Dosing profiles of concurrent opioid and benzodiazepine use associated with overdose risk among US Medicare beneficiaries: Group-based multi-trajectory models

Abstract

Background and aims:

One-third of opioid (OPI) overdose deaths involve concurrent benzodiazepine (BZD) use. Little is known about concurrent opioid and benzodiazepine use (OPI-BZD) most associated with overdose risk. We aimed to examine associations between OPI-BZD dose and duration trajectories, and subsequent OPI or BZD overdose in U.S. Medicare.

Design:

Retrospective cohort study.

Setting:

US Medicare.

Participants:

Using a 5% national Medicare data sample (2013–2016) of fee-for-service beneficiaries without cancer initiating OPI prescriptions, we identified 37,879 beneficiaries (age ≥65=59.3%, female=71.9%, White=87.6%, having OPI overdose=0.3%).

Measurements:

During the 6 months following OPI initiation (i.e., trajectory period), we identified OPI-BZD dose and duration patterns using group-based multi-trajectory models, based on average daily morphine milligram equivalents (MME) for OPIs and diazepam milligram equivalents (DME) for BZDs. To label dose levels in each trajectory, we defined OPI use as very-low-(<25 MME), low-(25–50 MME), moderate-(51–90 MME), high-(91–150 MME), and very-high-(>150 MME) dose. Similarly, we defined BZD use as very-low- (<10 DME), low- (10–20 DME), moderate- (21–40 DME), high- (41–60 DME), and very-high- (>60 DME) dose. Our primary analysis was to estimate the risk of time to first hospital or emergency department visit for OPI overdose within 6 months following the trajectory period using inverse-probability-of-treatment-weighted Cox proportional hazards models.

Findings:

We identified 9 distinct OPI-BZD trajectories: Group (A): Very-low OPI (early discontinuation)-Very-low declining BZD (n=10,598; 28.0% of the cohort); (B): Very-low OPI (early discontinuation)-Very-low stable BZD (n=4,923; 13.0%); (C): Very-low OPI (early discontinuation)-Medium BZD (n=4,997; 13.2%); (D): Low OPI-Low BZD (n=5,083; 13.4%); (E): Low OPI-High BZD (n=3,906; 10.3%); (F): Medium OPI-Low BZD (n=3,948; 10.4%); (G): Very-high OPI-High BZD (n=1,371; 3.6%); (H): Very-high OPI-Very-high BZD (n=957; 2.5%); and (I): Very-high OPI-Low BZD (n=2,096; 5.5%). Compared with Group (A), 5 trajectories (32.3% of the study cohort) were associated with increased 6-month OPI overdose risks: (E) Low OPI-High BZD HR=3.27, 95%CI=1.61–6.63); (F): Medium OPI-Low BZD (HR=4.04, 95%CI=2.06–7.95); (G): Very-high OPI-High BZD (HR=6.98, 95%CI=3.11–15.64); (H): Very-high OPI-Very-high BZD (HR=4.41, 95%CI=1.51–12.85); and (I): Very-high OPI-Low BZD (HR=6.50, 95%CI=3.15–13.42).

Conclusions:

Patterns of concurrent opioid and benzodiazepine use most associated with overdose risk among fee-for-service US Medicare beneficiaries initiating opioid prescriptions include very-high-dose opioid use (MME>150), high-dose benzodiazepine use (DME>40), or medium-dose opioid with low-dose benzodiazepine use.

The number of opioid (OPI) overdose deaths in the United States (US) was more than six times greater in 2019 than in 1999 (1), with ~35% involving prescription OPIs (2) and 31% to 61% involving benzodiazepines (BZDs) (3–7). Over 90% of BZD overdose deaths involved OPIs, highlighting the dangers of concurrent OPI and BZD use (hereafter OPI-BZD use) (8). OPI-BZD use can cause synergistic respiratory depression and can substantially increase overdose risk. For example, compared with OPI use alone, OPI-BZD use is associated with a 2 to 10-fold increased OPI overdose risk (4, 9–15). The number of patients receiving these classes concomitantly remains substantial in the U.S. (e.g., approximately 25% of long-term opioid users concomitantly use BZDs (16). Yet, little is known about the OPI-BZD dose combination patterns over time most associated with overdose risk.

Although clinical guidelines and U.S. Food and Drug Administration (FDA) black-box warnings caution that OPI-BZD use is a risk factor for adverse health outcomes (17–19), OPI-BZD use definitions vary substantially in the literature and focus on arbitrary duration thresholds (e.g., ≥1 overlapping day) or dose alone (3, 4, 9, 10, 15, 20–22). One 2019 Centers for Medicare & Medicaid Services (CMS) OPI safety measure included a concurrent OPI-BZD use measure (i.e., ≥30 cumulative days of overlapping OPI-BZD in a year among those having ≥2 OPI and ≥2 BZD fills), without specific dosing recommendations for either drug (23). However, OPI-BZD-use patterns may change over time and vary across patient subgroups due to patient conditions and clinical needs (e.g., pain, anxiety, other conditions). Applying arbitrary thresholds with unknown or suboptimal predictive values indistinctly across all patients poses challenges in clinical care and may lead to overly broad and ineffective interventions addressing OPI-BZD co-prescribing.

Identifying distinct OPI-BZD dose change patterns over time from prescription fill/refill claims data may more effectively guide clinical care and policies around OPI-BZD use. Group-based multi-trajectory models can simultaneously examine dose and duration change patterns and identify subgroups with similar concurrent-use patterns over time (24, 25). We aimed to apply group-based multi-trajectory modeling to first identify distinct OPI-BZD dose change patterns over 6 months in real-world settings, and then examine their associations with subsequent OPI-related overdose risk among Medicare beneficiaries initiating OPI prescriptions. We measured the OPI-BZD trajectories using a 6-month period because this duration allows sufficient time for identifying distinct patterns of OPI-BZD use over time (>30% of older adults take BZDs for ≥120 days (14)), and prior work has shown that in general overdose risk increases as duration of concurrent use of OPIs with other CNS medications increases (22). We chose this population as Medicare beneficiaries have high prevalence of pain, anxiety, and insomnia conditions, and concurrent OPI-BZD use (≥25% among those with OPI prescriptions), and Medicare claims data allow for national, longitudinal follow-up (14, 26). Furthermore, Medicare Part D plans will soon be required to implement programs specifically targeting individuals at high risk for aberrant OPI-related behaviors (27–30).

Methods

Data Source, Design, and Sample

The University of Florida Institutional Review Board approved this study. This study’s reporting complies with STROBE guidelines (31). Although the study protocol was developed prior to conducting the study, we did not pre-register it on a publicly available platform and therefore the results should be considered exploratory.

Our retrospective cohort study included a 5% nationally representative sample of U.S. Medicare beneficiaries from 2013 to 2016, given Medicare started covering BZDs in Medicare Part D in 2013 (32, 33). A 5% random sample of Medicare claims data, provided by Centers for Medicaid and Medicare Services, is commonly used by researchers for health services and pharmacoepidemiologic studies. Medicare is the national health insurance program in the U.S. for individuals aged 65 and older or aged under 65 with end-stage renal disease or certain disabilities (32). Medicare beneficiaries choose either federally administered fee-for-service plans that pay providers for each service performed or Medicare Advantage plans that provide capitated payments for all covered services. We limited our analysis to beneficiaries with fee-for-service plans; because, those plans have more complete data. Of the 64 million Medicare beneficiaries enrolled in 2019, two-thirds were enrolled in fee-for-service plans (34). The fee-for-service system includes Part A (hospital), Part B (medical), and Part D (prescription drug) insurance. Datasets used for this analysis included Medicare master beneficiary summary files, Part D drug event files, and medical claims of inpatient, outpatient, carrier, home health, skilled nursing facility, hospice, and durable medical equipment.

As shown in Figure 1, we first excluded beneficiaries who were non-US residents, had malignant cancer diagnoses (except non-melanoma skin cancer; Appendix Table 1) (35), received hospice services during the study period, or were enrolled in Medicare Advantage plans during the study period. Among beneficiaries who initiated non-injectable OPI prescriptions (excluding buprenorphine formulations approved for addiction treatment), we excluded those who: (1) only used OPI or BZD prescriptions, (2) were not continuously enrolled in Parts A, B, and D from 6 months before to 6 months after the index date (i.e., first OPI-prescription date), (3) had the first OPI prescription before July 1, 2013 or after July 1, 2016 (insufficient time for baseline and trajectory period), (4) had OPI- or BZD-overdose diagnoses (Appendix Table 1) during the 6 months before the index OPI prescription and 6-month trajectory measurement period (<0.08%), and (5) filled OPIs or BZDs likely for acute conditions (i.e., with sporadic exposure defined as filling only 1 OPI or BZD prescription, or with <15 days’ supply of OPIs or BZDs during the 6-month trajectory measurement period, based on CMS’ OPI-BZD risk measure used by Part D plan partners for quality improvement) (23).

Figure 1. Sample Size Flowchart.

* Two or more prescription claims for opioids filled on at least two separate days, for which the sum of the days’ supply was ≥15 during the 12-month measurement period

Exposures: Trajectories of Concurrent Opioid and Benzodiazepine (OPI-BZD) Use

Our exposure of interest was beneficiary’s membership in a distinct trajectory of OPI-BZD use. We identified these trajectories by (1) constructing daily measures of average standardized daily dose (SDD) separately for OPIs and BZDs during the 6 months after OPI initiation, and (2) applying group-based multi-trajectory models with SDD as the model’s outcome to identify distinct dose and duration patterns of OPI-BZD use (Appendix Figure 1).

To calculate SDD for OPIs, we calculated daily morphine milligram equivalents (MME) using dispensing dose, date and days’ supply and a conversion factor provided by the Centers for Disease Control and Prevention (CDC) (36, 37). For BZDs, we calculated diazepam milligram equivalents (DME), based on relevant clinical resources for equivalent BZD doses (Appendix Table 2) (38–40). Extreme outliers of daily MME and DME values were substituted with the 99th percentile of the MME and DME values in the study cohort. Then, we identified distinct OPI-BZD-use patterns based on doses used over time using group-based multi-trajectory models (24, 25, 41, 42). The group-based multi-trajectory models used censored normal distributions and included the average daily MME for OPIs and DME for BZDs as longitudinal, continuous dependent variables (y₁ and y₂) for each day in the 6 months after initiating opioids (i.e., a time variable) (25). Each group-based multi-trajectory model used the most flexible functional form (up to the fifth order polynomial function) of time to allow the trajectories to emerge from the data. The Appendix Methods provide additional analytical details for identifying OPI-BZD-use trajectories and the a priori rule used to select the final number of trajectories. Briefly, we selected the final trajectory model based on a combination of Bayesian Information Criterion (BIC) (larger value indicating better fitting model), Nagin’s criteria (43), and each trajectory group having at least 2 overdose cases (to obtain valid risk estimates in Cox models)(44), with a sufficient number of beneficiaries in each trajectory group and clinical relevance of trajectory patterns (i.e., each identified trajectory reflects a distinct and clinically meaningful OPI and BZD dose pattern over time, and a preference for a smaller number of trajectories to minimize complexity and maintain interpretability). Once the final trajectory groups were identified, to facilitate the labeling of OPI and BZD dose levels for each trajectory, we defined OPI use as: very-low- (SDD <25 MME), low- (25–50 MME), moderate- (51–90 MME), high- (91–150 MME), and very-high-dose (>150 MME) (45). Similarly, we defined BZD use as very-low- (<10 DME), low- (10–20 DME), moderate- (21–40 DME), high- (41–60 DME), and very-high-dose (>60 DME) (37–40).

Outcome Variables: OPI overdose and BZD overdose

The a priori primary outcome was defined as time to the first OPI overdose event captured in the claims data occurring in the 6 months following the 6-month OPI-BZD trajectory measurement period. Similar to prior studies using a validated algorithm in claims data (46, 47), we used the International Classification of Diseases codes (ICD-9/ICD-10; Appendix Table 1(46)) to identify overdose events from prescription or other OPIs including heroin from inpatient or ED settings. We examined two secondary outcomes including time to first diagnosis of (1) BZD overdose, and (2) a composite outcome of either OPI or BZD overdose from inpatient or ED settings during the 6 months following the 6-month OPI-BZD trajectory measurement period (46). We chose BZD overdose as a secondary outcome due to >90% of BZD overdose deaths involving OPIs and a lack of validated algorithms using claims data (8).

Covariates

Based on prior literature (3, 9, 10, 15, 20, 21, 48), we measured a series of covariates during the 6 months prior to OPI initiation, including age, sex, race/ethnicity (White, Black, Hispanic, and others), disability status indicating original reason for Medicare eligibility, and receipt of low-income subsidy (LIS) and dual Medicaid eligibility (with LIS and dual eligibility, with only LIS or dual eligibility, and no LIS or dual eligibility). LIS covers most prescription drug costs for eligible beneficiaries who have limited income and resources (49). Beneficiaries with dual eligibility receive both Medicaid and Medicare benefits (50). Health status factors included Elixhauser Comorbidity Index (excluding metastatic cancers and solid tumors with or without metastasis; range 0 to 27), alcohol use disorder, opioid use disorder (OUD), non-OPI substance use disorders, anxiety, mood, and sleep disorders and musculoskeletal and individual pain conditions (Appendix Table 1) (51). Health services use factors included any hospitalization, ED visit counts (0, 1, and ≥2), outpatient visit counts (0, 1, 2–5 and ≥5), and any urine drug tests. We measured a series of medication-use related variables during the 6 months prior to initiating OPIs including any use of BZDs, gabapentinoids, muscle relaxants, antidepressants, antipsychotics, and naltrexone and polypharmacy with ≥3 different medications not mentioned above.

We also described several characteristics including type of medications (e.g., short-acting, long-acting), unique medication ingredients, prescriber specialty (primary care providers, ED, surgery, psychiatry/psychology or physiatry, and others) for the first prescription, average days supplied and average number of OPI and BZD fills for OPI and BZD treatment episodes in the 6-month trajectory measurement period, separately.

Statistical Analysis

Given that identified trajectory groups were likely to differ by patient characteristics and disease complexity, we included all covariates measured in the 6 months prior to OPI initiation to estimate inverse probability of treatment weights (IPTW) for each beneficiary using gradient boosting machine. Gradient boosting machine has advantages over multinomial logistic regression for generating propensity scores, especially when including multiple comparison groups and variables with skewed distributions (52). IPTW was defined as the inverse probability of an individual’s likelihood to be placed in a specific trajectory group. Weighting created a sample in which treatment assignment was independent of measured covariates and thus, minimized confounding in the comparison of trajectory-specific OPI overdose risk (53–55). We excluded beneficiaries (n=32) with extreme IPTWs (>10) using trimming methods to increase validity of treatment effect estimates. We compared characteristics across trajectory groups before and after weighting subjects with IPTW (SMD>0.1 considered as non-negligible differences) using standardized mean differences (SMD).

We then used Cox proportional hazards models with IPTW in the weight function to compare time-to-event (i.e., OPI overdose, BZD overdose, OPI/BZD overdose) during the 6 months following the 6-month OPI-BZD trajectory measurement period across different OPI-BZD trajectories, adjusting for covariates with non-negligible differences (mean SMD>0.1) after IPTW weighting. These models treated beneficiaries switching to Medicare Advantage plans, died, or without any outcomes of interest during the 6 months following the 6-month trajectory period as censored observations. We assessed the proportional hazards assumption’s validity using Schoenfeld residuals (56).

To ensure the robustness of the findings, we conducted several additional analyses. First, we included beneficiaries who had OPI or BZD overdose during the 6-month trajectory period in the analysis. Second, we compared our trajectory analysis with three simple binary measures that had been used to identify high-risk individuals or OPI-BZD utilization behavior in Medicare, including: (1) any overlapping OPI-BZD day, (2) having ≥15 overlapping OPI-BZD overlapping days, and (3) having ≥30 overlapping OPI-BZD overlapping days during the 6-month trajectory period. Finally, to assess unmeasured confounders’ potential influences, we calculated the “E-value,” which is defined as the minimum strength of association an unmeasured confounder would need to have with the treatment and outcome to account for a specific treatment-outcome association, conditional on measured covariates (i.e., covariates included in the IPTW in our study due to no covariates with mean SMD>0.1 after IPTW) (57). Large E-values imply significant or substantial unmeasured confounding would be needed to account for effect estimates, whereas small E-values imply little unmeasured confounding would be needed (57).

Group-based multi-trajectory models were estimated using STATA 16.0 (Stata-Corp LP, College Station, TX) and the TRAJ macro (free download at http://www.andrew.cmu.edu/user/bjones). SMDs were calculated using R packages, tableone and survey. All other analyses were performed using SAS version 9.4 (SAS Inc., Cary, NC, USA).

Results

Overall Beneficiary Characteristics

Among 37,879 eligible beneficiaries, most were age ≥65 years (59.3%), female (71.9%) White (87.6%), and metropolitan county residents (72.3%), and had LIS and/or dual eligibility (55.5%) (Table 1). The average Elixhauser score was 2.3(standard deviation [SD]=2.8) and 49.8% of beneficiaries had musculoskeletal conditions (Table 1). The average weekly standardized daily dose in the 6 months following OPI initiation was 37.2 MME(SD=2.6) for OPIs and 18.7 DME(SD=0.9) for BZDs (Appendix Figure 2).

Table 1.

Characteristics of Medicare Beneficiaries by Opioid and Benzodiazepine Trajectory Group^†

Characteristics*	Overall	A: Very-low OPI (early discontinuation)-Very low declining BZD	B: Very-low OPI (early discontinuation)-low stable BZD	C: Very-low OPI (early discontinuation)-Medium BZD	D: Low OPI-Low BZD	E: Low OPI-High BZD	F: Medium OPI-Low BZD	G: Very-high OPI-High BZD	H: Very-high OPI-Very-High BZD	I: Very-high OPI-Low BZD	SMD‡ before IPTW	SMD‡ after IPTW
No. beneficiaries	37,879	10,598	4,923	4,997	5,083	3,906	3,948	1,371	957	2,096
% of the overall cohort	100	28.0	13.0	13.2	13.4	10.3	10.4	3.6	2.5	5.5
Socio-demographics
Age ≥65 years, %	59.3	77.6	83.7	65.7	53.4	32.5	45.1	21.3	13.7	30.2	0.71	0.08
Female, %	71.9	75.1	79.4	72.6	71.1	67.8	71.0	63.1	56.9	59.9	0.19	0.06
Race/ethnicity group, %
White	87.6	85.6	89.2	88.6	87.4	87.6	88.8	89.4	86.2	89.1	0.05	0.05
Black	7.7	8.3	6.4	6.3	8.9	8.7	7.3	6.4	9.1	7.3	0.05	0.03
Others	4.7	6.1	4.4	5.1	3.7	3.8	3.9	4.2	4.7	3.6	0.04	0.06
Disability status, %	54.4	33.8	31.2	49.0	63.6	80.7	70.4	89.4	94.9	82.7	0.68	0.08
LIS/Dual eligibility, %
No LIS/no dual eligibility	44.5	58.2	54.8	46.9	37.0	27.6	37.7	25.9	20.2	31.7	0.34	0.05
LIS or dual eligibility	5.5	4.3	3.8	4.9	6.5	6.5	6.3	7.3	8.4	8.3	0.08	0.02
LIS and dual eligibility	50.0	37.6	41.4	48.2	56.5	65.9	56.0	66.8	71.5	60.0	0.29	0.06
Metropolitan residence	72.3	75.0	72.9	71.6	66.6	70.8	71.1	72.0	79.0	74.6	0.09	0.06
Health status
Elixhauser index, mean (SD)	2.3 (2.8)	3.2 (2.9)	3.1 (2.9)	2.9 (2.8)	1.4 (2.3)	1.9 (2.5)	1.5 (2.4)	0.9 (1.8)	1.0 (1.8)	0.9 (1.9)	0.45	0.05
Opioid use disorder, %	1.0	0.5	0.3	0.9	0.7	1.8	1.2	2.9	2.8	2.0	0.10	0.04
Alcohol use disorders, %	1.3	1.5	1.1	1.7	0.8	2.1	0.9	0.7	0.8	0.2	0.07	0.07
Other non-opioid SUD, %	1.5	1.2	0.5	1.8	1.0	3.4	1.6	1.9	3.3	1.3	0.08	0.06
Anxiety disorders, %	18.9	21.5	22.1	26.8	11.3	25.0	10.9	12.9	13.9	7.3	0.23	0.07
Mood disorders, %	18.4	18.8	19.0	27.1	11.4	28.0	12.8	14.3	14.3	9.0	0.21	0.04
Sleep disorders, %	33.9	30.5	31.1	35.8	34.8	36.7	37.2	37.1	37.0	36.1	0.06	0.05
Musculoskeletal conditions, %	49.8	64.4	58.9	56.5	33.7	41.4	39.4	31.8	32.3	32.5	0.31	0.07
Pain conditions, %
Osteoarthritis	29.1	41.7	37.8	34.6	17.1	19.4	19.8	14.2	11.8	14.2	0.31	0.10
Back pain	25.3	30.5	27.1	26.8	18.2	22.6	23.4	20.4	21.4	21.1	0.11	0.07
Neck pain	9.3	11.8	9.6	10.4	6.1	7.9	8.5	7.1	5.9	7.6	0.09	0.05
Chest pain	10.1	15.6	12.3	12.0	4.9	8.2	6.0	3.6	4.0	2.9	0.20	0.05
Abdominal pain	11.3	16.2	13.5	14.3	5.6	9.8	7.5	5.0	5.0	4.9	0.18	0.03
Rheumatoid arthritis	2.6	3.3	3.1	2.6	1.7	1.8	2.4	2.3	1.8	2.1	0.04	0.03
Pelvic pain	1.8	2.6	2.5	2.3	0.9	1.7	0.9	0.8	0.8	0.3	0.08	0.06
Headache/migraine	6.7	8.5	7.4	8.8	3.5	7.8	4.6	3.9	3.3	3.1	0.12	0.04
TMJ	0.2	0.3	0.3	0.4	0.1	0.2	0.2	0	0.1	0	0.04	0.04
Others	26.8	36.9	33.9	31.2	15.6	22.3	19.1	13.0	12.1	14.8	0.27	0.06
Health services use
Any hospitalization, %	11.0	17.1	13.8	12.2	4.8	8.6	6.9	4.2	4.4	4.2	0.19	0.05
ED visits, %
1	11.9	16.2	14.4	14.8	6.9	11.5	7.6	4.9	6.1	6.2	0.17	0.05
≥2	6.1	8.5	7.5	7.4	2.8	6.2	4.0	2.5	3.4	2.1	0.13	0.04
Outpatient visits, %
1	16.7	18.6	17.4	18.4	4.9	15.7	15.4	13.1	13.7	12.9	0.07	0.06
2–5	20.0	28.3	26.1	25.0	10.5	17.4	11.9	7.7	7.6	7.8	0.27	0.07
>5	7.8	11.4	11.7	11.4	3.1	6.1	3.6	1.5	1.8	1.7	0.22	0.04
Urine drug test, %	3.4	2.2	2.0	3.3	3.4	5.2	4.3	4.6	5.5	5.5	0.09	0.05
Medication use related variables at baseline
Any benzodiazepines, %	32.5	47.1	66.0	72.3	24.9	51.3	25.1	30.9	26.0	17.1	0.52	0.09
Any antidepressants, %	32.8	34.8	37.8	46.8	23.4	39.0	24.5	21.9	20.2	18.0	0.27	0.08
Any antipsychotics, %	9.6	7.5	8.5	16.5	6.9	9.9	5.9	6.5	7.5	3.5	0.19	0.03
Any gabapentinoids, %	11.3	11.6	12.5	13.9	9.8	12.1	10.3	8.7	6.5	8.7	0.09	0.02
Any muscle relaxants, %	8.4	7.1	7.0	102	7.0	11.7	8.2	11.7	10.4	7.7	0.08	0.02
Any naltrexone, %	0.1	0.0	0.1	0.1	0.0	0.2	0.1	0.0	0.0	0.0	0.03	0.03
Had polypharmacy (≥3 medications not listed above), %	16.9	14.1	19.1	30.7	10.2	29.3	10.0	12.0	9.9	6.0	0.27	0.08
First opioid prescriber specialty, %
PCP	59.1	57.8	62.9	60.3	65.1	59.7	55.7	52.8	52.4	52.1	0.12	0.07
ED	5.4	7.1	6.2	6.5	3.8	5.7	3.6	2.5	3.3	2.0	0.04	0.02
Surgery	5.9	8.9	6.2	7.2	2.7	5.2	4.2	2.9	2.3	1.8	0.14	0.07
Psychiatry/psychology	0.1	0.1	0.0	0.2	0.1	0.3	0.1	0.2	0.6	0.1	0.11	0.07
Physical /rehabilitation	3.6	2.5	2.1	1.8	3.5	2.9	6.3	8.5	7.3	8.7	0.15	0.05
Others	25.9	23.6	22.6	24.0	24.9	26.2	30.1	33.1	34.1	35.3	0.13	0.07
First benzodiazepine prescriber specialty, %
PCP	64.3	68.1	70.7	62.6	67.2	53.9	63.2	55.9	51.0	60.8	0.17	0.02
ED	2.7	3.2	2.2	2.2	2.9	1.9	2.9	2.3	2.3	2.7	0.03	0.05
Surgery	1.2	2.1	0.3	0.4	0.9	0.5	1.9	1.0	0.7	1.3	0.07	0.04
Psychiatry/psychology	1.6	1.0	1.1	2.2	1.2	3.7	1.2	2.3	3.1	1.0	0.08	0.03
Physical /rehabilitation	1.1	1.1	0.4	0.4	0.7	0.6	2.1	3.1	2.4	3.6	0.11	0.03
Others	29.1	24.5	25.4	32.1	27.2	39.5	28.7	35.4	40.4	30.6	0.15	0.03

Abbreviations: ED, emergency department; IPTW, inverse probability of treatment weighting; LIS, low-income subsidy; No., number of; NSAID, nonsteroidal anti-inflammatory drug; PCP, primary care providers; SD, standard deviation; SMD: standardized mean difference; TMJ: temporomandibular disorder pain.

^*We described the characteristics of beneficiaries in each trajectory group with means and standard deviations (SD) for continuous variables and frequencies and percentages for categorical variables.

^†To facilitate the labeling of opioid and benzodiazepine dose levels for each trajectory, we defined opioid dosage use as: very-low- (SDD <25 MME), low- (25–50 MME), moderate- (51–90 MME), high- (91–150 MME), and very-high-dose (>150 MME). Similarly, we defined BZD dosage use as very-low- (<10 DME), low- (10–20 DME), moderate- (21–40 DME), high- (41–60 DME), and very-high-dose (>60 DME).

^‡Average SMD of 36 SMDs from group comparisons (the number of 2-combinations from 9 trajectories: $C_2^9=36;($ e.g., group A vs B, A vs C). Appendix Table 3 includes the maximum and minimum SMDs.

Trajectories of Concurrent OPI and BZD use

Appendix Table 3 shows the BIC values, condition of having a minimum 2 overdose cases (yes/no), and percent of the study cohort for each trajectory for 2- to 10-group trajectory solutions. Appendix Table 4 shows the Nagin’s diagnostic criteria for the final model with 9 distinct trajectories for OPI-BZD use (BIC = - 5,887,148.51). Figure 2 illustrates actual and predicted daily dose utilization patterns for OPI and BZD use in the 6-month period following OPI initiation. The 9 distinct OPI-BZD trajectories included: Group (A): Very-low OPI-Very-low BZD (declining BZD) (n=10,598; 28.0% of the cohort; <25 MME with decreasing <10 DME); (B): Very-low OPI-Very-low BZD (stable BZD) (n=4,923; 13.0%; <25 MME with decreasing <10 DME); (C): Very-low OPI-Medium BZD (n=4,997; 13.2%; <25 MME with 21–40 DME); (D): Low OPI-Low BZD (n=5,083; 13.4%; 25–50 MME with 10–20 DME); (E): Low OPI-High BZD (n=3,906; 10.3%; 25–50 MME with 41–60DME); (F): Medium OPI-Low BZD (n=3,948; 10.4%; 51–100 MME with 10–20 DME); (G): Very-high OPI-High BZD (n=1,371; 3.6%; >150 MME with 41–60 DME); (H): Very-high OPI-Very-high BZD (n=957; 2.5%; >150 MME with >60 DME); and (I): Very-high OPI-Low BZD (n=2,096; 5.5%; >150 MME with 10–20 DME).

Figure 2. Trajectories of Opioid and Benzodiazepine Utilization Patterns among Medicare Beneficiaries (n=37,879)*.

Abbreviations: BZD, benzodiazepine; DME, diazepam milligram equivalent; MME, morphine milligram equivalent; OPI, opioid; SDD, standardized daily dose.

* Black dotted lines indicate the actual trajectories and solid color lines indicate model-based trajectories. We calculated SDDs for OPIs using MME and for BZDs using DME. To facilitate the labeling of opioid and benzodiazepine dose levels for each trajectory, we defined opioid dosage use as: very-low- (SDD <25 MME), low- (25–50 MME), moderate- (51–90 MME), high- (91–150 MME), and very-high-dose (>150 MME). Similarly, we defined BZD dosage use as very-low- (<10 DME), low- (10–20 DME), moderate- (21–40 DME), high- (41–60 DME), and very-high-dose (>60 DME).

Characteristics by Trajectory Group

There were significantly different characteristics across trajectory groups before IPTW (Table 1). For example, compared to the overall study cohort, individuals in the consistent very-high OPI trajectories (i.e., Groups G, H, I) were more likely to be younger (age ≥65 years: 13.7%–30.2% vs. 59.3% overall); have a disability (82.7%–94.9% vs 54.4%); have an OUD diagnosis (2.0%–2.9% vs. 1.0%); have lower proportions of antidepressant (18.0%–21.9% vs. 32.8%) and gabapentinoid (6.5%–8.7% vs. 11.3%) use; and have their first OPI or BZD prescribed by physiatry providers or other specialties. After accounting for each beneficiary’s IPTW, all characteristics were considered comparable across trajectories (with mean SMDs <0.1; Table 1). Appendix Table 5 presents the minimum and maximum SMDs across the 36 group comparisons ($C_2^9$; e.g., A vs B, B vs C).

During the 6-month trajectory measurement period, compared with the overall study cohort, individuals in the consistently very-high-dose OPI trajectories were more likely to use long-acting OPIs such as transdermal fentanyl (17.5%–31.4% vs. 5.4% overall; Appendix Table 6), to have longer days’ supply for their OPI prescriptions (27.0–27.5 vs. 23.0 days), higher average daily MME (155.7–192.4 vs. 37.2), and more OPI fills (12.3–13.3 vs. 6.3). Similarly, high or very-high-dose BZD users (i.e., Groups E, G, H) were more likely to use only short-acting BZDs such as alprazolam (50.4%–69.8% vs. 38.8%) or BZDs with mixed action durations (Appendix Table 7).

Inverse Probability Treatment Weighted Multivariable Cox Proportional Hazards Model for Overdose Risk

Primary outcome: OPI overdose

Among eligible beneficiaries, 118 (0.31%), 99 (0.26%), and 175 (0.46%), respectively, experienced OPI overdose, BZD overdose, and OPI/BZD overdose in the subsequent 6 months after the 6-month trajectory measurement period (Appendix Table 8). As shown in Figure 3, compared with the Group (A): Very-low OPI-BZD (declining BZD) trajectory (crude rate: 1.3 per 10,000 person-months), there were substantially increased OPI overdose risks among individuals in the following five trajectories: Groups (E): Low OPI-High BZD (adjusted HR [aHR]=3.27, 95%CI=1.61–6.63); (F): Medium OPI-Low BZD (aHR=4.04, 95%CI=2.06–7.95); (G): Very-high OPI-High BZD (aHR=6.98, 95%CI=3.11–15.64); (H): Very-high OPI-Very-high BZD (aHR=4.41, 95%CI=1.51–12.85), and (I): Very-high OPI-Low BZD (aHR=6.50, 95%CI=3.15–13.42). The 5 high-risk OPI-BZD trajectories accounted for 32% of the overall cohort and captured 75% of the OPI overdoses.

Figure 3. Trajectories of Opioid and Benzodiazepine Utilization Patterns and Risk of Opioid Overdose, Benzodiazepine Overdose, and Opioid or Benzodiazepine Overdose among Medicare Beneficiaries (n=37,879).

Abbreviations: BZD, benzodiazepine; CI, confidence intervals; DME, diazepam milligram equivalent; HR, hazard ratio; MME, morphine milligram equivalent; OPI, opioid; SDD, standardized daily dose

To facilitate the labeling of opioid and benzodiazepine dose levels for each trajectory, we defined opioid dosage use as: very-low- (SDD <25 MME), low- (25–50 MME), moderate- (51–90 MME), high- (91–150 MME), and very-high-dose (>150 MME). Similarly, we defined BZD dosage use as very-low- (<10 DME), low- (10–20 DME), moderate- (21–40 DME), high- (41–60 DME), and very-high-dose (>60 DME).

Secondary outcomes: (1) BZD overdose; and (2) OPI or BZD overdose

As shown in Figure 3 and Appendix Table 8, the risks of BZD overdose and a composite outcome of OPI or BZD overdose were similar with the primary OPI overdose outcome findings. For example, compared with the Group (A): Very-low OPI-BZD (declining BZD) trajectory (crude rate for Group A: 0.9 per 10,000 person-months), the same five trajectories (Groups E to I) were associated with 1.1 to 3.7 times the BZD overdose risk.

Sensitivity analysis

The results from the sensitivity analyses that included beneficiaries with OPI or BZD overdose during the 6-month trajectory period were consistent with the main findings (Appendix Figure 3). Appendix Table 9 shows that simple OPI-BZD measures were not as effective as our trajectory analysis to identify beneficiaries at high risk of overdose. For example, 50.7% of beneficiaries with OPI-BZD overlapping ≥30 days had an adjusted HR of 1.56 (95%CI, 1.06–2.27) compared to those with OPI-BZD overlapping <30 days. Appendix Table 10 shows our findings’ robustness to potential influences of unmeasured confounders. E-values indicated that estimated HRs for trajectory groups with consistent very-high-OPI users (Groups G to I), Low OPI-High BZD (Group E), or Medium OPI-Low BZD (Group F) were robust to unmeasured confounders. For example, the observed OPI overdose risk (aHR=6.50) for very-high-dose OPI with low-dose BZD users (Group I) would require an unmeasured confounder associated with this trajectory group and OPI overdose with a HR of 5.74-fold each to negate the effect, which is a value beyond any currently measured confounders.

Discussion

We applied group-based multi-trajectory models to identify distinct patterns of concurrent OPI and BZD use most associated with overdose risk among fee-for-service Medicare beneficiaries initiating OPI prescriptions. We identified 9 distinct trajectories of OPI-BZD use during the 6-month period following OPI initiation. Trajectories characterized by either any very-high-dose OPI use (MME >150), any high-dose BZD use (DME>40), or medium-dose OPI with low-dose BZD use were associated with 3 to 7 times increased OPI overdose risks compared to the lowest-dose OPI-BZD trajectory. Some concurrent use levels (e.g., Very-low OPI-Medium BZD) posed much lower, or even no, additional overdose risk. Identifying OPI-BZD trajectory groups characterized by longitudinal dose and duration with distinct risk magnitudes provides more clinically actionable information than current approaches (e.g., examining any overlapping prescription) for detecting unsafe OPI-BZD use.

Our results add to the 8 previous studies suggesting concurrent OPI-BZD use is associated with increased adverse health-outcome risks (e.g., ED visit, inpatient admission for overdose) (3, 4, 9, 10, 15, 20–23). However, widely varying OPI-BZD definitions have been used in previous studies, including any current or former overlapping use (4); any overlapping use in the past 1 (10), 6 (20), or 12 months (9, 15); ≥7 consecutive overlapping days during a 1-year period (21); ≥30 cumulative overlapping days during a 1-year period (23); having any overdose diagnoses (3); or having different OPI-BZD duration exposures (e.g., 90 days) (22). In reality, treatment sequelae including dose and duration of concurrent use vary by patient characteristics and by presence and severity of comorbid conditions. Consequently, it may not be clinically appropriate to avoid co-prescribing in certain patients (e.g., co-occurring anxiety and chronic pain), even with known risks, and OPI-BZD use risk-benefit ratios will vary. Building on prior studies, our analyses applied group-based multi-trajectory models to identify dose and duration patterns of concurrent OPI-BZD use most often associated with OPI overdose risk, rather than using single gross measure (e.g., any overlapping use) over a fixed time period that may mask heterogeneity in concurrent use and corresponding risk.

The OPI-BZD use patterns we identified as being associated with much higher risk have important clinical implications. The current CDC OPI guidelines recommend clinicians avoid concurrent OPI-BZD use, without specifications on dose level (17). Our study was the first to identify OPI-BZD dose levels associated with elevated overdose risks, including any very-high OPI dose (>150 MME) or any high BZD dose (DME>40). Beneficiaries with medium-dose OPI (51–100 MME) with low-dose BZD use (10–20 DME) also had an increased risk, which is aligned with the CDC guideline indicating careful reassessment of the benefits and risks when considering increasing dosage to ≥50 MME/day (18). Notably, individuals with these high-risk OPI-BZD use patterns were more likely to receive their first OPI and BZD prescriptions from prescribers with the specialty of physical or rehabilitation compared to other patients. Patients seen by these specialists may benefit from education and interventions to promote awareness of potential risks associated with OPI-BZD use.

These treatment trajectory subgroups may be valuable for identifying individuals with the highest overdose risk to monitor and mitigate unsafe concurrent use. For example, those in the high-risk groups might be referred to case manager follow up. More restrictive interventions like pharmacy lock in programs might be able to be excluded from most (68%) OPI-BZD use groups with minimal or much lower overdose risks. Targeting the high-risk trajectories can capture over 70% individuals with overdose by focusing only on ~30% of the OPI-BZD users. Other interventions suggested for clinical practice include implementing auto-alert electronic health systems, and providing risk stratification and risk-informed monitoring of individuals with OPI-BZD use (58). Given the low overdose incidence, additional screening and assessment are needed to reduce false positives for overdose in high-risk trajectory subgroups. Although certainly not perfect, these high-risk OPI-BZD trajectories are more useful than simple criteria (e.g., overlapping ≥30 days) that have significantly more false-positives.

Our findings also have an important policy implication. Concurrent OPI-BZD measures included in current CMS OPI safety measures focus on the duration of OPI-BZD exposures. Application of this safety measure might disincentivize or discourage insurance carriers and clinicians from co-prescribing OPIs and BZDs for more than 30 days. However, our findings indicate that restricting the length of OPI-BZD exposure without considering levels of dose exposures may not effectively decrease OPI-related overdose risk. Therefore, designing performance measures or policy interventions for insurance carriers and clinicians should evolve to integrate dose and duration of OPI-BZD exposure in a longitudinal manner rather than at single time points. When initiation of OPI-BZD use is medically necessary, it is recommended to limit the dose and duration to the minimum required to achieve therapeutic goals (18).

This study has several limitations. First, our claims-based analyses have limited clinical and socio-behavioral information, such as BZD indications, pain severity, and relief with medication use that may influence OPI-BZD use. Second, we could not determine whether individuals took OPIs and BZD prescriptions as prescribed or whether they obtained additional medications through other payment types (e.g., cash) or sources (e.g., diversion). Although unmeasured confounders could not be ruled out, our E-value results showed that risk estimates for high-risk trajectory groups were robust to the potential presence of unmeasured confounders. Third, we were not able to link to death certificate data to capture fatal overdoses not receiving medical attention, nor could we distinguish fatal from non-fatal overdoses in our analysis. Therefore, our OPI/BZD overdose rates would be underestimated (e.g., 0.31% OPI overdose in our study vs. 0.47% from other studies including death certificates data) (59, 60). The low overdose incidence also led to wide 95% CI estimates. Fourth, the identified OPI-BZD trajectories represented average OPI-BZD dose use patterns over time. Although a prior study showed that group-average estimates generally represented most of the individual patient’s use patterns over time (61), some patients within a trajectory group may have different use patterns from the group-average estimates. Lastly, our findings have limited generalizability to Medicare beneficiaries with sporadic OPI-BZD use (i.e., only 1 prescription and <15 days’ supply) that are likely prescribed for acute conditions or to other populations (e.g., Medicaid). Nonetheless, our study showed substantial variability in the patterns of concurrent OPI-BZD use and associated patient characteristics, demonstrating an important nuance that is not reflected in the current guidance and policies. Our findings have relevance for Medicare patients given established and newly emerging quality measures around OPI-BZD use. Future research is warranted to evaluate these findings’ validity and generalizability using more recent data and other settings.

Conclusions

OPI overdose risk varied substantially by different OPI-BZD trajectory groups among fee-for-service Medicare beneficiaries. Concurrent use with either consistent OPI dose >150 MME (regardless of BZD dose levels), medium-dose OPI with low-dose BZD, or low-dose OPIs with >40 DME BZDs, was associated with the highest overdose risk, and concurrent use with lower dose ranges posed much lower, or even no, additional risk. These high-risk OPI-BZD trajectories may strengthen current OPI measures and guidelines to better guide OPI-BZD prescribing in clinical practice. When co-administration of OPIs and BZDs is medically necessary, clinicians should regularly monitor patient’s benefit-risk profiles for continuation of OPI-BZD use and adjust OPI and BZD doses accordingly. These trajectory models may have value in risk stratifying among patients with concurrent OPI-BZD use.