Relationship between depression on cocaine craving over time amongst patients in a trial evaluating the use of buprenorphine on cocaine use

Abstract

Aims

This study aimed to examine within and between effects of the relationship between depression, using the Beck Depression Inventory (BDI), and cocaine craving, using the visual analog craving scale (VAS), over time.

Methods

Data from the NIDA Clinical Trial Network Study Cocaine Use Reduction with Buprenorphine (CTN-0048) were used in a secondary analysis (CTN-0148). Random-effects regression modelling was used to examine relationships between participants’ depressive symptoms and their cocaine craving over time.

Results

A total of 301 participants with past-year DSM-IV criteria for cocaine and opioid use disorder were analyzed (21.6% female, 10.3% Hispanic, 66.4% Black) being treated with placebo or buprenorphine + naloxone. Craving significantly decreased over time (B = −1.11, 95% CI [-1.21, −1.02], p < 0.001). Depression emerged as a significant within-person predictor of craving over time (B = 0.93, 95% CI [0.76, 1.09], p < 0.001), indicating that when a person’s BDI score increased by one point from their own mean, their craving increased by 0.71 units. Between-person differences in average BDI did not have significant effects (p > 0.05), indicating that depression scores across participants did not significantly predict differences in craving.

Conclusions

These findings highlight depression as a dynamic, time-varying clinical marker of heightened craving risk and suggest that monitoring and addressing increases in depressive symptoms during treatment may help mitigate craving spikes and potentially reduce vulnerability in returning to use.

Highlights

• •Depression emerged as a significant within-person predictor of craving over time.
• •When a person’s BDI score increased by one point from their own mean, their craving increased by 0.71 units.
• •Depression scores across participants did not significantly predict differences in craving.
• •Depression is a dynamic, time-varying clinical marker of heightened craving risk.
• •Monitoring and addressing increases in depressive symptoms during treatment may help mitigate craving spikes.

1. Introduction

Cocaine craving, commonly defined as a subjective desire or urge to use cocaine, is central to how cocaine use disorder (CUD) is diagnosed, monitored, and treated (Hasin et al., 2013). In the context of CUD treatment research, cocaine craving is considered a particularly meaningful treatment indicator as it is often experienced as the immediate driver of cocaine seeking behaviors and has been closely tied to return to use (Hartz et al., 2001, Tiffany and Wray, 2012). In fact, craving has been identified as a central symptom in substance use disorder showing connectivity in diagnostic criteria and an important treatment target (Gauld et al., 2023).

A strong empirical literature supports craving as a predictor of subsequent cocaine use and return to use (Toulami et al., 2025). For example, craving assessed during treatment has been shown to predict time to return to use in patients who use cocaine, (Paliwal et al., 2008) while a recent meta-analysis conducted across different drug types found craving indicators to be prospectively associated with later drug use or return to use (Vafaie and Kober, 2022). Notably, craving is highly sensitive to conditioned drug cues and context (Perry et al., 2014). Classic cue-reactivity work demonstrates that exposure to drug-related stimuli reliably elicits craving and related responses across substances, supporting conditioning-based models of return to use risk (Carter and Tiffany, 1999). This cue sensitivity is especially relevant in real-world settings where individuals encounter triggers (i.e., people, places, paraphernalia, stressors) that can rapidly amplify subjective urge and reduce inhibitory control (Pike et al., 2013). As a result, understanding craving in CUD requires moving beyond static, cross-sectional assessments to models that treat craving as a dynamic, time-varying process (Serre et al., 2015, Vafaie and Kober, 2022).

Beyond cue exposure and context, depressive symptoms and negative affect are especially important drivers of craving dynamics (Sinha et al., 1999, Sinha et al., 2000). Depressive states are common among individuals with CUD and has been linked to heavier substance involvement and worse treatment outcomes (Brown et al., 1998). At a broader level, meta-analytic evidence indicates that depressive symptoms among people who use cocaine is associated with greater substance use and impairment, supporting clinical concern that comorbid mood symptoms may maintain or exacerbate cocaine-related challenges (Conner et al., 2008).

Importantly, a diagnosis of depression, like craving, is not simply a stable characteristic, but is rather a mood state that varies over time (Eddington et al., 2017, Martikkala et al., 2025). As a result, depression and depressive states may influence craving in time-varying, within-person ways, meaning that changes in a person’s mood from their own typical level may be more predictive of craving spikes than stable differences between people (Jenkins et al., 2021). That is, on days when depressive symptoms intensify, craving may rise accordingly, increasing near-term risk of return to use (Moore et al., 2014). Secondary analyses of clinical trials may offer a strong platform to evaluate time-varying processes. The Cocaine Use Reduction with Buprenorphine (CURB) trial (Ling et al., 2016, Mooney et al., 2013) provides longitudinal craving and depressive symptom data among individuals seeking CUD treatment, supporting questions about how craving changes during treatment and how depressive symptomatology interact with those changes. Furthermore, by separating within-person effects (visit-to-visit deviations in depressive symptoms predicting deviations in craving) from between-person effects (stable differences in average depression predicting average craving), analyses may provide a better understanding of whether depression functions primarily as a static vulnerability marker or may also act as a fluctuating proximal trigger for craving escalation. Within this context, this study aimed to examine both between- and within-person effects of the relationship between depression and cocaine craving over time.

2. Methods

2.1. Secondary data source

The original data set comes from a previously reported clinical trial, A randomized study to test the safety and effectiveness of buprenorphine in the presence of naltrexone for the treatment of cocaine dependence (CTN0048; ClinicalTrials.gov record ID: NCT01402492). This trial recruited 302 subjects from 11 sites total in Washington DC, New York, Ohio, Georgia, Texas, Colorado, Washington, Oregon, and California. This double-blind, randomized controlled trial included three treatment arms wherein all participants received extended-release injectable naltrexone. One arm also received 8 weeks of placebo, while two arms received 8 weeks of buprenorphine + naloxone at doses of 4 or 16 mg/day of buprenorphine. All participants were offered weekly cognitive-behavioral return to use prevention psychotherapy. A full description of this trial can be found at the protocol and the primary outcome publications (Ling et al., 2016, Mooney et al., 2013).

2.2. Participants

Of the 302 participants originally recruited, 301 were included in this secondary analysis. One participant was removed due to a duplicated ID number. All participants were between the ages of 18 and 65, met DSM-IV diagnosis criteria for past-year cocaine and opioid use disorders, and were seeking treatment for CUD. The secondary analysis of these data was deemed non-human research by the Washington State University Institutional Review Board.

2.3. Measures

2.3.1. Craving Visual Analog Scale (VAS)

The VAS measures craving with a single self-administered question inquiring about a person’s intensity of their strongest craving in the past 24 h. It is scored using an analog scale from 0 to 100 where 0 identifies no craving in past 24 h and 100 signifies an extremely strong craving for cocaine in the past 24 h (McMillan and Gilmore-Thomas, 1996). This analysis utilized measurements collected at baseline and once per week throughout the duration of the 8-week treatment for a total of 9 repeated assessments (Mooney et al., 2013).

2.3.2. Beck’s Depression Inventory (BDI)

The BDI is a 21-item self-administered assessment of depression severity. When interpreting the BDI, answers can range from 0 to 3 for each question with 0 indicating no depressive symptoms and 3 indicating severe depressive symptomology and are used to determine an overall depression score. The highest overall score possible on this assessment is 63 (severe depression) and the lowest is 0 (indicating no depression). Categorizations that fall within this scoring include minimal depression (score of 0–13), mild (score of 14–19), moderate (score of 20–28), and severe (score of 29–63) (Beck et al., 1988). The BDI was collected at the same time-points as the VAS.

2.4. Data analyses

Our intention was to examine relationships between depression (BDI score) with cocaine craving over time. Covariates included treatment assignment, age, sex, ethnicity, and race. We conducted Hausman tests (robust sigmamore) to compare random- and fixed-effects specifications. This test indicated a significant difference in coefficients (X²(2) = 8.29, p = 0.02), suggesting that the assumptions of a conventional random-effects model may be violated because predictors may be correlated with unobserved individual-level effects. While this is often addressed using a fixed-effects model, it would preclude our ability to examine time-invariant covariates, a key interest in this study. We thus elected to use a hybrid (within-between) model (Schunck, 2013, Schunck and Perales, 2017), (xthybrid command in STATA), estimating a hybrid generalized linear mixed model (GLMM) with a Gaussian family and identity link (Schunck and Perales, 2017). This approach decomposes time-varying covariates into within-person deviations from each individual’s mean and the individual’s person-specific mean. Thus, time-varying covariates were effectively person-mean centered rather than grand-mean centered, whereas covariates (all time-invariant) were entered as time-invariant level−2 predictors and were not decomposed into within- and between-person components. In this parameterization the within-person coefficient is equivalent to the fixed-effects estimate for time-varying predictors, while also allowing the inclusion of time-invariant covariates (Schunck, 2013, Schunck and Perales, 2017). We note, however, that this protection applies most directly to the within-person coefficients. Interpretation of between-person effects and time-invariant covariates still assumes that after conditioning on the person-specific means of the time-varying covariates, the remaining individual-level random effect is not correlated with the observed level−2 covariates (Schunck, 2013, Schunck and Perales, 2017).

Our outcome, craving, was modeled as a function of time (linear and quadratic terms), depression severity (BDI score), treatment assignment, and demographic variables, with repeated observations clustered within patient. Depression severity and craving were modeled as contemporaneous repeated measures at each assessment, such that the estimated within- and between-person effects reflect associations at concurrent timepoints, rather than lagged prospective effects. No lagged predictors were included in the model. Additionally, no “randomslope” option was specified in our model and thus it included a random intercept for patient only and did not estimate random slopes. Baseline covariates (e.g., treatment group, demographic indicators) were treated as level 2 predictors. Visual inspection of exploratory plots of the relationship between depression and craving suggested potential non-linearity, therefore a quadratic term was included at the between-person level to test for potential curvilinear effects.

Analyses were conducted using STATA 17.0 (StataCorp, 2025). An α-level of.05 and two-tailed tests of significance were used for all inferential analyses. Unstandardized coefficients are presented to facilitate interpretation of the results in the units of the original measurement, thus promoting direct clinical application of the results. In xthybrid, the estimation sample is defined by excluding observations with missing values on variables that are included in the model. Thus, missingness on a time-varying outcome or covariate results in omission of that observation, whereas missingness on a time-invariant covariate can exclude all observations for the participant. Person-specific means and within-person deviations are then computed using the remaining nonmissing observations. Due to the minimal amount of missing data (<1% on baseline variables; 8%, range 7–13% across timepoints on craving), no additional missing data management was employed.

3. Results

3.1. Descriptive statistics

Data from n = 301 participants were included in the analyses. Most participants identified as Black (66.4%), 21.6% were female, and 10.3% were Hispanic. Mean VAS score at baseline was 56.39 (SD = 30.81) and mean BDI score at baseline was 14.10 (SD = 9.54). See Table 1 for full sample descriptive statistics.

Table 1.

Full Sample Descriptives with Multilevel Model Results.

Predictor	Estimate	SE	z	95% CI	P > |z|
Within-person effects
Depression (person-mean centered)	0.93	0.08	10.93	[0.76, 1.09]	< 0.001
Time (linear)	−1.11	0.05	−22.85	[−1.21, −1.02]	< 0.001
Between-person effects
Depression (person-mean centered) Linear Effects	0.75	0.44	1.70	[−0.11, 1.61]	0.089
Depression (person-mean centered) Quadratic effects	0.01	0.02	0.60	[−0.02, 0.04]	0.552
Covariates (between; modeled as time invariant)
Treatment group (reference = Control)
2	0.84	2.75	0.31	[−4.54, 6.23]	0.759
3	2.02	2.72	0.74	[−3.31, 7.35]	0.458
Age	0.03	0.14	0.23	[−0.24, 0.30]	0.822
Sex (reference = Male)	2.12	2.70	0.78	[−3.17, 7.41]	0.433
Ethnicity (reference = Non-Hispanic)	−0.77	4.67	−0.16	[−9.93, 8.40]	0.870
Race (reference = black)
American Indian	1.38	5.10	0.27	[−8.63, 11.38]	0.787
Asian	−14.43	18.96	−0.76	[−51.60, 22.73]	0.447
White	−3.07	2.61	−1.18	[−8.19, 2.05]	0.240
Other	−0.53	6.33	−0.08	[−12.94, 11.87]	0.933

Note: Quadratic terms are not included for within-person effects due to potential interaction with centering and random slopes

Model fit for our GLMM was statistically significant (Wald X²(15) = 1044.23, p < 0.0001), with adequate fit to the data (Log likelihood = −12411.16). There was notable variability in craving across (random intercept variance = 317.88, 95% CI [265.14, 381.10]), and within (residual variance = 362.64, 95% CI [342.96, 383.44]) participants, supporting the use of a mixed-effects approach. The intraclass correlation coefficient (ICC) was 0.47 (SE = 0.024, 95% CI [0.42, 0.52]), indicating that nearly half (46.7%) of the total variance in craving was attributable to between-person differences, with the remaining variance reflecting within-person fluctuations over time. Additionally, the likelihood ratio vs. linear model test indicated that the mixed-effects model improved fit over a linear model (X² = 1084.72, p < 0.0001).

3.2. Between-person effects

As can be seen in Table 1, between-person differences in average BDI did not have significant linear (p = 0.09) or quadratic (p = 0.55) effects, indicating that depression scores across participants did not significantly predict differences in craving. Effects were not significant for treatment group, age, sex, ethnicity, or race (all p > 0.05).

3.3. Within-person effects

Craving significantly decreased over time (B = −1.11, 95% CI [−1.21, −1.02], p < 0.001), indicating a decrease of 1.11 units in craving on the VAS for each consecutive week of treatment. Depression score emerged as a significant within-person predictor of craving (B = 0.93, 95% CI [0.76, 1.09], p < 0.001), indicating that when a person’s BDI score increased by one point from their own mean, their craving increased by 0.93 units. See Table 1 for additional detail on effects.

4. Discussion

To our knowledge, this work was the first to examine both between- and within-person effects of the relationship among depression and cocaine craving over time, while controlling for the treatment assignment group and demographic data. Results remained significant after controlling for treatment group, indicating that these relationships were not attributed to the effects of medication.

Overall, this secondary analysis observed three main findings: 1. cocaine craving decreased over the 8-week treatment period; 2. week-to-week changes in depressive symptoms were strongly associated with concurrent changes in craving within individuals; and 3. between-person differences in average depression severity did not significantly explain differences in craving across participants.

As observed, craving declined by approximately 1.11 VAS units per week, suggesting a cumulative and meaningful improvement over time during the duration of treatment. This sustained downward trajectory aligns with findings showing that evidence-based interventions for CUD are effective in reducing craving (Johnson et al., 2020). At the same time, our model indicated substantial within-person variability in craving, emphasizing that reductions in mean craving do not necessarily preclude clinically meaningful craving spikes. This finding aligns with broader evidence showing that craving is time-limited and context-sensitive and supports shifting from static, cross-sectional conceptualizations toward time-varying models of craving risk that better reflect the lived experience of CUD and the episodic nature of return to use vulnerability (Vafaie and Kober, 2022).

The results indicated that between-person effects were not statistically significant, suggesting that mean differences across individuals did not explain variability in outcomes. Clinically, this finding implies that cross-sectional comparisons of depressive symptom severity may be insufficient for capturing stable individual differences. In particular, within the context of stimulant use, depressive symptoms may contribute to dynamic fluctuations in negative affect, reflecting depressive symptoms that vary over time rather than a stable trait-like disposition. This temporal variability underscores that depressive symptomatology in this population may not align with the relatively enduring patterns observed in major depressive disorder or chronic depressive conditions. Instead, both within-person variability and short-term fluctuations appear to be more prominent. Consequently, aggregating data across individuals may obscure meaningful, individual patterns. Notably, this limitation persists even when individuals present with substantial heterogeneity in baseline depressive symptom severity at treatment initiation.

Perhaps the most clinically informative results from this analysis were the robust within-person association between depressive symptoms and cocaine craving and the lack of a significant effect of the between-person association between depressive symptoms and cocaine craving over time. Such findings suggest that depressive states may play a significant role as a proximal trigger for craving escalation and has the potential to impact imminent return to use, rather than a static predictor of clinical profile severity and treatment prognosis. Such a hypothesis is aligned with novel theoretical models such as the Addiction Neuroclinical Assessment Framework that suggests that depressive symptoms may increase craving via negative reinforcement processes (using cocaine to escape dysphoria), reduced reward sensitivity (increasing reward salience), and/or diminished executive control and self-regulation during low mood states (Koob and Le Moal, 2008, Koob and Schulkin, 2019, Koob and Volkow, 2016, Kwako et al., 2017, Kwako et al., 2016).

These results have practical implications for CUD treatment and monitoring. If increases in depressive symptoms are coupled with increases in craving within the same individual, then routine, repeated mood measurement may provide clinically actionable information that static baseline screening cannot. This notion supports measurement-based care approaches in which mood symptoms are monitored over time and used to inform stepped care decisions or adaptive supports (Lewis et al., 2019). Because craving is prospectively linked to return to use risk across substances and settings, (Brown et al., 1998) identifying mood-linked craving spikes may help anticipate near-term risk. It also points to the potential value of integrating mood-focused strategies (e.g., behavioral activation, coping skills for dysphoria, stress management) into CUD treatment interventions (Conner et al., 2008).

4.1. Strengths and limitations

This study has several strengths, including a large sample size (n = 301) with participants from multiple sites across the United States. The original study had rigorous methodology in its randomized, double-blinded, placebo-controlled clinical trial design. Another strength lies in the fact that both the BDI and the VAS assessments were collected at the same time-points.

This study has limitations typical of secondary analyses. First, the analysis is observational with respect to the depression–craving coupling; therefore, causal direction cannot be established (e.g., increased craving could also worsen mood). Second, both depression and craving were measured weekly; higher-frequency assessments might capture more granular lagged effects and clarify temporal ordering. Third, observations with missing values on variables included in the model were excluded from the estimation sample; although missingness was modest, bias is possible if missingness is related to symptom severity or other unobserved factors. Fourth, results may not generalize beyond treatment-seeking adults with co-occurring cocaine dependence and opioid use history enrolled in a clinical trial with intensive contact and pharmacotherapy. Additionally, our analyses modeled depression severity and craving contemporaneously, so findings characterize concurrent associations and do not establish whether changes in depression preceded subsequent changes in craving.

5. Conclusion

Depression severity emerged as a strong within-person predictor of craving over time, suggesting that mood fluctuations, rather than stable differences in depression severity between individuals, may be particularly important for identifying when craving risk is elevated. These findings support time-varying, personalized models of craving and highlight depression as a clinically actionable signal for targeted intervention during CUD treatment.

Future studies should test lagged within-person models (e.g., whether increases in depression predict next-week craving, or vice versa), evaluate moderators (stress exposure, cue exposure, sleep, etc.), and examine whether mood-linked craving spikes mediate outcomes related to return to use. Given evidence that stress and negative affect can experimentally increase cocaine craving, (Conner et al., 2008, Sinha et al., 1999) studies integrating laboratory paradigms with intensive longitudinal assessments may be especially informative. Lastly, from an intervention standpoint, adaptive approaches that intensify support during periods of rising depressive symptoms (i.e., initial abstinence) may help reduce craving surges and improve outcomes.

Fig. 1.

Regression Coefficients by Effect Type.

Disclosures

Given his role as Associate Editor, Sterling McPherson, PhD had no involvement in the peer-review of this article and has no access to information regarding its peer-review. Additionally, given their role as an Editorial Board Members, Andrew Saxon, MD and Andre Miguel, PhD had no involvement in the peer-review of this article and have no access to information regarding its peer-review. Full responsibility for the editorial process for this article was delegated to Editor-in-Chief, Teri Franklin.

CRediT authorship contribution statement

Sterling M. McPherson: Writing – review & editing, Supervision, Funding acquisition. Andrew J. Saxon: Writing – review & editing, Supervision, Data curation. Andre C. Miguel: Writing – review & editing, Writing – original draft, Methodology, Conceptualization. Rachel Ryan: Writing – review & editing, Writing – original draft. Smith Crystal L.: Writing – review & editing, Writing – original draft, Visualization, Software, Methodology, Formal analysis, Data curation, Conceptualization. Rodin Nicole: Writing – review & editing, Writing – original draft, Project administration, Methodology, Conceptualization.

Funding

This work was supported by the National Drug Abuse Treatment Clinical Trials Network Pacific Northwest Node (UG1 DA013714; CTN−0148, MPIs Hatch, Roll) and 5UG1DA013714–21 Supplement (MPI: Roll, Hatch-Maillette; Supplement MPI: Smith, McPherson, Saxon).

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.