Probability of fentanyl adulteration in cocaine selectively decreases cocaine demand

Abstract

Fentanyl-related, cocaine-overdose deaths have drastically increased, yet research on how people who use cocaine perceive fentanyl adulteration is limited. This study developed the novel Adulterated Cocaine Purchasing Task (ACocPT), a modification of the original Cocaine Purchasing Task, to quantify how people respond to fentanyl adulteration in cocaine. In the ACocPT, participants indicated how much cocaine they would purchase when cocaine had no (0%) vs. some (10%) probability of fentanyl adulteration. Study aims were: 1) determine how possible fentanyl adulteration affects cocaine demand, and 2) determine which individual characteristics predict continued demand for cocaine despite fentanyl adulteration. This Amazon Mechanical Turk study included self-reported cocaine purchasers (N = 64), who completed self-report questionnaires (demographics, substance use history, depression/PTSD symptoms, fentanyl knowledge quiz), and the ACocPT. Results showed 1) greater probability of fentanyl adulteration (10%) lowered cocaine demand, but only for intensity (Q₀; amount of cocaine consumed when free; p < .001); 2) no effect on other demand indices (O_max, P_max, essential value, breakpoint); 3) significantly more zero-responders with 10% probability of fentanyl adulteration than 0%, p < .001; and 4) opioid co-use, depression, age, PTSD, fentanyl knowledge, and cocaine use severity did not moderate the relationship between fentanyl adulteration and intensity. Overall, fentanyl adulteration reduced cocaine demand but only for volume preferred at minimal cost, not general motivational drive for use, illustrating the dangerous insensitivity to toxic contamination. The internal validity of the paradigm provides proof-of-concept for this approach to identify individuals at risk from fentanyl adulterated cocaine.

Introduction

Cocaine-related deaths have significantly increased since 2015 (National Institute on Drug Abuse, 2024). Numerous cocaine overdose deaths are driven by intentional co-use of opioids and stimulants (van Amsterdam & van den Brink, 2025). However, research suggests some overdoses may occur due to unintentional consumption of fentanyl-adulterated stimulants, such as cocaine (Bazazi et al., 2024; Nolan et al., 2019; Wagner et al., 2023). Consistent with this, the DEA reported in some states there was a 112% increase in cocaine samples containing fentanyl between 2016 and 2017 (Drug Enforcement Administration, 2018), and there have been instances of fentanyl-adulterated cocaine documented across the United States (Canning et al., 2021; DiSalvo et al., 2021). Yet, fentanyl test strip use and naloxone distribution are more common among people who use opioids, highlighting the need to extend these efforts to those who use cocaine. Although, to inform interventions, we must first understand how people who use cocaine perceive and respond to fentanyl adulteration of cocaine.

Although fentanyl poses a major threat to people who use cocaine, there is limited research on how people who use both opioids and cocaine perceive the safety of their cocaine, or how people who only use cocaine perceive fentanyl at all (Reed et al., 2021). People who use crack cocaine have qualitatively reported concern about fentanyl and reported willingness to use drug checking technologies (Reed et al., 2021). Other studies similarly highlight individuals who report crack and powder cocaine use express concern about fentanyl in the drug supply (Reed et al., 2024). Previous findings also show people with no current/past history of opioid use are less aware about fentanyl in the drug supply and less prepared for possible overdose (i.e., fewer carried naloxone) compared to those with opioid history (Hughto et al., 2021). Additional studies have shown increasing levels of fentanyl awareness over time. Among participants who reported past year cocaine use, agreement with the statement that cocaine may contain fentanyl significantly increased over the years (Palamar, 2023). Studies on the general use of fentanyl further present mixed findings, with some suggesting cocaine use is associated with increased likelihood of intentional fentanyl use (Chandra et al., 2021) and others finding no significant relationship (Ferguson et al., 2022). It is also unclear how different individual characteristics (e.g., opioid history, cocaine use severity, mental health conditions, fentanyl knowledge, age) impact decisions related to use of adulterated cocaine. Overall, additional studies that quantify how people who use cocaine respond to fentanyl in cocaine, and that examine these individual characteristics are needed to better identify who may be at risk of fentanyl overdose.

Behavioral economics may provide a sensitive and valid way to quantify how people who use cocaine respond to fentanyl adulteration. Hypothetical purchase tasks use a common “real-world” decision, i.e., purchasing drugs at varying prices. These tasks measure drug demand and capture the relationship between drug consumption as a function of increasing unit price under conditions of constraint (e.g., time constraints, personal consumption, etc.). Overall, purchase tasks help quantify the reinforcing value of drugs and the likely impact of different conditions on drug use (Bickel et al., 2014; MacKillop, 2016). While qualitative studies have highlighted fentanyl adulteration concerns among people who use cocaine, behavioral economics provides a quantifiable proxy for likely use behavior. Cocaine demand has been associated with self-reported cocaine use (Yoon et al., 2020, 2021), cocaine use severity (Nunez et al., 2024; Zvorsky et al., 2019), substance use treatment response, and relapse risk (Yoon et al., 2020, 2021). Further, perceived quality affects demand for other drugs (Vincent et al., 2017), and studies on heroin have found heroin use likelihood decreased with increased impurity or overdose risk (Dolan et al., 2021), suggesting demand will be sensitive to adulteration. More broadly, there is a growing literature demonstrating behavioral economic purchase tasks are sensitive to experimental manipulations to understand dynamic effects on value preferences (Acuff et al., 2020), suggesting similar methods could be used for evaluating adulteration.

The Present Study

This study developed a novel Adulterated Cocaine Purchasing Task (ACocPT) to quantify how people use cocaine respond to fentanyl adulteration. The original Cocaine Purchasing Task asks individuals to indicate how much cocaine they would purchase across escalating prices for personal consumption over a 24-hour period (Bruner & Johnson, 2014). Our ACocPT manipulated perceived quality of cocaine by having participants complete the same task when informed the cocaine has no (0%) vs. some (10%) probability of fentanyl adulteration. A 10% probability was set based on rates seen in cocaine seizures (Park et al., 2021; Zibbell et al., 2019) and is similar to rates found in studies on drug checking (Wagner et al., 2023). The study addressed the following aims: 1) determine how possible fentanyl adulteration affects cocaine demand, 2) determine which individual characteristics predict continued demand for cocaine despite fentanyl adulteration, to identify who may be most at risk from fentanyl adulterated cocaine. For aim 1, we hypothesized greater probability of fentanyl adulteration would lower cocaine demand. For aim 2, we hypothesized individuals with less opioid use experience would be particularly likely to show continued demand for fentanyl adulterated cocaine.

Methods

Transparency and Openness

We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study, and we follow JARS (Appelbaum et al., 2018). Data, analysis code, and research materials can be provided upon request. Data were analyzed using R (Version 4.0.1; RStudio Team, 2015), and the main package for analyses was lme4 (Version 1.1-25; Bates et al., 2015b). This study was not preregistered.

Overall Design

This study (Protocol Number: CTN-0130) was part of the National Institute on Drug Abuse’s Clinical Trial Network, which aims to improve addiction treatment. This was a two-part online study conducted on Amazon Mechanical Turk (MTurk), that followed established procedures for this platform (Bruner & Johnson, 2014), using CloudResearch (formerly TurkPrime; Litman et al., 2017). In the screening, participants completed self-report questionnaires on demographics, substance use history, and depression. In the main survey, participants completed measures of post-traumatic stress disorder (PTSD) symptoms, a fentanyl knowledge quiz, and the ACocPT.

Participants

The literature on perceived quality of drugs shows a wide range of effects from small to large (d = 0.27 to 1.26) on demand for drugs. If the effect size for fentanyl adulteration is in the medium range (d = 0.5), a minimum sample size of N = 54 would be required to achieve 95% power at α = .05, but if it is on the small end of that range (d = 0.27), a minimum sample size of N = 174 would be required to achieve 95% power at α = .05. Given the resource limitations of this initial minimally funded study, we selected the smaller sample (assuming a medium effect) as our target.

The study included participants who reported purchasing powder cocaine in the past year. Participants were recruited from May 2022 – June 2022 and December 2022 – April 2023. During the recruitment gap, study changes were made, such as adding survey questions and other administrative changes. Eligibility criteria included: past year self-reported cocaine purchasing, 90% or higher approval rating on all previously submitted MTurk tasks, over 100 approved tasks on MTurk, current residence in the United States, able to read and complete forms in English, and CloudResearch data quality approved. Participants were compensated $0.15 for completing the screening survey, $3.00 for completing the main survey, and a $1.00 bonus if they passed three attention checks. Participants provided written informed consent, and procedures were conducted in accordance with the Declaration of Helsinki and approved by the University of Illinois Chicago Institutional Review Board.

Measures

Demographics Survey

Participants reported demographics, including gender, age, race, years of education.

Substance Use History

Participants reported past year drug use. They were asked the following to determine cocaine purchasing qualifications: “what is your typical unit of purchase for cocaine (e.g., grams, ounces, etc.),” “how much do you typically pay for each unit (e.g., each gram, ounce etc.) of cocaine (in U.S. dollars),” and “when you purchase cocaine, how long does that amount usually last you (e.g., 24-hours, a weekend, one week, etc.)?” Substance use for several substances (e.g., cocaine, opioids) was assessed further with the Kreek-McHugh-Schluger-Kellogg Scale (KMSK; Kellogg, 2003). This scale assessed the frequency, amount, and duration of use for each substance, which produces a drug use lifetime severity score for each substance.

Patient Health Questionnaire (PHQ-8)

The PHQ-8 (Kroenke et al., 2009) assessed depression symptoms over the past two weeks. Participants indicated how often they have been bothered by each symptom. Total scores range from 0 to 24 points, and higher scores suggest greater severity of depression symptoms.

Life Events Checklist for DSM-5 (LEC-5) Extended Version

The LEC-5 (Weathers et al., 2013) assessed for lifetime exposure of different traumatic events. Participants identified the worst event (i.e., event that bothers them the most) and completed additional questions (e.g., “was someone’s life in danger?”). These responses were used to determine whether the event met The Diagnostic and Statistical Manual of Mental Health Disorders (5^th ed.; DSM-5; (American Psychological Association, 2013) Criterion A for PTSD.

PTSD Checklist for DSM-5 (PCL-5)

The PCL-5 (Blevins et al., 2015), a 20-item self-report questionnaire, assessed for the presence and severity of PTSD symptoms in the past month. Participants completed the PCL-5 based on the worst event they identified from the LEC-5. Total scores range from 0 to 80, and higher scores indicate greater PTSD symptom severity.

Fentanyl Knowledge Quiz

This seven-item quiz included short response and multiple-choice questions about fentanyl’s potency, effects, overdose risks, and other fentanyl facts. These questions have been used in previous studies, and the fentanyl facts have been presented on MedlinePlus, an online information service produced by the United States National Library of Medicine (Krieger et al., 2018; Moallef et al., 2019; Persaud & Jennings, 2020). Participants received a point for each correct answer. Total scores range from 0 to 7, with higher scores indicating greater fentanyl knowledge.

Adulterated Cocaine Purchasing Task (ACocPT)

The ACocPT is modification of the Cocaine Purchasing Task (Bruner & Johnson, 2014). Participants indicate how much cocaine they would purchase across escalating prices ($0.00 to $10,000 per unit, with a unit being defined as 1 rock for crack cocaine and 0.1 grams for powder cocaine) for personal consumption over a 24-hour period. The prices (Yoon et al., 2020) and units for powder cocaine (0.1 grams) (Strickland, Lile, et al., 2016; Strickland, Reynolds, et al., 2016) and for crack cocaine (rocks) (Yoon et al., 2020) on the ACocPT were based on previous research. Before the task, participants reported whether they typically use powder cocaine or crack cocaine and then completed the corresponding task. Participants completed the ACocPT under each probability of adulteration (0% and 10%), with probabilities presented in counterbalanced order. Participants were informed of the adulteration probabilities (0% or 10%) prior to each task.

Five behavioral economic metrics from the ACocPT were generated at each probability of fentanyl adulteration: essential value (change in purchasing as a function of price), intensity (Q₀; amount of cocaine purchased when it is free), breakpoint (first price with no cocaine purchased), O_max (maximum expenditure for cocaine), and P_max (price that maximum expenditure for cocaine occurs).

Procedure

Participants had access to the screening recruitment materials on MTurk if they met the MTurk qualifications previously described. Participants were provided a brief study description stating it was investigating peoples’ decisions related to drug use, so they were not aware we were specifically recruiting people who use cocaine. Participants self-selected to complete the screening survey (~3 minutes) to assess for additional eligibility criteria for the main survey (i.e., past year cocaine purchasing). Cocaine purchasing responses were reviewed to determine whether participants reported the typical unit purchased, amount paid per unit, and how long that amount lasts. If participants passed an attention check and met eligibility criteria, they were assigned a custom qualification that provided access to the main survey (~17 minutes). Eligible participants had access to the informed consent form before starting the main survey.

Analytic Approach

All analyses were mixed effects models. All continuous independent variables were mean centered and categorical independent variables were contrast coded.

Data Quality

Participants who failed any of the three attention checks were excluded from the main analyses, as often done with MTurk questionnaires. Purchase task was evaluated for systematicity (i.e., whether the data followed expected patterns) with Bruner and Johnson’s criteria (Bruner & Johnson, 2014) to identify and remove nonsystematic demand data. We used these criteria instead of other criteria such as Stein et al. (2015) due to the presence of zero-responders (i.e., participants who did not purchase cocaine at any price, even when free). Although we used the beezdemand package to analyze our purchase task data, we did not use its function for identifying non-systematic data since it uses Stein et al., (2015) criteria. Instead, non-systematic data was evaluated separately. Based on Bruner and Johnson’s criteria, demand data was considered nonsystematic if 1) the units of cocaine purchased at a given price were at least 20% greater than the preceding price, or 2) the units of cocaine purchased at the final price were not less than the first price by at least 10%. An exception was made to the second rule for zero-responders, which is consistent with practices applied in previous studies (Yoon et al., 2021). Participants were excluded if they did not have systematic data for both tasks (0% and 10%) prior to fitting the data to our demand model. Demand data were fit to Equation 1 (the exponentiated demand model; Koffarnus et al., 2015) below with the beezdemand package (Version 0.1.0; Kaplan et al., 2019).

$$Q=Q0\ast {10}^{k(e^{-\alpha Q0C}-1)}$$ (1)

The exponentiated model has advantages, such as the inclusion of zero consumption values and improved fit compared to the exponential model (Koffarnus et al., 2015; Strickland et al., 2016). In the exponentiated model, Q is the amount of consumption at commodity price C. Q₀ represents consumption levels at or near 0 price, also referred to as intensity. k represents the range of consumption in logarithmic units and set to 4.5 for both the 0% and 10% tasks. α is the rate of change in slope of the demand function, which is inversely proportional to essential value according to Equation 2 below (Hursh & Roma, 2016):

$$EssentialValue=\frac{1}{(100\cdot \alpha \cdot k^{1.5})}$$ (2)

Other demand indices include breakpoint, O_max, and P_max. For these demand indices (Q₀, O_max, P_max) observed values were obtained from the raw data, and derived values were obtained after fitting the raw data to the exponentiated model. Zero-responders were handled per procedures here (Heckman et al., 2019; Yoon et al., 2021), such as defining Q₀, O_max, P_max, and breakpoint as 0. Since α is undefined when there is zero consumption, we defined essential value as 0, the lowest possible value. Participants who purchased at the highest price point (i.e., $10,000) were assigned a breakpoint at that price (n = 2) (Coelho et al., 2024; Murphy & Mackillop, 2006; Yurasek et al., 2024).

Hypothesis Testing

The first aim was to determine how possible fentanyl adulteration affects cocaine demand. Hypothesis one was tested with mixed effects models (one per demand index) with probability of adulteration (0% and 10%) as a within-subject fixed effect and random effects calculated per Bates et al., (2015a). Our primary outcome was intensity (Q₀). This outcome was selected based on previous literature; Q₀ has been associated with real-world outcomes, including self-reported cocaine use and treatment outcomes (Yoon et al., 2020, 2021). Exploratory analyses were conducted with the other demand indices (O_max, P_max, essential value, breakpoint).

The second aim was to determine which individual characteristics predict continued demand for cocaine despite fentanyl adulteration. The individual characteristics tested were opioid co-use, age, depression symptoms, PTSD symptoms, fentanyl knowledge, and cocaine use severity. Hypothesis two was assessed using mixed effects models (one per characteristic). We only included demand indices that were significantly related with probability in Aim 1. Each characteristic was entered as a covariate in each mixed effects model, along with the covariate x probability of adulteration interaction. In recognition of the fact that this was a hypothesis-generating analysis, any interaction at p < 0.05 between a covariate and adulteration was considered to indicate a significant moderator.

Results

Participants and Data Quality

7,566 participants completed the screening, 131 qualified, and 103 completed the main survey. 25 participants (18 powder cocaine, 7 crack cocaine) had nonsystematic data and were excluded. For powder cocaine, 18 participants failed criterion 1 (nine on the 0% task, three on the 10% task, six for both 0% and 10%), and one of these participants also failed criterion 2 (on both 0% and 10% task). For crack cocaine, seven participants failed criterion 1 (two on the 0% task, two on the 10% task, three on both 0% and 10%), and three of these participants also failed criterion 2 (one on 0% task, two on 10% task). An additional 4 participants were excluded (1 no demand data, 1 no cocaine use, 2 failed attention checks). 10 participants who reported crack cocaine use were excluded. These individuals were initially included in the study since they are of interest. However, it is unknown whether people who use powder vs. crack cocaine respond differently on the demand task, especially since the units (grams vs. rocks) differ. Due to the small sample of people who use crack, it was not feasible to assess systematic differences or effectively control for type of cocaine used, so they were excluded. Our analyses only included people who reported powder cocaine use.

Participants (N = 64) predominately identified as White, were in their 20–40’s, with some college education, and about half were male (see Table 1).

Table 1.

Descriptive statistics

Variable	M(SD) or n (%)
Socio-demographic Characteristics
Age (years)	35.14 (8.23)
Gender
Female	30 (46.88%)
Male	33 (51.56%)
Non-binary	1 (1.56%)
Race/Ethnicity
White	51 (79.69%)
Black, African, or African American	3 (4.69%)
Hispanic, Latino, or Spanish	2 (3.13%)
Asian or Asian American	2 (3.13%)
More than one race	6 (9.38%)
Education
High School/GED	16 (25.00%)
Associate degree or some college	24 (37.50%)
Bachelor’s degree	19 (29.69%)
Master’s degree	4 (6.25%)
Other advanced degrees	1 (1.56%)
Income	54,756.22 (34,496.05)
United States regiona
Midwest	7 (10.94%)
Northeast	5 (7.81%)
South	13 (20.31%)
West	7 (10.94%)
Type of areaa
Living in a city (urban area with > 25,000 people)	15 (23.44%)
Living in a rural area (not near a city or town)	5 (7.81%)
Living in a town (urban area with < 25,000 people)	5 (7.81%)
Living in the suburbs (just outside a city or town)	7 (10.94%)
Homelessness (% yes)	18 (28.13%)
Fentanyl knowledge quiz total	5.11 (1.24)
Drug Use History
Cocaine use (# days in past month)	4.16 (7.27)
Cocaine use severity score	10.16 (3.37)
Past year opioid use (% yes)	24 (37.5%)
Mental Health variables
PHQ-8 total	10.23 (6.28)
PCL-5 total	22 (17.73)

Note. N = 64

^aVariable was added mid-study, data missing for 32 participants enrolled prior to its inclusion.

Internal validity of cocaine demand task

The exponentiated demand model effectively described the demand data, with an average r² = 0.95. Spearman rank order correlations examined the relationship between observed and derived demand indices (Q₀, O_max, P_max) for the 0% and 10% tasks to further assess the utility of the exponentiated demand model. For the 0% task, results demonstrated positive, strong correlations for Q₀ (r = 0.96, p < .001), O_max (r = 0.92, p < .001), and P_max (r = 0.68, p < .001). For the 10% task, results also showed positive, strong correlations for Q₀ (r = 0.99, p < .001), O_max (r = 0.98, p < .001), and P_max (r = 0.88, p < .001). These results indicate the exponentiated model provided a good fit to our demand data.

The relationships between cocaine demand indices and self-reported cocaine use were also assessed with spearman rank order correlations (see Table 2).

Table 2.

Comparison of baseline demand indices with self-report measures of cocaine use

	Cocaine Use (# days in past month)		KMSK Lifetime Score
Demand indices	r	p	r	p
0% Demand Task
Q0	0.16	0.20	0.43	< .001
Omax	0.25	< .05	0.41	< .001
Pmax	0.16	0.20	0.08	0.52
Essential value	0.25	< .05	0.41	< .001
Breakpoint	0.18	0.15	0.10	0.42
10% Demand Task
Q0	0.11	0.38	0.23	0.07
Omax	0.15	0.25	0.27	0.03
Pmax	0.16	0.19	0.14	0.28
Essential value	0.15	0.25	0.27	< .05
Breakpoint	0.21	0.10	0.92	0.47

Note. Bolded cells indicate significant associations

Aim 1

Cocaine Purchasing

Two participants were zero-responders on the 0% task, and twenty were zero-responders on the 10% task. A McNemar’s test indicated there was a significant change in zero-responder status between the two tasks, such that there were more zero-responders in the 10% task than in the 0% task, χ²(1) = 16.06, p < .001. See Table 3.

Table 3.

Participant purchasing behavior at free and paid price points

	0% Demand Task	10% Demand Task
Price	n (%)	n (%)
Free	62 (96.88%)	44 (68.75%)
Paid	60 (93.75%)	43 (67.19%)

Note. This table includes the proportion of participants who purchased at least one unit (0.1 gram) at free and paid price points. “Free” refers to $0.00 per unit. “Paid” refers to the subsequent price point, $1.00 per unit.

Although it could be argued that zero-responders (i.e., non-purchasers) on the 0% probability task should be excluded in the main analysis, as certain characteristics may lead to zero motivation for cocaine use (e.g., recently initiated a recovery attempt), we included these participants. We included zero-responders on the 0% task, because there is the possibility that fentanyl may still be appealing to some participants, and we did not want to rule out the possibility that demand could increase under the 10% probability task. The main analyses were conducted with and without these participants.

Intensity

The first model included our primary outcome, intensity of demand (Q₀). There was a significant effect of fentanyl adulteration on Q₀ (p < .001), such that higher probability of fentanyl adulteration in cocaine (10%) significantly decreased the amount of cocaine individuals reported they would consume when cocaine was free. See Figure 1 for demand curves. There was also a significant effect of task order (p < .05); however, the interaction was not significant (p = 0.70) (see Table 4). The results did not change when excluding the two zero-responders from the 0% task.

Figure 1. Demand Curves for 0% and 10% demand tasks.

Note. Demand curves for 0% and 10% probability of adulteration tasks derived from the exponentiated model. Symbols represent the mean number of units purchased at each price point for the two probabilities (0% and 10%). The error bars represent ±1 standard error of the mean. Cocaine purchasing in units (0.1 grams) is presented on the y-axis, and unit price is presented on the x-axis. Note that the x-axis is non-continuous and is on a log scale. The ‘0’ price point represents Q_0, consumption at or near zero; zero price points were replaced with 0.1 for plotting.

Table 4.

Mixed-effects models of probability of fentanyl adulteration on demand indices

Demand indices	Fentanyl Adulteration	Task Order	Fentanyl Adulteration x Task Order
Q0	B = −5.34, SE = 1.43 t(62) = −3.73, p < .001	B = 7.87, SE = 3.02 t(62) = 2.60, p < .05	B = 0.56, SE = 1.43 t(62) = 0.39, p = 0.70
Omax a	B = 347.8, SE = 359.2, t(63) = 0.97, p = 0.34	—	—
Pmax a	B = 16.46, SE = 19.19, t(63) = 0.86, p = 0.39	—	—
Essential valuea	B = 9.75, SE = 10.06, t(63) = 0.97, p = 0.34	—	—
Breakpointa	B = 16.75, SE = 41.34, t(63) = 0.41, p = 0.69	—	—

Note. Analyses were conducted with and without the two zero-responders from the 0% probability task, excluding the two zero-responders did not change the results. The findings reported are based on the final sample of 64 participants.

^aTask order and interaction were not significant; final model includes only fentanyl adulteration.

Other demand indices

Mixed effects models tested the effect of fentanyl adulteration on other demand indices, one model per demand index. There was no significant effect of fentanyl adulteration on O_max (p = 0.34), P_max (p = 0.39), essential value (p = 0.34), or breakpoint (p = 0.69) (see Table 4). Excluding the two zero-responders from the 0% task did not change the results for any of our demand indices.

Exploratory analyses

Exploratory analyses tested the effect of fentanyl adulteration on demand among the participants who purchased cocaine (i.e., would use at all) in the 10% task (n = 44) using mixed effects models (one model per demand index). There was no significant effect of fentanyl adulteration on Q₀ (p = 0.13), which was different from our previous results, suggesting that lower relative adulterated purchasing was driven by zero-responders. Similar to our previous findings, there was no significant effect of fentanyl adulteration on O_max (p = 0.33), P_max (p = 0.33), essential value (p = 0.32), or breakpoint (p = 0.56).

Analyses were conducted to compare people who purchased at 10% (n = 44) to those who did not purchase at 10% (n = 20) on cocaine use (number of days used in past month), cocaine use severity, and past year opioid use. Results indicated no significant difference between adulterated cocaine purchasers and non-purchasers for cocaine use (adulterated cocaine purchasers: M = 2.40, SD = 5.61; non-purchasers: M = 4.95, SD = 7.83; p = 0.20) and cocaine use severity (adulterated cocaine purchasers (M = 10.23, SD = 3.32; non-purchasers: 10.00, SD = 3.55; p = 0.81). Purchase status at 10% was not significantly related to past year opioid use ($\chi 2$ (1, N = 64) = 0.31, p = 0.58).

Aim 2

Since Aim 1 results showed there was a significant effect of fentanyl adulteration on intensity (Q₀), Aim 2 used Q₀ as the outcome, testing one model for each characteristic.

Opioid co-use was not a moderator between fentanyl adulteration and Q₀ (B = −0.63, SE = 1.47, t(62) = −0.43, p = 0.67), suggesting people who used both cocaine and opioids were not significantly more likely to reduce the amount of adulterated cocaine they would use when cocaine was free. However, adulteration remained significantly related to Q₀ (B = −5.46, SE = 1.47, t(62) = −3.70, p < .001), consistent with our findings in Aim 1. Similarly, the interactions between age and adulteration (B = 0.22, SE = 0.17, t(62) = 1.25, p = 0.22), depression and adulteration (B = 0.14, SE = 0.23, t(62) = 0.61, p = 0.54), PTSD and adulteration (B = −0.04, SE = 0.08, t(62) = −0.48, p = 0.64), fentanyl knowledge and adulteration (B = −1.23, SE = 1.15, t(62) = −1.07, p = 0.29), and cocaine severity and adulteration (B = −0.50, SE = 0.42, t(62) = −1.18, p = 0.24) were not significant. However, adulteration remained significantly related to Q₀ in the age, depression, fentanyl knowledge, and cocaine severity models (all p < .001).

Discussion

Overall, results suggested our new ACocPT may be a useful tool for identifying people at risk of fentanyl overdose. Consistent with our primary hypothesis, we found greater probability of fentanyl adulteration (1 in 10 chance) reduced cocaine demand, but only for intensity (Q₀; amount of cocaine consumed when free). Qualitatively, this seems to represent that some individuals respond to fentanyl adulteration by reducing their purchasing to zero, indicating they would not consume at any price. Notably, 68.75% of participants reported they would still consume cocaine despite there being a 10% probability of adulteration, suggesting this group is at increased risk of overdose. Additionally, adulteration did not impact other demand indices. Possible explanations for this are intensity may be more sensitive than the other demand indices, or it may be that it is easier to change excess or optional cocaine consumption.

Contrary to our secondary hypotheses, opioid co-use did not significantly moderate the relationship between fentanyl adulteration and demand. Opioid co-use was selected as a moderator based on studies suggesting people who use opioids expressed greater concern for fentanyl adulteration in their drugs (Rouhani et al., 2019). Although some people may be concerned about fentanyl adulteration, they may change their behavior in other ways not captured in this study, such as using smaller amounts of the drug, using drugs more slowly (Rouhani et al., 2019), or they may conclude they could test their cocaine with fentanyl test strips at a later time. It is also possible some people who use opioids may have a history of fentanyl use or feel more comfortable with the idea of fentanyl adulteration in their drugs. We also did not identify other strong moderators (age, depression, PTSD symptoms, fentanyl knowledge, and cocaine use severity) of the effect of adulteration. This may be due to the small sample size and should be confirmed in larger studies.

Additionally, even though this study used hypothetical cocaine purchasing, the results between demand indices and cocaine use variables show the potential utility of demand for measuring likely real-world behavior. Demand measures for both tasks (0% and 10%) were significantly related with self-report cocaine use measures, which is consistent with previous research (Nunez et al., 2024; Yoon et al., 2020, 2021; Zvorsky et al., 2019) and further supports the validity of the purchase task. The results show the utility of using hypothetical purchase tasks to measure cocaine demand (Bruner & Johnson, 2014) through online platforms (Strickland, Reynolds, et al., 2016).

This study had limitations. Since this was an online study, we could not verify participants’ self-reported cocaine purchases or biologically verify cocaine use. However, we implemented strategies to improve the reliability of the sample. Participants were not aware the study was recruiting people who use cocaine, and we carefully reviewed the cocaine purchasing responses during the screening to increase the likelihood that the participants were actual cocaine purchasers. We also evaluated the purchase task data to ensure that the responses were logical. Additionally, online studies may help reduce underreporting bias related to substance use (Strickland & Stoops, 2019). Nonetheless, future studies may consider conducting procedures in person to verify cocaine use with drug screen tests. Second, the study included predominately self-identified White participants, which might limit the generalizability of these findings. Additional studies may consider recruiting more diverse samples, especially groups who have high cocaine overdose rates, such as Non-Hispanic Black males (Fleming et al., 2020; Han et al., 2019; Nolan et al., 2019). Third, our sample consisted of participants who had low cocaine use rates. There were two non-consumers (i.e., no consumption, even when unadulterated and free) and four non-purchasers (i.e., no consumption when paying) on the 0% task, and overall participants reported on average using cocaine 4.16 days in the past month, which is much lower than the averages reported in studies among people with cocaine use disorder (Bentzley et al., 2021). It is possible this low cocaine use may reflect lower motivation for cocaine use and reduced risk-taking, which may contribute to decreased demand under the 10% probability of adulteration. This could differ in populations with more severe cocaine use. In sum, our findings may not generalize to people with cocaine use disorder; thus, it may be helpful to conduct future studies with this population to examine how they respond on the ACocPT. Fourth, we did not investigate other adulteration probabilities. Future studies may explore the impact of higher probabilities of adulteration on cocaine demand, which may illuminate further relationships not observed at 10% probability. Fourth, although we tested whether some individual characteristics were related to the use of fentanyl adulterated cocaine, there are several other demographic variables (e.g., overdose history) that might contribute. Finally, and relatedly, this study consisted of a small sample. A larger sample will be needed to establish the effect of these potential multiple related variables without an undue increase in Type I Error.

In summary, fentanyl adulteration reduced demand for cocaine in some but not in all participants, suggesting the ACocPT may be a useful tool for identifying people at risk of fentanyl overdose. Additional studies are needed that relate the ACocPT to real-world risk, such as overdose history, explore its sensitivity as an indicator of treatment response (e.g., educational interventions), and assess if it is predictive of subsequent fentanyl exposure, to further validate this new measure. It will also be critical to explore other individual characteristics that might put individuals at increased risk for use of fentanyl adulterated cocaine, which may inform future interventions. Without a doubt, people who use cocaine are a group who is at risk of overdose, and harm reduction efforts (e.g., overdose education, fentanyl education) should address this group. Future studies may also explore whether harm reduction efforts decrease demand for fentanyl adulterated cocaine, as a potential early indicator of public-health-level effectiveness.

Public Health Significance:

People who use cocaine are at risk of fentanyl overdose due to fentanyl adulteration in the drug supply, so it is important to understand how they perceive and respond to fentanyl adulteration to inform future interventions. Our findings suggest that probability of fentanyl adulteration in cocaine decreases cocaine consumption for some but not all individuals.