25 July 2019 — 0734 mdt

Preliminary notes on the Washington Post’s
dump of DEA opioid data for Montana

Update, 25 July. The post below was generated by a project that got away from me. The post may be closer to a rough draft than a polished article, but it’s time to share what I have and get on with other things. I’ll return to the issue of opioids in the future. My thanks to people who know who they are for their input.

♦ ♦ ♦ ♦ ♦

24 July. Last week the Washington Post published a nationwide county-by-county breakdown of manufacturing and sales data, for two prescription opioids, for 2006–2012, that were obtained from the Drug Enforcement Agency, which tracks every transaction involving legal narcotics. The WP developed a metric — pills per person per year averaged over the seven-year recordd — that provides a rough, very rough, measure of where the distribution of the drugs was concentrated, and presented it graphically in an interactive national map, a tour de force of reporting.

The WP has made the data available to everyone, a tremendous public service that improves the public’s understanding of the issue, and not incidentally turns the data loose in the wild, ensuring there’s no practical way it can be recalled by the DEA, a federal court, or even the Almighty’s edict.

I’ve been reviewing a 749,512-record data set for Montana. How useful the metric and the data may be in understanding why deaths from opioid overdoses have skyrocketed in some states and counties and declined in others remains to be seen. The PPY rates differ widely from county-to-county, and for various reasons. In eastern Montana, some of the sparsely populated counties PPYs are well under ten, while the state’s largest counties have PPYs an order of magnitude higher. Lincoln County, at 61, seems to be the highest. I calculated annual PPYs for Montana and Flathead County, and a seven-year average PPY for Montana, and added them to the following table of WP PPY values:

Gallatin County’s PPY of 21 is considerably lower than the PPYs for Montana’s other large counties. I don’t know why.

Mushy math. In at least two ways, the WP’s seven-year PPY statistic is a mush metric. Alone, it suppresses trends over time (in at least one iteration of the interactive map, the PPY value was accompanied by a column graph displaying annual values). The number of pills is only a rough approximation of the painkilling power, usually expressed as morphine milligram equivalents (MME), involved.

First causes. The data do not answer the chicken and egg question: did the supply of prescription opioids drive the demand for 2006–2012, or did the demand drive the supply? And if demand drove the supply, what was it about high prescription opioid death rate states and counties that differed from states and counties with low death rates? (I don’t think Big Pharma is guilty of setting up pill mills to create a market for opioids, but it sure seems guilty as hell of filling outlandish orders with no questions asked and eyes averted from red flags.)

Montana’s place in the opioid universe

Compared to the rest of the nation, Montana has comparatively low rates of death from drugs and opioids. Deaths from opioid overdoses are declining, while deaths from non-opioid drugs such as methamphetamine are increasing:

     Double size      PDF for printing      Download data

Exercise caution when comparing the end points of the record. Annual values and trends over time are much more useful than a ratio of bookends. Proportionally, Montana’s opioid death rate for 2017 is notably higher than for 2001, but the number of opioid deaths for the beginning of the record, when reporting was not as efficient as it is today, may have been underreported. Put another way, 2001–2002 may not be a useful baseline.

Both legal and illegal opioids are included in the death rates. The dot chart below depicts the annual distribution of death rates for the 50 states and D.C., and displays how the death rate has increased in some states for 1999–2017. Not all states, incidentally, have done a good job of reporting deaths from opioid overdoses. For example, there are no data for many years for North Dakota.

     Double size      PDF for printing      Download data

The data in the dot plot above also can be presented as a notched box and whiskers plot that quantifies the annual distributions of state opioid overdose death rates. To avoid clutter, I did not overlay this plot with the values for Montana.

     Double size      PDF for printing      Download data

Maps provide yet another way of displaying the death rate data. Below, the Kaiser Family Foundation’s map for 2017 shows that the largest increases in deaths from opioid overdoses are found in rust belt, Appalachian, and northeastern and mid-Atlantic coastal states. KFF has maps for 1999–2017.

These plots and map make a very strong case that the conventional wisdom notwithstanding, Montana is not an opioid epidemic state. But the level of belief that Montana does have an epidemic or crisis convinces me that Montana probably is experiencing an opioid panic. That’s deeply concerning, for panics invariably produce bad policy.

Denied one drug, abusers switch to another

Bootleg fentanyl, not legally obtained prescription opioids, is the leading cause of opioid overdose deaths today. According to the Centers for Disease Control and Prevention:

The CDC reports (Table 1) that in 2017, prescription opioids caused 17,029 of 47,600 deaths from opioid overdoses. In 2016, the respective numbers were 17,087 and 42,249.

Preliminary observations, somewhat random

The WP’s cleaned data set for Montana probably can be opened in Excel, but I haven’t tried doing that. I've been using Filemaker Pro, a relational database. Because the dumped data are formatted as tab separated plain text (tsv extension), the files can be opened in a powerful text editor, such as BBEdit, which might suffice for a cursory search.

• The WP’s dump of DEA data is just the tip of the government's huge iceberg of data on opioids and drugs. Not all the data are available to the public, sometimes for valid reasons, sometimes to protect agency agendas and budgets, sometimes to help Big Pharma escape independent oversight.


• Some of the transactions involved veterinary clinics. Running the search term “DVM” produced 1,147 records and total of 185,500 pills. The respective percentages of the database are ≈ 0.08 and 0.2.


• Oxycontin, the Hillbilly Heroin that some believe triggered the uptrend in overdose deaths, accounted for approximately four percent of the pills and just under ten percent of the transactions in Montana.


• Some of the field labels are cryptic. “DOSAGE UNIT” means the number of pills. Some fields violate best practices by containing more than one data item. “Product Name” for example sometimes contains the brand name of the medication, such as Oxycontin, and the amount of the drug (example: 80mg). As a general rule, a field should contain a single value, and the unit should be specified in the field’s label.


• The “STRENGTH” field, where I expected to find the amount of the painkiller in each pill, contains either “null,” “0000,” or “1000,” and does not specify units.


• The field “TRANSACTION DATE” displays the date in the old mainframe style: an eight-digit number, such as 04012009 for April 1, 2009. The eight-digit number can be parsed easily, of course — I converted it to three fields, day, month, and year — but it’s not friendly for most people. I don’t know whether the eight-digit number is the DEA’s or the WP’s.


• In isolation, the WP’s PPY metric is interesting, but not definitive. Other metrics, such as prescribing rates, also must be considered when analyzing opioid overdoses.


• Unrealistic expectations. Politicians are fond of saying “If the death rate is above zero, there’s room for improvement.” This statement — technically true — fails to acknowledge how little room for improvement practicalities impose; fails to acknowledge that effort becomes disproportionate to effect as zero is approached. An opioid overdose death rate of zero is perfection, an aspirational objective that's not attainable in the real world. There always will be suicides by medications. There always will be fatal self-dosage errors. [Rewritten on 25 July.]


• I hope the WP will make the data set more user friendly by providing a metadata summary, by deleting some of the fields that contain no useful data, and by reformatting the transaction date field. When I finish reworking my data set for Montana, I’ll make it available for download.