Sociodemographic factors, prescription history and opioid overdose deaths: a statewide analysis using linked PDMP and mortality data

Abstract

Background:

Opioid overdose deaths have continued to rise in Tennessee (TN) with fentanyl emerging as a major contributor. Current data are needed to identify at-risk populations to guide prevention strategies. We conducted a large statewide observational study among TN adult decedents (2013–2016) to evaluate the association of sociodemographic factors and prescribing patterns with opioid overdose deaths.

Methods:

Among drug overdose decedents identified using death certificate data (n=5,483), we used logistic regression to estimate adjusted odds ratios and 95% confidence intervals for characteristics associated with prescription opioid (PO) (excluding fentanyl), fentanyl, and heroin alone overdoses. Among decedents linked to TN’s Prescription Drug Monitoring Database using deterministic algorithms, we obtained prescription history in the year before death (n=3,971), which was evaluated by type of overdose using descriptive statistics.

Results:

Younger, non-White decedents had lower odds of PO overdose, while females and benzodiazepines as a contributing cause were associated with increased odds of PO overdose. Younger age, Non-Hispanic Black race/ethnicity, greater than high school education, and cocaine/other stimulants as a contributing cause were associated with increased odds of fentanyl or heroin overdoses. Over 55% of PO, 39.2% of fentanyl, and 20.7% of heroin overdoses had an active opioid prescription at death. For PO, fentanyl, and heroin decedents, respectively, 46.0%, 30.5%, and 26.2% had an active prescription for benzodiazepines at death.

Conclusions:

Prescription opioid overdose deaths were associated with different sociodemographic profiles and prescribing history compared to fentanyl and heroin overdose deaths in TN. Data can guide prevention strategies to reduce opioid overdose mortality.

1. Introduction

Drug overdose deaths in the US have been increasing since 1999, with an age-adjusted rate of 6.1 per 100,000 in 1999 and 16.3 per 100,000 in 2015 (Hedegaard et al., 2017; Rudd et al., 2016). Increases in all drug overdose deaths have occurred across all age groups, with the highest rates among ages 45–54 years (Hedegaard et al., 2017). Opioid overdose deaths have also increased, with substantial elevations since 2013 for heroin and synthetic opioids (excluding methadone) regardless of sex, age, and race/ethnicity (Rudd et al., 2016).

In Tennessee (TN), the age-adjusted rate of overdose deaths increased from 18.2 per 100,000 in 2010 to 24.6 per 100,000 in 2016. Fentanyl, a synthetic opioid known to be manufactured illegally (illicitly manufactured fentanyl (IMF)) (Peterson et al., 2016), has emerged as a major contributor to opioid overdose deaths in TN, with 7.0% and 24.8% of opioid deaths involving fentanyl in 2013 and 2016, respectively. Similarly, heroin deaths rose from 8.4% to 21.9% of opioid deaths in 2012 and 2016, respectively. The age-adjusted rate of opioid and benzodiazepine deaths has continued to rise (3.1 per 100,000 in 2010; 8.1 per 100,000 in 2016), which is particularly concerning given the known harmful interaction of combined opioid and benzodiazepines on respiration (Garg et al., 2017; Volkow and McLellan, 2016),

Prescription Drug Monitoring Programs (PDMPs), now implemented in all 50 states, collect data on controlled substance prescriptions, prescribers, and dispensers. Some studies have shown the implementation of PDMPs has resulted in reduced opioid mortality and prescribing, although data are not consistent with comparisons hampered by differences in state-specific policies (Pardo 2017; Patrick et al., 2016; Suffoletto et al., 2018). Several decedent-only/crosssectional studies have linked PDMP and mortality data to understand characteristics of opioid deaths (Austin et al., 2017; Hall et al., 2008; Lev et al., 2015; Lev et al., 2016; Mercado et al., 2018; Paulozzi et al., 2009; Slavova et al., 2017a; Weimer et al., 2011). Further, many ecological and descriptive studies have evaluated the role of prescribing practices/behaviors and sociodemographic factors in overdose deaths (Cerda et al., 2013; Green et al., 2011; King et al., 2014). However, studies included data from earlier years, which may not reflect current trends as the drug epidemic in the US has shifted (Kolodny et al., 2015; Rudd et al., 2016), or were limited by small sample size and/or did not consider sociodemographic factors and prescribing history by contributing drugs. Data on sociodemographic factors and prescribing patterns by type of drug overdose death (e.g., prescription opioids, fentanyl, and heroin) are needed to guide development and implementation of education and intervention efforts. These include improved awareness of risk factors for prescription and illicit opioid overdoses to inform timely state policies and harm reduction strategies for clinical practice, and identification of high risk populations to target for community-based prevention programs that include overdose prevention training and take-home naloxone kits (Coffin et al., 2016; Fairbairn et al., 2017; Raffel et al., 2018; Starrels et al., 2011).

We conducted a statewide observational study among all drug overdose decedents aged ≥18 years during 2013–2016 in TN. Our specific objectives were: (1) to evaluate the associations of sociodemographic factors and multiple drug use with opioid overdose deaths and (2) to evaluate opioid and benzodiazepine prescription history by type of opioid overdose death.

2. Materials and Methods

2.1. Data Sources

2.1.1. TN Death Certificates (2013–2016).

TN’s death certificates provide data on cause of death (COD) and decedent characteristics for the TN death statistical files at the TN Department of Health (TDH). Deaths are coded according to the ICD-10 guidelines by the National Center for Health Statistics with an underlying COD (UCD) code and up to 20 additional multiple COD (MCOD) codes. Additional data collected on the death certificate includes name, address, date of birth (DOB), date of death, and sociodemographic information (e.g., sex, race, ethnicity, education, and marital status). The percent of deaths with unknown underlying COD (R99) was <1.0% during the study years.

2.1.2. TN’s Controlled Substance Monitoring Database (CSMD) (2012–2016).

In accordance with the Controlled Substance Monitoring Act of 2002, TDH established the CSMD in 2006 to monitor the dispensing of Schedule II, III, IV & V controlled substances with required reporting within seven days of dispensing effective January 1^st, 2013 based on the TN Prescription Safety Act of 2012. This was changed to required reporting within 1 business day in the TN Prescription Safety Act of 2016 (Tennesee Department of Health, 2017). Controlled substances administered directly to a patient, as part of a narcotic treatment program registered by the U.S. Drug Enforcement Administration (DEA), or samples of a schedule IV (for treatment ≤ 3 days) or schedule V controlled substances (for treatment ≤ 14 days) do not have to be reported. Selected data items collected include patient name, address, and DOB, and prescriptions characteristics (e.g., national drug code, quality, days’ supply, date filled, payment type, prescriber DEA number, dispenser DEA number). Prescription data collected follow the American Society for Automation in Pharmacy standards for PDMPs, which are used by all states.

2.2. Study Design and Study Population

We conducted an observational population-based study among TN adult residents aged ≥18 years who died of a drug overdose between 2013 and 2016 (n=5,483). Eligible decedents had drug poisoning as their underlying COD (ICD-10 codes X40–X44; X60-X64; X85; Y10-Y14) (Rudd et al., 2016). To obtain prescription history in the year before death (Hall et al., 2008; Lev et al., 2015), decedents with ≥ one prescription record and at least one days’ supply were identified in the CSMD (n=3,971). Figure 1 provides information on the study population and exclusions. The study was approved by the TDH IRB.

Figure 1.

Study Design and Population: Data Linkage Methods, Exclusions, and Sample Size.

2.3. Drug Overdose Death Definitions

ICD-10 MCOD codes T40.0 to 40.6 were used for contributing CODs related to opioids, including any opioids (T40.1-T40.4, T40.6), prescription opioids excluding fentanyl (T40.2-T40.4), heroin (T40.1), fentanyl (identified via literal text searches in the MCOD text fields for the term “FENTAN” (Trinidad et al., 2016)), methadone (T40.3) and benzodiazepines (T42.4) (Rudd et al., 2016; Trinidad et al., 2016; Warner et al., 2016). Cocaine (T40.5), stimulants (T43.6), and alcohol (T51.0, T51.9) were included as contributing causes (Hedegaard et al., 2017; Kandel et al., 2017). Supplementary Table S1 describes the ICD-10 codes¹. Drug overdose deaths may involve multiple contributing drugs and categories are not mutually exclusive unless indicated.

2.4. Sociodemographic Characteristics

Characteristics available in TN death certificate data included the following, categorized based on standard definitions and sample size considerations: age in years (18–24, 25–34, 35–44, 45–54, 55–64, 65–74, ≥75), race/ethnicity (Non-Hispanic White, Non-Hispanic Black, other Non-Hispanic, Hispanic), gender (male, female), marital status (married, divorced/separated, never married, widowed), education (<high school, high school, some college, ≥college), and place of death (TN hospital, TN jail, non-medical facility, private residence, out-of-state hospital, hospice, other residence, and nursing home) (Green et al., 2011; Siegler et al., 2014).

2.5. Data Linkage Strategy

Identifiers available for linkage in the CSMD included patient name and DOB. Data cleaning and standardization methods were applied to names in the CSMD (e.g., removing extra spaces, special characters, identifying and extracting prefixes and suffixes, and identifying names/text to be parsed into separate columns to enable linkage (e.g., “Samantha (Sam)”), described in detail here: (Golladay and Nechuta, 2018). Names in the death certificate file were similarly cleaned and standardized. A multi-step deterministic approach was used to link eligible overdose deaths to the CSMD patient file using multiple variations of first and last names (a patient may have >one record in the CSMD with a name variation), and DOB (see Figure 1).

We developed an approach to identify potential false positives (i.e., incorrect matches) by patient in the CSMD using manual review for a subset of cases with more than one name, including multiple first or last names or middle initial across patient records in the CSMD (n=2,598 records, 26.1%). The address was used to identify potential false positives among matched patient recordsets. The address was geocoded and standardized for both death and CSMD files using ArcGIS (ESRI, Redlands, CA). Among the 2,598 records, 76.3% matched on address across all available records for a patient, and of the remaining, 20.9% matched on full name via manual review, leaving 71 records with differing address information or non-match on full name (<1% of records), as potential false positives. As patients may move and address changes across a patient’s record in the CSMD are common over time, without additional identifying information (e.g., social security number), the possibility that these matches are incorrect cannot be excluded.

2.6. Prescription History in the CSMD

Prescription history in the year before death was collected for decedents linked to the CSMD. Information on class of drugs (opioids, benzodiazepines) was obtained from the CDC Conversion Reference Table, based on national drug code number (National Center for Injury Prevention and Control, 2016). Dichotomous variables were created for specific drug types in the CSMD for three-time frames (active at death, within 30 days of death, and within 90 days of death). Specific drug types included any opioid, fentanyl, hydrocodone, oxycodone, buprenorphine for medication-assisted treatment (MAT), and benzodiazepines. Active prescriptions at death were defined as having overlapping days’ supply with date of death (Slavova et al., 2017a). Total morphine milligram equivalents (MME) were calculated using strength × oral MME conversion factor × quantity. Number of prescribers and dispensers were calculated using prescriber and dispenser DEA number (Lev et al., 2016; Paulozzi et al., 2012). Payment type was collapsed into categories of cash, Medicare, and Medicaid to evaluate number of payments for an opioid prescription by these payment types.

2.7. Statistical Analysis

First, we calculated descriptive statistics for sociodemographic factors and contributing drugs among all drug overdose decedents (n=5,483, Figure 1). Based on literature review, we selected age, race/ethnicity, gender, place of death, marital status, and education level as characteristics potentially associated with drug overdose deaths (Green et al., 2011; Hall et al., 2008; King et al., 2014; Siegler et al., 2014). We estimated unadjusted and adjusted ORs and corresponding 95% CIs using binary logistic regression. We created three separate models for each of the primary outcomes: 1. deaths due to prescription drugs excluding fentanyl; 2. deaths due to fentanyl; and 3. deaths due to heroin). We also evaluated deaths due to a combination of opioids and benzodiazepines. The reference for all models was all other types of opioid deaths. Covariates, selected a priori and using statistical considerations (sample size, multicollinearity), included year of death, age, race/ethnicity, sex, and education (when not main effects of interest). Missing data for covariates (frequencies shown in Table 1) were excluded in adjusted models, results from a sensitivity analyses to include missing data via indicator variables were similar (Smith-Warner et al., 2006).

Table 1.

Sociodemographic characteristics among TN adult drug overdose decedents by contributing drugs, 2013–2016 (n=5,483)^a

	All Overdoseb (n=5,483)		Opioid Overdosec (n=3,818)		Prescription Opioid Overdose (excluding fentanyl)d (n=2,594)		Fentanyl Overdosee (n=583)		Heroin Overdosef (n=674)		Methadone Overdoseg (n=303)		Opioid and Benzodiazepines Overdoseh (n=1,656)
	n	%	n	%	n	%	n	%	n	%	n	%	n	%
Age at death
18–24	322	5.9	242	6.3	118	4.6	66	11.3	68	10.1	10	3.3	99	6.0
25–34	1,030	18.8	833	21.8	477	18.4	182	31.2	224	33.2	66	21.8	373	22.5
35–44	1,271	23.2	948	24.8	654	25.2	140	24.0	172	25.5	99	32.7	428	25.9
45–54	1,610	29.4	1063	27.8	793	30.6	119	20.4	132	19.6	76	25.1	465	28.1
55–64	944	17.2	603	15.8	461	17.8	60	10.3	71	10.5	45	14.9	250	15.1
65–74	204	3.7	106	2.8	75	2.9	15	2.6	6	0.9	7	2.3	36	2.2
≥75	102	1.9	23	0.6	16	0.6	1	0.2	1	0.2	0		5	0.3
Race/ethnicity
Non-Hispanic White	4,804	89.1	3425	91.2	2398	94.0	495	85.1	536	82.7	295	98.3	1504	91.6
Non-Hispanic Black	493	9.2	264	7.0	124	4.9	72	12.4	89	13.7	5	1.7	109	6.6
Other Non-Hispanic	24	0.5	18	0.5	8	0.3	4	0.7	6	0.9	0		6	0.4
Hispanic	68	1.3	49	1.3	21	0.8	11	1.9	17	2.6	0		23	1.4
Missing	94		62		43		1		26		3		14
Gender
Male	3,090	56.4	2195	57.5	1381	53.2	378	64.8	490	72.7	170	56.1	896	54.1
Female	2,393	43.6	1623	42.5	1213	46.8	205	35.2	184	27.3	133	43.9	760	45.9
Place of death
Hospital	2,186	40.0	1356	35.6	931	35.9	185	31.8	191	28.4	96	31.8	560	33.9
Jail	4	0.1	3	0.1	3	0.1		0		0		0	1	0.1
Non-medical facility	466	8.5	337	8.8	206	8.0	46	7.9	127	18.9	20	6.6	131	7.9
Private residence	2,360	43.1	1775	46.6	1246	48.1	290	49.9	263	39.1	166	55.0	822	49.7
Out-of-state hospital	60	1.1	33	0.9	20	0.8	NA		10	1.5	1	0.3	6	0.4
Hospice	12	0.2	6	0.2	2	0.1	0		2	0.3		0	1	0.1
Other residence	363	6.6	288	7.6	176	6.8	56	9.6	78	11.6	19	6.3	126	7.6
Nursing home	21	0.4	13	0.3	7	0.3	4	0.7	1	0.2			7	0.4
Missing	11		7		3		2		2		1		2
Marital status
Married	1476	27.7	999	27.0	748	29.8	134	23.7	103	16.1	77	26.1	462	28.8
Divorced/separated	1,878	35.3	1290	34.9	935	37.2	150	26.6	187	29.2	118	40.0	535	33.4
Never married	1,618	30.4	1205	32.6	673	26.8	261	46.2	330	51.6	77	26.1	521	32.5
Widowed	348	6.5	200	5.4	158	6.3	20	3.5	20	3.1	23	7.8	86	5.4
Missing	163		124		80		18		34		8		52
Education
<High school diploma	1,339	24.8	936	24.9	686	26.8	108	18.8	135	20.5	92	30.7	422	25.8
High school	2,516	46.6	1754	46.6	1177	45.9	264	45.9	314	47.7	143	47.7	776	47.4
Some college	1,127	20.9	810	21.5	513	20.0	160	27.8	167	25.4	49	16.3	329	20.1
≥College graduate	418	7.7	265	7.0	186	7.3	43	7.5	42	6.4	16	5.3	111	6.8
Missing	83		53		32		8		16		3		18
Cocaine i
No	4,768	87.0	3388	88.7	2395	92.3	486	83.4	501	74.3	288	95.1	1469	88.7
Yes	715	13.0	430	11.3	199	7.7	97	16.6	173	25.7	15	5.0	187	11.3
Other stimulants i
No	5,034	91.8	3538	92.7	2436	93.9	516	88.5	611	90.7	287	94.7	1536	92.8
Yes	449	8.2	280	7.3	158	6.1	67	11.5	63	9.4	16	5.3	120	7.3
Alcohol i
No	4,781	87.2	3276	85.8	2260	87.1	506	86.8	512	76.0	279	92.1	1443	87.1
Yes	702	12.8	542	14.2	334	12.9	77	13.2	162	24.0	24	7.9	213	12.9
Benzodiazepines i
No	3,664	66.8	2162	56.6	1359	52.4	347	59.5	435	64.5	169	55.8
Yes	1819	33.2	1656	43.4	1235	47.6	236	40.5	239	35.5	134	44.2	1656	100.0
Intentionality
Unintentional	4,789	87.3	3516	92.1	2351	90.6	559	95.9	656	97.3	284	93.7	1508	91.1
Suicide	452	8.2	179	4.7	153	5.9	9	1.5	6	0.9	8	2.6	86	5.2
Homicide	5	0.1	5	0.1	4	0.2		0	1	0.2	1	0.3	3	0.2
Undetermined	237	4.3	118	3.1	86	3.3	15	2.6	11	1.6	10	3.3	59	3.6

^aFrequencies and percents exclude missing.

^bICD-10 codes X40–X44; X60-X64; X85; Y10-Y14

^cICD-10 codes X40–X44; X60-X64; X85; Y10-Y14 and T40.0-T40.4, T40.6.

^dICD-10 codes X40–X44; X60-X64; X85; Y10-Y14 and T40.2-T40.4, excluding fentanyl deaths (2 fentanyl deaths were identified via literal text searches, but were not classified via ICD-10 T codes for synthetic opioids).

^eICD-10 codes X40–X44; X60-X64; X85; Y10-Y14 and literal text searches for fentanyl.

^fICD-10 codes X40–X44; X60-X64; X85; Y10-Y14 and T40.1.

^gICD-10 codes X40–X44; X60-X64; X85; Y10-Y14 and T40.3.

^hICD-10 codes X40–X44; X60-X64; X85; Y10-Y14 and T40.0-T40.4, T40.6 and T42.4.

ⁱOther contributing drug besides opioids identified with ICD-10 MCD T codes (cocaine, T40.5, other stimulants T43.6, alcohol T51.0 and T51.9, benzodiazepines T42.4).

In the second analysis, we evaluated associations between prescription history characteristics and type of opioid death (prescription opioids, fentanyl, heroin) using mutually exclusive outcome categories. Frequencies and proportions were calculated for categorical variables and medians with ranges were calculated for continuous variables. Bivariable statistical tests of association were calculated for categorical (general chi-square tests) and continuous (Kruskal-Wallis tests) variables. This analysis was limited to decedents linked to the CSMD with days’ supply ≥ 1 (n=3,971, Figure 1). All analyses were conducted in SAS (SAS Institute Inc., Cary, North Carolina). P-values <0.05 were considered statistically significant.

3. Results

A total of 5,483 TN residents aged ≥18 years died between 2013 and 2016 of a drug overdose. The majority of overdose deaths involved an opioid (n=3,818, 69.6%) and 47.3% were opioids typically obtained through a prescription. Fentanyl was involved in 10.6% (n=583) of deaths, heroin in 12.3% (n=674), buprenorphine in 3.4% (n=186), and methadone in 5.5% (n=303). Additional major drugs involved included benzodiazepines (33.2%, n=1,819), cocaine (13.0%, n=715), other stimulants (8.2%, n=449), and alcohol (12.8%, n=702).

3.1. Sociodemographic Factors and Multiple Drug Use among Overdose Decedents (Table 1)

Heroin and fentanyl decedents were younger, while prescription opioid decedents were older, with the highest number of overdose deaths in the age group of 45–54 years. Most prescription opioid and methadone deaths occurred among Non-Hispanic Whites. The highest proportion of Non-Hispanic black decedents and males was observed in heroin and fentanyl deaths. Heroin deaths had the highest proportion of cocaine and alcohol as contributing substances. The contribution of benzodiazepines was high across all types of deaths.

3.2. Polysubstance Use by Year and Type of Opioid-Related Overdose Death (Figure 1)

A high proportion of opioid deaths involved multiple contributing substances, with changing patterns in specific drugs over time by type of death. The contribution of heroin to prescription opioid deaths and fentanyl deaths increased from 1.2% and 0% in 2013 to 6.5% and 22.5% in 2016, respectively. The proportion of heroin deaths also involving fentanyl increased from 0% in 2013 to 25.4% in 2016. Benzodiazepines were involved in about 50% of prescription opioid deaths.

3.3. Factors Associated with Type of Opioid Death in Multivariable Analyses (Table 2)

Table 2.

ORs for the associations of sociodemographic factors by type of opioid overdose death among TN adult opioid overdose decedents 2013–2016 (n=3,818)^a

	Prescription Opioid Overdose (excluding fentanyl) (n=2,594)						Fentanyl Overdose (n=582)						Heroin Overdose (n=674)
	Unadjusted Modelb			Adjusted Modelc			Unadjusted Modelb			Adjusted Modelc			Unadjusted Modelb			Adjusted Modelc
	n	OR	(95% CI)	OR	(95% CI)		n	OR	(95% CI)	OR	(95% CI)		n	OR	(95% CI)	OR	(95% CI)
Age at death
<25	118	0.37	(0.28– 0.49)	0.44	(0.34–0.59)		66	2.54	(1.86–3.46)	2.18	(1.57–3.04)		68	2.19	(1.62–2.98)	1.68	(1.21–2.32)
25–34	477	0.52	(0.44–0.62)	0.57	(0.48–0.68)		181	1.88	(1.52–2.32)	1.72	(1.38–2.15)		224	2.07	(1.70–2.51)	1.83	(1.49–2.26)
35–54	1447	1.00	(reference)	1.00	(reference)		259	1.00	(reference)	1.00	(reference)		304	1.00	(reference)	1.00	(reference)
55–64	461	1.27	(1.02–1.56)	1.27	(1.02–1.58)		60	0.75	(0.56–1.01)	0.72	(0.53–0.98)		71	0.75	(0.57–0.99)	0.75	(0.56–1.01)
≥65	91	0.93	(0.63–1.38)	0.97	(0.64–1.46)		16	0.96	(0.56–1.64)	0.91	(0.52–1.58)		7	0.32	(0.15–0.70)	0.23	(0.09–0.57)
Race/ethnicity
Non-Hispanic White	2398	1.00	(reference)	1.00	(reference)		494	1.00	(reference)	1.00	(reference)		536	1.00	(reference)	1.00	(reference)
Non-Hispanic Black	124	0.38	(0.30–0.49)	0.41	(0.31–0.53)		72	2.23	(1.67–2.97)	1.94	(1.43–2.63)		89	2.74	(2.09–3.60)	2.57	(1.93–3.43)
Other	29	0.33	(0.20–0.53)	0.44	(0.26–0.74)		15	1.71	(0.96–3.06)	1.27	(0.68–2.37)		23	2.82	(1.69–4.70)	2.06	(1.18–3.62)
Gender
Male	1381	1.00	(reference)	1.00	(reference)		377	1.00	(reference)	1.00	(reference)		490	1.00	(reference)	1.00	(reference)
Female	1213	1.74	(1.51–2.01)	1.60	(1.38–1.85)		205	0.70	(0.58–0.84)	0.77	(0.64–0.94)		184	0.45	(0.37–0.53)	0.47	(0.38–0.57)
Never Married	673	1.00	(reference)	1.00	(reference)		260	1.00	(reference)	1.00	(reference)		330	1.00	(reference)	1.00	(reference)
Not Married	1093	2.18	(1.85–2.56)	1.36	(1.12–1.66)		170	0.47	(0.38–0.58)	0.77	(0.59–1.00)		207	0.43	(0.35–0.52)	0.74	(0.58–0.93)
Married	748	2.36	(1.96–2.83)	1.55	(1.26–1.90)		134	0.56	(0.45–0.71)	0.89	(0.68–1.16)		103	0.31	(0.24–0.39)	0.47	(0.36–0.61)
Education
<High school	686	1.00	(reference)	1.00	(reference)		108	1.00	(reference)	1.00	(reference)		135	1.00	(reference)	1.00	(reference)
High school	1177	0.74	(0.62–0.89)	0.78	(0.65–0.94)		264	1.36	(1.07–1.73)	1.27	(0.99–1.63)		314	1.29	(1.04–1.61)	1.25	(0.99–1.58)
>High School	699	0.68	(0.56–0.82)	0.67	(0.55–0.83)		202	1.77	(1.38–2.28)	1.76	(1.35–2.29)		209	1.43	(1.13–1.81)	1.53	(1.19–1.97)
Cocaine d
No	2395	1.00	(reference)	1.00	(reference)		485	1.00	(reference)	1.00	(reference)		501	1.00	(reference)	1.00	(reference)
Yes	199	0.36	(0.29–0.44)	0.45	(0.36–0.56)		97	1.74	(1.36–2.23)	1.34	(1.02–1.75)		173	3.88	(3.13–4.81)	2.86	(2.26–3.63)
Other stimulants d
No	2436	1.00	(reference)	1.00	(reference)		515	1.00	(reference)	1.00	(reference)		611	1.00	(reference)	1.00	(reference)
Yes	158	0.59	(0.46–0.75)	0.65	(0.50–0.84)		67	1.85	(1.38–2.47)	1.65	(1.22–2.25)		63	1.39	(1.04–1.87)	1.16	(0.85–1.59)
Alcohol d
No	2260	1.00	(reference)	1.00	(reference)		505	1.00	(reference)	1.00	(reference)		512	1.00	(reference)	1.00	(reference)
Yes	334	0.72	(0.60–0.87)	0.80	(0.66–0.98)		77	0.91	(0.70–1.18)	0.80	(0.61–1.06)		162	2.30	(1.87–2.83)	2.02	(1.61–2.52)
Benzodiazepines d
No	1359	1.00	(reference)	1.00	(reference)		347	1.00	(reference)	1.00	(reference)		435	1.00	(reference)	1.00	(reference)
Yes	1235	1.73	(1.51–2.00)	1.76	(1.51–2.04)		235	0.87	(0.72–1.04)	0.85	(0.70–1.03)		239	0.67	(0.56–0.80)	0.71	(0.59–0.86)

^aAnalysis is among TN adults who died due to an opioid overdose death (T40.1, T40.2, T40.3, T40.4, and T40.6). One fentanyl death that was not classified as an opioid death based on ICD-10 T codes is excluded. The outcome reference group for each model is other type of opioid deaths (e.g., heroin overdose deaths vs. other types of opioid deaths).

^bMissing data are excluded for race/ethnicity (n=62), marital status (n=124), and education (n=53).

^cAdjusted for year of death, age, race/ethnicity, gender, and education (when not main effects of interest). Adjusted analysis excludes 101 decedents missing data on age, race/ethnicity, or education (models for marital status exclude an additional 93 decedents).

^dOther contributing drugs besides opioids identified with ICD-10 MCD T codes (cocaine (T40.5), other stimulants (T43.6), alcohol (T51.0, T51.9), benzodiazepines T42.4). Reference group is absence of the drug

We evaluated sociodemographic factors and multiple drug use among opioid overdose decedents (study population 1 in Figure 1). In adjusted models, younger age (<25 and 25–34 years) was inversely associated with prescription opioid overdose deaths, compared to decedents aged 35–54 years, while decedents aged 55–64 years had increased odds of prescription opioid overdose deaths (ref, other opioid deaths). Female gender, having ever been married, and presence of benzodiazepines as a contributing cause were also associated with increased odds of prescription opioid overdoses. Non-White race/ethnicity, ≥high school education, and involvement of cocaine, stimulants, or alcohol at death were associated with reduced odds of prescription opioid overdoses.

In adjusted models, younger age, Non-Hispanic Black race/ethnicity, >high school education, and presence of cocaine or other stimulants were associated with increased odds of a fentanyl death (ref, other opioid deaths), while female gender was inversely associated with fentanyl. A similar risk factor profile was observed for heroin deaths (ref, other opioid deaths), with the exception of an association between alcohol and increased odds of heroin overdose deaths.

Decedents aged ≥65 years, compared to 35–54 years, had reduced odds of an opioid and benzodiazepine combined overdose death (Table S2²), while females had increased odds of death. Race/ethnicity, marital status, and other contributing causes were not significantly associated with a combined opioid and benzodiazepine overdose (Table S2²).

3.4. Opioid Prescription History by Type of Opioid Death (Table 3)

Table 3.

Opioid prescription history by type of opioid-related overdose death among individuals who died in TN of a prescription opioid overdose, fentanyl overdose, or heroin overdose and filled ≥1 prescription within 365 days before death (n=2,686)

	Prescription Opioid Overdosea (n=1,965)	Fentanyl Overdoseb (n=393)	Heroin Overdosec (n=328)	P-valued
Number of opioid prescriptions within 365 days before death, n (%)
0	126 (6.4)	43 (10.9)	54 (16.5)
1–2	306 (15.6)	94 (23.9)	83 (25.3)
≥3	1,533 (78.0)	256 (65.1)	191 (58.2)	<0.0001
Median (range)	12 (0–88)	7 (0–83)	4 (0–102)	<0.0001
Opioid prescription 90 days before death, n (%)
No	507 (25.8)	154 (39.2)	155 (47.3)
Yes	1,458 (74.2)	239 (60.8)	173 (52.7)	<0.0001
Opioid prescription 30 days before death, n (%)
No	746 (38.0)	211 (53.7)	222 (67.7)
Yes	1,219 (62.0)	182 (46.3)	106 (32.3)	<0.0001
Active opioid prescription at death, n (%)
No	882 (44.9)	239 (60.8)	260 (79.3)
Yes	1,083 (55.1)	154 (39.2)	68 (20.7)	<0.0001
Active hydrocodone prescription at death, n (%)
No	1681 (85.6)	349 (88.8)	305 (93.0)
Yes	284 (14.5)	44 (11.2)	23 (7.0)	0.0005
Active oxycodone prescription at death, n (%)
No	1300 (66.2)	322 (81.9)	301 (91.8)
Yes	665 (33.8)	71 (18.1)	27 (8.2)	<0.0001
Active buprenorphine for MAT prescription at death, n (%)
No	1,922 (97.8)	375 (95.4)	319 (97.3)
Yes	43 (2.2)	18 (4.6)	9 (2.7)	0.0246
Active fentanyl prescription at death, n (%)
No	1,955 (99.5)	335 (85.2)	0
Yes	10 (0.5)	58 (14.8)		NA
Total MME for opioid analgesics in 90 days before death e , n (%)
≤2395.0	264 (18.6)	74 (32.7)	92 (57.5)
2395.0-<12787.5	356 (25.1)	46 (20.4)	31 (19.4)
12787.5-<39352.5	400 (28.2)	33 (14.6)	25 (15.6)
≥39352.5	399 (28.1)	73 (32.3)	12 (7.5)	<0.0001
Total MME for opioid analgesics in 30 days before death e , n (%)
≤4992	237 (19.9)	58 (33.5)	59 (60.2)
4992-<16350	297 (25.0)	26 (15.0)	19 (19.4)
16350-<46800	331 (27.8)	33 (19.1)	12 (12.2)
≥46800	325 (27.3)	56 (32.4)	8 (8.2)	<0.0001
Number of prescribers for opioids f , n (%)
1	424 (23.1)	118 (33.7)	78 (28.5)
2	360 (19.6)	65 (18.6)	66 (24.1)
3–4	527 (28.7)	76 (21.7)	70 (25.6)
≥5	528 (28.7)	91 (26.0)	60 (21.9)	0.0001
Median (range)	3 (1–34)	2 (1–21)	2 (1–15)	0.0002
Number of dispensers for opioids f , n (%)
1	688 (37.4)	143 (40.9)	106 (38.7)
2	476 (25.9)	86 (24.6)	75 (27.4)
3–4	471 (25.6)	78 (22.3)	60 (21.9)
≥5	204 (11.1)	43 (12.3)	33 (12.0)	0.6465
Median (range)	2 (1–13)	2 (1–12)	2 (1–10)	0.6098
Cash payment for an opioid prescription f , n (%)
0		218 (62.3)	147 (53.7)
1–2		78 (22.3)	82 (29.9)
3–4	129 (7.1)	18 (5.1)	22 (8.0)
≥5	272 (15.0)	36 (10.3)	23 (8.4)	0.0011
Medicare used for an opioid prescriptions f , n (%)
0	1,320 (72.6)	288 (82.3)	249 (90.9)
1–2	100 (5.5)	11 (3.1)	4 (1.5)
≥3	398 (21.9)	51 (14.6)	21 (7.7)	<0.0001
Medicaid used for an opioid prescriptions f , n (%)
0	1,411 (77.6)	265 (75.7)	212 (77.4)
1–2	119 (6.6)	38 (10.9)	21 (7.7)
≥3	288 (15.8)	47 (13.4)	41 (15.0)	0.0655

^aDefined as prescription opioid overdoses excluding fentanyl and heroin overdoses.

^bAll fentanyl overdose deaths based on literal text searches.

^cDefined as heroin overdoses excluding fentanyl.

^dChi-square test for categorical variables and Kruskal-Wallis test for continuous variables.

^eLimited to decedents with analgesics opioid use in the last 30 days (n=1465) or 90 days (n=1811), cutpoints based on quartiles in each group.

^fExcludes deaths with no opioid prescription (n=223); payment variables additional exclude missing (n=21).

Overall, prescription opioid use was highest among prescription opioid overdoses, followed by fentanyl and heroin deaths. Within each overdose category, the proportion of decedents without a prescription was higher the closer to date of death the measurement. For example, 60.8% of fentanyl decedents had filled a prescription 90-days before death, 46.3% 30 days before death, and 39.2% had an active prescription at death. Oxycodone was the most frequently filled prescription opioid, in particular among prescription overdose and fentanyl deaths. Only 14.8% of fentanyl decedents had an active prescription for fentanyl at death.

Most decedents filled opioid prescriptions from more than one prescriber in the year before death, and close to 60% of prescription opioid decedents filled prescriptions from more than two prescribers. Across type of death, about 60% of decedents received an opioid from more than one pharmacy. Use of cash payment for at least one opioid prescription was most common among heroin deaths (46.3%), followed by prescription opioid (43.9%) and fentanyl (37.7%) deaths. Use of Medicare for more than one opioid prescription was highest among prescription opioid deaths; use of Medicaid did not vary by type of death.

3.5. Benzodiazepine Prescription History by Type of Opioid Death (Table 3)

Prescription opioid decedents had the highest benzodiazepine use in the 90 days before death (55.7%), followed by fentanyl (40.2%) and heroin (35.7%). About 46% of prescription opioid overdoses had an active benzodiazepine prescription, compared to ~1/3 of fentanyl and heroin overdoses. About half of decedents filled their prescriptions for benzodiazepines from only one provider or one dispenser, with limited variation by type of death.

4. Discussion

Data on sociodemographic and prescribing patterns by contributing substances for overdose deaths are needed to guide targeted and appropriate prevention strategies to reduce opioid mortality. In this large population-based analysis of TN drug overdose decedents, sociodemographic profiles, non-opioid contributing substances, and opioid and benzodiazepine prescribing patterns differed by type of opioid overdose death (i.e., prescription opioid, fentanyl, and heroin deaths). For prescription opioid overdose deaths (compared to other opioid overdose deaths), younger age (<25 and 25–34 years), Non-Hispanic black race/ethnicity, and higher education were associated with reduced odds of overdose, while decedents who were female, married or previously married, or had benzodiazepines as a contributing cause were at increased odds of prescription drug overdose. For fentanyl overdose deaths (compared to other opioid overdose deaths), female gender was associated with reduced odds of overdose, while younger ages (< 25 and 25–34 years), Non-Hispanic Black race/ethnicity, higher education, and cocaine or stimulants as a contributing cause were associated with increased odds of overdose. A similar risk factor pattern was observed among heroin deaths, with the exception of increased odds of overdose death for alcohol use.

Two previous studies reported associations between female gender and presence of benzodiazepines at death with prescription overdose deaths, including a study of 2,231 decedents in Connecticut (1997–2007) (Green et al., 2011) and 2,888 decedents in New York (1990–2006) (Cerda et al., 2013). The authors of the Connecticut study also evaluated characteristics of heroin deaths and reported associations for younger age groups and male gender, similar to our findings. In contrast to our findings, heroin deaths were more likely to occur among white decedents and not involve multiple non-opioid substances. The first study of linked PDMP and mortality data, conducted in West Virginia in 2006 among 295 unintentional prescription overdose deaths, reported decedents who were never married/divorced, of lower education or income, or male had higher death rates, and few decedents had filled a prescription in the year before death (Hall et al., 2008). Mercado et al. evaluated factors associated with illicit fentanyl deaths compared to all overdose deaths in a study of 536 decedents in Rhode Island (2013–2014), and similar to our study, younger age was associated with illicit fentanyl overdose. However other sociodemographic factors were not associated with deaths (Mercado et al., 2017). A study using data from Kentucky found that the highest rates of fentanyl and heroin overdoses were among 25–34-year old’s (Slavova et al., 2017b).

We found that regardless of the type of opioid-related death, a large number involved polysubstance, as observed previously (Hall et al., 2008; King et al., 2014; Mercado et al., 2017; Warner et al., 2016). From 2013–2016, fentanyl involvement increased from 0% to 25.4% among heroin deaths, and cocaine increased 9.4% to 18.8% among fentanyl deaths. Benzodiazepines were involved in nearly half of prescription opioid and ~30–40% of fentanyl and heroin deaths across years, similar to findings from a study of national trends in polysubstance use among overdose deaths (Kandel et al., 2017). This study also reported that heroin increased (1.2% to 24.5%) among deaths due to synthetic opioids other than methadone (a category that includes fentanyl overdoses).

In our study, opioid prescription use was highest for prescription opioid overdose deaths, followed by fentanyl and then heroin only overdose deaths, regardless of opioid use timing or type. Few fentanyl decedents had an active prescription at death (<15%), very similar to the proportion reported in a study of overdose deaths during 2013–2014 in Kentucky (Slavova et al., 2017a), and suggests that the majority of fentanyl deaths in TN are due to illicit use. Potentially risky prescription-related behaviors, including visiting multiple doctors and pharmacies and cash payments for prescriptions, were highly prevalent in the year before death. Of particular concern was the high amount of benzodiazepine use regardless of the type of death, known to have high-risk when co-prescribed with opioids (Dasgupta et al., 2016; Dilokthornsakul et al., 2016; Jones and McAninch, 2015; Park et al., 2015; Volkow and McLellan, 2016).

Our findings support the need to consider multiple susceptible populations and targeted education and intervention efforts to reduce prescription opioid, heroin, and fentanyl overdose deaths. Community-based programs that provide naloxone take-home kits and overdose prevention training are key public health interventions that can be modified and directed to high-risk populations (Straus et al., 2013). To be successful, these strategies should consider younger age groups, males, Non-Hispanic Blacks, and individuals with alcohol and other illicit drug use history, addressing barriers to receiving and using educational and intervention programs. Targeted programs can benefit from the fact that high risk individuals may be more likely to utilize the programs and should consider both community and clinical-based naloxone distribution to maximize effectiveness (Barocas et al., 2015; Coffin et al., 2016; Raffel et al., 2018; Siegler et al., 2014). Our data show that over half of fentanyl and heroin overdoses had a filled opioid prescription in the 90-days before death, suggesting a missed opportunity for prevention. Continued efforts to educate healthcare professionals on evidence-based practices and policies including the Tennessee Chronic Pain Guidelines and resources for implementation, increased availability of medication-assisted treatment as part of an individually-tailored comprehensive treatment program based on best practices, and increased awareness of risky behaviors and adverse outcomes is needed (Cochran et al., 2016; Tennessee Department of Health, 2017; Kolodny et al., 2015; Kresina and Lubran, 2011; Starrels et al., 2011; Volkow and McLellan, 2016). The role of benzodiazepines in the opioid public health crisis needs to be integrated into community and health-care based educational and intervention efforts, and for multiple susceptible populations including illicit users, prescription opioid users, and individuals with opioid use disorder.

4.1. Limitations

The CSMD, similar to other PDMPs (Hartung et al., 2017; Mercado et al., 2017), has limited identifiers available for data linkage. We evaluated the potential for false positives based on our deterministic approach in a subset of patients with multiple names across records and found a very low number of potential false positives (<1% of patient records), consistent with the assumption that matching on full name and DOB accurately identifies an individual if identifiers are reasonably accurate (McCoy et al., 2013). While our approach resulted in few potentially incorrect matches, we may have missed both decedents and prescription records, which would lead to underestimation of the role of prescription history in overdose deaths in the present analysis. Another limitation is the use of DEA number to identify prescribers and dispensers, which may result in misclassification in these variables.

Additional limitations inherent in using PDMP data include: 1) lack of information on drug indication, 2) incomplete opioid prescription history as not all controlled substances are captured by the CSMD (e.g., out of state prescriptions, drugs dispensed directly to the patient, some short-term supply samples, opioids for MAT in federal treatment centers, and drugs obtained illegally), and 4) lack of medical history, including psychiatric or substance abuse history, known predictors of overdose death (Brady et al., 2017; King et al., 2014). The time frame of our study (one year before death) may miss the full trajectory from the first prescription to overdose, for example, as many heroin users may have started with prescription opioids (Muhuri PK et al., 2013).

Using death certificate data to classify overdose deaths has several known limitations including lack of specific drug types, toxicology results, and reporting variations across time and jurisdictions (Moyer et al., 1989; Slavova et al., 2015; Stone et al., 2017; Warner et al., 2016). In particular, heroin deaths may be underestimated as the death certificate may list morphine without providing the source (e.g., metabolized heroin or prescription morphine), which can be difficult to determine without the presence of 6-acetylmophine, a metabolite of heroin (Ellis et al., 2016; Mertz et al., 2014). Further, it is not possible to distinguish fentanyl from fentanyl analogs (e.g., carfentanil) using only death certificate data. Combining toxicology and death certificate data could help address these limitations in the future (Slavova et al., 2017a). It is encouraging that reporting on specific drugs involved in overdose deaths is improving in TN (Rudd et al., 2016). Finally, we did not exclude intentional overdoses or overdoses of undermined intent as misclassification is a known concern (Rockett et al., 2015; Stone et al., 2017).

4.2. Conclusions

Our findings provide a comprehensive description of opioid and benzodiazepine prescribing history in the year before death. These data help to identify the most susceptible populations at risk for opioid mortality. Results regarding benzodiazepine involvement in opioid-related deaths, regardless of type, as well as the high proportions of patients on opioid analgesics receiving prescriptions for benzodiazepines highlight that more work is needed to address risky prescribing practices contributing to opioid deaths. TN has invested substantial effort in legislation, law enforcement, surveillance, population-based analytics, professional education, and community outreach to reduce opioid-related morbidity and mortality. Our findings can support targeted education and intervention efforts, building on existing infrastructure, and help improve utilization of TN’s PDMP.

Figure 2.

Major Contributing Causes of Death by Type of Opioid Overdose Death in TN, 2013-2016.

Table 4.

Benzodiazepine prescription history by type of opioid-related overdose death among individuals who died in TN of a prescription opioid overdose, fentanyl overdose, or heroin overdose and filled ≥1 prescription within 365 days before death (n=2,686)

	Prescription Opioid Overdosea (n=1,965)	Fentanyl Overdoseb (n=393)	Heroin Overdosec (n=328)	P-valued
Number of benzodiazepine prescriptions, n (%)
0	662 (33.7)	197 (50.1)	168 (51.2)
1–2	176 (9.0)	36 (9.2)	43 (13.1)
≥3	1,127 (57.4)	160 (40.7)	117 (35.7)	<0.0001
Benzodiazepine prescription 90 days before death, n (%)
No	870 (44.3)	235 (59.8)	211 (64.3)
Yes	1,095 (55.7)	158 (40.2)	117 (35.7)	<0.0001
Benzodiazepine prescription 30 days before death, n (%)
No	1,026 (52.2)	270 (68.7)	238 (72.6)
Yes	939 (47.8)	123 (31.3)	90 (27.4)	<0.0001
Prescription for both opioid and benzodiazepines 30 days before death, n (%)
No	1,225 (62.3)	305 (77.6)	283 (86.3)
Yes	740 (37.7)	88 (22.4)	45 (13.7)	<0.0001
Active benzodiazepine prescription at death, n (%)
No	1,062 (54.1)	273 (69.5)	242 (73.8)
Yes	903 (46.0)	120 (30.5)	86 (26.2)	<0.0001
Number of prescribers for benzodiazepinese, n (%)
1	608 (46.7)	98 (50.0)	87 (54.4)
2	371 (28.5)	51 (26.0)	37 (23.1)
3–4	265 (20.3)	34 (17.4)	29 (18.1)
≥5	59 (4.5)	13 (6.6)	7 (4.4)	0.3858
Median (range)	2 (1–13)	1.5 (1–8)	1 (1–6)	0.2834
Number of dispensers for benzodiazepinese, n (%)
1	654 (50.2)	110 (56.1)	84 (52.5)
2	362 (27.8)	51 (26.0)	42 (26.3)
3–4	242 (18.6)	26 (13.3)	22 (13.8)
≥5	45 (3.5)	9 (4.6)	12 (7.5)	0.0678
Median (range)	1 (1–10)	1 (1–7)	1 (1–9)	0.3259

^aDefined as prescription opioid overdoses excluding fentanyl.

^bAll fentanyl overdose deaths based on literal text searches.

^cDefined as heroin overdoses excluding fentanyl.

^dChi-square test for categorical variables and Kruskal-Wallis Test for continuous variables.

^eExcludes deaths with no benzodiazepine prescription (n=1,503).

Highlights.

• Large study of prescription history before overdose death by contributing drugs.

• Sociodemographic risk factor profiles differed by type of opioid death.

• >50% of heroin and fentanyl deaths had an opioid prescription 90 days before death.

• >1/3 of opioid decedents had an active prescription for a benzodiazepine at death.

• Risky prescription-related behaviors were common in the year before death.

Abbreviations:

COD

Cause of Death

CSMD

Controlled Substance Monitoring Database

DOB

Date of birth

DEA

Drug enforcement administration

IMF

Illicitly manufactured fentanyl

MAT

Medication-assisted treatment

MME

Morphine milligram equivalents

MCOD

Multiple causes of death

NCHS

National Center for Health Statistics

PDMPs

Prescription Drug Monitoring Programs

TN

Tennessee

TDH

Tennessee Department of Health

UCD

Underlying cause of death