that the war on drugs is the real problem here and the situation would improve without it. It also runs counter to the alternate theories about substance use that some people see are identifying remediable problems like poverty and disparity. I see those theories as being equally stigmatizing and inaccurate.
For all of these reason I was very interested in the recent paper in Science (1) that looks at accidental drug overdoses as a chronic problem rather than a discrete series of events. The authors analyzed a total of 599,255 deaths from 1979 through 2016 from the National Vital Statistics System, specifically the Mortality Multiple Cause Micro-data Files. Their main finding is illustrated in Figure 1 above and that is the aggregate overdose mortality rate increases exponentially for a period of 38 years (panel B). The individual drugs are broken out in panel A. Only accidental deaths due to drugs were included and that determination adds some element of uncertainty to the numbers.
This is probably the best place to comment on the methodology of this research and further visualizations of the data. Inspection of the individual drug shows that there are a total of 7 drugs. Common drugs of concern in overdose situations like benzodiazepines and z-drugs are not mentioned specifically but there is a category of unspecified drugs and unspecified options. The supplementary material lists accidental poisoning due to antiepileptic, sedative-hypnotic, antiparkinsonism and psychotropic drugs - not elsewhere classified, hallucinogens - not elsewhere classified, other drugs acting on the autonomic nervous system, and unspecified drugs, medicaments and biological substances. By that description benzodiazepines, z-drugs, and other neurological and psychiatric medications are probably in that category. Poisoning by solvents or intentional use was not included. Intentional drug poisoning - both suicide and homicide was not included. The coding of the cause of death changed substantially in 1999 providing more detail and allowing for the separation of synthetic and semi-synthetic opioids after that time. Mortality rates were calculated by drug type, age, sex, and urbanicity.
A few points from the initial graphs. Mortality curves have generally increased for all of the drugs of interest since 2010 except methadone and the unspecific drugs. By the authors definition the range of the unspecified drugs is so broad that it may conceal clear trends within subcategories like benzodiazepines and z-drugs - both important in polydrug overdoses. The authors suggest that the variability in some of these lines like the dip in the prescription opioid line after 2010 may have been related to attempts to reduce the number of prescriptions, laws making mandatory checking for prescriptions for controlled substances prior to prescribing, and the production of a long acting and less abusable form of oxycodone in 2010. It seems as likely that that a lot of prescription opioid users switched to heroin at that point reflected by the rapid increase in heroin death rates from 2010 to the end of the study period. There is a time lag to 2013 and at that point the fentanyl death rates begin an even steeper curve to the end of the study period. There is no associated decrement in the heroin or prescription opioid curve at that point suggesting that the fentanyl rates reflects a different problem from the baseline opioid death rates. Those problems could include the fact that fentanyl is much more toxic, it could be an adulterant, it could be sold as heroin by diluting it, and it could be sought out as a way to pursue a high when tolerance has developed to the original opioid being used. The authors points out that increased access to fentanyl was documented by increased seizures by law enforcement. The decrease in methadone deaths may have been due to its removal from pain formularies and CDC initiative on discouraging use for pain medication due to excessive toxicity.
In panel B, the authors shade the area up to 1998 to show when the opioid epidemic begins. They make the point "of particular interest is the observation that the first half of this long smooth exponential growth curve predates the opioid epidemic." While that may be true, it is obvious graphically that the 1979-1998 curve could also be extended without the inflection point in 1998 without the superimposed opioid epidemic and significantly higher mortality rates.
The authors also examined what they referred to as drug specific subepidemics, building on their assumptions that the exponential mortality curve that they describe is due to subepidemics. Their methods of analysis included heat mapping and geospatial hotpot analysis. I elected to license the latter image and include it below.
For the above geospatial hotspot analysis. the authors looked at 8 drug categories, during 4 times frames and all of the time frames are form the opioid epidemic. The only drug that showed a peak intensity and spatial distribution followed by a decline was methadone. All of the prescription and nonprescription opioids show a progressive increase during this time frame. That patterns are also remarkable for a spread from metro areas to non-metro areas. There are some interesting geographic observations including a relatively cold spot of overdose mortality in the north central states.
Heatmap analysis (not shown) showed a bimodal distribution of mortality in 20-40 yr olds and 40 to 60 yr olds. Heroin, synthetics, white race, male gender, and urban counties were over represented in the younger group. The older group deaths were predominately white women in rural counties using prescription opioids. Prescription drug mortality rates four times higher in younger men than women were attributable largely to synthetic opioids.
The authors main point in the paper is that the combination of epidemics came together to compose the exponential curve after the inflection point and that there is not clear way to figure out how that happened. They emphasize the importance of understanding these forces and cite a number of possibilities including supply side components of more efficient manufacturing of drugs, better supply chains, high purities and lower prices. These are all well known factors in why people stop using prescription opioids and fentanyl from illegal sources. They discuss sociological factors like fragmentation of communities, despair and a lack of purpose. They discuss public health interventions like community surveillance for patterns of drug use and availability addiction treatment for secondary and tertiary prevention.
As I read the article, I thought about my model of addiction and that is biological vulnerability + access leads to addiction. Certainly there are people who will say that the sociological concerns described by these authors can lead to the biological vulnerability, but the protective factors in some of those communities are generally ignored. The authors seem to have at least some of the data here to show that socioeconomic status and race do not determine drug epidemics. Availability determines drug epidemics and there is no better example than some of the data they present here. In addiction to race, urbanization is also an example of vulnerability + access with the spread of overdose mortality out into the country side. There is an addition dimension of data here that could be used to look at these mortality rates and that is US Census data provided tract-level measures of poverty, education, crowding, and race/ethnicity. In other words how does the mortality correlate with these factors. A recent study of alcohol retail density showed a high correlation with these factors. In other words, urban minority populations face a higher level of retail alcohol outlet density and exposure than white populations in urban or rural zip codes. The only difference from the study in reference 2 is that there are no clear measures of opioid exposure.
This is an important study with a unique approach to the problem of progressive drug epidemics. Mortality rate from overdoses is not the same as measuring the total drug exposure and resulting addiction but there is no clear way to determine that. I would also not consider the heroin and fentanyl mortality rates to be independent of the original increase due to prescription opioids. My rationale is that there are very few people that start using either compound. Once an addiction to opioids starts there are progressively larger number of people each year competing for the pool of illicit opioids. They are looking for less expensive alternatives. May are not risk averse and are consciously looking for more potent opioids. Although there is no data to support the progression from prescription opioids to fentanyl and heroin - clinical experience suggests it is the likely explanation and it could account for the mortality curve in Figure 1 as a single rather than multiple epidemics.
These authors have come up with a unique contribution to the literature and I encourage anyone interested in the epidemiology of drug epidemics to read the full text of this paper.
George Dawson, MD, DFAPA
1: Jalal H, Buchanich JM, Roberts MS, Balmert LC, Zhang K, Burke DS. Changing dynamics of the drug overdose epidemic in the United States from 1979 through 2016. Science. 2018 Sep 21;361(6408). pii: eaau1184. doi: 10.1126/science.aau1184. PubMed PMID: 30237320.
2: Berke EM, Tanski SE, Demidenko E, Alford-Teaster J, Shi X, Sargent JD. Alcohol retail density and demographic predictors of health disparities: a geographic analysis. Am J Public Health. 2010 Oct;100(10):1967-71. doi: 10.2105/AJPH.2009.170464. Epub 2010 Aug 19. PubMed PMID: 20724696.
Both Figure 1 and Figure 2 above are used with permission from the American Academy for the Advancement of Science (AAAS) the copyright holders. The figures are both from reference 1 above used with permission per the following license number 4441420359702 obtained on October 3, 2018. The AAAS also requires the following notice based on this content:
"Readers may view, browse, and/or download material for temporary copying purposes only, provided these uses are for noncommercial personal purposes. Except as provided by law, this material may not be further reproduced, distributed, transmitted, modified, adapted, performed, displayed, published, or sold in whole or in part, without prior written permission from the publisher."