Highlights
-
•
Four distinct stimulant use trajectories were identified over 12 months.
-
•
High stimulant use was associated with recent overdose and being unhoused.
-
•
Powder cocaine showed a unique declining pattern of use.
-
•
Diagnosed psychosis was associated with more days of recent stimulant use (p < 0.05).
-
•
Findings may inform the design of tailored harm reduction interventions.
Keywords: MeSH, Stimulants, Trajectory, Longitudinal Studies, Drug Overdose, Harm Reduction, United States
Abstract
Overdoses in the United States are increasingly driven by co-occurring stimulant and fentanyl use, yet limited research has examined individual-level stimulant use patterns over time. We assessed self-reported stimulant use among 505 people who use drugs enrolled in the Rhode Island Prescription and Illicit Drug Study (RAPIDS) from 2020 to 2024. Group-based trajectory modeling identified distinct patterns of past-month stimulant use over 12 months of follow-up. We examined associations between trajectories and sociodemographic characteristics, other substance use patterns, and mental health diagnoses. Four distinct stimulant use trajectories emerged: low (12%), low/moderate (52%), moderate/high (19%), and high (17%) use. The high-use group exhibited the highest proportion of people who were unhoused (82%, p < 0.01) and had injected drugs in the past month (42%, p < 0.01) at baseline. The high-use group also reported the highest proportions of a lifetime diagnosis of bipolar disorder (46%, p = 0.05), anxiety disorder (57%, p < 0.01), and psychosis (25%, p < 0.01) at baseline. Trajectory-based analyses by specific stimulant type revealed that powder cocaine decreased at three months compared to baseline, while other stimulant use (i.e., crack cocaine, crystal methamphetamine, extra-medical prescription stimulants) remained relatively stable over follow-up. Findings underscore the need for prevention and harm reduction-focused interventions in settings that serve people who are unhoused (e.g., living in homeless shelters, transitional programs) and those with diagnosed psychosis (e.g., receiving behavioral and mental health services).
1. Introduction
Stimulants are a diverse class of psychoactive compounds that activate the central nervous system, increasing alertness and motor activity (Shetty et al., 2021). This class of drugs includes prescription drugs such as amphetamines, methylphenidate, and other unregulated drugs such as cocaine, methamphetamine, and other synthetic cathinones (Tanz et al., 2025). Health risks associated with stimulant use are physical (e.g., cardiovascular complications, overdose) and psychological (e.g., anxiety, psychosis) (Ciccarone, 2021, Shetty et al., 2021). Stimulant-involved overdose deaths have risen rapidly in the United States (US) (Ciccarone, 2021), often used in combination with opioids (Kariisa et al., 2021, Philbin et al., 2020, Tanz et al., 2025), with methamphetamine and cocaine driving most of the stimulant-involved deaths (Medley et al., 2016).
Stimulant-related overdose disproportionally affects marginalized communities (e.g., people who are unhoused) and varies by geography (Tanz et al., 2025). Recent rises in stimulant-involved deaths have been largest among American Indian and Alaskan Native populations as well as Black or African American people; these increases are largely driven by deaths involving both stimulants and opioids (Tanz et al., 2025). Transgender individuals and men who have sex with men also have higher rates of stimulant use and stimulant-related overdose compared to the general population (Medley et al., 2016, Philbin et al., 2020, Rivera et al., 2021).
Previous research has highlighted various pathways to initiation and continuation of stimulant use in the fentanyl era, including the rise of polysubstance use and co-occurring opioid and stimulant use (Ciccarone, 2021). Recent data suggest that 83% of those with diagnosed opioid use disorder had used stimulants at any point in their lifetime (Ellis et al., 2021). Prior nationally representative analyses identified three typologies (e.g., conservative initiation, nondiscriminatory experimentation) representing distinct patterns of stimulant initiation and experimentation (Black et al., 2023). One qualitative study assessed motivations for changing stimulant use patterns in rural areas of the US (Fredericksen et al., 2024). Key emergent themes included changes to drug markets and the cost of drugs, a desire to balance the effects of other substances (e.g., high sedation, withdrawal), and functional goals such as increased energy and pain management (Fredericksen et al., 2024). Few studies, however, have attempted to identify trajectories of stimulant use by type of stimulant (e.g., non-medical prescription stimulants, cocaine, methamphetamine) over time, and scarce research has assessed the relationship between distinct stimulant use trajectories and overdose in the fentanyl era.
To address this gap, we employed group-based trajectory modeling (GBTM) to chart individual-level trajectories of stimulant use and identify factors predicting membership in distinct trajectory patterns. We also examined trajectories of specific classes of stimulants to inform future treatment and harm reduction interventions. Our hypotheses were twofold; first, given the rise in overdoses involving both fentanyl and stimulants, we hypothesized that trajectories characterized by more frequent and consistent use of stimulants would have a higher likelihood of self-reported overdose and more frequent use of fentanyl. Second, we hypothesized that self-reported stimulant use would increase over time, mirroring national trends (Ciccarone, 2021, Tanz et al., 2025). This longitudinal analysis may inform future work that tests whether individuals following different stimulant use trajectories might respond differently to substance use disorder treatment or the timing of harm reduction interventions.
2. Materials and methods
2.1. Study design and participants
The present analysis used data from the Rhode Island Prescription and Illicit Drug Study (RAPIDS), a randomized clinical trial that aimed to assess the efficacy of a fentanyl test strip intervention among people who use drugs in Rhode Island from 2020 to 2023 (Jacka et al., 2020). Residents of Rhode Island were eligible to participate in the RAPIDS study if they were ≥ 18 years of age, able to complete interviews in English, able to provide informed consent, and reported prior 30-day use of heroin, unregulated stimulants, counterfeit prescription pills (i.e., pills purchased on the street), or any drug by injection in the past 30 days. At baseline and at 5 follow-up visits over 12 months (1, 2, 3, 6, and 12 months post-baseline), participants completed a standardized, interviewer-administered questionnaire (Jacka et al., 2020). Prior RAPIDS analyses found that 30% of participants used fentanyl, 24% used prescription opioids, 61% crack cocaine, and 24% powder cocaine at least weekly at baseline assessment (Shaw et al., 2025). RAPIDS was approved by the Brown University Institutional Review Board.
2.2. Primary outcome measures
The primary outcome of interest was the number of days of self-reported stimulant use in the past 30 days at each of the 6 study visits. Participants were asked to self-report days in the prior month when they snorted, smoked, swallowed, injected, or otherwise used stimulants without a prescription or not as directed by a clinical provider. Extra-medical use of prescription stimulants (i.e., Adderall, Ritalin, Focalin, Concerta, Dexedrine, taken without a prescription or not as the clinical provider directed), crystal methamphetamine, powder cocaine, and crack cocaine were assessed both separately and as an overall stimulant use outcome (see section 2.4.1). Importantly, although this method would accurately count days of poly-stimulant use, it undercounts days of use for participants who use multiple stimulants on different days.
2.3. Covariates
Covariates of interest were included based on a priori knowledge, previous literature, and data collected as part of the study. All data were self-reported. All covariates were derived from the baseline instrument and were used to predict membership in stimulant use trajectories based on longitudinal follow-up data (see section 2.4.2).
2.3.1. Sociodemographics
We presented age in years, sex and gender (cisgender man, cisgender woman, transgender/genderqueer/other), and sexual orientation (straight vs. gay/lesbian/bisexual/something else/I don’t know). Sexual orientation was operationalized in this way due to small sample sizes (see Table 1 footnotes) and is aligned with prior RAPIDS work (Shaw et al., 2025). We also presented information on race/ethnicity (white non-Hispanic, Black non-Hispanic, other/multi-racial non-Hispanic, Hispanic/Latine of any race). Individual monthly take-home earnings (≤$500, $501–1,500, >$1,500 USD) and being unhoused in the prior month (no, yes) were assessed.
Table 1.
Baseline characteristics of the study sample (N = 505) stratified by stimulant use trajectories developed based on the maximum number of days of self-reported stimulant use in the past 30 days (n, % except as noted). Percentages reflect assigned proportions based on the maximum posterior probability.
| Baseline Characteristic |
Overall (N = 505) |
Low ((n = 60, 11.9%) |
Low/Moderate (n = 263, 52.1%) |
High/Moderate (n = 94, 18.6%) |
High (n = 88, 17.4%) |
Global P-value |
|---|---|---|---|---|---|---|
| Study Arm | 0.56 | |||||
| Intervention | 252 (49.9) | 32 (53.3) | 134 (51.0) | 48 (51.1) | 38 (43.2) | |
| Control | 253 (50.1) | 28 (46.7) | 129 (49.1) | 46 (48.9) | 50 (56.8) | |
| Sociodemographics | ||||||
| Age in years, median (IQR) | 43 (35, 53) | 42 (34, 56) | 44 (35, 53) | 42 (35, 51) | 45 (37, 54) | 0.65 |
| Sex and gender | ||||||
| Men (cisgender) | 321 (63.8) | 42 (70.0) | 164 (62.8) | 62 (66.0) | 53 (60.2) | 0.45 |
| Woman (cisgender) | 163 (32.4) | 17 (28.3) | 85 (32.6) | 27 (28.7) | 34 (38.6) | |
| Transgender/Other | 19 (3.8) | 1 (1.7) | 12 (4.6) | 5 (5.3) | 1 (1.1) | |
| Sexual orientation | 0.97 | |||||
| Straight | 422 (83.6) | 49 (81.7) | 221 (84.0) | 79 (84.0) | 73 (83.0) | |
| Lesbian, gay, bisexual, somethinga else, or I don’t know | 83 (16.4) | 11 (18.3) | 42 (16.0) | 15 (16.0) | 15 (17.1) | |
| Race/ethnicity | ||||||
| Non-Hispanic white | 260 (51.7) | 32 (53.3) | 138 (52.9) | 51 (54.3) | 39 (44.3) | 0.33 |
| Non-Hispanic Blackb | 83 (16.5) | 13 (21.7) | 43 (16.5) | 14 (14.9) | 13 (14.8) | |
| Non-Hispanic other/multiracial | 56 (11.1) | 7 (11.7) | 29 (11.1) | 12 (12.8) | 8 (9.1) | |
| Hispanic/Latine | 104 (20.7) | 8 (13.3) | 51 (19.5) | 17 (18.1) | 28 (31.8) | |
| Monthly income | ||||||
| $0 − $500 | 226 (45.2) | 23 (39.0) | 109 (41.9) | 50 (53.8) | 44 (50.0) | 0.31 |
| $501 − $1500 | 234 (46.8) | 31 (52.5) | 126 (48.5) | 37 (39.8) | 40 (45.5) | |
| >$1501 | 40 (8.0) | 5 (8.5) | 25 (9.6) | 6 (6.5) | 4 (4.6) | |
| Unhoused, past month | ||||||
| No | 208 (41.2) | 39 (65.0) | 124 (47.2) | 29 (30.9) | 16 (18.2) | <0.01 |
| Yes | 297 (58.8) | 21 (35.0) | 139 (52.9) | 65 (69.2) | 72 (81.8) | |
| Study recruitment | ||||||
| Number of visits completed, mean (SD) | 3.4 (1.8) | 3.9 (1.6) | 3.5 (1.9) | 3.6 (1.8) | 2.8 (1.7) | <0.01 |
| Year of enrollment | ||||||
| 2020 and 2021 | 244 (48.3) | 30 (50.0) | 144 (54.8) | 41 (43.6) | 29 (33.0) | <0.01 |
| 2022 and 2023 | 261 (51.7) | 30 (50.0) | 119 (45.3) | 53 (56.4) | 59 (31.8) | |
| Substance use history | ||||||
| Stimulant use, average number of days in past month (SD) | ||||||
| Any stimulant usec | 14.3 (11.4) | 2.8 (5.2) | 9.3 (8.4) | 20.8 (8.4) | 29.8 (0.9) | <0.01 |
| Extra-medical prescription stimulants | 3.4 (7.2) | 1.1 (2.2) | 2.3 (4.4) | 6.2 (10.6) | 5.0 (9.5) | <0.01 |
| Crystal methamphetamine | 4.3 (7.9) | 1.9 (6.0) | 2.4 (4.8) | 7.0 (9.7) | 6.8 (10.7) | <0.01 |
| Powder cocaine | 3.9 (7.5) | 0.9 (1.5) | 2.7 (5.1) | 5.8 (8.8) | 7.3 (11.3) | <0.01 |
| Crack cocaine | 13.1 (11.7) | 1.9 (4.4) | 8.1 (8.5) | 18.2 (9.9) | 27.7 (7.0) | <0.01 |
| Other substance use and related practices | ||||||
| Injection drug use, past month | <0.01 | |||||
| No | 368 (72.9) | 54 (90.0) | 198 (75.3) | 65 (69.2) | 51 (58.0) | |
| Yes | 137 (27.1) | 6 (10.0) | 65 (24.7) | 29 (30.9) | 37 (42.1) | |
| Smoking, swallowing, or inhaling of drugs, past month |
<0.01 |
|||||
| No | 60 (11.9) | 17 (28.3) | 40 (15.2) | 3 (3.2) | 0 (0.0) | |
| Yes | 445 (88.1) | 43 (71.7) | 223 (84.8) | 91 (96.8) | 88 (100.0) | |
| Fentanyl use frequency, past month | <0.01 | |||||
| Never | 212 (44.8) | 30 (52.6) | 122 (50.8) | 38 (41.3) | 22 (26.2) | |
| Once or a couple times | 121 (25.6) | 20 (35.1) | 60 (25.0) | 18 (19.6) | 23 (27.4) | |
| At least every week | 55 (11.6) | 4 (7.0) | 30 (12.5) | 11 (12.0) | 10 (11.9) | |
| Every day | 85 (18.0) | 3 (5.3) | 28 (11.7) | 25 (27.2) | 29 (34.5) | |
| Preference for fentanyl or drugs that contain fentanyl |
0.05 |
|||||
| No | 426 (84.9) | 53 (89.8) | 228 (87.4) | 78 (83.0) | 67 (76.1) | |
| Yes | 76 (15.1) | 6 (10.2) | 33 (12.6) | 16 (17.0) | 21 (23.9) | |
| Overdose, ever | 0.21 | |||||
| No | 230 (45.9) | 28 (47.5) | 128 (49.2) | 42 (44.7) | 32 (36.4) | |
| Yes | 271 (54.1) | 31 (52.5) | 132 (50.8) | 52 (55.3) | 56 (63.6) | |
| Overdose, past month | 0.04 | |||||
| No | 230 (85.5) | 30 (96.8) | 115 (88.5) | 42 (80.8) | 43 (76.8) | |
| Yes | 39 (14.5) | 1 (3.2) | 15 (11.5) | 10 (19.2) | 13 (23.2) | |
| Mental health diagnoses | ||||||
| ADD/ADHD | 114 (22.6) | 15 (25.0) | 50 (19.0) | 27 (28.7) | 22 (25.0) | 0.22 |
| OCD | 43 (8.5) | 7 (11.7) | 22 (8.4) | 4 (4.3) | 10 (11.4) | 0.28 |
| Eating disorders | 14 (2.8) | 4 (6.7) | 4 (1.5) | 4 (4.3) | 2 (2.3) | 0.10 |
| Depressive disorder | 221 (43.8) | 31 (51.7) | 107 (40.7) | 42 (44.7) | 41 (46.6) | 0.42 |
| Bipolar disorder | 176 (34.9) | 14 (23.3) | 91 (34.6) | 31 (33.0) | 40 (45.5) | 0.05 |
| Anxiety disorder | 221 (43.8) | 34 (56.8) | 105 (39.9) | 32 (34.0) | 50 (56.8) | <0.01 |
| Psychosis | 75 (14.9) | 1 (1.7) | 32 (12.2) | 20 (21.3) | 22 (25.0) | <0.01 |
Notes:
a) Participants identifying as gay (n = 19), lesbian (n = 3), bisexual (n = 49), other (n = 10), or those who responded ‘I don’t know’ (n = 2) were combined and are compared to participants who identified as straight (n = 422).
b) Includes self-identified African, Haitian, and Cape Verdeans.
c) Any stimulant use, including extra-medical stimulants, crystal methamphetamine, powder cocaine, and crack cocaine, is included. Trajectories based on the most conservative estimate of days using stimulants.
2.3.2. Study recruitment
We assessed the mean number of visits completed over the 12-month study period. We also assessed the year of enrollment (dichotomized as 2021–2022 vs. 2022–2023), in alignment with prior RAPIDS research (Shaw et al., 2024), and RAPIDS study arm (intervention, control).
2.3.3. Substance use history
We assessed self-reported average number of days of use in the past month of any stimulant, as well as extra-medical prescription stimulants (i.e., use without a prescription or not as a clinical provider directed), crystal methamphetamine, powder cocaine, and crack cocaine. Categories are not mutually exclusive; a participant could report 5 days of powder cocaine use and 5 days of crack cocaine use. We also presented information on past-month injection drug use (yes, no), past month smoking, swallowing, or inhaling of drugs (no, yes), past month frequency of known or suspected fentanyl use (never, once or a couple of times, at least every week, every day), preference for fentanyl or drugs that contain fentanyl (no, yes), lifetime experience of overdose (no, yes), past-month experience of overdose at baseline (no, yes).
2.3.4. Mental health diagnoses
At baseline, we asked participants about any past lifetime diagnosis of the following mental health conditions: attention-deficit/hyperactivity disorder (ADHD), obsessive–compulsive disorder (OCD), any eating disorder, depressive disorder, bipolar disorder, anxiety disorder, and psychosis. Thus, these diagnoses are not measures of current symptomatology; it is important to note that temporal ordering relative to stimulant use cannot be determined.
2.4. Statistical analysis
2.4.1. Trajectory modeling
We used group-based trajectory modeling (GBTM) to estimate the trajectories for self-reported stimulant use over the 12-month study period. GBTM allows us to evaluate whether there are distinct longitudinal patterns across repeated measurements, even when the overall average change is small or near zero. The outcome, days of use, was modeled using a censored normal distribution. This choice was motivated by the bounded nature of the variable (range: 0–30 days) and the substantial clustering at both the lower (0 days) and upper (30 days) limits, indicating floor and ceiling effects. The censored normal specification appropriately accounts for these features by modeling an underlying continuous latent variable subject to censoring at the bounds. We examined the cumulative number of days of stimulant use in the past month, with a 30-day maximum, reported at each study visit. If participants reported multiple types of stimulant use, we used the maximum value reported with a 30-day maximum. For example, if a participant reported 10 days of methamphetamine use and 20 days of powder cocaine use at baseline, we used 20 as the value for any stimulant use for the study visit. If a participant reported 5 days of methamphetamine use and 5 days of powder cocaine use at baseline, then we used 5 as the value for any stimulant use for the study visit. We combined all modalities of use; thus, injecting, smoking, inhaling, and swallowing are all included in the composite measure of stimulant use. As noted above, we assessed overall stimulant use in the primary analysis, and then examined each type (e.g., extra-medical use of prescription stimulants, crystal methamphetamine, powder cocaine, and crack cocaine) separately in secondary analyses.
For each stimulant use outcome, models with two to five trajectory groups were sequentially estimated. For each candidate model, polynomial orders of 0 (constant), 1 (linear), 2 (quadratic), and 3 (cubic) were systematically evaluated for each trajectory group. Bayesian information criterion (BIC) was the primary criterion for model selection. We selected the model with the lowest BIC value. However, several other criteria were also considered: 1) a preference for a parsimonious model that best fitted the data, 2) average posterior probability value > 0.7, 3) adequate number of participants in each group (e.g., the smallest group had to be > 5% of the sample); and 4) the odds of correct classification based on the posterior probabilities of group membership > 5 (Nagin, 2014). While GBTM assumed that missing outcome data are missing at random conditional on latent class membership, we also incorporated a “dropout” option in the model to relax this assumption and address the potential not-at-random attrition. Non-random attrition over time was modeled by including key predictors of loss to follow-up, which were determined based on prior work, and included gender identity, past-month injection drug use, lifetime history of incarceration, and lifetime history of drug selling. Recent work has explored retention in the RAPIDS trial, with loss to follow-up more common among men and those with a history of drug selling; there were no meaningful differences by recent stimulant use or preference for stimulants (Goldman et al., 2025).
2.4.2. Associations between participant characteristics and trajectory membership
We used Chi-Square tests to compare baseline characteristics for each stimulant-use trajectory group as an outcome. Kruskal-Wallis tests and ANOVA were used to compare the median and mean, respectively, for continuous variables. A two-sided p-value of ≤ 0.05 was considered statistically significant. Because our objective was to descriptively characterize differences in baseline characteristics across empirically derived trajectory groups rather than to estimate adjusted causal effects, we limited our analysis to bivariate associations and did not fit multivariable models. This approach is consistent with the exploratory nature of the analysis, and although some key variables (e.g., being unhoused, monthly income) are likely correlated, exploring these relationships was outside the scope of the present analysis.
2.5. Sensitivity analyses
Two sensitivity analyses were performed to supplement the primary analysis. First, we restricted the sample to participants who completed the baseline assessment and at least one subsequent follow-up visit to assess any systematic differences between those who completed only the baseline assessment (and subsequently became lost to follow-up) and those who completed multiple study visits (Supplementary Table 2, Supplementary Fig. 1). Second, rather than using the maximum value of days of stimulant use, we present results using the cumulative number of days of stimulant use (Supplementary Fig. 2). For example, a given participant reporting 10 days of methamphetamine use and 20 days of powder cocaine use at baseline would be assigned a value of 30 days of stimulant use at baseline. In this way, the cumulative approach is less conservative than the primary analysis using maximum days.
3. Results
3.1. Characteristics of the sample
Among 505 eligible participants, the median age was 43 years, 64% identified as men, and 52% identified as non-Hispanic and white. The majority of the sample (59%) reported being unhoused in the past month. The mean number of study visits completed was 3 (SD = 2). Forty-eight percent of participants were recruited between 2020 and 2021, and 52% between 2022 and 2023. Follow-up declined over time, as reflected in a mean of approximately three completed visits per participant, with some variation in attrition across trajectory groups (Supplemental Table 1).
At baseline, participants reported a mean of 14 days (SD = 11) of any stimulant use in the past month. Crack cocaine use averaged the highest number of days used in the past month (13 of 30 days), and other substances (extra-medical prescription stimulants, crystal methamphetamine, and powder cocaine) ranged from an average of 3–4 days of past-month use. Twenty-seven percent of participants reported past-month injection drug use, and 88% reported past-month smoking, swallowing, or inhaling drugs. Fifteen percent of participants reported a preference for fentanyl or drugs that contain fentanyl. Over half (54%) of participants reported a lifetime experience of overdose, and 15% of participants had overdosed in the past month. Participants reported a high burden of diagnosed mental health conditions, with 44% reporting a depressive disorder, 35% reporting bipolar disorder, and 15% reporting a diagnosis of psychosis.
3.2. Description of stimulant use frequency trajectories
Table 1 shows baseline sociodemographic, study recruitment, substance use-related behaviors, and mental health diagnosis characteristics of the study sample stratified by the estimated stimulant use trajectories. As described above, groups were based on the maximum number of days of past-month self-reported stimulant use, with a 30-day maximum value. We fit a model for stimulant use in the past 30 days, yielding a solution with four groups, all characterized by cubic parameters and an average posterior probability greater than 0.7. Groups were characterized by self-reported low use (12% of the sample), low/moderate use (52%), moderate/high use (19%), and high use (17%), as shown in Fig. 1. Average days of stimulant use in the past month were largely stable over follow-up. Even though we found that trajectories were largely “flat”, GBTM provides a principled approach to characterizing longitudinal patterns of use (i.e., intensity levels over time).
Fig. 1.
Trajectories based on the number of days of any stimulant use in the past month. ‘Days of use’ was defined as the maximum number of days of any type of stimulant use in the past month, with a 30-day maximum (N = 505). Percentages reflect model-estimated class proportions.
The low-use group reported about 3 days of any stimulant use in the past month over the observation period. This group comprised the largest proportion of men (70%), those identifying as non-Hispanic and Black (22%), and the highest mean number of completed study visits (4; SD = 2). Compared to other groups, this group had the lowest proportion of people who were unhoused in the past month (35%). At baseline, this group reported an average of 1 day of extra-medical prescription stimulant use, 2 days of crystal methamphetamine use, 1 day of powder cocaine use, and 2 days of crack cocaine use. This group also had the lowest rate of injection drug use in the past month (10%) and the lowest baseline rate of overdose in the past month (3%) of all groups.
The low/moderate use group included over half of the sample (52%). This group showed a slight, steady decline in reported stimulant use, sustained over the study period (Fig. 1). This group had the highest proportion of individuals in the highest income group (10%) and the lowest average number of days of crystal methamphetamine use (2, SD = 6). The low/moderate use group also had the lowest proportion of having ever experienced an overdose (51%).
The moderate/high use group comprised 19% of participants. This group had the highest rate of non-cisgender individuals (5%) and the highest proportion of individuals in the lowest income bracket (54%). Its trajectory featured the highest proportion of extra-medical stimulant use (11%) and crack cocaine use (11%).
The high-use group included 17% of participants and averaged near-daily stimulant use in the past month over the study period. This group comprised individuals with the highest median age (45; IQR = 37, 54). This group had the highest proportion of women (39%) and the highest proportion of people who were unhoused in the past month (82%). This group also had the lowest proportion of mean study visits completed (2.8; SD = 1.7). Additionally, the high-use group had the highest proportion of individuals reporting past-month injection drug use (42%); daily fentanyl use (35%); preference for drugs containing fentanyl (24%); ever having experienced an overdose (64%); and past-month overdose (23%) at baseline. The high-use group also reported the highest burden of mental health disorder diagnoses (78%), notably psychosis (25%).
We then fit separate models for four stimulant use trajectories by stimulant type (crack cocaine, powder cocaine, crystal methamphetamine, and extra-medical prescription stimulants) based on the number of days of use in the past month (Fig. 2). Four distinct categories of crack cocaine use emerged (Fig. 2a). These trajectories closely mirrored those for any type of stimulant use. Trajectories for powder cocaine, however, revealed 3 distinct groups: very low (62%), low (32%), and high (7%), and showed decreasing days of use for the high group only. The observed steep drop occurred from baseline to three months and then leveled out for the remainder of the study period. Crystal methamphetamine use revealed two distinct trajectories: low use (90%) and high use (10%). Both groups remained relatively stable throughout the study period. Extra-medical prescription use also revealed two distinct trajectories, low (88%) and high (12%), both of which also remained relatively static throughout the study period.
Fig. 2.
Panel of use trajectories based on the number of days of use in the past month by stimulant type. ‘Days of use’ was defined as the maximum number of days of any type of stimulant use in the past month, with a 30-day maximum (N = 505). Percentages reflect model-estimated class proportions.
3.3. Sensitivity analyses
Results from the sensitivity analyses are presented in the Supplementary Materials. We restricted the sample to participants who completed the baseline assessment and at least one follow-up visit (N = 391). Results did not differ meaningfully in this restricted sample (Supplemental Table 2), although the low/moderate class was proportionally smaller than in the main analysis (44% in the restricted sample vs. 52% in the main analysis). Less conservative estimates of past-month stimulant use (using the cumulative number of days rather than the maximum number of days) did not result in meaningful differences in trajectory modeling (Supplemental Fig. 2); results still show four distinct groups that are all relatively “flat.”.
4. Discussion
In our cohort of 505 adults who use drugs in Rhode Island, we identified four unique, heterogeneous trajectories of recent stimulant use patterns over 12 months of follow-up. Thus, trajectories highlight four groups, largely characterized by different intensities of stimulant use. We also identified distinct use trajectories by stimulant type: crack cocaine, powder cocaine, crystal methamphetamine, and extra-medical use of prescription stimulants. We found evidence supporting our primary hypothesis: more frequent stimulant use would be associated with a higher likelihood of recent overdose and a higher likelihood of frequent fentanyl use. We did not find evidence to support our secondary hypothesis; most trajectories of stimulant use did not increase over time.
We observed that trajectory group membership did not vary meaningfully with the individual’s sociodemographic and psychosocial characteristics (e.g., age, gender, sexual orientation). This finding runs counter to the literature supporting the rising prevalence of misuse of prescription stimulants, cocaine, and methamphetamine among adolescents and young adults relative to older age groups (LaBossier & Hadland, 2022). Further, more pronounced use of methamphetamine as a “club drug” at younger ages (<30 years old) is well documented (Park et al., 2018). However, a recent national study estimating the relative odds of past-year cocaine and methamphetamine use by age, stratified by gender and sexual identity, found that gay/lesbian men and women and bisexual men were more likely to use cocaine at later ages, and that heterosexual and gay men ages 26–34 were more likely to report past-year methamphetamine use than their younger counterparts (Philbin et al., 2023). Future work should focus on individuals who use stimulants but not opioids; due to the co-occurrence of these substances in this sample, we did not pursue this analysis.
We also found that participants who were unhoused in the last month were much more likely to belong to the higher use groups; participants who were recently unhoused represented only 35% of the lowest stimulant use group, yet 82% of the highest stimulant use group, with a clear gradient across the four groups. This finding is not surprising given prior research documenting associations between being unhoused and stimulant use (Baggett et al., 2013, Baggett et al., 2015, Bauer et al., 2016, Fine et al., 2022, Fine et al., 2023). Our results highlight the need for prevention and harm reduction-focused interventions in settings such as homeless shelters and transitional programs. Offering services such as drug checking and promoting safer use practices (i.e., avoid using alone) in those settings (e.g., homeless shelters, transitional programs) could help prevent stimulant use-related overdose and death.
Our findings suggest a positive association between higher stimulant use and diagnosis of psychosis. These findings are congruent with prior work documenting an association between mental health and substance use, particularly those involving stimulants (Goodwin et al., 2002, Raines et al., 2021). Prior work has also shown that methamphetamine use may increase the risk of psychosis (Glasner-Edwards & Mooney, 2014). Integrating stimulant use prevention education and services for those already using stimulants into behavioral and mental health services, especially programs serving younger adults, is necessary to prevent adverse outcomes for those with co-occurring mental health diagnoses.
All of the use trajectories were relatively stable over follow-up, except the powder cocaine group (see Fig. 2), which experienced a sharp decline around month three (from an average of 26 days of use to an average of 6 days of use), and then subsequently leveled off. This drop is unlikely due to the study intervention, as individuals who used powder cocaine were not specifically targeted for the intervention. While few studies have analyzed the use of powder cocaine over time, this finding is consistent with a yearlong study examining the natural trajectories of powder cocaine use that found that men who reported using the drug to avoid physical discomfort (i.e., fatigue) or as a social lubricant were also more likely to decrease their frequency of use over time (Palamar et al., 2008). Changes to the supply chain during COVID-19 (Swaich et al., 2024) could also have restricted use, although this connection is fairly speculative. National epidemiologic data also showed a rise in substance use, counterfeit pill availability and use, and overdose rates across the US (O’Donnell et al., 2023).
This study has several limitations. Our study did not assess the age of stimulant use initiation or follow participants for more than 12 months. Our data, therefore, does not inform critical time periods for intervention based on group trajectory membership; trajectories are latent statistical constructs and do not necessarily represent valid clinical categories. No causal inferences should be drawn from these models, and p-values should be interpreted with caution. Additionally, our findings related to mental health diagnoses are complicated; differential access to healthcare may result in some participants being more likely to have received a diagnosis than others. Relatedly, all data are self-reported; without biological verification, some stimulant use may be misclassified. We were unable to directly measure healthcare access, nor did we collect information on the treatment (including medication) of mental health diagnoses of interest. Lastly, the RAPIDS cohort is not necessarily representative of all people who use stimulants; generalizability should be exercised with caution.
5. Conclusions
Variations in use patterns by stimulant type and trajectories that were largely found to be stable over the course of follow-up highlight the need for tailored interventions in settings where people may be using stimulants, specifically aimed at those who are unhoused and those with co-occurring mental illness.
CRediT authorship contribution statement
Leah C. Shaw: Writing – review & editing, Writing – original draft, Investigation, Formal analysis, Conceptualization. Yu Li: Writing – review & editing, Methodology, Formal analysis, Data curation, Conceptualization. Julia E. Noguchi: Writing – original draft, Conceptualization. Carolyn J. Park: Writing – review & editing, Methodology, Formal analysis, Data curation, Conceptualization. Katie B. Biello: Writing – review & editing, Project administration, Methodology, Formal analysis. Scott E. Hadland: Writing – review & editing, Conceptualization. Susan G. Sherman: Writing – review & editing, Conceptualization. Alexandria Macmadu: Writing – review & editing, Supervision, Project administration, Conceptualization. Brandon D.L. Marshall: Writing – review & editing, Writing – original draft, Supervision, Project administration, Conceptualization.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Footnotes
Supplementary data to this article can be found online at https://doi.org/10.1016/j.abrep.2026.100709.
Contributor Information
Leah C. Shaw, Email: leah_shaw@brown.edu.
Yu Li, Email: yu_li1@brown.edu.
Julia E. Noguchi, Email: julia.noguchi@yale.edu.
Carolyn J. Park, Email: carolyn_park@brown.edu.
Katie B. Biello, Email: katie_biello@brown.edu.
Scott E. Hadland, Email: SHADLAND@mgh.harvard.edu.
Susan G. Sherman, Email: ssherman@jhu.edu.
Alexandria Macmadu, Email: alexandria_macmadu@brown.edu.
Brandon D.L. Marshall, Email: Brandon_marshall@brown.edu.
Appendix A. Supplementary data
The following are the Supplementary data to this article:
Supplemental materials include patterns of retention, sensitivity analysis, and the consistency of trajectory group assignment.
Data availability
Data will be made available on request.
References
- Baggett T.P., Chang Y., Singer D.E., Porneala B.C., Gaeta J.M., O’Connell J.J., Rigotti N.A. Tobacco-, alcohol-, and drug-attributable deaths and their contribution to mortality disparities in a cohort of homeless adults in Boston. American Journal of Public Health. 2015;105(6):1189–1197. doi: 10.2105/AJPH.2014.302248. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Baggett T.P., Hwang S.W., O’Connell J.J., Porneala B.C., Stringfellow E.J., Orav E.J., Singer D.E., Rigotti N.A. Mortality among Homeless adults in Boston: Shifts in Causes of death over a 15-year period. JAMA Internal Medicine. 2013;173(3):189–195. doi: 10.1001/jamainternmed.2013.1604. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Bauer L.K., Brody J.K., León C., Baggett T.P. Characteristics of Homeless adults who died of Drug Overdose: A Retrospective Record Review. Journal of Health Care for the Poor and Underserved. 2016;27(2):846–859. doi: 10.1353/hpu.2016.0075. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Black J.C., Burkett H.L., Rockhill K.M., Olson R., Dart R.C., Iwanicki J. Initiation patterns and Transitions among adults using Stimulant drugs: Latent transition Analysis. Journal of Medical Internet Research. 2023;25 doi: 10.2196/46747. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Ciccarone D. The rise of illicit fentanyls, stimulants and the fourth wave of the opioid overdose crisis. Current Opinion in Psychiatry. 2021;34(4):344–350. doi: 10.1097/YCO.0000000000000717. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Ellis M.S., Kasper Z.A., Scroggins S. Shifting Pathways of Stimulant Use among individuals with Opioid Use Disorder: A Retrospective Analysis of the last Thirty Years. Frontiers in Psychiatry. 2021;12 doi: 10.3389/fpsyt.2021.786056. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fine D.R., Dickins K.A., Adams L.D., De Las Nueces D., Weinstock K., Wright J., Gaeta J.M., Baggett T.P. Drug Overdose Mortality among People Experiencing Homelessness, 2003 to 2018. JAMA Network Open. 2022;5(1) doi: 10.1001/jamanetworkopen.2021.42676. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fine D.R., Dickins K.A., Adams L.D., Horick N.K., Critchley N., Hart K., Gaeta J.M., Lewis E., Looby S.E., Baggett T.P. Mortality by Age, Gender, and race and Ethnicity in People Experiencing Homelessness in Boston. Massachusetts. JAMA Network Open. 2023;6(8) doi: 10.1001/jamanetworkopen.2023.31004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fredericksen R.J., Baker R., Sibley A., Estadt A.T., Colston D., Mixson L.S., Walters S., Bresett J., Levander X.A., Leichtling G., Davy-Mendez T., Powell M., Stopka T.J., Pho M., Feinberg J., Ezell J., Zule W., Seal D., Cooper H.L.F., Tsui J.I. Motivation and context of concurrent stimulant and opioid use among persons who use drugs in the rural United States: A multi-site qualitative inquiry. Harm Reduction Journal. 2024;21:74. doi: 10.1186/s12954-024-00986-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Glasner-Edwards S., Mooney L.J. Methamphetamine Psychosis: Epidemiology and Management. CNS Drugs. 2014;28(12):1115–1126. doi: 10.1007/s40263-014-0209-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Goldman J.E., Park C.J., Shaw L.C., Noguchi J.E., Hadland S.E., Marshall B.D.L. Predictors of Retention in a Clinical Trial of a Fentanyl Overdose Prevention intervention: Implications for supporting Participants’ Retention in Harm Reduction Research. Substance Use & Misuse. 2025:1–10. doi: 10.1080/10826084.2025.2570496. [DOI] [PubMed] [Google Scholar]
- Goodwin R.D., Stayner D.A., Chinman M.J., Wu P., Tebes J.K., Davidson L. The relationship between anxiety and substance use disorders among individuals with severe affective disorders. Comprehensive Psychiatry. 2002;43(4):245–252. doi: 10.1053/comp.2002.33500. [DOI] [PubMed] [Google Scholar]
- Jacka B.P., Goldman J.E., Yedinak J.L., Bernstein E., Hadland S.E., Buxton J.A., Sherman S.G., Biello K.B., Marshall B.D.L. A randomized clinical trial of a theory-based fentanyl overdose education and fentanyl test strip distribution intervention to reduce rates of opioid overdose: Study protocol for a randomized controlled trial. Trials. 2020;21:976. doi: 10.1186/s13063-020-04898-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kariisa M., Seth P., Scholl L., Wilson N., Davis N.L. Drug overdose deaths involving cocaine and psychostimulants with abuse potential among racial and ethnic groups—United States, 2004-2019. Drug and Alcohol Dependence. 2021;227 doi: 10.1016/j.drugalcdep.2021.109001. [DOI] [PubMed] [Google Scholar]
- LaBossier N.J., Hadland S.E. Stimulant misuse among youth. Current Problems in Pediatric and Adolescent Health Care. 2022;52(9) doi: 10.1016/j.cppeds.2022.101265. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Medley, G., Lipari, R. N., Bose, J., Cribb, D. S., Kroutil, L. A., & McHenry, G. (2016). Sexual Orientation and Estimates of Adult Substance Use and Mental Health: Results from the 2015 National Survey on Drug Use and Health (NSDUH Data Review). SAMHSA. https://www.samhsa.gov/data/sites/default/files/NSDUH-SexualOrientation-2015/NSDUH-SexualOrientation-2015/NSDUH-SexualOrientation-2015.htm.
- Nagin D.S. Group-based trajectory modeling: An overview. Annals of Nutrition & Metabolism. 2014;65(2–3):205–210. doi: 10.1159/000360229. [DOI] [PubMed] [Google Scholar]
- O’Donnell J., Tanz L.J., Miller K.D., Dinwiddie A.T., Wolff J., Mital S., Obiekwe R., Mattson C.L. Drug Overdose Deaths with evidence of Counterfeit Pill Use—United States, July 2019-December 2021. MMWR. Morbidity and Mortality Weekly Report. 2023;72(35):949–956. doi: 10.15585/mmwr.mm7235a3. [DOI] [PubMed] [Google Scholar]
- Palamar J.J., Mukherjee P.P., Halkitis P.N. A longitudinal investigation of powder cocaine use among club-drug using gay and bisexual men. Journal of Studies on Alcohol and Drugs. 2008;69(6):806–813. doi: 10.15288/jsad.2008.69.806. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Park J.N., Weir B.W., Allen S.T., Chaulk P., Sherman S.G. Fentanyl-contaminated drugs and non-fatal overdose among people who inject drugs in Baltimore. MD. Harm Reduction Journal. 2018;15(1):34. doi: 10.1186/s12954-018-0240-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Philbin M.M., Greene E.R., LaBossier N.J., Martins S.S., McCrimmon T., Mauro P.M. Age-related patterns of cocaine and methamphetamine use across the life course in the United States: Disparities by gender and sexual identity among adults. Addictive Behaviors. 2023;137 doi: 10.1016/j.addbeh.2022.107539. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Philbin M.M., Greene E.R., Martins S.S., LaBossier N., Mauro P.M. Medical, Nonmedical, and Illegal Stimulant Use by Sexual Identity and Gender. American Journal of Preventive Medicine. 2020;59(5):686–696. doi: 10.1016/j.amepre.2020.05.025. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Raines A.M., Ennis C.R., Allan N.P., McGrew S.J., Walton J.L., Rogers A.H., Zvolensky M.J., Schmidt N.B., Laurel Franklin C. Anxiety sensitivity and substance use: Differential levels across individuals primarily using opioids, cannabis, or stimulants. Addictive Behaviors. 2021;116 doi: 10.1016/j.addbeh.2020.106791. [DOI] [PubMed] [Google Scholar]
- Rivera A.V., Harriman G., Carrillo S.A., Braunstein S.L. Trends in Methamphetamine Use among men who have sex with men in New York City, 2004-2017. AIDS and Behavior. 2021;25(4):1210–1218. doi: 10.1007/s10461-020-03097-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Shaw L.C., Biello K.B., Buxton J.A., Goldman J.E., Hadland S.E., Sherman S.G., Li Y., Macmadu A., Marshall B.D.L. How substance use preferences and practices relate to fentanyl exposure among people who use drugs in Rhode Island, USA. Drug and Alcohol Dependence Reports. 2024;12 doi: 10.1016/j.dadr.2024.100280. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Shaw L.C., Kumar A., Park C.J., Li Y., Lenox C.A., Collins A.B., Sherman S.G., Marshall B.D.L., Macmadu A. Gender-based differences in harm reduction practices among people who use drugs in Rhode Island: A latent class analysis. Harm Reduction Journal. 2025;22(1):140. doi: 10.1186/s12954-025-01295-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Shetty S., Malik A.H., Ali A., Yang Y.C., Briasoulis A., Alvarez P. Characteristics, trends, outcomes, and costs of stimulant-related acute heart failure hospitalizations in the United States. International Journal of Cardiology. 2021;331:158–163. doi: 10.1016/j.ijcard.2021.01.060. [DOI] [PubMed] [Google Scholar]
- Swaich A., Ignatieff D., Milloy M.-J., Choi J., Kerr T., Hayashi K. Impacts of the COVID-19 pandemic on the availability and retail price of unregulated drugs in Vancouver, Canada: An interrupted time-series analysis, 2018-2022. International Journal of Drug Policy. 2024;134 doi: 10.1016/j.drugpo.2024.104633. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Tanz, L. J., Miller, K. D., Dinwiddie, A. T., Gladden, M., Asher, A., Baldwin, G., Nesbit, B., & O’Donnell, J. (2025). Drug Overdose Deaths Involving Stimulants—United States, January 2018-June 2024. (74(32);491–499; Morbidity and Mortality Weekly Report). CDC. https://www.cdc.gov/mmwr/volumes/74/wr/mm7432a1. [DOI] [PMC free article] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental materials include patterns of retention, sensitivity analysis, and the consistency of trajectory group assignment.
Data Availability Statement
Data will be made available on request.