Repeated measures latent class analysis identifies unique daily cocaine-use patterns and associated clinical characteristics: A timeline followback-based analytic approach Authors: Alixandria T. Mascarin, William Burstein, Lana R. Grasser, Mark K. Greenwald Published: 2026 ABSTRACT BackgroundCocaine use poses significant health risks and treatment challenges. Individual patterns of multi-daily use and their association with clinical characteristics remain poorly understood. Novel approaches that differentiate cocaine-use patterns may improve clinical understanding and guide treatments. This study aimed to identify distinct patterns of cocaine use with a timeline follow-back (TLFB) approach and to evaluate relationships with demographics, cocaine-use severity, and negative emotionality.MethodsData were collected from 109 (88 male) non-treatment seeking people who use cocaine (mean + SD age 45.6+7.1 years) through past 15-day TLFB interviews tracking daily cocaine use and amounts. Bias-adjusted 3-step repeated-measures latent class analysis identified distinct patterns of daily binary (yes/no) cocaine use. Predictors of class membership including age, income, negative emotionality (depression and anxiety symptoms), and cocaine-use severity were evaluated via regression models as part of the last step of the 3-step RM-LCA approach.ResultsThree cocaine-use patterns emerged: high variability across days (Class 1, n = 36), infrequent use (Class 2, n = 52), and consistent use across days (Class 3, n = 21). Consistent use class reported higher daily cocaine-use amounts and cocaine craving than high-variability and infrequent use classes; consistent use class also reported higher income than infrequent use. Infrequent use class reported lower income compared to high variability use.DiscussionThis study identifies distinct patterns of cocaine use associated with different behavioral and demographic characteristics, providing a foundation for tailored treatment development and stratification. Future work should expand these methods and findings to explore the broader context of polysubstance use. Introduction Despite decades of intensive research, cocaine use remains a significant public health concern. An estimated 4.3 million people in the United States report past-year cocaine use (Substance Abuse and Mental Health Services Administration, 2025), and opioid overdoses involving cocaine continue to climb, with an almost five-fold increase from 2011 to 2023 (Spencer et al., 2023, Garnett and Miniño, 2024). Treatment options for people who use cocaine are insufficient, as there are no FDA-approved medications and standard psychosocial and behavioral approaches commonly show high attrition rates and low efficacy in yielding sustained abstinence (Chan et al., 2019, Kampman, 2019, Bentzley et al., 2021). Treatment efficacy may be limited by heterogeneity in cocaine-use patterns and polysubstance use; thus, identifying subgroups of people who use cocaine could facilitate targeted treatment approaches (Chan et al., 2019, Kampman, 2019, Bentzley et al., 2021, Palazon-Llecha et al., 2024). In the present study, we first explore heterogenous cocaine-use patterns only. Then, in a companion study (reported separately), we expand this analysis to heterogenous cocaine, cannabis, and alcohol polysubstance use. Across species, cocaine-use patterns vary widely among individuals. In animal models, heterogeneous cocaine self-administration patterns correlate with differences in cocaine-induced locomotor sensitization, cocaine preference, and incubation of cocaine-seeking (Ferland and Winstanley, 2017, O'Connor et al., 2021, Rakela et al., 2024). Cocaine use patterns, rather than total use, predict return-to-use outcomes in rat models, suggesting within-subject variability as an important consideration for risk assessment (Mainville-Berthiaume et al., 2026). Variation in human cocaine use has been linked to demographic and clinical outcomes—including gender, current age and age at use onset, criminal justice referrals, alcohol-use severity, cocaine use-related consequences across psychological, familial, employment, and legal domains (Myers et al., 1995, Palij et al., 1996, Shearer et al., 2007, Roos et al., 2019)—and to psychiatric and neurobiological features (Brucar et al., 2025). These findings highlight the importance of considering individual use patterns during interventions. However, optimal methods to meaningfully characterize use patterns remain unclear. In an early exploration of cocaine-use patterns, Myers et al. (1995) employed cluster analysis to identify subgroups of people who use cocaine that differed on quantity and frequency of use and preferred route of administration. Alternatively, Palij et al. (1996) used a regression-based approach to identify predictors of daily cocaine-use patterns. Results from this study found considerable variation across days and highlighted the need for approaches that include time-related variables that account for within-subject variability. Roos et al. (2019) examined daily cocaine-use patterns and found distinct subgroups: abstinence, low-frequency use, and persistent frequent use. These use profiles were differentially associated with clinical outcomes. Notably, all three of these studies examined individuals in treatment for cocaine use. Active treatment protocols may have influenced outcomes, and these results may not generalize to all people who use cocaine, most of whom are not engaged in care (Mustaquim et al., 2021). While these reports demonstrate the predictive value of assessing heterogenous cocaine use patterns, they also highlight the need for approaches that capture within-subject variability in use in samples unconfounded by treatment effects. To address these limitations, person-centered approaches that capture day-level variability across time are needed. Timeline followback (TLFB), paired with repeated measures latent class analysis (RM-LCA), is an advantageous approach, because it accommodates within-subject changes and enables detection of empirically different subgroups in ecologically valid, non-treatment-seeking samples. Therefore, this study employed RM-LCA to identify subgroups of participants endorsing recent cocaine use with distinct daily cocaine-use patterns assessed via TLFB. We examined relations between cocaine-use patterns, demographic factors, and clinical characteristics, with the goal of identifying meaningful indicators of phenotypically distinct cocaine-use profiles that could ultimately inform targeted treatment development and stratification. Methods The local Institutional Review Board approved study procedures, which were conducted according to the Declaration of Helsinki. A certificate of confidentiality was obtained. All volunteers provided written informed consent. Alcohol-free breath samples (<0.02%; Alco Sensor III breathalyzer) and no evidence of cognitive impairment defined by total verbal IQ score > 80 on the Shipley Institute of Living Scale (Zachary, 1991) were required for informed consent. Participant selection Individuals aged 18–55 years, regardless of racial/ethnic background, self-reporting past-month cocaine use, and not seeking treatment for their substance use, were recruited for one of two laboratory-based cocaine self-administration studies, registered on www.clinicaltrials.gov as NCT00946660 (Greenwald et al., 2014) and NCT01392092 (Woodcock et al., 2021). Participants were recruited via advertisements and word-of-mouth referral throughout the Detroit/metropolitan area, initially screened by telephone interview, and underwent a comprehensive in-person screening assessment. All data reported herein are derived from this in-person screening assessment. Measures Demographic data Participants reported their age, race/ethnicity, biological sex at birth, and years of education they had completed. Body mass index (kg/m2) was computed from height and weight. The Cocaine Purchasing and Use Questionnaire (CPUP; Greenwald and Steinmiller, 2014) was administered to estimate total (legal and illegal) past-month income. Drug History and Use Questionnaire (DHUQ) The DHUQ (available on request) is a comprehensive assessment of lifetime and current use of psychoactive substances. For cocaine, lifetime variables included duration of regular cocaine use, number of cocaine use-related adverse consequences, and number of lifetime quit attempts. Participants also reported the number of past-month days of using cocaine, alcohol, and cannabis, and the number of tobacco cigarettes/day used. Cocaine-use severity Participants completed the Cocaine Selective Severity Assessment (CSSA; Kampman et al., 2002), which has 18 items (each rated on 0–7 scale) that measure cocaine use-related consequences (e.g., increased/decreased appetite, craving, increased/decreased sleep, excess anxiety, inattention, paranoid ideation, anhedonia, depression, suicidality, irritability). Timeline follow-back (TLFB) assessment In prior studies of people who use cocaine, TLFB methodology has varied considerably in terms of the temporal window (e.g., one week to one year), type and number of substances assessed (one vs. multiple), and granularity of data gathered such as incidence (use vs. no use), amount per day, route of administration, and within-day order or timing of co-use (Gossop et al., 2006, Hoeppner et al., 2010, Robinson et al., 2014, Martin-Willett et al., 2020, Fitzgerald et al., 2024). In the present study, each participant completed a past 15-day (from −14 [distant] to 0 [day of screening]) TLFB assessment of daily use amounts of cocaine. Based on predominant substance use patterns in the local area, we defined cocaine “use amounts” in terms of multiples of $10 dose units. Urine drug screening (UDS) Volunteers provided a urine sample into a Clinical Laboratory Improvement Amendments (CLIA)-waived multi-test cup with temperature strip. The sample (required temperature range: 92–96º F) was tested for presence of cocaine metabolites, benzodiazepines, opioids, methadone (positive cutoffs >300 ng/ml), amphetamines (positive cutoff >1000 ng/ml), barbiturates (positive cutoff >200 ng/ml) and THC (positive cutoff >50 ng/ml). Negative emotionality and impulsivity Two measures of negative emotionality were explored in relation to cocaine-use patterns. Past two-week depression symptoms were measured using the Beck Depression Inventory-II (BDI-II; Beck et al., 1996). Trait anxiety symptoms were measured using the State-Trait Anxiety (STAI) Y2 scale (Spielberger et al., 1970). Impulsivity was measured using the Barratt Impulsiveness Scale (BIS) version 11 (Patton et al., 1995). Data analyses Repeated measures latent class analysis (RM-LCA) was conducted in MPlus (Muthén and Muthén, 1998) based on the cocaine daily use binary data (use vs. no use), because sufficient free parameters were not available to model continuous daily cocaine use amounts (i.e., a larger sample size would be needed). Model fit indices, entropy, null hypothesis testing, cell size, and theory were used to evaluate the optimum solution. We used the log-likelihood value, Akaike information criterion (AIC), standard Bayesian information criterion (BIC), sample-size adjusted BIC, and chi-square test as model fit indices. The sample-size adjusted BIC was used given the modest sample size. For log-likelihood, AIC, and BIC values, a smaller value indicates a better-fitting model, while for the chi-square test a non-significant p-value indicates that the fitted model and the observed data do not significantly differ. Entropy is a measure of precision of classification; values range from 0 to 1, and higher values indicate better classification accuracy. The Vuong-Lo-Mendell-Rubin (VLMR) Likelihood Ratio Test (LRT) and Lo-Mendell Rubin Adjusted (LMRA) LRT tests were used to determine whether a n-class or n + 1 class solution was more appropriate, testing the null hypothesis that the n-class solution was a better fit of the observed data than the n + 1 class solution. Significant VLMR-LRT and LMRA-LRT values suggest that the n + 1 class solution was a better fit than the n-class solution. Finally, we screened models for cell size and evaluated the number of classes in accordance with theories about use patterns to choose the optimum solution. Using the optimum solution, a 3-step RM-LCA was performed in MPLus to assess psychopathologic and demographic factors associated with cocaine-use class membership, for each individual predictor. A Bonferroni-corrected alpha value of < .0021 was applied to account for multiple (24) comparisons. Figures were generated using GraphPad Prism 10. Results Participant characteristics As shown in Table 1, participants were predominantly Black (79.8%) males (80.7%), with a mean age of 45.6 years and 13 years education. On average, participants reported long histories of regular cocaine use (18 years), having tried to quit cocaine numerous times (Mattempts=13.6), and having used cocaine on 16 of the past 30 days. Current primary route of cocaine administration was smoked (‘crack’, 87.6%); some individuals reported insufflation (‘snorting’, 10.5%) and few injected cocaine (1.9%). On average, participants reported total past-month total income of $1531.Table 1Participant characteristics¶ by cocaine-use latent class membership.MeasureOverall sample (N=109)Class 1 (n=36): Variable UseClass 2 (n=52): Infrequent UseClass 3 (n=21): Consistent UseSex (% male)80.7%88.9%75.0%76.2%Race (% black)79.8%83.3%84.6%66.7%Age (years)45.6 (7.1)43.8 (6.1)47.1 (6.9)45.0 (8.6)Education (years)13.0 (1.6)13.0 (1.1)13.0 (1.9)13.2 (1.7)Estimated Verbal IQ (Shipley)103.0 (11.6)102.7 (9.9)103.9 (12.0)101.2 (13.5)Age at First Cocaine Use (years)24.9 (7.3)23.2 (6.0)26.0 (7.4)25.3 (8.6)Age at Regular Cocaine Use (years)27.6 (7.2)25.1 (6.2)29.6 (7.1)27.0 (7.9)Years Since First Cocaine Use20.6 (8.3)20.6 (7.6)21.1 (8.9)19.7 (8.4)Years Since Regular Cocaine Use18.0 (8.2)18.7 (7.6)17.5 (8.4)18.0 (8.9)Weight (lbs)181.2 (38.2)176.1 (30.6)189.2 (42.3)170.1 (36.5)Height (inches)69.0 (3.3)69.5 (2.8)69.0 (3.3)68.5 (4.1)Body Mass Index (kg/m2)26.7 (5.5)25.7 (4.4)27.9 (6.2)25.5 (5.0)Past-month Income (USD)£1531 (1180)1694 (1376)1112 (758)2294 (1268)Cocaine Use Severity (CSSA)18.6 (11.9)20.7 (13.1)15.3 (9.9)23.1 (12.7)Cocaine Craving (range: 0–7; CSSA)4.6 (1.8)4.6 (1.8)4.0 (1.7)5.9 (1.2)Cocaine-related Consequences (DHUQ)4.8 (4.2)6.1 (4.3)4.3 (4.2)4.2 (3.6)Lifetime Cocaine Quit Attempts13.6 (25.2)13.9 (21.9)14.6 (28.5)10.5 (22.2)Cocaine Urine Drug Screen (% positive)78.5%82.9%68.6%95.2%Tobacco Use (cigarettes/day)8.9 (8.3)9.6 (6.9)7.6 (7.4)10.9 (11.9)Depression Symptoms (BDI-II)11.1 (9.1)12.7 (8.1)8.8 (8.5)13.9 (11.1)Trait Anxiety Symptoms (STAI)38.7 (11.2)41.4 (10.7)36.4 (10.7)39.6 (12.3)Attentional Impulsivity (BIS-11)14.7 (4.2)15.0 (4.3)14.3 (4.2)15.4 (4.3)Motor Impulsivity (BIS-11)23.0 (5.1)23.9 (5.2)21.9 (5.3)24.3 (4.1)Non-Planning Impulsivity (BIS-11)26.0 (5.8)26.5 (6.0)25.1 (5.6)27.2 (6.0)¶ Mean (SD) or percentage values.£ Analyses performed on log10-transformed values. Table 1Participant characteristics¶ by cocaine-use latent class membership.MeasureOverall sample (N=109)Class 1 (n=36): Variable UseClass 2 (n=52): Infrequent UseClass 3 (n=21): Consistent UseSex (% male)80.7%88.9%75.0%76.2%Race (% black)79.8%83.3%84.6%66.7%Age (years)45.6 (7.1)43.8 (6.1)47.1 (6.9)45.0 (8.6)Education (years)13.0 (1.6)13.0 (1.1)13.0 (1.9)13.2 (1.7)Estimated Verbal IQ (Shipley)103.0 (11.6)102.7 (9.9)103.9 (12.0)101.2 (13.5)Age at First Cocaine Use (years)24.9 (7.3)23.2 (6.0)26.0 (7.4)25.3 (8.6)Age at Regular Cocaine Use (years)27.6 (7.2)25.1 (6.2)29.6 (7.1)27.0 (7.9)Years Since First Cocaine Use20.6 (8.3)20.6 (7.6)21.1 (8.9)19.7 (8.4)Years Since Regular Cocaine Use18.0 (8.2)18.7 (7.6)17.5 (8.4)18.0 (8.9)Weight (lbs)181.2 (38.2)176.1 (30.6)189.2 (42.3)170.1 (36.5)Height (inches)69.0 (3.3)69.5 (2.8)69.0 (3.3)68.5 (4.1)Body Mass Index (kg/m2)26.7 (5.5)25.7 (4.4)27.9 (6.2)25.5 (5.0)Past-month Income (USD)£1531 (1180)1694 (1376)1112 (758)2294 (1268)Cocaine Use Severity (CSSA)18.6 (11.9)20.7 (13.1)15.3 (9.9)23.1 (12.7)Cocaine Craving (range: 0–7; CSSA)4.6 (1.8)4.6 (1.8)4.0 (1.7)5.9 (1.2)Cocaine-related Consequences (DHUQ)4.8 (4.2)6.1 (4.3)4.3 (4.2)4.2 (3.6)Lifetime Cocaine Quit Attempts13.6 (25.2)13.9 (21.9)14.6 (28.5)10.5 (22.2)Cocaine Urine Drug Screen (% positive)78.5%82.9%68.6%95.2%Tobacco Use (cigarettes/day)8.9 (8.3)9.6 (6.9)7.6 (7.4)10.9 (11.9)Depression Symptoms (BDI-II)11.1 (9.1)12.7 (8.1)8.8 (8.5)13.9 (11.1)Trait Anxiety Symptoms (STAI)38.7 (11.2)41.4 (10.7)36.4 (10.7)39.6 (12.3)Attentional Impulsivity (BIS-11)14.7 (4.2)15.0 (4.3)14.3 (4.2)15.4 (4.3)Motor Impulsivity (BIS-11)23.0 (5.1)23.9 (5.2)21.9 (5.3)24.3 (4.1)Non-Planning Impulsivity (BIS-11)26.0 (5.8)26.5 (6.0)25.1 (5.6)27.2 (6.0)¶ Mean (SD) or percentage values.£ Analyses performed on log10-transformed values. Participant characteristics¶ by cocaine-use latent class membership. ¶ Mean (SD) or percentage values. £ Analyses performed on log10-transformed values. Optimum solution Fit indices, entropy values, null hypothesis testing, cell size, and theory indicated a three-class solution as the model of best fit (see Table 2). Class 1 (n = 36) was described by a highly variable day-to-day pattern of cocaine use; Class 2 (n = 52) was described by infrequent cocaine use; and Class 3 (n = 21) was described by a pattern of consistent cocaine use. Representative images of individual subject cocaine use trajectories for each class are shown in Figs. 1C, 1D, and 1E. As shown in Figs. 1A and 1B respectively, classes differed on the number of days of cocaine use during the TLFB period and average daily cocaine use, which were both lowest in Class 2 (M [SD] days used=3.94 [1.92]; M [SD] average daily use=21.18 [21.53]), intermediate in Class 1 (M [SD] days used=8.47 [1.84]; M [SD] average daily use=48.87 [28.57]), and highest in Class 3 (M [SD] days used=13.38 [2.33]; M [SD] average daily use=78.33 [47.04]). Class 2 had a lower percentage of cocaine + UDS (68.6%) than Class 1 (82.9%) or Class 3 (95.2%) (Table 1).Table 2Model fit indices.Classes in SolutionLog-likelihoodAICBICN-Adjusted BICEntropyVLMR LRTLMRA LRTC1 CountC2 CountC3 CountC4 CountC5 CountC6 Count2-940.391942.782026.211928.260.95Ϯ279.23; Ϯ3079--------3-906.711907.412033.911885.390.87Ϯ66.48; Ϯ365221------4-889.961905.912075.471876.400.92NS33.06; NS53171425----5-870.781899.552112.171862.540.92NS37.86; NS1924292116--6-854.801899.592155.271855.080.92NS31.54; NS16143032125AIC: Akaike information criterion. BIC: Bayesian information criterion. N-Adjusted BIC: sample-size adjusted Bayesian information criterion. VLMR LRT: Vuong-Lo-Mendell-Rubin Likelihood Ratio Test. LMRA LRT: Lo-Mendell Rubin Adjusted Likelihood Ratio Test. Cells with bolded text represent the chosen (three-class) solution. Ϯ: significant LRT value. NS: not significant.Fig. 1Cocaine use classes differ on magnitude and variability of cocaine use. (A) days of cocaine use across 15-day TLFB period. (B) average number of daily $10 units of cocaine used. Error bars represent standard error of the mean (SEM). (C, D, E) Representative images (i.e., one participant) of cocaine use trajectories for Classes 1, 2, and 3, respectively. *: p < .05, **: p < .01, ***: p < .001, ****: p < .0001. Table 2Model fit indices.Classes in SolutionLog-likelihoodAICBICN-Adjusted BICEntropyVLMR LRTLMRA LRTC1 CountC2 CountC3 CountC4 CountC5 CountC6 Count2-940.391942.782026.211928.260.95Ϯ279.23; Ϯ3079--------3-906.711907.412033.911885.390.87Ϯ66.48; Ϯ365221------4-889.961905.912075.471876.400.92NS33.06; NS53171425----5-870.781899.552112.171862.540.92NS37.86; NS1924292116--6-854.801899.592155.271855.080.92NS31.54; NS16143032125AIC: Akaike information criterion. BIC: Bayesian information criterion. N-Adjusted BIC: sample-size adjusted Bayesian information criterion. VLMR LRT: Vuong-Lo-Mendell-Rubin Likelihood Ratio Test. LMRA LRT: Lo-Mendell Rubin Adjusted Likelihood Ratio Test. Cells with bolded text represent the chosen (three-class) solution. Ϯ: significant LRT value. NS: not significant. Model fit indices. AIC: Akaike information criterion. BIC: Bayesian information criterion. N-Adjusted BIC: sample-size adjusted Bayesian information criterion. VLMR LRT: Vuong-Lo-Mendell-Rubin Likelihood Ratio Test. LMRA LRT: Lo-Mendell Rubin Adjusted Likelihood Ratio Test. Cells with bolded text represent the chosen (three-class) solution. Ϯ: significant LRT value. NS: not significant. Cocaine use classes differ on magnitude and variability of cocaine use. (A) days of cocaine use across 15-day TLFB period. (B) average number of daily $10 units of cocaine used. Error bars represent standard error of the mean (SEM). (C, D, E) Representative images (i.e., one participant) of cocaine use trajectories for Classes 1, 2, and 3, respectively. *: p < .05, **: p < .01, ***: p < .001, ****: p < .0001. Table 1 reports descriptive statistics by the full sample and by class. We performed a 3-step RM-LCA with three latent classes for each demographic and psychological variable in Table 1. Notably, individuals who had a UDS positive for cocaine were more likely to be classified in the consistent use class compared to the low use class, B= 2.34, SE= 1.20, p = .050, however this finding did not survive correction for multiple comparisons. Psychopathology differentiates cocaine-use classes Total self-reported cocaine-use severity significantly differed by class (Fig. 2A). Individuals in Classes 1 and 3 reported greater cocaine-use severity compared to individuals in Class 2 (B=0.06, SE=0.03, p = .038 and B=0.07, SE=0.03, p = .012, respectively). Cocaine-use severity did not differ between Class 1 and Class 3 (B=-0.01, SE=0.02, p = .610). More specifically, individuals in Class 3 reported greater craving than individuals in Class 2 (B=0.83, SE=0.21, p < .001), while individuals in Class 1 reported less craving than individuals in Class 3 (B=-0.58, SE=0.21, p = .006) (Fig. 2B). Craving did not differ between Classes 1 and 2 (B=0.25, SE=0.18, p = .086). Self-reported depression symptom severity also differed by class (Fig. 2C). Higher severity scores were reported by individuals in Class 3 compared to Class 2 (B=0.07, SE=0.04, p = .049), however scores did not differ between Classes 3 and 1 (B=0.01, SE=0.03, p = .753) or 1 and 2 (B=0.06, SE=0.03, p = .053). Finally, motor impulsivity (but not attention or non-planning) as measured by the BIS significantly differed by class (Fig. 2D). Individuals in Class 2 had lower motor impulsivity scores compared to Class 3 (B=-0.10, SE=0.05, p = .045), but not compared to Class 1 (-0.09, SE=0.05, p = .077). Classes 1 and 3 did not differ (B=-0.01, SE=0.05, p = .891). Severity of self-reported anxiety symptoms did not differentiate by class. Following multiple comparisons correction, the difference in cocaine craving between Class 3 and Class 2 was the only surviving significant effect.Fig. 2Cocaine use classes differ on clinical symptoms of cocaine-use severity, craving, and depression. (A) Cocaine Selective Severity Assessment (CSSA) total score. (B) CSSA craving item measure. (C) Beck Depression Inventory-II (BDI-II) total symptom score. Error bars represent standard error of the mean (SEM). *: p < .05, **: p < .01, ***: p < .001, ****: p < .0001. Cocaine use classes differ on clinical symptoms of cocaine-use severity, craving, and depression. (A) Cocaine Selective Severity Assessment (CSSA) total score. (B) CSSA craving item measure. (C) Beck Depression Inventory-II (BDI-II) total symptom score. Error bars represent standard error of the mean (SEM). *: p < .05, **: p < .01, ***: p < .001, ****: p < .0001. Demographic factors differentiate cocaine-use classes Current age, age at onset of regular cocaine use, and past-month total income were the demographic factors that significantly differentiated cocaine-use class. Older individuals were grouped in Class 2 compared to Class 1, B= 0.08, SE= 0.04, p = .029 (Fig. 3A). Age did not differentiate Class 1 from Class 3 (B=-0.03, SE=0.04, p = .525) nor Class 2 from Class 3 (B=0.05, SE=0.05, p = .276). As shown in Fig. 3B, individuals in Class 2 were older than individuals in Class 1 when they first started using cocaine regularly (B=0.12, SE=0.05 p = .014); no other significant differences in age at regular cocaine use were identified. As shown in Fig. 3C, Income significantly differentiated across classes. Individuals in Classes 1 and 3 reported greater income than individuals in Class 2 (B=3.39, SE=1.42, p = .017 and B=5.54, SE=1.66, p = .001, respectively). Reported income did not differ between Classes 1 and 3, B= -2.16, SE= 1.35, p = .111. The finding regarding the difference in income by Classes 3 and 2 was the only result that survived correction for multiple comparisons. There were no other demographic factors that significantly differentiated cocaine-use class. Sex distribution did not differ by class. Table 3 indicates means and standard deviations by sex across the full sample.Fig. 3Cocaine use classes differ on demographic factors. (A) Age at time of TLFB screening, in years. (B) Age at onset of regular cocaine use, in years. (C) Total past-month income (log10 transformed), in US dollars. Error bars represent standard error of the mean (SEM). *: p < .05, **: p < .01, ***: p < .001, ****: p < .0001.Table 3Descriptive statistics by sex across full sample.Females (n=21)Males (n=88)MSDMSDAge (Years)43.298.0546.136.79Education (Years)13.001.6113.021.63Estimated Verbal IQ (Shipley)101.1410.84103.4211.75Age at First Cocaine Use (Years)23.388.3825.316.97Age at Regular Cocaine Use (Years)25.868.0828.076.96Years Since First Cocaine Use19.9010.2820.827.80Years Since Regular Cocaine Use17.439.7218.137.82Weight (lbs)170.5736.48183.7038.31Height (Inches)66.052.7869.763.00Body Mass Index (kg/m2)27.475.5926.535.47Past-month Income (USD)1307.39777.221581.661251.33Cocaine Dependence Severity (CSSA)23.2414.0817.5011.16Cocaine Craving (CSSA)5.431.434.381.83Cocaine-related Consequences (DHUQ)5.705.334.623.84Lifetime Cocaine Quit Attempts12.0024.0613.9925.59Cocaine Urine Drug Screen (% positive)£90.5%75.6%Tobacco Use (cigarettes/day)11.3310.228.277.74Depression Symptoms (BDI-II)12.438.4810.769.30Trait Anxiety Symptoms (STAI)38.9510.1538.6011.48Attentional Impulsivity (BIS-11)14.933.2714.684.42Motor Impulsivity (BIS-11)23.125.4822.995.05Non-Planning Impulsivity (BIS-11)25.836.2226.015.77£Data for cocaine UDS was missing for 2 of the 109 participants (both male). Percentages are based on the available UDS data for males (68) and females (21). Cocaine use classes differ on demographic factors. (A) Age at time of TLFB screening, in years. (B) Age at onset of regular cocaine use, in years. (C) Total past-month income (log10 transformed), in US dollars. Error bars represent standard error of the mean (SEM). *: p < .05, **: p < .01, ***: p < .001, ****: p < .0001. Table 3Descriptive statistics by sex across full sample.Females (n=21)Males (n=88)MSDMSDAge (Years)43.298.0546.136.79Education (Years)13.001.6113.021.63Estimated Verbal IQ (Shipley)101.1410.84103.4211.75Age at First Cocaine Use (Years)23.388.3825.316.97Age at Regular Cocaine Use (Years)25.868.0828.076.96Years Since First Cocaine Use19.9010.2820.827.80Years Since Regular Cocaine Use17.439.7218.137.82Weight (lbs)170.5736.48183.7038.31Height (Inches)66.052.7869.763.00Body Mass Index (kg/m2)27.475.5926.535.47Past-month Income (USD)1307.39777.221581.661251.33Cocaine Dependence Severity (CSSA)23.2414.0817.5011.16Cocaine Craving (CSSA)5.431.434.381.83Cocaine-related Consequences (DHUQ)5.705.334.623.84Lifetime Cocaine Quit Attempts12.0024.0613.9925.59Cocaine Urine Drug Screen (% positive)£90.5%75.6%Tobacco Use (cigarettes/day)11.3310.228.277.74Depression Symptoms (BDI-II)12.438.4810.769.30Trait Anxiety Symptoms (STAI)38.9510.1538.6011.48Attentional Impulsivity (BIS-11)14.933.2714.684.42Motor Impulsivity (BIS-11)23.125.4822.995.05Non-Planning Impulsivity (BIS-11)25.836.2226.015.77£Data for cocaine UDS was missing for 2 of the 109 participants (both male). Percentages are based on the available UDS data for males (68) and females (21). Descriptive statistics by sex across full sample. £Data for cocaine UDS was missing for 2 of the 109 participants (both male). Percentages are based on the available UDS data for males (68) and females (21). Discussion The present study identified three distinct patterns of daily cocaine use measured over the prior two weeks: high variability (Class 1), infrequent (Class 2), and consistent use (Class 3). These classes differed on some clinical and demographic variables, supporting the idea that meaningful heterogeneity in cocaine use exists and could be relevant as a baseline phenotype to guide treatment development and stratification. Prior work using LCA to examine cocaine use patterns (Roos et al., 2019) also found distinct subgroups; however, these groups were differentiated only by their frequency of use, whereas within-subject variability in use was central to class composition in our sample. Roos et al. (2019) reported associations between cocaine-use patterns both during treatment and 12 months post-treatment, supporting this proposed future application. In our present sample, cocaine-use patterns were related to severity of cocaine use consequences (although this finding did not survive correction for multiple comparisons) and craving for cocaine, measures that may predict treatment outcomes for cocaine use disorder (Myers et al., 1995, Rohsenow et al., 2007, Palazon-Llecha et al., 2024). Poorer long-term outcomes are related to higher cocaine-use severity at treatment admission (Simpson et al., 2002), and cocaine craving predicts treatment attrition and return to use (Palazon-Llecha et al., 2024). Daily cocaine-use patterns in the present study were also related to depression symptoms and motor impulsivity; however, these findings did not survive correction for multiple comparisons. Given the demonstrated relation between cocaine use patterns and treatment outcomes (Myers et al., 1995, Palij et al., 1996, Shearer et al., 2007, Roos et al., 2019, Mainville-Berthiaume et al., 2026), our results identifying distinct subgroups of cocaine use patterns in non-treatment-seeking individuals may have implications for targeted treatment interventions. While cocaine craving was higher in those who consistently used cocaine compared to those whose cocaine use was variable and low, cocaine craving did not differentiate participants whose cocaine use was variable versus low. Rather, age and age at regular use were the only factors that differentiated participants whose cocaine use was variable versus low (although these findings did not survive correction for multiple comparisons). The latter finding agrees with previous findings that age of cocaine-use onset differs by cocaine-use pattern (Myers et al., 1995); here, older age at onset of regular cocaine use was associated with infrequent cocaine use in this sample. Similarly, income has been repeatedly established as a predictor of cocaine use (Greenwald and Steinmiller, 2014, Petry, 2000) and differentiated low from variable and consistent cocaine use in this study. Daily cocaine-use patterns reveal unique features that may contribute to variable, low, and consistent use patterns that may not otherwise be elucidated when examining group-averaged data, with important implications for personalized treatment. Examining cocaine use in a non-treatment-seeking sample provides more generalizable information on baseline cocaine use patterns and their relations to clinical and demographic variables. Conducting TLFB interviews is likely too burdensome to employ in treatment settings. However, an understanding of related variables that can be more readily assessed in the clinical context may enable the identification of individuals that may require more extensive treatment management approaches to successfully address their cocaine use. Some limitations of this work should be considered. First, the sample size was relatively small for a latent class analysis; therefore, we focused our approach to include only binary (yes/no) daily cocaine use, rather than amount, data in classifying participants. Although we mitigated this by analyzing quantitative use data by class, including it in the initial RM-LCA would have been ideal. Sample size also restricted a) inclusion of multiple or all predictors into the same model, b) tests of interactions between predictors, and c) statistical power. Results should be interpreted with caution with these limitations in mind, and these limitations may seed future research addressing current gaps in knowledge. Second, our sample included mostly male and Black participants, all recruited from one geographical area. A more diverse sample would have better enabled us to assess the impacts of sex and race on cocaine-use patterns and improved generalizability of results. Third, cocaine use was based on two-week retrospective self-report data. This time frame may limit our ability to capture individual differences in longer-term use patterns, but it was deliberately chosen to minimize recall bias that may have impacted longer TLFB periods. Regardless, participants may still have over- or under-reported use for other reasons in addition to potential recall bias. Finally, this analysis defers examination of polysubstance use as a potential correlate of cocaine-use patterns. To start more simply, we elected to focus here on cocaine; then, in our companion paper (Grasser et al., 2026), we build greater complexity into our analysis. In conclusion, we conducted these analyses with the longer-term goal of informing tailored treatment development and stratification. Initially, we focus only on cocaine use in this sample, despite mounting evidence that polysubstance use is prevalent and negatively impacts outcomes for people who use substances (Williamson et al., 2006, Kedia et al., 2007, de la Fuente et al., 2014, Liu et al., 2019, Ferretti et al., 2025). Future work should therefore investigate the relations of polysubstance use patterns with relevant clinical and demographic variables. CRediT authorship contribution statement Grasser Lana R: Writing – review & editing, Writing – original draft, Visualization, Formal analysis, Conceptualization. William Burstein: Writing – review & editing, Conceptualization. Mark K. Greenwald: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. Mascarin Alixandria T: Writing – review & editing, Writing – original draft, Visualization, Formal analysis, Conceptualization. Ethics declaration Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed. This study adhered to relevant laws, regulatory frameworks and guidelines where the research took place. This study was approved by the Wayne State University Institutional Review Board. (Approval No. 040209M1F and 104810MP4F (two studies)) Role of funding source This work was supported by NIH grant R01 DA026861 (MKG) from the 10.13039/100000026National Institute on Drug Abuse, the Gertrude Levin Endowed Chair in Addiction and Pain Biology (MKG), and 10.13039/100009931Michigan Department of Health and Human Services (Helene Lycaki/Joe Young, Sr. Funds). Funding sources had no role in designing or conducting this study, nor in preparing this manuscript. Declaration of Competing Interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Mark Greenwald reports financial support was provided by National Institute on Drug Abuse. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.