Opioid deaths involving concurrent benzodiazepine use: assessing risk factors through the analysis of prescription drug monitoring data and postmortem toxicology

Abstract

Background:

A high proportion of opioid drug deaths involve concurrent benzodiazepine use. To reduce the risk of drug overdose, various prescription drug monitoring programs have been implemented. This study examined the impact of concurrent benzodiazepine use on opioid-related deaths, and the utility of the Michigan Automated Prescription System (MAPS) in predicting risk of opioid death.

Methods:

Wayne County Medical Examiner’s Office cases from 2018 were examined in terms of MAPS data and MAPS-derived drug risk scores, as well as postmortem toxicology. Opioid death cases with concurrent benzodiazepine use were compared to non-drug deaths.

Results:

For cases with a MAPS history for 6 months preceding death, the incidence of opioid prescriptions filled did not differ between groups. In contrast, significantly more opioid death cases had filled a benzodiazepine prescription; alprazolam prescription was the single best predictor of opioid drug death. Groups differed in MAPS-calculated drug risk scores, though these were less predictive of opioid death than some individual measures of prescription drug use. In terms of postmortem toxicology, fentanyl was the best discriminator between cohorts, with significant associations seen for morphine, benzodiazepine, or cocaine use. Similar results were obtained in the subset of subjects filling a prescription within a month of death, except that MAPS risk scores no longer predicted drug deaths.

Conclusion:

MAPS scores did not adequately predict risk of opioid-related death. Contrary to expectations, prescription opioid use was not correlated with opioid-related death, whereas concurrent use of opioids and benzodiazepines represented a highly significant risk factor.

1. Introduction

Drug overdose is the leading cause of accidental death in the United States, with the bulk of these deaths involving prescription or illicit opioids (CDC, 2020). Although the opioid epidemic was initially linked to widespread use of prescription opioids, this ongoing crisis has been sustained by a resurgence in heroin abuse and, more recently, ready access to illicit fentanyl (Kolodny et al., 2015; LaRue et al., 2019; Rudd et al., 2016; Saad et al., 2018). The propensity for polydrug abuse in individuals with substance use disorder represents an additional, challenging aspect of the opioid epidemic (LaRue et al., 2019; Saad et al., 2018).

Benzodiazepines have been commonly co-prescribed with opioids for a variety of reasons, ranging from treatment of anxiety accompanying pain to adjunctive therapy for medication-assisted treatment of opioid dependence (Guy et al., 2019; Liang and Shi, 2019; Macleod et al., 2019; Park et al., 2015). Although co-prescription of benzodiazepines and opioids has been associated with increased risk of drug overdose death (Macleod et al., 2019; Park et al, 2015), rates of concurrent prescribing have risen substantially during the past several decades (Guy et al., 2019; Liang and Shi, 2019). Recent CDC guidelines cautioning against concurrent use of opioids and benzodiazepines have had only a modest impact on this trend (Jeffery et al., 2019).

As part of the effort to combat the current drug overdose crisis, prescription drug monitoring programs have been implemented throughout much of the United States. Although efforts are underway to use data from these monitoring programs to identify individuals most at risk for overdose deaths, a recent study concluded that access to, or mandatory use of, such programs has not significantly impacted benzodiazepine- and/or opioid-related overdose deaths (Liang and Shi, 2019). In order to better understand the contributions of prescribed and illicit opioids, and of benzodiazepines, to opioid-related deaths, and to determine the utility of a new state-wide prescription monitoring program in predicting risk of unintentional drug overdose, the present study compared data from the Michigan Automated Prescription System (MAPS) with autopsy findings (e.g. cause of death, postmortem toxicological data) from a major metropolitan area (Wayne County, MI).

2. Methods

2.1. Setting and data

The Wayne County Medical Examiner’s Office (WCMEO) conducts investigations to determine the cause and manner of death based on death scene investigations, autopsies (i.e. pathological and toxicological findings) and reviews of medical, police, and other records, including routine searches of prescription history in MAPS (Saad et al., 2018). WCMEO death investigations, mandated by MI law, are not subject to Institutional Review Board review. Cases were de-identified for this study. Comprehensive toxicological analyses (>400 drugs and metabolites) were performed by NMS labs (www.nmslabs.com), with quantitative data for each drug transformed into binary variables (i.e. presence or absence).

As the purpose of the study was to examine the impact of concurrent benzodiazepine use on the likelihood of opioid-related deaths, and to assess the utility of the Michigan Automated Prescription System (MAPS) in predicting the risk of these opioid deaths based solely on prescription histories, we utilized the pool of all 1898 WCMEO cases from 2018 with blood levels of at least 1 drug, without regard to evidence of medical or psychiatric co-morbidity (for which documentation varied across medical examiner cases) or the manner of death (the majority of which were accidental, but also included natural, homicide, suicide, and indeterminant manners of death). From this larger group, 252 of cases with an opioid-related cause of death and a positive toxicology for both an opioid and a benzodiazepine were identified. A control group of 252 cases with a positive blood toxicology for any drug, but a non-drug cause of death, was then matched to the opioid-death group using the optimal method of matching based on propensity scores with the R package “MatchIt” (R version 3.6.1)(Hansen and Klopfer, 2006; Ho et al., 2011). The assembled non-drug death group was 42.2±15.0 years of age, 75.4% white/19.4% black, and 41.7% female, while the opioid death group was 41.9±13.2 years of age, 74.6% white/19.4% black, and 42.1% female (not different by Pearson’s Chi-square; P = 0.831, P = 0.926, P = 9.28, respectively).

2.2. Statistical analyses

Statistical analyses were conducted using IBM SPSS Statistics (version 26). Variables organized into a matrix included subjects’ cause of death, demographic data, presence/absence of each drug in blood, and data from the statewide MAPS prescription monitoring program (MAPS, 2020), including prescriptions for each drug (+/−), number of prescribers, dispensers, and dispensations, and various risk scores (Overdose Risk Scores, Narcotics scores, and Sedative Scores) generated by MAPS using a proprietary algorithm (NarxCare, 2020). Pearson Chi-Square analyses determined differences between categorical variables comparing the opioid death and non-opioid death groups, whereas 2-sided t-tests were used to determine differences for numeric variables (e.g. age, number of prescribers, dispensers, or dispensations, and Overdose Risk, Narcotic, or Sedative Scores). P < 0.05 was considered statistically significant.

3. Results

3.1. Cases with a 180 day MAPS history

3.1.1. Prescriptions for opioids and benzodiazepines

Cases with an opioid-related cause of death and a positive postmortem toxicology for an opioid and benzodiazepine (N=252) were examined in terms of prescription history (schedule 2–5 drugs). Two-thirds of these cases (167/252) had a history of scheduled medications dispensed within 180 days prior to death. In comparison, significantly fewer non-drug death cases had a 180 day MAPS history (109/252; P = 2.10×10⁻⁷). The two groups were well-matched in terms of gender, ethnicity, and age (Table 1).

Table 1.

Analysis of 2018 WCMEO Cases with a 180 Day MAPS History

Characteristic	Non-Drug Death	Opioid Death	P value
Number of Cases	109	167
Agea	44.4 (1.20)	43.4 (0.99)	0.52
Genderb
Female	53 (49)	77 (46)	0.68
Male	56 (51)	90 (54)
Ethnicityb, c
White	87 (80)	125 (75)	0.63
Black	17 (16)	32 (19)
Rx Historyb
Any Opioid	85 (78)	133 (80)	0.74
Hydrocodone	47 (43)	62 (37)	0.32
Oxycodone	24 (22)	36 (22)	0.93
Codeine	15 (14)	29 (17)	0.42
Morphine	9 (8)	8 (5)	0.24
Buprenorphine/Naloxone	6 (6)	29 (17)	0.004
Methadone	0 (0)	10 (6)	0.009
Any Benzodiazepine	60 (55)	129 (77)	1.04×10−4
Opioid Plus Benzodiazepine	42 (39)	99 (59)	0.001
Alprazolam	25 (23)	67 (40)	0.003
MAPS Dataa
Overdose Risk Score	236 (12.3)	280 (11.25)	0.01
Narcotic Score	113 (9.01)	134 (6.60)	0.05
Sedative Score	98.6 (7.86)	127 (6.30)	0.006
Number of Prescribers	3.47 (0.27)	3.86 (0.22)	0.26
Number of Prescriptions	16.6 (1.44)	23.6 (1.32)	0.001
Number of Pharmacies	2.08 (0.13)	3.05 (0.16)	2.7×10−5
Positive Toxicologyb
Any Opioid	55 (50)	167 (100)	3.60×10−24
Any Benzodiazepine	40 (37)	167 (100)	1.65×10−32
Fentanyl	16 (15)	119 (71)	3.86×10−20
Morphine	16 (15)	74 (44)	2.84×10−7
Alprazolam	17 (16)	104 (62)	2.17×10−14
Cocaine	5 (5)	54 (32)	3.87×10−8

^aData are presented as mean (SE) of variable.

^bData are presented as number (%) of cases.

^cRemaining cases were of other ethnicities.

Unexpectedly, the proportion of individuals with any prescription history who had filled a prescription for an opioid was the same (~80%) in the opioid death and non-drug death groups (Table 1). This observation held for commonly prescribed individual opioid drugs, including hydrocodone, oxycodone, codeine, and morphine (Table 1). The exceptions were buprenorphine/naloxone combination and methadone, opioid drugs commonly used for medication-assisted therapy for substance abuse; filled prescriptions for these did predict an opioid-related drug death (Table 1), consonant with the notion that these opioids may serve as markers for a relapse into substance abuse with subsequent drug death.

Contrary to the equivalent rate of opioid prescriptions, significantly more cases within the opioid death group had filled a prescription for a benzodiazepine, or a benzodiazepine plus an opioid during the 6 months preceding death (Table 1). In fact, a filled prescription for the benzodiazepine alprazolam was the single best drug predictor of subsequent opioid drug death (Table 1).

3.1.2. MAPS-calculated risk scores

The MAPS suite (MAPS, 2020) includes a proprietary model (NarxCare, 2020) that utilizes various data (e.g. the number of pharmacies and providers, morphine milligram equivalents) to calculate individual Overdose Risk Scores, purported to predict the risk of unintentional overdose death. Overdose Risk Scores differed between the opioid death and non-drug death groups when calculated using assigned individual numerical values (Table 1), though not when individual subject scores were first binned into groups, as is done by the MAPS suite itself (P = 0.3). Related Narcotics Scores and Sedative Scores also differed between the two groups (Table 1). These MAPS-generated scores were, nevertheless, less predictive of opioid-related drug death than filling single opioid and benzodiazepine prescriptions within 6 months of death (Table 1), or than the total number of prescriptions filled or pharmacies used by a given subject (Table 1).

3.1.3. Postmortem toxicology

Postmortem toxicological reports provided data regarding which prescribed drugs were used near the time of death and, conversely, which drugs contributing to death were not prescribed (i.e. illicit use). Consistent with the stark rise in fentanyl-related overdose deaths seen in Wayne County (Saad et al., 2018) and nationally (CDC, 2020; LaRue et al., 2019) fentanyl was, by far, the best toxicological discriminator between cohorts: 71% of subjects in the opioid death group were positive for illicit fentanyl compared to 15% fentanyl-positive cases in the non-drug death group (Table 1). Morphine-positive toxicology was the next most significant opioid discriminator (Table 1). Although postmortem morphine can derive from either recent medical use of morphine or rapid in vivo metabolism of illicit heroin (itself rarely detected in blood)(Karch, 2002; Saad et al., 2018), fully 92% of the large number of morphine-positive opioid deaths (as well as two-thirds of the smaller number of morphine-positive non-drug deaths) did not have a morphine prescription prior to (not shown), indicating that morphine-positive postmortem toxicology was largely reflective of recent heroin use, as previously described (Saad et al., 2018).

Postmortem detection of benzodiazepines varied greatly as a function of the specific drug: whereas lorazepam, chlordiazepoxide, and oxazepam were infrequently detected, clonazepam, temazepam, and diazepam were more prevalent in benzodiazepine-positive opioid deaths (P = 2.30×10⁻², P = 3.50×10⁻², P = 3.00×10⁻³, respectively). Alprazolam-positive toxicology was highly predictive of inclusion in the drug death group (Table 1): fully 40% of cases had an alprazolam prescription, and another 25% had no prescription but were nevertheless alprazolam-positive, whereas only 16% of non-drug deaths were alprazolam-positive. Aside from detection of alprazolam and fentanyl, the presence of cocaine in postmortem blood was the next best discriminator of an opioid-related drug death (Table 1), consistent with other evidence of extensive co-abuse of these drugs and high rates of associated mortality (LaRue et al., 2019; Saad et al., 2018).

3.2. Cases with active MAPS history at death

3.2.1. Prescriptions for opioids and benzodiazepines

The proportion of subjects with an active MAPS history at death (i.e. filled a scheduled drug prescription within the 30 days preceding death) was higher for the opioid death cases than the non-drug death group (53% compared to 27%; P = 3.36×10⁻⁹). Examination of this subset of subjects (Table 2) provided additional insights. As seen with the 6 month MAPS dataset, the proportions of subjects filling an opioid prescription within a month of death did not differ between the opioid death and non-drug death groups; nor did prescriptions for individual opioids (again, with the exception of opioids used for medication-assisted therapy) (Table 2). On the other hand, the proportion of subjects filling a prescription for alprazolam (or any benzodiazepine) within a month of death was significantly higher in the opioid death group (Table 2).

Table 2.

Analysis of 2018 WCMEO Cases with a 30 Day MAPS History

Characteristic	Non-Drug Death	Opioid Death	P value
Number of Cases	68	133
Rx Historyb
Any Opioid	57 (84)	109 (82)	0.74
Hydrocodone	31 (46)	54 (41)	0.50
Oxycodone	18 (26)	33 (25)	0.80
Codeine	8 (12)	22 (17)	0.37
Morphine	8 (12)	6 (5)	0.06
Buprenorphine/Naloxone	5 (7)	24 (18)	0.04
Methadone	0 (0)	8 (6)	0.04
Any Benzodiazepine	44 (65)	108 (81)	0.01
Opioid Plus Benzodiazepine	36 (53)	86 (65)	0.12
Alprazolam	19 (28)	57 (43)	0.04
MAPS Dataa
Overdose Risk Score	262 (15.9)	293 (13.0)	0.15
Narcotic Score	134 (12.2)	141 (7.1)	0.59
Sedative Score	118 (10.5)	137 (6.9)	0.13
Number of Prescribers	3.88 (0.38)	4.14 (0.26)	0.58
Number of Prescriptions	21.5 (1.93)	26.7 (1.47)	0.04
Number of Pharmacies	2.21 (0.17)	3.22 (0.19)	0.001
Positive Toxicologyb
Any Opioid	39 (57)	133 (100)	3.90×10−16
Any Benzodiazepine	35 (51)	133 (100)	1.53×10−18
Fentanyl	6 (9)	93 (70)	2.44×10−16
Morphine	10 (15)	53 (40)	2.77×10−4
Alprazolam	15 (22)	82 (62)	1.07×10−7
Cocaine	4 (6)	41 (31)	6.00×10−5

^aData are presented as mean (SE) of variable.

^bData are presented as number (%) of cases.

3.2.2. MAPS-calculated risk scores and postmortem toxicology

The number of pharmacies used and prescriptions filled within the last month of life were significantly higher for the opioid death group yet, surprisingly, the MAPS-calculated Overdose Risk Scores, Narcotic Scores, and Sedative Scores were not predictive of drug death for these cases (Table 2). In this subset of subjects with an active prescription history, the strongest predictors of drug death were, by far, postmortem detection of fentanyl, morphine, alprazolam, and/or cocaine (Table 2).

4. Discussion

Access to prescription opioids in the six months preceding death was not a predictor of subsequent opioid death, with the exception of prescriptions for opioids indicative of attempted medication-assisted therapy for substance use disorder (Tables 1 and 2). MAPS-generated scores purported to predict risk of drug overdose were either moderately predictive (6 month MAPS dataset; Table 1) or not predictive (1 month MAPS dataset; Table 2) in discriminating which subjects succumbed to opioid death. Indeed, as a single variable, the number of pharmacies used for obtaining controlled substances better predicted opioid drug deaths than did the MAPS algorithm (Tables 1 and 2). It is not clear why this is the case (as the MAPS algorithm is proprietary)(MAPS, 2020; NarxCare, 2020), but these data indicate that further refinements are needed in the application of drug prescription monitoring program data for prediction of risk of overdose and drug-related death.

Despite a growing appreciation of the dangers inherent in concurrent use of opioids and benzodiazepines, rates of co-prescription, co-abuse, and association with drug deaths remain alarmingly high (Guy et al., 2019; Liang and Shi, 2019; Macleod et al., 2019; Park et al., 2015; Saad et al., 2018). It was remarkable that filling a prescription for alprazolam or other benzodiazepines was a stronger predictor of opioid death than filling an opioid prescription (Tables 1 and 2). The nature of the relationship between benzodiazepine use and opioid death, however, remains unclear: although this combination of drugs may produce a synergistic depression of respiration, it is also possible that alprazolam serves as a marker for more risky patterns of substance abuse, for adjunctive treatment for substance use disorder, and/or co-morbidity for other psychiatric disorders. Further studies are required to help distinguish from among these possibilities.

5. Conclusions

The current data and other studies indicate that concurrent use of benzodiazepines and opioids constitutes a highly risky practice. Risk assessment models based on prescription drug monitoring programs need to better reflect this heightened risk. Medical professionals should avoid co-prescription of benzodiazepines and opioids whenever possible.

Highlights.

• Opioid deaths with concurrent benzodiazepine use were compared to non-drug deaths

• These deaths were associated with prescriptions for benzodiazepines but not opioids

• The Michigan Automated Prescription System predicted drug death in some instances

• Individual measures of prescription activity were better predictors of drug death

• Positive toxicology for fentanyl or alprazolam best predicted drug-related death