Longitudinal patterns of use of stimulants and opioids in the AIDS Linked to the IntraVenous Experience cohort, 2005–2019

Abstract

Background.

Overdoses involving opioids and stimulants are on the rise, yet few studies have examined longitudinal trends in use of both substances. We sought to describe use and co-use of opioids and stimulants, 2005–2019, in the AIDS Linked to the Intravenous Experience (ALIVE) cohort – a community-based cohort of people with a history of injection drug use living in or near Baltimore, MD.

Methods.

We included 2,083 ALIVE participants, who had at least two visits during the study period. Our outcome was based on self-reported use of opioids and stimulants in the prior 6 months. We estimated prevalence of 4 categories of use (neither stimulants nor opioids, only stimulants, only opioids, stimulants and opioids), using a non-parametric multi-state model, accounting for the competing event of death and weighting for informative loss to follow-up. All analyses were stratified by enrollment cohort, with the main analysis including participants who enrolled prior to 2015 and a sub-analysis including participants who enrolled 2015–2018.

Results.

In the main analysis, prevalence of using stimulants and opioids decreased from 38% in 2005 to 12% 2013 but stabilized from 2014 onwards (13–19%). The prevalence of using only stimulants (7–11%) and only opioids (5–10%) was stable across time. Participants who reported using both were more likely to report homelessness, depression, and other substance use (e.g., marijuana and heavy alcohol use) than participants in the other use categories. On average, 65% of visits with use of both were followed by a subsequent visit with use of both; of participants transitioning out of using both, 13% transitioned to using neither.

Conclusions.

While use of stimulants and opioids declined in the cohort through 2013, a meaningful proportion of participants persistently used both. More research is needed to understand and develop strategies to mitigate harms associated with persistent use of both stimulants and opioids.

INTRODUCTION

Overdose is the leading cause of death for people who inject drugs globally, with rising rates over the last twenty years in many regions including North America.^1,2 In the United States (US), overdose mortality has also been escalating, particularly in the past decade due to synthetic opioids other than methadone (e.g., fentanyl) and due to an increasing number of overdoses involving both opioids and stimulants.^3–10 Persons with opioid use disorder commonly engage in the use of multiple (illicit) substances, including stimulants such as cocaine and methamphetamine.^3,11–14 There are several potential reasons an individual might intentionally use more than one substance. Individuals may seek to experience the effects of both drugs simultaneously, to experience a more potent drug effect than would be experienced when using the drugs separately, to reduce feelings of withdrawal, or to self-medicate for mental or physical health conditions.^{3,4,14,28–30}

Use of multiple substances is also associated with increased rates of non-fatal overdose,^11,13–15 HIV-related risk behaviors, poor mental health, incarceration, homelessness, and unemployment.^13,16–21 Despite these trends, most interventions to prevent overdose focus on opioid use disorder and may be less effective in the context of use of opioids with other substances.³ Individuals who use multiple substances tend to have worse initiation of and retention in opioid use treatment programs and higher rates of relapse, and naloxone is less effective for reversing overdose among those who use other substances in addition to opioids.^3,4,6,16,22

Given the rise in overdose rates and challenges with effective interventions in the face of polysubstance use, additional research is needed to understand how use of multiple substances has changed over time. Current evidence on use of multiple substances mostly derives from cross-sectional data, general population samples, or treatment-based cohorts.^{5,6,8,11–13,15,18,20,23} Cross-sectional studies on substance use fail to capture how frequency of substance use, drug preferences, and drug markets change over time in the community. Additionally, findings from studies conducted in general population and treatment-based samples may not reflect substance use as it occurs in localized communities of people who use drugs. Studies conducted in general population samples report a lower prevalence of use of multiple substances than is found among people who use drugs.^12,24 Moreover, the subset of individuals in these samples who report substance use may not be representative of the community of people who use drugs due to the exclusion of individuals who are incarcerated or experiencing homelessness and due to differential participation by drug use.^25,26 Treatment-based samples are a select group of individuals who elect to start treatment, making them distinct from individuals in the community who do not engage (or inconsistently engage) in drug treatment.²⁷

Here, we describe use and co-use of opioids and stimulants in the period 2005–2019 in the AIDS Linked to the IntraVenous Experience (ALIVE) cohort, a community-based study of people with a history of injection drug use (IDU), living in or near Baltimore, Maryland. We focus specifically on the longitudinal patterns of use and co-use of opioids and stimulants, as the combination of heroin and cocaine has historically been common in this cohort.²⁴

METHODS

Study sample

Enrollment into the ALIVE cohort began in 1988, with subsequent enrollment waves in 1994–1995, 1998, 2005–2008, and 2015–2018. Participants attend bi-annual study visits, at which they respond to survey questions regarding their substance use, drug treatment, comorbidities, quality of life, and other lifestyle factors; provide clinical measurements; and receive testing for HIV/HCV. The Johns Hopkins Bloomberg School of Public Health institutional review board approved the study, and all participants provide written informed consent.

Participants were eligible for inclusion in this analysis if they had more than one study visit between July 1, 2005 and December 31, 2019 (n=2,634). We ended follow-up in 2019 due to the disruption in regular study visits caused by the COVID-19 pandemic. We partitioned follow-up into 29 visit intervals, each spanning 6 months between either January and June or July and December of a given year. If a participant contributed two study visits during any visit interval, we summarized across the visits by taking the maximum value for all variables (all variables were continuous, binary, or ordered categorical). Figure S1 and Table S1 describe selection of the study sample and loss to follow-up. The 551 participants who were excluded due to contributing only one visit interval to the analysis were younger, less likely to report Black race and have HIV, and more likely to report using illicit substances, experiences of homelessness, and depressive symptoms than participants who were included (Table S2).

Measures

We defined any illicit opioid use based on use of heroin via any route of administration (i.e., injection or smoking); use of speedball (injection of heroin and cocaine together); and non-medical use of prescription opioids. Non-medical use of prescription opioids was measured in different ways across the study period. From 2005–2013, participants were asked about methadone, buprenorphine, oxycontin, or percocet “purchased on the street taken orally.” From 2014–2019, participants were asked about injection of painkillers (specifically, oxycontin, percocet, codeine, darvon, percodan, dilaudid, demerol, or buprenorphine) or painkillers (same list with the addition of methadone) obtained from any source other than a doctor or program prescription taken orally. This definition did not include self-reported use of fentanyl, which was not collected by the study until 2017 and was subsequently not consistently collected. We defined any illicit stimulant use based on use of crack, cocaine via any route of administration, speedball, methamphetamine (2005–2013: crystal methamphetamine via injection; 2014–2019: crystal methamphetamine via injection or taken orally, methadrine obtained from any source other than a doctor or program taken orally), and other stimulants (2014–2019: preludin, benzedrine, uppers, speed, ritalin, dexedrine, or adderall obtained from any source other than a doctor or program taken orally). Unless otherwise specified, all measures were based on participant self-report of the last 6 months.

In each visit interval, we defined a 4-level outcome: use of neither stimulants no opioids, use of only stimulants, use of only opioids, or use of both. Use of both stimulants and opioids was defined as use of speedball (representing simultaneous co-use) or reporting both opioid use and stimulant use (representing either simultaneous or sequential use within the time period). Note that use of “only stimulants” means that an individual did not report use of opioids; they could have used other substance types. We also assessed mortality as a competing event. Date and cause of death was obtained through linkage to the National Death Index. Deaths were classified as being overdose-related if the underlying cause had an International Classification of Diseases (ICD) 9^th or 10^th edition code related to non-alcohol poisonings (i.e., excluding ICD-10 codes F101, F109, X45).^31,32

We examined a number of participant characteristics to describe our study sample or to control for potential informative loss-to-follow-up, selected based on background knowledge of the ALIVE cohort. Time-fixed characteristics measured at enrollment into the study included sex (female/male), race (Black/Other), completion of a high school education (yes/no), and marital history (ever/never). Characteristics measured at each study visit included age (continuous), HIV status (yes/no), frequency of injection drug use (none/less than daily/daily), current cigarette smoking (yes/no), current alcohol use (none/moderate/severe, based on the Alcohol Use Disorders Identification Test),³³ marijuana use (yes/no), any medication for opioid use disorder (MOUD; yes/no, based on having a prescription for methadone, buprenorphine, or naltrexone), residence within Baltimore city limits (yes/no), employment (yes/no), experience of homelessness (yes/no), 6-month income <$5,000 (yes/no), incarceration for 1 week or more (yes/no), seeing the same doctor 90% of the time (yes/no), depressive symptoms (yes/no, indicated by a score of ≥23 on the Center for Epidemiological Studies-Depression questionnaire), diagnosis of anxiety or depression (ever/never), and severity of general body pain (none/very mild/mild/moderate/severe/very severe, from the Medical Outcome Study-HIV questionnaire).

Once the data were set up into the 29 visit intervals, we used random forest imputation to impute missing values in the outcomes and covariates, using the missForest package in R.³⁴ The imputation model included any opioid use, any stimulant use, and all covariates above. We repeated the imputation process 5 times; all point estimates and standard errors were summarized across the imputation data sets.³⁵ We further handled missing data by, as described below, censoring participants when they missed 2 or more consecutive visit intervals. Further details on the missing data are provided in the Appendix.

Statistical analyses

For all analyses, we stratified our sample by enrollment cohort. In our main analysis, we included participants who enrolled prior to 2015. We then repeated the analysis among participants who enrolled 2015–2018, due to the shorter follow-up time and known differences in cohort characteristics.³⁶

We first characterized the study sample by comparing the distribution of participant characteristics by reported co-use across all visit intervals. For continuous variables, we estimated the median and interquartile range (IQR); for categorical variables, we estimated proportions.

We then estimated prevalence of each level of the outcome in each visit interval using a non-parametric, multi-state model (using discrete time and a counting process format).^37,38 This approach is an extension of survival methods for competing events, extended to allow individuals to move into and out of states more than once over time. The possible states and state transitions are visualized in Figure S2. We accounted for the competing events of overdose-related and non-overdose-related mortality by including both types of death as absorbing states in the model (i.e., once a participant died they left the sample and could not return). The visit interval in which a participant died was classified as being in the “dead” state, even if a visit occurred in the same visit interval as the death. Participants were allowed late entry, if their first observed visit occurred after the first visit interval. Participants were classified as right censored if they had no study visits for 2 or more consecutive visit intervals. The visit interval that followed the last interval where a participant was observed was given the censored state. We did not allow participants to return to the analysis once they were censored.

In the crude model, we assume that the experience of those who remained under observation reflected what would have been seen from those who were right censored, i.e. that loss-to-follow-up was random. This assumption is unlikely to hold in practice; thus, we constructed inverse probability of censoring weights to account for the possibility of non-random right censoring by controlling for variables potentially related to both censoring and co-use. The construction of these weights is described in the Appendix. We then repeated the analysis, applying the weights to the multi-state model. For the weighted model, we obtained 95% confidence intervals using the standard deviation from 500 bootstrap resamples.

In addition to prevalence estimates, we also estimated crude and weighted transition probabilities between each state from one visit interval to the next, defined as the probability of being in state Y during visit interval t, conditional on having been in state X during visit interval t-1. To smooth across small cells, we summarized these probabilities annually by taking the average of the visit interval transition probabilities measured in each year. Given our particular interest in the use of both stimulants and opioids, we compared participant characteristics among those who remained using both to those who transitioned out of using both or died.

We conducted three sensitivity analyses. First, we redefined our outcome to examine only use of heroin and cocaine, excluding use of other opioids and stimulants. Second, we examined an outcome defined solely by self-reported speedball use, as our only measure of simultaneous co-use of opioids and stimulants. These analyses were done in part because use of speedball, heroin and cocaine was measured consistently across the full study period and because of the differences in risk conveyed by simultaneous use relative to use of both substances within a 6-month period. The ALIVE questionnaires were restructured in 2014, resulting in changes to how the substance use variables were collected. Data on non-prescription painkillers and non-prescription stimulants were only collected after 2014, while data on non-medical methadone were only collected 2005–2013. Third, we assessed whether results of the main analysis would change if we censored participants upon missing 4 consecutive visit intervals, rather than only two.

All analyses were run using R version 4.3.1 (The R Foundation, Vienna, Austria).

RESULTS

We analyzed data from 1,615 study participants, contributing 21,676 visit intervals, who enrolled in the cohort prior to 2015. The median number of visit intervals was 11 (IQR: 4, 23). Participants who reported use of stimulants and opioids were marginally younger than those reporting single use or use of neither (median age: 49.7 years; Table 1). Visit intervals with use of both had the highest prevalence of daily injection (39.9%), heavy alcohol use (43.8%), marijuana use (29.7%), homelessness (22.2%), incarceration (12.2%), and depressive symptoms (33.7%) and the lowest prevalence of current employment (11.8%). Participants reporting use of neither opioids nor stimulants tended to be older (median age: 53.9 years) and less socioeconomically disadvantaged. Visit intervals with use of neither had the lowest prevalence of other substance use and depressive symptoms. Participants reporting only opioid use were the least likely to be female (28.9%) or to have HIV (19.5%). Compared to participants reporting only opioids, participants reporting only stimulants were more likely to report care provider consistency (82.3% vs. 72.6%). Participants reporting only stimulants also had the highest prevalence of MOUD (39%). Otherwise, participants using either stimulants or opioids (but not both) were similar and fell between the ranges of those using both and those using neither in terms of other substance use and socioeconomic status.

Table 1.

Distribution of participant characteristics across all visit intervals and averaged across imputation data sets, among participants who enrolled prior to 2015

Variable	Overall (v=21,408)	Neither (v=12,300)	Stimulants Only (v=1,924)	Opioids Only (v=1,815)	Co-use (v=5,368)
Female sex, %	34.2	34.0	37.1	28.9	35.5
Black race, %	91.1	93.3	90.4	86.6	87.9
Age, median (IQR)	52.6 (46.9, 57.8)	53.9 (48.6, 59.0)	52.4 (46.9, 57.8)	51.5 (46.1, 56.8)	49.7 (43.4, 55)
Enrolled 2005–2008, %	38.1	28.4	41.2	49.8	55.4
HIV, %	30.9	33.7	34.4	19.5	27.2
Frequency of injection,a %
None	70.9	97.5	76.4	41.5	17.8
< Daily	15.8	1.7	17.5	31.2	42.4
Daily	13.3	0.8	6.0	27.3	39.9
Cigarette smoking,a %	78.3	70.4	86.9	85.2	91.0
Alcohol use,a %
None	51.2	67.0	30.4	33.1	28.5
Moderate	21.9	16.6	32.1	30.0	27.7
Heavy	26.9	16.4	37.5	36.9	43.8
Marijuana use,a %	16.3	8.5	24.3	20.4	29.7
Any MOUD,a %	28.7	27.5	36.0	26.5	29.6
Never married, %	65.1	64.9	60.4	69.4	65.9
High school education, %	41.8	41.7	44.7	44.1	40.1
Lives outside city,a %	9.7	9.9	8.0	11.3	9.2
Employed,a %	18.9	23.3	14.2	14.9	11.8
Income <$5,000,a %	71.8	67.6	73.5	76.2	79.2
Any homelessness,a %	10.5	4.8	11.8	12.9	22.2
Incarcerated ≥1 week,a %	5.5	2.3	5.5	7.4	12.2
Consistent provider,a %	79.0	84.4	82.3	72.6	67.6
Depressive symptoms,a %	21.7	15.4	25.2	25.0	33.7
Ever depression/anxiety, %	45.2	41.1	52.0	46.1	51.6
General body pain,a %
None	25.3	27.8	23.9	22.1	21.0
Very mild	25.1	25.4	21.7	26.6	25.2
Mild	14.3	13.8	17.3	15.6	13.7
Moderate	23.8	22.1	24.7	25.2	27.0
Severe/very severe	11.6	10.9	12.4	10.6	13.1

Abbreviations: IQR, interquartile range; MOUD, medication for opioid use disorder; v, number of person-visits

^aIn the past 6 months

In the weighted analysis (Figure 1, Table S3), the prevalence of use of stimulants and opioids decreased from 39% in July-December 2005 to a low of 13% in January-June 2014. From 2014 onwards, the prevalence of use of both stabilized, with a range of 13–19%. The prevalence of using only stimulants and using only opioids was steady across calendar time, with a range of 6–10% and 6–10%, respectively. Overall cumulative incidence of overdose-related and non-overdose related mortality was 4% and 10% by the end of follow-up. See Figure S3 for the crude results.

Figure 1.

Weighted prevalence of the outcome defined by use of stimulants and opioids and incidence of the competing event of death across visit intervals, among participants who enrolled prior to 2015

In general, participants reporting use of stimulants and opioids were mostly likely to remain using both in the next visit interval (Figure 2), with an average transition probability across follow-up of 65%. Among participants who transitioned out of using both, most transitioned into using neither stimulants nor opioids (13%). The average probability across follow-up of transitioning into using stimulants alone (9%) was marginally higher than the average probability of transitioning into using opioids alone (7%). Compared to participants who remained using both (Table 2), participants who transitioned into using neither stimulants nor opioids in the next visit interval were less likely to report heavy alcohol use (34% vs. 46%), marijuana use (21% vs. 31%), and depressive symptoms (26% vs. 35%) and more likely to report provider consistency (75% vs. 65%).

Figure 2.

Weighted probability of transitioning from the use of both stimulants and opioids state by calendar year, among participants who enrolled prior to 2015

Table 2.

Characteristics of participants reporting co-use averaged across imputation data sets, stratified by their state during the subsequent visit interval

Variable	Neither (v=680)	Stimulants Only (v=427)	Opioids Only (v=377)	Co-use (v=3,427)	Overdose Mortality (v=19)	Non-Overdose Mortality (v=46)	Drop Out (v=318)
Female sex, %	33.0	40.6	34.4	35.6	26.3	25.4	34.2
Black race, %	90.2	87.6	83.1	88.9	84.2	91.2	77.1
Age, median (IQR)	49.4 (44.2, 55.9)	50.3 (44.3, 55.3)	49.2 (42.1, 54.6)	49.9 (43.4, 54.8)	54.4 (44.2, 57.4)	53.3 (47.8, 56.6)	46.6 (40.4, 52.1)
Enrolled 2005–2008, %	49.3	54.1	60.0	55.5	52.6	49.6	65.2
HIV, %	34.1	35.8	20.5	24.9	36.8	57.5	29.5
Frequency of injection,a %
None	15.7	33.3	15.0	16.9	15.8	11.8	12.2
< Daily	49.2	45.8	38.8	40.6	42.1	55.7	45.1
Daily	35.1	20.9	46.2	42.5	42.1	32.5	42.7
Cigarette smoking,a %	86.5	93.4	91.1	91.3	94.7	96.1	94.3
Alcohol use,a %
None	37.5	28.3	27.7	27.0	22.1	39.4	24.4
Moderate	28.4	33.8	26.5	26.5	31.6	33.8	27.5
Heavy	34.1	37.9	45.8	46.4	46.3	26.8	48.1
Marijuana use,a %	21.2	30.9	28.5	31.2	36.8	28.5	32.9
Any MOUD,a %	32.2	40.0	21.9	28.4	26.3	32.5	24.6
Never married, %	67.5	63.6	70.8	65.1	78.9	59.6	67.9
High school education, %	40.3	45.5	43.3	38.9	42.1	30.7	42.8
Lives outside city,a %	9.2	9.2	9.7	8.7	10.5	3.9	15.3
Employed,a %	11.9	12.1	11.3	11.7	0.0	15.4	13.4
Income <$5,000,a %	78.0	75.8	81.9	79.7	83.2	70.2	80.4
Any homelessness,a %	22.1	16.3	22.4	21.8	30.5	26.3	36.4
Incarcerated ≥1 week,a %	11.8	7.5	14.9	11.9	15.8	11.2	21.9
Consistent provider,a %	74.6	78.1	66.4	65.1	63.2	88.2	60.9
Depressive symptoms,a %	26.3	31.9	33.0	34.8	36.8	31.6	38.1
Ever depression/anxiety, %	52.9	53.7	49.6	50.5	68.4	48.3	57.9
General body pain,a %
None	25.4	20.6	22.2	20.4	15.8	17.1	21.7
Very mild	25.2	24.4	25.7	26.0	21.1	12.7	19.0
Mild	12.9	14.5	11.7	13.5	15.8	21.9	16.0
Moderate	23.0	26.5	26.7	27.6	21.1	25.0	28.3
Severe/very severe	13.4	14.0	13.8	12.4	26.3	23.2	15.1

Abbreviations: IQR, interquartile range; MOUD, medication for opioid use disorder; v, number of person-visits

^aIn the past 6 months

Prior to 2014, the probability of overdose-related death was relatively stable regardless of prior use (Figure 3A), with similar sized peaks for use of both stimulants and opioids and single use of stimulants or opioids. The probability was lower in those who used neither substance. In 2015, there was a peak in mortality (3%) from those who reported use of both. A smaller peak was seen for those transitioning from using only opioids in 2014–2015 (2%) and among those using only stimulants in 2017–2019 (1–2%). While those reporting use of neither stimulants nor opioids generally had the lowest probability of dying due to overdose, we observed a slight rise in the probability of dying due to an overdose in this group between 2015–2017. When looking at non-overdose-related mortality (Figure 3B), we saw a peak in mortality between 2012–2013 (2–3%) among participants using only opioids. The annual transition probabilities were relatively stable for all other groups. The weighted annual transition probabilities for all other states are summarized in Figures S4-7.

Figure 3.

The weighted probability of transitioning into the (A) overdose-related mortality and (B) non-overdose-related mortality state by calendar year, among participants who enrolled prior to 2015. Note that the scale of the y-axes differ by panel, to better highlight trends over time.

In the sensitivity analysis with the outcome defined based on use of cocaine and heroin, the prevalence estimates were nearly identical to the main analysis (Figure S8). Use of speedball also followed the overall trend in use of stimulants and opioids, with an initial decrease followed by a stabilizing of the prevalence (Figure S9). We also noted that the proportion of overall use of both stimulants and opioids involving speedball decreased, from 73% in 2005 to 40% in 2019. Results were nearly identical to the main analysis when we censored participants if they missed 4 consecutive visit intervals (Table S4).

The results for 2015–2018 enrollees are presented in Appendix 3. These participants were substantially different from earlier enrollees in terms of demographics and substance use behaviors (Table S5). Use of both stimulants and opioids was more common, with a prevalence of 85% in 2015 and 46% in 2019 (Figure S10).

DISCUSSION

To our knowledge, this study is among the first longitudinal analyses of the trends in use and co-use of stimulants and opioids among people with a history of IDU. Prevalence of use of both substances in this community-based, urban cohort initially decreased across calendar time, with some stabilization after 2014 among earlier recruited participants. We further saw higher use of both among participants who enrolled after 2015. In contrast, the prevalence of single use of either substance remained constant across follow-up. These trends were seen even after accounting for the competing event of death and informative loss to follow-up.

Previous studies in community-based cohorts have reported that the prevalence of injection drug use decreases with length of time enrolled in the study.^39–41 In this analysis, while the prevalence of substance use overall and particularly use of both substances declined through 2013, it plateaued from 2014 onwards among pre-2015 enrollees. This plateau may represent a natural trend that occurs in drug use as individuals age out of use and may partially reflect some effect of being enrolled in a research study. However, it is important to highlight that even while some aged out of using both substances, there remained a non-trivial subset of the population who persistently used both. Such a subgroup has been previously reported in the context of longitudinal heroin use and overall injection drug use.^39–42

It is further notable that the group using stimulants and opioids experienced higher levels of other social and structural challenges including homelessness, low employment, and incarceration. Prior work has similarly found an association between polysubstance use and these factors.¹³ Moreover, literature suggests that social and economic disadvantage can in turn impact drug use practices, for example by limiting the substances an individual can afford or by decreasing access to treatment services.⁴³ The intersection of (poly)substance use, social and economic disadvantage, and stigma poses serious challenges for substance use treatment. These challenges reinforce the idea that effective treatment interventions will need to target the use of more than one substance, as well as upstream structural factors.^44,45

It is worth noting that the plateau in use of stimulants and opioids was concurrent with the emergence of fentanyl in Baltimore, MD in late 2013, although we cannot say definitively whether the two trends were related.⁴⁶ Qualitative work has reported an increase in the use of stimulants to directly counteract fentanyl’s potency;⁴⁷ fentanyl has also been associated with increased frequency of injection and other injection risk practices.^47–50 Studies examining whether fentanyl is associated with worse drug treatment outcomes have largely found no difference when compared to individuals using other opioids, although some studies have reported more severe withdrawal symptoms when using buprenorphine after fentanyl.^50–52

This analysis also provides novel insight on transitions from using both stimulants and opioids to using one or neither substance. While most visits with use of both were followed by a subsequent visit with use of both, a sizable proportion of participants using both transitioned to using neither (13%). These visits were characterized by more stable engagement in health care (i.e., seeing the same provider) compared to visits where participants remained using both (75% vs. 65%) and lower reporting of depressive symptoms (26% vs. 35%). This underscores the importance of ensuring high retention in care to address substance use and comorbidities (e.g., depression), as well as integration of care for mental health and substance use disorders. It is further worth noting that the prevalence of using alcohol and marijuana was much lower among participants who were using neither stimulants nor opioids, indicating that most participants who ceased using stimulants and opioids did not substitute these illicit substances with other, licit substances. Additionally, participants transitioning from using both into using neither stimulants nor opioids had a higher prevalence of self-reported MOUD (32%) compared to those who remained using both (28%) and to all visits where use of neither was reported (27%). A similar trend was observed for transitions from using both into using only stimulants (40% prior to transition vs. 36% among all visits with only stimulant use). While people engaging in use of more than one substance tend to have worse drug treatment initiation and outcomes, these findings suggest that at least some ALIVE participants who were using stimulants and opioids were being linked to MOUD and either stopping use of opioids or of both substances.^4,6,16,22

We saw that overdose-related mortality increased in all groups after 2013, with the largest spike occurring among those using stimulants and opioids. This trend was seen despite no increase in the prevalence of self-reported stimulant and opioid use during this time period. An increase in the rate of drug-related mortality among ALIVE participants, 2010–2015, has been previously reported.³¹ The increase in mortality observed in our study coincided with the third wave of overdose epidemic, which began with the increase in fentanyl-related overdoses in 2013.^46,53 Synthetic opioids like fentanyl now contribute to a larger proportion of overdose deaths than other opioids, including prescription opioids and heroin.⁵ While the ALIVE study lacked consistent data on fentanyl use during this time period, it is likely that many of those reporting use of heroin in this analysis were also either intentionally or unintentionally using fentanyl.^5,7,8,54,55 We further saw a rise in overdose-related mortality, 2017–2019, among participants using only stimulants. This could be related to the national rise in overdoses involving stimulants, as well as potential adulteration of stimulants with fentanyl.^8,10

Our study had several limitations. First, our outcome was defined based on self-reported use of stimulants and opioids in the last 6 months. In addition to the potential for incomplete disclosure, this would result in us missing any unintentional substance use, such as might occur due to adulteration of the stimulant supply with fentanyl. We also expect that the drug use practices of those classified as using both stimulants and opioids could be quite heterogeneous, with some individuals using both substances simultaneously, sequentially, or potentially weeks apart. Nevertheless, the overall trends were similar when examining specifically co-use via speedball, and we saw that a large proportion of use of stimulants and opioids involved speedball, especially at the start of the study period. Second, we lacked consistent data on intentional fentanyl use.⁴⁹ Not only does this impact our ability to assess the impact of fentanyl on trends in this period, but our estimate of the prevalence of opioid use was likely an underestimate during the period after fentanyl became the primary illicit opioid on the market and after participants would have been able to know whether they were using fentanyl (i.e., after the distribution of fentanyl test strips).⁵⁶ Indeed, the slight increase in overdose-related mortality, 2015–2017, among individuals reporting use of neither stimulants nor opioids may be evidence of misclassification due to incomplete disclosure or lack of data on fentanyl. Third, there may have been misclassification of cause of death. We noted an increase in non-overdose-related mortality among those using only opioids, 2012–2014. We have no reason to expect such a spike, and it is possible that these were misclassified fatal overdoses. Fourth, our data only covered the pre-COVID period. There was a disruption in study visits during the pandemic, and participants who returned to the study were substantially different from those engaged prior to the pandemic. Thus, we cannot speak to how the pandemic may have impacted use and co-use of stimulants and opioids in our cohort.

Finally, our results pertain to the context of the ALIVE study, which has a participant sample that is predominately urban, non-Hispanic Black, male, and middle aged. Our findings may thus not be applicable to other settings and communities of people who use drugs. Furthermore, our findings may not be fully representative of the experiences of people who use drugs in Baltimore who are not enrolled in ALIVE but are otherwise similar, due to the impacts of long-term enrollment in a cohort study.

Despite these limitations, our study is among the few to examine trends in use of multiple substances longitudinally.^16,57–59 In general, these studies have similarly reported a decrease in the use of some substances over time, although the different time periods, locations, and definitions of polysubstance use make direct comparisons challenging. These studies have also reported that participants using stimulants and opioids were more likely to have experienced homelessness and incarceration, compared to participants with low or no substance use.⁵⁸

Here, we documented the prevalence of and transitions in use of stimulants and opioids in a community-based sample of adults with a history of IDU, across 15 years of data that spanned the pre- and post-fentanyl periods. We saw that use of stimulants and opioids largely decreased over time; however, there remained a persistent group comprising 13–19% of the pre-2015 enrollees that continued to use both substances. This is concerning given we observed that those who were using both tended to have a higher prevalence of other substance use, experiences of homelessness and incarceration, and current depressive symptoms and that this group tended to have the highest probability of fatal overdose in the post-fentanyl period. More research needs to be done to understand transitions in and out of use of multiple substances at the individual level and how we can successfully engage individuals using more than one substance in MOUD or other overdose prevention services.