Mortality risk among individuals who smoke opioids compared with those who inject: A propensity score-matched cohort analysis of United States national treatment data

Abstract

Background and Aims:

Opioid smoking is becoming more common in the United States. The aim of this analysis was to estimate relative mortality risk among those who primarily smoke opioids compared with those who inject.

Design:

Retrospective propensity score-matched cohort analysis.

Setting:

2006–2021 US treatment episode data from SAMHSA TEDS-D.

Participants:

We matched 287 481 individuals in a substance use treatment program reporting smoking opioids as their primary substance use to an equal weighted number of individuals in a substance use treatment program reporting injecting opioids as their primary substance use. The majority of individuals reporting smoking were male (62.6%), 21–29 years old (47.9%), white (65.7%), independently housed (54.3%) and in the West Census Region (70.3%). Cohort characteristics were closely balanced after matching.

Measurements:

The outcome of interest was death during a treatment episode. Variables used for matching were year, opioid category, gender, race/ethnicity, age category, census region, housing status, employment status, number of prior treatment admissions, variables associated with opioid use severity (opioid use frequency, treatment setting intensity, age at first opioid use, use of medication-assisted treatment) and other reported substance use (methamphetamine, alcohol, benzodiazepines, cocaine).

Findings:

The mortality rate was 6.5 [95% confidence interval (CI) = 5.9–7.1] per 1000 person-years in the smoking cohort and 9.7 (95% CI = 8.8–10.8) per 1000 person-years in the injection cohort, with a mortality rate ratio of 0.67 (95% CI = 0.58–0.77).

Conclusions:

Among individuals in substance use treatment in the United States, those who usually smoke opioids appear to have a lower all-cause mortality risk than those who usually inject.

INTRODUCTION

Opioid smoking, previously thought to be rare in the United States (US), [1] has now surpassed injection as the most common route of administration implicated in opioid-related fatal overdoses [2]. Death counts, however, reflect both the underlying prevalence, as well as the specific risk, of each route of administration. Recent local studies [3–6] have documented widespread switching to opioid smoking among individuals who previously injected, whereas an analysis of national treatment admissions has found increases in opioid smoking prevalence that are especially large in the western US and a corresponding decrease in injection use [7]. The increasing prevalence of opioid smoking relative to injection may, therefore, drive rising opioid smoking-related deaths even if the mortality risk of opioid smoking is lower than that of injection.

Although rising deaths indicate that opioid smoking likely carries a substantial overdose risk, the risks of opioid smoking relative to injection use are poorly understood in the US context, especially with respect to mortality. As of yet, only one analysis—a single-state observational study—has explored the health repercussions of this recently identified trend in US opioid use patterns. The authors find evidence that, relative to injection, smoking is associated with a decrease in self-reported skin and soft tissue infections, hospitalizations and non-fatal overdoses [8]. There are no studies to our knowledge, however, that have estimated relative mortality. We analyzed US national treatment data to estimate the all-cause mortality risk among those who primarily smoke opioids compared with those who primarily inject.

METHODS

Data

Our data source was the Substance Abuse and Mental Health Services Administration’s Treatment Episode Data Set-Discharges (SAMHSA TEDS-D) from 2006 to 2021. This time period spans the multiple waves of the US opioid epidemic starting from its origins in rising prescription opioid-related deaths in the early 2000s followed by rising heroin deaths in 2010 and synthetic opioid deaths in 2013 [9].

Each observation in TEDS-D represents a single treatment period (admission through discharge) in an inpatient, residential, or outpatient substance use treatment program reported to state governments in the United States. The majority of US states require reporting from publicly funded substance use treatment facilities, and some states additionally require reporting from private facilities. Because each observation is a treatment episode, it is possible for a single individual to contribute multiple observations.

Collected variables include self-reported primary substance, usual route of administration, discharge disposition including death and treatment duration. Because treatment duration is recorded as a continuous variable until 30 days and then reported in intervals, we used the median of each interval.

Study population

Our study cohort included observations of individuals who, at the time of admission, self-reported opioids as their primary substance use and injecting or smoking as the primary reported usual route of opioid administration, excluding inpatient hospital admissions. Opioids included the TEDS categories of ‘heroin’ and ‘other opioids or synthetics’. TEDS does not discretely code fentanyl; however, for most of the study period fentanyl was considered an adulterant of heroin and, therefore, likely resulted in observations in the ‘heroin’ category [10]. As fentanyl replaced heroin in many locales, the probability that an individual would report fentanyl as their primary substance has likely increased, resulting in observations in the ‘other opiates and synthetics’ category [9].

Outcome

The outcome was death during treatment within 1 year of admission.

Exposure

The exposure was smoking or injection as the usual route of administration.

Matching

We conducted generalized full matching on propensity scores using the MatchIt package in R [11, 12]. We required exact matching on year, opioid category, age category, census region, number of prior treatment admissions, methamphetamine use and variables associated with opioid use severity including opioid use frequency and treatment setting intensity. Propensity for smoking was estimated from gender, race/ethnicity, other substance use (alcohol, benzodiazepines and cocaine) and additional factors associated with opioid use severity including housing, employment, age at first opioid use and use of medication-assisted treatment. Co-use of opioids with other substances is both common and may impact mortality, motivating the inclusion of additional substance use variables. We assessed the quality of the matched sample with absolute standardized mean differences.

Statistical analysis

We calculated mortality rates and rate ratios from a Poisson regression model with treatment duration as an offset to account for variable time under observation. We conducted several sensitivity analyses, including subgroup analyses stratified before or after 2014 (the inflection point in synthetic opioid—predominately fentanyl—overdoses). Because TEDS does not provide individual identifiers to allow episode linkage, we also conducted a sensitivity analysis limited to individuals with no prior admissions to eliminate the possibility of a single individual contributing more than one observation. Additional sensitivity analyses assessed robustness to differences in underlying use severity: one limited to those reporting daily use and one limited to residential or intensive outpatient treatment, rather than less intensive treatment. Statistical analysis was conducted with Stata version 18 (StataCorp) and R version 4.3.2. All analyses were conducted with matching weights. The primary research question and analysis plan were not publicly pre-registered and, therefore, the results should be considered exploratory.

RESULTS

From 2006 to 2021, there were 3 634 160 treatment episodes of individuals who reported opioid smoking (7.9%) or injecting (92.1%) as their primary substance use. Matching retained 99.6% of the smoking cohort. Cohort characteristics pre- and post-matching are provided in Table 1. The majority of matched individuals reporting smoking were male (62.6%), 21 to 29 years old (47.9%), white (65.7%), independently housed (54.3%) and in the West Census Region (70.3%). Cohort characteristics were closely balanced after matching.

TABLE 1.

Smoking and injection cohort characteristics before and after matching.

	Before matching		After matching
	Injection	Smoking	Injection	Smoking
n	3 345 474	288 686	287 481*	287 481
Year (mean)	2014.2	2015.2	2015.2	2015.2
Opioid category, %
Heroin	91.6	79.5	79.7	79.7
Other opioids and synthetics	8.4	20.5	20.3	20.3
Gender, %
Female	35.5	37.4	38.6	37.3
Male	64.5	62.6	61.4	62.6
Age, %
12–20	4.1	10.7	9.1	10.7
21–29	38.2	47.9	46.2	47.9
30–39	32.2	26.1	27.4	26.1
40–49	15.6	9.3	10.1	9.3
50 or older	9.8	6.0	7.2	6.0
Race/ethnicity, %
White, NH	74.5	65.7	64.0	65.7
Black, NH	6.3	6.4	5.2	6.4
Indigenous, NH	1.2	3.3	4.4	3.3
API, NH	0.4	1.8	1.5	1.8
Other, NH	2.5	3.9	4.2	3.9
Hispanic	14.4	17.3	18.9	17.3
Census region, %
Northeast	45.2	12.9	12.9	12.9
Midwest	15.0	8.5	8.5	8.5
South	19.8	8.3	8.3	8.3
West	19.8	70.3	70.3	70.3
Employment, %
Full-time	9.1	12.6	12.7	12.6
Part-time	5.2	7.4	7.9	7.4
Unemployed	42.7	40.9	39.6	41.0
Not in labor force	40.5	32.8	32.8	32.8
Housing, %
Homeless	17.8	14.4	14.4	14.4
Dependent living	17.6	24.3	24.3	24.3
Independent living	61.8	54.1	54.3	54.3
Age at first use, %
20 or younger	52.4	48.4	47.8	48.4
21–29	33.7	36.1	36.0	36.1
30 or older	13.2	14.9	15.6	15.0
Prior admissions, %
0	21.7	43.1	43.1	43.1
1 or more	71.6	53.3	53.5	53.5
Past month opioid use, %
None	19.4	21.8	21.8	21.8
Some	14.9	25.8	25.8	25.8
Daily	63.6	50.4	50.6	50.6
Medication assisted treatment
No	66.8	68.2	65.0	68.1
Yes	29.4	28.6	31.5	28.7
Treatment setting, %
Residential	48.2	38.9	38.9	38.9
Ambulatory, intensive	10.3	10.3	10.2	10.2
Ambulatory, non-intensive	41.5	50.8	50.9	50.9
Substances used, %
Alcohol	15.4	12.2	12.5	12.2
Cocaine/crack	27.1	11.2	10.5	11.2
Methamphetamine	11.1	24.8	24.8	24.8
Benzodiazepines	9.9	4.0	4.0	3.9

^*Unweighted controls after matching = 3,258,318. Generalized full matching weights controls by 1/N, where N is equal to the total number of controls matched to a specific exposed unit (for example, if an exposed unit was matched to 3 controls, each of those controls would receive a weight of 1/3).

Missing responses for race/ethnicity (<1.8%), housing status (<7.2%), age at first use (<0.7%), prior admission (<6.7%), use frequency (<2.1%), employment (<7%), and medication assisted treatment (<3.8%) are not presented to conserve space. Puerto Rico represented <0.2% of observations and has also been removed from the table. NH: non-Hispanic, API: Asian/Pacific Islander. Indigenous includes American Indian, Alaskan Native, and Native Hawaiian.

There were 415 deaths in the smoking and 699 in the injection cohorts during treatment periods. The mortality rate was 6.5 (95% CI = 5.9–7.1) per 1000 person-years in the smoking cohort and 9.7 (95% CI = 8.8–10.8) per 1000 person-years in the injection cohort, with a mortality rate ratio of 0.67 (95% CI = 0.58–0.77). All sensitivity analyses produced similar estimates of the mortality rate ratio between smoking and injecting (Table 2).

TABLE 2.

Smoking versus injection mortality rate ratio (MRR) in primary analysis and sensitivity analyses.

Model	1	2	3	4	5	6
MRR (95% CI)	0.67 (0.58, 0.77)	0.68 (0.52, 0.90)	0.66 (0.51, 0.84)	0.64 (0.51, 0.80)	0.68 (0.57, 0.81)	0.59 (0.45, 0.76)

Note: 1 = Primary model; 2 = intensive treatment setting only; 3 = daily use frequency only; 4 = year 2013 or earlier; 5 = year 2014 or later; 6 = no prior admissions.

DISCUSSION

In this propensity score-matched cohort analysis, individuals reporting opioid smoking experienced a lower all-cause mortality risk during treatment than those reporting injection. A combination of reduced risk of fatal overdose and a reduction of other injection-specific morbidity, including infectious complications, may mediate decreased mortality. Although mortality rates were less than 9.7 per 1000 person-years in both groups, there are more than 3.5 million individuals estimated to inject drugs in the United States, the majority of whom inject opioids [13].

Our results agree with a recent study among a California-wide cohort of individuals using fentanyl that found that the incidence risk ratio (IRR) for self-reported non-fatal overdose in the 3 months preceding interview was 1.41 (95% CI = 1.03–1.93) for those injecting compared with those who only smoked. This is equivalent to an IRR of 0.71 (95% CI = 0.52–0.97) if the comparison is reversed (i.e. comparing smoking vs. injection) and is similar to our estimated mortality rate ratio of 0.67 (95% CI = 0.58–0.77). Our results also match the perception among individuals using opioids—documented in multiple local US studies [3–6]—that smoking reduces the risk of overdose relative to injection. Field observations [3, 4] have documented numerous techniques facilitated by smoking that may reduce overdose risk that are difficult to implement when injecting. These include starting with smaller initial doses with further titration based on observed response as well as splitting doses across administrations over the course of several hours rather than the administration of a large amount in a single bolus dose.

Our results add to the evidence that opioid smoking may be associated with reduced harm relative to injection use, including reduced mortality risk. Further studies are needed to confirm our findings as well as detail potential mechanisms of risk reduction. Additional research should also investigate the acceptability and feasibility, as well as the broader public health implications, of the substitution of smoking for injection among individuals already injecting opioids. Such ‘reverse transitions’ have previously been documented to occur organically [14] as appears to be occurring currently in the United States, predominantly in its western region. The recognition of injection-specific health risks has also previously prompted several European initiatives to encourage a switch from injection to smoking—including in The Netherlands, England and Germany—that could guide similar efforts in the United States [15]. Further research is also needed on the specific morbidities associated with opioid smoking, which has previously been linked to increased chronic respiratory disease among other potential health risks [16].

Relative risks notwithstanding, the rising number of opioid smoking-related overdoses in the United States indicate that opioid smoking likely carries a substantial overdose risk. Even if smoking confers some overdose risk reduction relative to injection use, overestimating the safety of smoking could induce individuals to lower their vigilance against overdose, offsetting or reversing any potential risk reduction. It is critically important, therefore, that clinicians and harm reduction workers assess the route of administration preferences of those they serve, their perceptions regarding associated relative risks and how these perceptions affect their drug use practices. Individuals should be counseled that although smoking may reduce harms relative to injection, opioid smoking is not safe. Furthermore, if perceptions of safety reduce other harm-reduction practices, such as carrying naloxone and avoiding using alone—both well-established protective factors against overdose—risks could paradoxically increase. Individuals should also be warned that sharing smoking equipment can lead to inadvertent fentanyl resin exposure and overdose, especially in opioid naïve individuals [4, 17]. Harm reduction organizations should be supported in their efforts to provide safe smoking supplies to mitigate these and other harms.

Limitations of our analysis include potential unmeasured confounding, lack of specific cause of death, lack of discrete coding for fentanyl use and generalizability beyond individuals in treatment. Different patterns of substance co-use may contribute to differential mortality risk; however, our sample was well matched on the co-use of several substances commonly used with opioids. There might be other characteristics of individuals who usually smoke or inject that were not balanced through matching, including use severity. To minimize this possibility, we balanced cohorts across several factors associated with severity and conducted sensitivity analyses limited to those who likely have the most severe use based on opioid use frequency or treatment setting intensity. Differences in socio-economic status could also contribute to mortality differences. Although we balanced the cohorts across housing and employment status, it is possible that there are residual differences in socio-economic status that we were unable to account for.

Our study has additional limitations. Route of administration was reported at admission only; however, crossover between smoking and injection would likely bias our estimate toward the null by making the cohorts more similar. Some deaths may have been misclassified as treatment drop out, leading to an underestimation of mortality. The close balance achieved across study variables, however, reduces the risk for important differential misclassification. Additionally, because treatment is associated with reduced mortality, our estimates likely underestimate the mortality risk in the untreated population. Our analysis does not account for a theoretical differential interaction between treatment and route of administration preference.

CONCLUSION

Individuals that report opioid smoking appear to have a lower risk of death compared to those injecting. Nevertheless, the rising number of opioid smoking-related overdoses in the United States suggests that smoking still carries a substantial overdose risk. Additional research is needed regarding the harm reduction potential of route of administration transition interventions in the United States. Individuals who elect to smoke opioids should be cautioned regarding the persistent overdose risk, the continued importance of taking precautions against overdose including carrying naloxone and avoiding using alone, and the dangers of sharing smoking equipment.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are openly available at https://www.samhsa.gov/data/data-we-collect/teds-treatment-episode-data-set.