Preliminary Evidence of Inhibitory and Error-Related Neural Alterations Associated with Early Relapse in Cocaine Use Disorder: A Retrospective Go/No-Go Event-Related Potential Study

Abstract

Introduction

Cocaine use disorder (CUD) is associated with impairments in inhibitory control and error-monitoring, two neurocognitive mechanisms strongly linked to relapse. Characterizing how these processes differ at treatment entry between patients who later remain abstinent versus those who relapse may help clarify the neurophysiological factors associated with early recovery.

Methods

Thirty-one CUD patients undergoing a 15-day inpatient detoxification program were included and classified as abstainers (n = 16) or relapsers (n = 15) based on drug use reports 1 month after discharge. Participants completed an EEG Go/No-Go task approximately 4 to 7 days after admission, once acute withdrawal symptoms had subsided. Inhibitory (N2d, P3d) and error-related (ERN, PE) ERP components were extracted. Multivariate analyses controlled for psychotropic medication status.

Results

No behavioral differences were observed between groups. However, abstainers showed larger P3d and ERN amplitudes, reduced N2d and PE amplitudes, and longer P3d latencies than relapsers. This profile suggests more efficient inhibitory processing and sharper early error monitoring in abstainers, while relapsers exhibited attenuated error detection and greater reliance on later evaluative processes.

Conclusion

The findings indicate that CUD patients differ markedly in their inhibitory and error-monitoring neural responses already at treatment entry. ERP measures may therefore constitute clinically relevant indicators of cognitive functioning in early detoxification and could support more individualized therapeutic strategies.

Introduction

Cocaine use disorder (CUD) is a major global health problem defined as the compulsive use of cocaine despite its harmful social, professional, familial, and medical consequences. Since 2012, all around the world, a marked increase in weekly cocaine use has been observed in both adults and adolescents. In 2017, an estimated 0.4% of the adult population aged 15–64 years (around 18 million people) reported past‐year cocaine use [1]. This contributes, for instance, to a dramatic increase in the rate of overdose deaths, of emergency room visits due to overdose, accidents, or violent acts, or of the risk of contracting HIV and hepatitis C [2, 3]. However, despite multiple clinical trials, recent meta-analyses and narrative reviews have shown that no treatment (psychosocial and/or pharmacological) has demonstrated clear efficacy in treating CUD [4, 5], so that the relapse rate remains tremendously high [6].

Understanding which factors may help cocaine addicts recover and which ones may contribute to relapse risk is therefore still a major public health issue. Contemporary neurocognitive models conceptualize addiction as a dysregulation between reward-related systems and cognitive control networks, with impaired engagement of prefrontal inhibitory mechanisms contributing to compulsive drug use [7]. Among the multiple (genetic, environmental, etc.) causes contributing to the onset of CUD, a main investigated factor refers to behavioral disinhibition, i.e., the inability to refrain impulses or socially restricted actions [8]. This decreased impulse control, due to a hypoactivity of brain regions involved in the response inhibition circuit (right anterior cingulate cortex, right insula, right inferior parietal lobule, and right middle frontal gyrus [9]), would be partly responsible for the transition from single to daily (compulsive) drug usage [10].

At the experimental level, a classical way to study inhibitory control refers to the speeded motor Go/No-Go task. Participants must press as quickly as possible a button in response to one type of frequently presented stimuli (“Go”) and withhold that response to some rare and random stimuli of another type (“No-Go”). At the behavioral level, many studies have shown that CUD individuals have difficulties withholding the motor dominant response (hits to Go trials) when they are presented with a No-Go trial. Compared to nonusers, the number of commission errors (hits on No-Go trials) is therefore increased, while the speed of the response (reaction times [RTs]) to Go trials is decreased, as time pressure is known to generate a higher rate of false alarms on No-Go trials [11]. When successful, inhibition is a latent process, producing no overt measurable behavioral response (absence of hit on a No-Go trial). For this reason, neurophysiological approaches are particularly valuable, as they allow the investigation of covert cognitive processes that may not be reflected in overt performance. Event-related potentials (ERPs), derived from electroencephalography (EEG), offer millisecond-level temporal resolution, making them especially well suited to disentangle the rapid and sequential stages of cognitive control involved in inhibitory processing and error monitoring. In contrast to neuroimaging techniques such as functional magnetic resonance imaging or fNIRS, which provide superior spatial resolution but limited temporal precision, ERPs allow fine-grained characterization of cognitive processes. Moreover, EEG is noninvasive, relatively cost-effective, and well tolerated in clinical detoxification settings, which makes it particularly suitable for investigating relapse vulnerability in naturalistic inpatient contexts. ERPs have therefore been widely used to investigate the neural processes indexing successful inhibitory control in healthy controls [12]. Two main fronto-central components were disclosed: a No-Go N200, arising around 200–400 msec, indexing conflict-monitoring processes, and a No-Go P300, arising around 300–600 ms, reflecting motor inhibitory processes [13, 14]. Interestingly, ERP studies were also devoted to the investigation of neural correlates of inhibitory errors, as detecting errors and evaluating performance is a fundamental process of cognitive control allowing learning and flexible behaviors [15]. Two independent error-related ERPs waves emerge after committing errors: (1) a fronto-central error-related negativity (ERN), arising around 50–150 ms after an incorrect response, and representing the brain automatic detection of errors, and (2) a positive centro-parietal potential (error positivity [PE]), arising around 200–500 ms after an incorrect response, and being modulated by the degree of awareness of an error [16]. Importantly, error-related neural activity partially overlaps with more general response-monitoring processes observed during correct responses, often indexed by the correct-related negativity (CRN). To isolate error-specific processing, ERN analyses are therefore commonly interpreted relative to the CRN, using difference waveforms [16]. Importantly, alterations in ERN and PE amplitudes have been described as transdiagnostic markers of disrupted performance monitoring across psychiatric conditions, including addictive disorders [15]. Similar ERP studies applied to CUD patients have been particularly insightful, as monitoring cognitive control and error processing processes through No-Go N200/N2d, No-Go P300/P3d, ERN, and PE components has helped highlight alterations due to cocaine abuse and its neurotoxicity (reduced amplitudes were common results when compared to controls [17, 18]), but also to assess mechanisms predicting abstinence versus relapse. Indeed, recovered inhibitory functions indexed by recovered No-Go N2d and P3d amplitudes were good predictors of abstinence [19, 20], while reduced capacity to detect errors (by preventing the adjustment of motor response to task demand), indexed by reduced ERN and increased PE amplitudes, defined a vulnerability factor to relapse [21–23]. Such findings are quite common for addictive disorders in general, as recovered P3d components predict abstinence in patients detoxified for alcohol [24, 25] and heroin [26, 27], while reduced ERN amplitudes at the entrance of a detoxification program for alcohol use disorder also predict early relapse [28]. This reinforces the idea that recovering inhibitory resources is a key protective factor against relapse, while difficulties in error detection may trigger early relapse in addictive states. Training such functions could therefore be of the greatest relevance at the clinical level, and encouraging results have already been obtained for cocaine and substance use disorders in general [29, 30].

In this view, the main aim of this retrospective study was to replicate previous findings by investigating the N2d, P3d, ERN, and PE components. The present study extends prior work by providing an integrated examination of inhibitory (N2d, P3d) and error-related (ERN, PE) components within the same clinical sample at treatment entry, allowing a more comprehensive characterization of early relapse vulnerability. The study aimed to analyze available ERP data from CUD patients who participated in a 15-day detoxification program between 2022 and 2025 and to determine whether neural markers of inhibitory control measured at treatment entry differentiate future abstainers from future relapsers.

Specifically, we hypothesized that larger N2d and/or P3d amplitudes (indexing inhibitory engagement [25]) would characterize patients who maintain abstinence, whereas reduced ERN amplitudes and increased PE amplitudes (indicative of altered early error detection and post-error adjustment, respectively [27, 31]) would predict early relapse following discharge. By simultaneously examining inhibitory and error-monitoring dynamics in a naturalistic inpatient context, the present study extends prior work and seeks to provide a more comprehensive characterization of early relapse-related neural vulnerability. If confirmed, such results could have important clinical implications, enabling clinicians to identify, at the entrance of a detoxification program, CUD patients at high risk of relapse and guide them toward tailored, individualized, cognitive rehabilitation programs.

Methods

Participants

This study retrospectively analyzed data from patients diagnosed with CUDs according to the DSM-V criteria (APA, 2013), who underwent a 2-week inpatient detoxification program between 2020 and 2025. Inclusion and exclusion criteria were as follows: only patients with reliable EEG recordings (i.e., acceptable signal-to-noise ratio) who completed the anamnesis, provided informed consent for the use of their data in research, and agreed to be contacted for a 1-month follow-up regarding relapse status were considered.

We collected data from 40 patients based on several exclusion steps to ensure data quality. Of these, 31 were included in the study (aged between 25 and 65 years old, M_age = 39.74, SD = 7.14, 9.7% females). Patients were excluded if they had a history of neurological disorders, any major medical condition, current use of substances other than cocaine, alcohol, and benzodiazepines, any current DSM-V axis I diagnosis other than substance or alcohol dependence, a major medical condition, pregnancy, or poor ERP quality (i.e., low signal-to-noise ratio, n = 3). Poor ERP quality was defined as an insufficient number of artifact-free trials remaining after preprocessing to compute reliable averaged ERPs. Specifically, participants were retained only if they provided at least 10 artifact-free correct No-Go trials for the N2d/P3d quantification, and at least 5 commission error trials for the ERN/PE quantification, in line with established psychometric reliability thresholds for these components [32, 33]. Patients who could not be reached or declined to participate in the follow-up interview assessing relapse status were also excluded (n = 6). The remaining patients (n = 31) were categorized into two groups: those who relapsed (n = 15) and those who remained abstinent 1 month after discharge (n = 16).

Patients were assessed approximately 4 to 7 days after admission, once withdrawal symptoms subsided and a clinically stable state was reached. EEG assessments were scheduled based on clinical stability rather than a fixed time point. During the session, patients were confronted with an ERP Go-No-Go task, rated their craving for cocaine using the Cocaine Craving Questionnaire – brief (CCQ-brief [34]), and completed additional questionnaires to control for factors known to trigger relapse: the severity of alcohol dependence – the Alcohol Use Disorder Identification Test (AUDIT [35]); the severity of benzodiazepine dependence – Cognitive Scale of Attachment to Benzodiazepines (ECAB [36]); the severity of tobacco dependence – the Fagerstrom [37]; the severity of drug dependence – the Drug Abuse Screening Questionnaire (DAST [38]); the temporal experience of pleasure – Temporal Experience of Pleasure Scale (TEPS [39]); depression – the Beck Depression Inventory (BDI-II [40]); generalized anxiety – the State-Trait Anxiety Inventory (STAI [41]); metacognition – the metacognition questionnaire (MCQ-30 [42]); impulsivity – the Urgency Premeditation Perseverance and Sensation Seeking Impulsive Behavior Scale (UPPS [43]).

Medication data were collected at the time of the discharge. Among the patients, 93% of the relapsers were taking psychotropic medication while it was the case for only 56.2% of the abstinent patients. This difference was statistically significant, χ²(1) = 5.560, p = 0.018.

Follow-up data on drug use were collected 1 month after discharge to assess relapse. Relapse was defined as a return to drug dependence according to the DSM-V criteria. Patients were classified as “abstinent” only if they remained entirely drug-free throughout the follow-up period. Information was gathered through phone interviews with patients and confirmed by close relatives and general practitioners. The characteristics of the groups, categorized as abstainers or relapsers, are detailed in Table 1.

Table 1.

Demographic characteristics of abstainer and relapser groups

​	Abstainers (n = 16)	Relapsers (n = 15)	Statistical tests
Gender (male:female)	15:1	13:2	χ2(1) = 0.444, p = 0.505
Age	39.81±7.48	39.67±7.03	F(1, 29) = 0.003, p = 0.956
Consumption duration, years	9.06±6.98	11.20±8.27	F(1, 29) = 0.607, p = 0.442
AUDIT	16.06±9.8	11.36±10.10	F(1, 29) = 1.67, p = 0.207
ECAB	2±1.9	2.21±2.61	F(1, 29) = 0.067, p = 0.797
Fagerstrom	3.13±2.78	5±2.54	F(1, 29) = 3.680, p = 0.065
DAST	5.75±3.13	6.29±2.99	F(1, 29) = 0.227, p = 0.637
EETP-Cons	35.31±7.7	35±5.25	F(1, 29) = 0.016, p = 0.899
EETP-Ant	42.62±8.74	44.79±10.21	F(1, 29) = 0.39, p = 0.537
BDI	22.85±13.66	22.16±11.66	F(1, 29) = 0.019, p = 0.891
STAI-A	58.25±15.43	55.71±10.79	F(1, 29) = 0.264, p = 0.611
STAI-B	58.38±17.08	56.14±9.01	F(1, 29) = 0.192, p = 0.665
MCQ – positive beliefs	13.06±4.71	12.79±2.75	F(1, 29) = 0.037, p = 0.849
MCQ – negative beliefs	13.19±5.5	14.64±4.32	F(1, 29) = 0.637, p = 0.432
MCQ – cognitive confidence	11.75±4.03	11.14±4.33	F(1, 29) = 0.158, p = 0.694
MCQ – need to control thoughts	13.38±3.93	14.64±3.41	F(1, 29) = 0.878, p = 0.357
MCQ – cognitive self-consciousness	16.88±3.28	16.07±3.12	F(1, 29) = 0.468, p = 0.500
UPPS total	50.44±10.24	49.07±9.7	F(1, 29) = 0.139, p = 0.712
Craving	20.5±10.86	27.43±15.12	F(1, 29) = 2.117, p = 0.157
Psychotropic medication (yes:no)	9:7	14:1	χ 2 (1) = 5.560, p = 0.018
Antidepressants (yes:no)	6:10	11:4	χ 2 (1) = 4.014, p = 0.045
Anxiolytic/sedative medication (yes:no)	4:12	5:10	χ2(1) = 0.261, p = 0.609
Antipsychotic (yes:no)	3:13	9:6	χ 2 (1) = 5.552, p = 0.018
Mood stabilizers (yes:no)	1:15	2:13	χ2(1) = 0.444, p = 0.505

AUDIT, Alcohol Use Disorder Identification Test; ECAB, Cognitive Scale of Attachment to Benzodiazepines; DAST, Drug Abuse Screening Questionnaire; TEPS, Temporal Experience of Pleasure Scale; BDI, Beck Depression Inventory; STAI, State-Trait Anxiety Inventory; MCQ, metacognition questionnaire; UPPS, Urgency Premeditation Perseverance and Sensation Seeking Impulsive Behavior Scale.

The Go-No-Go Task

During the Go/No-Go task, all the participants were seated at 1 m from the screen (17-inch monitor), where the stimuli were centrally displayed in yellow capital letters (Arial, 500 × 400 mm in size) on a black screen; whenever the letter M (“Go” trial) was displayed, the participants had to press a button with the index finger of the right hand, as quickly and accurately as possible, and when the letter W (“No-Go” trial) was displayed, they had to withhold pressing the button. Stimuli were presented in a large, high-contrast format using highly familiar capital letters, which likely minimized potential difficulties related to mirrored-letter discrimination.

Participants completed two blocks of this task, each comprising 133 letters displayed in a semi-random order to avoid the presentation of two consecutive “No-Go letters.” Each block comprised 93 “Go” (70%) and 40 “No-Go” (30%). The characteristics of the stimuli presentation are shown in Figure 1.

Fig. 1.

Go-No-Go task.

EEG Recording

EEGs were recorded by 21 electrodes fixed in an electrode Quick-Cap with a linked mastoid physical reference (M1, M2). The electrode positions included the standard 10 to 20 system locations (Fpz, Fp1, Fp2, Fz, F3, F4, F7, F8, Cz, C3, C4, T7, T8, P3, Pz, P4, P7, P8, O1, Oz, O2). The amplification of the EEG was realized by a battery-operated A.N.T.^® amplifier with a gain of 30,000 and a band-pass of 0.01–100 Hz. The ground electrode (AFz) was positioned between FPz and Fz along the midline. The impedance of the electrodes was maintained below 10 kΩ during the experiments. The recording of the EEG was continuously made at a sampling rate of 256 Hz with A.N.T.^® Cognitrace software. A cutoff of 30 mV was used to define trials that were contaminated by either eye movement or muscular artifacts. They were corrected offline with a regression-based procedure as per Gratton et al. [44].

Once acquired, a band-pass filter from 0.3 to 30 Hz was applied. To generate the “Go” and “No-Go” ERP components, stimulus-locked epochs of 1,000 msec (200 msec before and 800 msec after the stimulus onset) were created with a baseline correction of −200 to 0 msec. Incorrect responses (i.e., missed hits for “Go,” or a hit for “No-Go”) were excluded from further analysis. Specific “inhibitory No-Go effect” tagging inhibitory control skills were obtained by computing the difference “No-Go minus Go” waveform, from giving rise to the N2d and the P3d (d for difference) waves [45]. As ERP latencies provide information about the processing speed, the individual latency has been quantified, according to the peak latency method [46]. The time after the stimulus onset at which the peak reaches the maximum negative amplitude for the N2d (within the 200 to 400 msec) and then the maximum positive amplitude for the P3d component (within the 400 to 600 msec) was identified in a cluster of classical frontal and central electrodes (Fz, F3, F4, and Cz) [28, 47]. The N2d (200–400 ms) and P3d (400–600 ms) quantification windows were deliberately defined as broader than those typically reported in healthy populations, in order to accommodate the latency prolongation of inhibition-related ERPs documented in substance use disorders; this is consistent with the P3d latencies observed here (e.g., 500 ± 44 ms in the abstinent group).

To generate the “ERN” and “CRN” waveform, response-locked epochs of 700 ms (200 msec before and 500 msec after the response onset) were created. Only trials related to correct hits for Go stimuli and hits for No-Go stimuli (commission errors) were included in these averages. ERP averages were computed separately for incorrect responses (commission errors on No-Go stimuli), yielding the ERN waveform, and for correct responses to Go stimuli, yielding the CRN waveform. Averages were calculated at the individual participant level. The CRN is thought to reflect general response-monitoring processes that are not specific to error commission, whereas the ERN reflects neural activity associated with error detection. To isolate error-specific neural activity and control for overlapping processes common to both correct and incorrect responses, we computed a difference waveform (ERN_subtract), obtained by subtracting the CRN from the ERN [48, 49]. Analyses of the ERN focused on this difference waveform.

The ERN component was quantified on the ERN_subtract waveform within a 50–150 msec time window following the response. The PE component was quantified on the ERN waveform within a 200–500 msec post-response window. Error-related ERP components were quantified as the mean activity across central and centro-parietal electrodes (Cz, C3, C4, and Pz). As the ERP’s latency and amplitude provide information about the processing speed and the intensity of processing, respectively, the individual latency and amplitude have been quantified according to the peak latency method.

Statistics

Analyses were performed using IBM SPSS Statistics © 30.0 (Armonk, NY, USA) statistical software. The behavioral and electrophysiological data were analyzed using multivariate analyses of covariance (MANCOVAs) investigating the effect of group with two levels: abstinent vs. relapser. Because the medication status at the end of the detoxification program differed significantly between the relapse and abstinent groups, we included it as a covariate in the analyses with two levels: psychotropic medication vs. no medication.

Homogeneity of variances was verified using Levene’s test. All dependent variables showed nonsignificant Levene’s statistics (p > 0.05), indicating equal variances across groups. This met the assumption of homogeneity required for MANCOVAs. The level of significance was set at p < 0.05.

At the behavioral level, we did not analyze Go trial accuracy because participants exhibited near-ceiling performance on these trials, resulting in insufficient variability for meaningful statistical analysis. Behavioral analyses therefore focused on commission error rates and correct Go RTs. We conducted the same analyses on the electrophysiological data (amplitudes and latencies of the No-Go N2d, No-Go P3d, ERN_subtract, and PE) in order to investigate the neurophysiological aspect of the behavioral effects.

Results

Behavioral Data

With regard to the commission errors and the RTs, the results of the MANCOVAs revealed no significant effect (p > 0.05). All values are presented in Figure 2.

Fig. 2.

Behavioral data for the average reaction times (msec) across groups (abstainers vs. relapsers) and average commission errors across groups (abstainers vs. relapsers).

Neurophysiological Data

Controlling for medication status, the results of the MANCOVA conducted on the amplitude of the N2d revealed a significant effect of group (F(1, 28) = 9.82, p = 0.004, ${\eta }_p^2$ = 0.26). Participants in the abstinent group (−2.16 ± 1.91 μV) showed a smaller amplitude compared to the relapser group (−3.58 ± 3.07 μV). Medication status was also a significant covariate (F(1, 28) = 12.03, p = 0.002, ${\eta }_p^2$ = 0.30).

No significant effect of group was observed for N2d latency (p > 0.05), and medication status did not significantly contribute to the model (p > 0.05). A significant effect of group was found for the P3d amplitude (F(1, 28) = 9.10, p = 0.005, ${\eta }_p^2$ = 0.25). Participants in the abstinent group exhibited larger amplitudes (6.74 ± 2.80 μV) than those in the relapser group (3.00 ± 3.77 μV). Medication status was not a significant covariate, p > 0.05.

A significant effect of group was observed for the P3d latency (F(1, 28) = 11.47, p = 0.002, ${\eta }_p^2$ = 0.29). Participants in the relapser group demonstrated shorter latency (422.40 ± 68.20 msec) than in the abstinent group (500.61 ± 44.11 msec). Medication status was not a significant covariate, p > 0.05.

A significant effect of group was found for the amplitude of the ERN_subtract (F(1, 28) = 4.38, p = 0.045, ${\eta }_p^2$ = 0.12). Participants in the relapser group (−7.90 ± 3.56 μV) showed a smaller amplitude compared to those in the abstinent group (−10.98 ± 4.53 μV). Medication status was not a significant covariate, p > 0.05.

No significant effect of group was observed for the latency of the ERN_subtract, p > 0.05. Medication was not a significant covariate, p > 0.05.

A significant effect of group was observed for the amplitude of the PE (F(1, 28) = 4.18, p = 0.05, ${\eta }_p^2$ = 0.13). Participants in the abstinent group (9.24 ± 4.57 μV) showed a smaller amplitude compared to those in the relapser group (13.12 ± 4.45 μV). Medication status was not a significant covariate, p > 0.05.

No significant effect of group was found for the latency of the PE, p > 0.05. Medication was not a significant covariate, p > 0.05. See Figure 3 and Table 2 for more details.

Fig. 3.

Grand-averaged ERP waveforms and scalp topographies for inhibition-related component (obtained at the Fz electrode) and error-related components (obtained at the Cz electrode) in the abstinent and relapser groups. a No-go N2d and No-go P3d. b The ERN_subtract waveform. c The ERN and PE components. Shaded areas indicate the predefined time windows used for component quantification. For visualization purposes only, ERP waveforms are shown at a representative electrode from each predefined electrode cluster. All statistical analyses were performed on mean amplitudes computed across the full electrode clusters, as described in the Methods. Scalp topographies illustrate mean amplitudes within the corresponding time windows.

Table 2.

Individual inhibitory ERP amplitudes (N2d, P3d) and error-related ERP amplitudes (ERN, PE) in matched patient pairs

Patient ID	Abstainers (n = 16)		Patient ID	Relapsers (n = 15)		Δ = abstainers − relapser
	N2d	P3d		N2d	P3d	N2d	P3d
a Individual inhibitory ERP amplitudes (N2d, P3d) in matched patient pairs
1	0.40	4.92	11	−3.11	6.45	3.51	−1.44
2	−3.56	12.17	14	−1.65	8.85	−1.91	3.32
4	−2.58	3.37	20	−12.48	−0.33	9.89	3.7
5	−1.49	7.17	16	−4.87	1.12	3.38	6.054
6	−4.01	7.32	3	−3.86	1.83	−0.15	5.49
7	−5.8	2.49	17	−6.79	−1.59	0.99	4.09
9	−4.3	6.04	30	−3.35	1.04	−0.95	4.997
13	−2.66	6.21	​	​	​	​	​
15	−1.76	8.28	10	−1.3	1.59	−0.453	6.69
18	1.44	10.71	29	−2.69	4.16	4.13	6.55
19	−1.56	5.73	​	​	​	​	​
21	−0.83	10.74	8	−1.80	8.91	0.966	1.83
25	−0.24	7.13	24	−0.67	4.28	0.430	2.84
26	−3.95	6.41	23	−1.68	8.34	−0.99	−1.921
27	−2.67	2.42	12	−1.71	2.59	−0.962	−0.16
28	−0.94	6.68	22	−1.47	1.32	0.523	5.36
​	​	​	31	−6.31	−3.43	​	​

Patient ID	AAbstainers (n = 16)		Patient ID	Relapsers (n = 15)		Δ = abstainers − relapsers
	ERN	PE		ERN	PE	ERN	PE
b Individual error-related ERP amplitudes (ERN, PE) in matched patient pairs
1	−10.2	1.61	11	−10.08	3.19	−0.1	−1.58
2	−17.65	9.54	14	−4.8	19.33	−12.9	−9.79
4	−17.14	6.59	20	−10.89	19.62	−6.2	−13.03
5	−13.67	10.97	16	0.25	22.90	−13.9	−11.93
6	−13.33	−1.23	3	−8.29	14.43	−5	−15.66
7	−13.16	−2.12	17	−10.89	5.67	−2.3	−7.78
9	−5.76	5.77	30	−12.87	−5.66	7.1	11.42
13	−9.06	11.95	​	​	​	​	​
15	−12.3	−0.57	10	−5.75	3.95	−6.6	−4.51
18	−12.93	14.46	29	−4.95	5.96	−8	8.5
19	−4.73	12.4	​	​	​	​	​
21	−3.57	11.47	8	−10.57	−6.61	7	18.08
25	−10.16	11.03	24	−6.91	0.98	−3.3	10.05
26	−3.65	5.75	23	−10.87	24.31	7.2	−18.55
27	−13.06	0.32	12	−5.27	−0.01	−7.8	0.33
28	−15.30	7.12	22	−11.16	18.36	−4.1	−11.23
​	​	​	31	−5.48	9.85	​	​

N2d, P3d, amplitudes (µV) for abstinent (n = 16) and relapsing (n = 15) patients. Each row corresponds to a matched pair composed of one abstainer and one relapser who were individually paired based on age and gender to allow direct comparison. The Δ columns represent the within-pair amplitude difference (Δ = abstainer − relapser), illustrating individual variability underlying group-level effects.

ERN and PE amplitudes (µV) for abstinent (n = 16) and relapsing (n = 15) patients. Each row corresponds to a matched pair composed of one abstainer and one relapser who were individually paired based on age and gender to allow direct comparison. The Δ columns represent the within-pair amplitude difference (Δ = abstainer − relapser), illustrating individual variability underlying group-level effects.

Discussion

The main objective of the present retrospective study was to verify whether distinct neural inhibitory and error-related processes at treatment entry differentiate future abstinent from relapsing patients. Between 2020 and 2025, 40 CUD patients were admitted to the detoxification program and agreed on a voluntary basis (without financial compensation) to be included in our research program. Relapse was defined as a return to cocaine dependence based on DSM-V criteria. The final sample was composed of 31 patients, separated into two subgroups: patients who remained abstinent (n = 16) versus those who relapsed (n = 15) 1 month after discharge.

The main results showed that, in the absence of any behavioral difference in performance and RTs, when future cocaine abstainers were compared to future relapsers, abstainers disclosed higher amplitudes of the P3d and ERN components, reduced amplitudes of the N2d and PE components, and longer P3d latency. Together, these findings suggest that abstinent and relapsing patients rely on qualitatively different inhibitory and error-monitoring strategies at the entrance of a detoxification program, extending previous work linking cognitive control deficits to relapse vulnerability [12].

Regarding inhibitory control, future abstainers exhibited a distinct electrophysiological profile characterized by reduced No-Go N2d amplitude and enhanced No-Go P3d amplitude. Contrary to our initial hypothesis predicting larger N2d amplitude in abstinent patients, we observed that abstainers showed a reduced N2d amplitude compared to relapsers. Rather than reflecting weaker inhibitory control, this result can be explained by a theoretical framework suggesting that, in the absence of behavioral differences, a reduced N2d may reflect a less effortful conflict-monitoring process. That is, abstinent individuals may rely on a more proactive and economical control strategy, requiring fewer early conflict-related resources to prepare inhibition effectively [50–52]. Conversely, the larger P3d component found in the present study in CUD future abstainers suggests a stronger engagement of inhibitory control mechanisms at later processing stages. The present finding is in line with previous data suggesting that stronger inhibitory resources reflect a more effective impulse control mechanism and constitute key protective factors against relapse in several forms of addictive disorders (cocaine [19, 20]; alcohol [24, 25]; heroin [26]; methadone [53]). The opposite modulation of N2d and P3d amplitudes may also suggest a functional interplay between early conflict-monitoring and later inhibitory implementation processes. Within sequential models of cognitive control, N2 has been associated with the detection of response conflict and the signaling of the need for control [45], whereas P3 reflects the subsequent engagement of inhibitory mechanisms [13]. In this framework, reduced reliance on early conflict-monitoring signals may be accompanied by stronger engagement of later inhibitory mechanisms, consistent with dynamic models of medial frontal control [54]. Although this compensatory interpretation cannot be directly tested within the present dataset, the pattern is consistent with dynamic models proposing functional interactions between sequential control stages.

In addition to larger P3d amplitudes, abstinent patients also exhibited longer P3d latencies compared to relapsers. While shorter P3d latencies are often interpreted as reflecting faster stimulus evaluation [13, 43, 44], the functional meaning of P3d latency in inhibitory tasks is complex and context-dependent. Longer latencies in the present context may instead reflect a more controlled and less impulsive mode of inhibitory processing. In clinical populations, prolonged No-Go P3 latencies have been associated with increased engagement of top-down control mechanisms and reduced impulsive responding [50–52]. Accordingly, the longer P3d latency observed in abstinent patients may index a more cautious and controlled inhibitory strategy at treatment entry, which could be protective against relapse. In contrast, the shorter P3d latencies observed in relapsers may reflect a more rapid but less adaptive inhibitory response, consistent with impulsive or prematurely terminated inhibitory processing. Nevertheless, the interpretation of No-Go P3 latency remains complex and likely reflects multiple interacting control processes rather than a simple index of processing speed.

Beyond the inhibitory profile, the present study reveals a coherent pattern of differences in the error-monitoring dynamics between future abstainers and future relapsers. As initially hypothesized, abstainers exhibited larger ERN amplitudes combined with smaller PE amplitudes than relapsers. Although the p values obtained for these two effects were close to the conventional significance threshold and should therefore be interpreted with caution, the medium effect sizes (η²p = 0.13 in both cases), the directional consistency with our a priori hypotheses, and the convergence with prior findings in independent samples [22, 23, 28] support a substantive interpretation of this pattern.

Conceptually, ERN and PE are thought to capture two functionally dissociable stages of performance monitoring [15, 16]. Within dominant theoretical frameworks, the ERN is considered to reflect a rapid and largely automatic error detection signal, operating as a reinforcement-learning-like prediction error signal that drives immediate behavioral recalibration [54, 55]; a robust ERN therefore indicates that errors are detected quickly and reliably, providing the input necessary for downstream control adjustments. The PE, by contrast, is associated with a more elaborate and effortful evaluative stage, linked to the conscious recognition of errors and to the recruitment of additional resources supporting post-error adjustment [16, 31]. Within this framework, the combination of a strong ERN with a reduced PE, observed here in abstainers, is consistent with an efficient monitoring system in which errors are detected quickly and accurately at an automatic stage, requiring only limited additional conscious appraisal to drive adaptive behavioral correction. Conversely, the blunted ERN paired with exaggerated PE observed in relapsers can be interpreted as reflecting a weakened automatic error detection process compensated by greater reliance on later, more conscious and effortful evaluation. Such a compensatory shift is functionally costly: when automatic error signaling is impaired, the timely recruitment of control mechanisms after an error is delayed, which compromises trial-to-trial behavioral adjustment and may prevent the rapid reengagement of inhibitory resources after a slip – a process that is central to relapse prevention.

Mechanistically, weakened automatic error monitoring in CUD has been linked to dysregulation of the mesencephalic-cingulate dopaminergic loop, which is thought to provide the prediction error signal underlying ERN generation [54, 56]. Chronic cocaine exposure is known to alter striato-cortical dopaminergic signaling, which in turn disrupts the reinforcement-learning machinery that the ERN is believed to index. The result is a system in which behaviorally relevant error signals are attenuated, prediction-error-driven learning is impoverished, and post-error behavioral adjustment becomes less efficient. Such impairments have been repeatedly associated with poorer treatment outcomes in cocaine users [22], with treatment drop-out [23], and with relapse in alcohol use [28] and broader substance use [15] populations, supporting the interpretation that the integrity of early error-monitoring mechanisms constitutes a protective factor against early relapse in CUD.

A coherent integration of these results is that enhanced error monitoring observed in abstainers may reduce the need to hyperactivate earlier conflict-monitoring processes. Although this interaction between ERN and N2d cannot be directly tested in the current retrospective dataset, the pattern is consistent with conflict-monitoring and reinforcement-learning models, according to which the ERN, commonly linked in the literature to ACC-related circuits, serves as a rapid and robust error detection signal, decreasing reliance on early monitoring mechanisms (N2d) for subsequent trials [55–57]. Moreover, source localization studies have shown that these two components exhibited highly similar scalp distribution and dipole source localization suggesting the same ACC generator for both peaks [56]. Poorly functioning error-monitoring systems in cocaine addicts may result in altered No-Go processing so that conflicting No-Go trials were not signaled “at time” for executive control and who need therefore to be hyperactivated to compensate for the poor error-monitoring system [21, 58]. Both reported data are also in line with previous neuroimaging studies (using, for instance, functional magnetic resonance imaging), showing hypoactive inhibitory and error detection neural networks in CUD [59, 60].

Regarding the amplitude of the PE, our results are consistent with neurofunctional models proposing that the amplitude of the PE is negatively related to the rostral ACC activation, whereas the ERN is associated with greater caudal ACC engagement [59]. According to this framework, a pattern combining strong ERN activity – typically interpreted as reflecting robust caudal ACC recruitment – and reduced PE amplitude, indicative of reduced compensatory post-error evaluation, has been associated with more adaptive performance monitoring. Conversely, weaker ERN responses and abnormally large PE amplitudes have been described as reflecting an inefficient, more effortful error-monitoring system, often associated with poorer treatment outcomes in substance use disorders [23].

This interpretation fits closely with the present results. Abstinent patients displayed larger ERN amplitudes and smaller PE amplitudes, suggesting more efficient error-monitoring functioning at treatment entry, consistent with models implicating ACC-related circuits. In contrast, relapsers showed blunted ERN responses paired with exaggerated PE amplitudes, a pattern consistent with reduced early error detection and a compensatory overreliance on later evaluative processes, in keeping with the cingulate models outlined above. Importantly, given the absence of source localization analyses in the present study, these anatomical interpretations remain inferential and should be considered in light of this methodological limitation. Such altered monitoring may prevent adequate behavioral adjustments after errors, thereby contributing to a higher vulnerability to early relapse.

Even if, of course, neural alterations could not be considered as primary and specific causes of addictions broadly, such results stress nevertheless the importance of finding reliable monitoring/predictive biomarkers of addictive states to help in their clinical management [60]. Indeed, despite the absence of overt behavioral differences between groups, robust electrophysiological differences were observed. The near-ceiling performance concerned correct Go responses, whereas commission errors and error-related responses showed sufficient variability for meaningful ERP analyses. Because the components of interest (N2d, P3d, ERN, and PE) index inhibitory and performance-monitoring processes rather than simple Go execution, the ceiling effect on Go accuracy is unlikely to explain the observed neural group differences. More fundamentally, successful inhibition is a covert process that produces no overt response, so behavioral indices provide only a coarse measure of inhibitory functioning; ERPs, by contrast, track the underlying neural dynamics directly, which explains why group differences can emerge at the electrophysiological level even in the absence of behavioral differences. Instead, this dissociation suggests that ERPs may capture subtle neurocognitive dynamics that are not directly reflected in overt behavioral performance, a pattern frequently reported in addiction research. Therefore, monitoring thanks to ERPs the neural dynamics of targeted cognitive mechanisms (well known to trigger addictive states, such as inhibition and error detection) appears to be of the greatest clinical importance, as it may help predict at the entrance of a treatment (detoxification or other), which patient is at high risk of relapse. In this way, clinicians may orient these high-risk patients, with a reduced P3d amplitude, shorter P3d latency, blunted ERN responses, and enhanced PE amplitudes, to tailored cognitive rehabilitation programs, and this is true for CUD [28] as well as for substance use disorders in general [30, 61]. Such programs, aiming for instance to increase inhibitory skills through cognitive training and/or neuromodulation tools [57], may be proposed to individual patients on the basis of their singular profile [30, 52]. Finding a way to implement such a procedure into clinical care units probably represents the future main challenge for neurophysiologists. Such a procedure may involve (1) the gathering of normative control ERP values, in order to guarantee access to normative data recorded on large samples by similar procedures; and (2) the use of ERP test-retest sessions, in order to index the progression of patients’ scores as a function of the treatment [62].

Obviously, we are totally aware that, if cognitive control is an important mechanism involved in addictive disorders, many other factors (cognitive, but also genetic, environmental, social, economic, or emotional) are at least as important, and that training and recovering inhibitory and error-related mechanisms is a necessary (but not sufficient) step toward abstinence. Also, if decreased impulse control appears as a central neurocognitive mechanism in addictive disorders in general, it is also clear that, due to the specific neuropharmacological action of different addictive substances (such as cocaine, alcohol, heroin, marijuana, etc.), each type of substance use disorder or even polydrug use may lead to specific cortical anomalies linked to different cognitive disturbances [63]. Therefore, due to this interindividual variability, and because different patients may present a singular cognitive profile requiring a specific form of rehabilitation, we are trying to move nowadays to a “stratified psychiatry,” which aims to use biomarkers in clinical practice in order to facilitate the orientation of individual patients to best “on-label” treatments [64].

Limitations

Also, it should be noted that, given the 21-channel EEG montage used in the present study, no source localization analyses were conducted. Therefore, interpretations regarding anterior cingulate involvement remain indirect and are based on prior literature rather than direct source reconstruction. Future studies using high-density EEG and source modeling approaches (e.g., sLORETA) would help refine these anatomical interpretations. In addition, the absence of pre-detoxification baseline measurements and longitudinal ERP assessments limits our ability to determine whether the observed neural differences reflect stable trait-like vulnerability markers or transient state-related effects associated with early detoxification. Although EEG recordings were conducted after acute withdrawal symptoms had subsided, and no behavioral differences were observed between groups, it remains possible that some electrophysiological alterations reflect short-term neurophysiological adaptations rather than enduring relapse-related characteristics. Future longitudinal studies including repeated ERP assessments across recovery phases would help clarify the temporal stability of these neural markers.

In addition, the modest sample size and the absence of an a priori power estimation limit the sensitivity of the study to small effects, and the analysis of several ERP variables entails an increased risk of type I error; the ERN and PE effects, whose p values were close to the conventional threshold, should be considered as preliminary and should be replicated in studies using larger samples. Finally, the sample was composed almost exclusively of male participants. Although this distribution reflects the clinical reality of inpatient detoxification programs for CUD – where men are disproportionately represented – it nevertheless limits the generalizability of the findings. Given documented sex differences in inhibitory control, error monitoring, and addiction trajectories, it remains unclear whether the present electrophysiological markers would operate similarly in female patients. Future studies including more balanced samples will be necessary to determine whether sex-specific neural predictors of relapse emerge. Recruiting data on larger samples will help, for instance, use machine learning algorithms in order to elucidate the most robust predictors involved in diagnosing clinical status, predicting addiction risk, and personalizing treatment [65]. In addition, although regression-based artifact correction was applied, the use of a 21-channel EEG montage limited the feasibility of robust ICA-based decomposition procedures. Future studies using high-density EEG would allow more advanced component separation and potentially more precise characterization of frontal control processes.

Conclusion

Despite the caveats outlined above, the present findings provide preliminary but coherent evidence that future abstainers and future relapsers differ already at treatment entry in the qualitative profile of their inhibitory and error-monitoring neural dynamics. If confirmed in larger and more balanced samples, such markers could contribute to the early identification of high-risk patients and inform individualized cognitive rehabilitation strategies within a stratified-psychiatry framework.

Statement of Ethics

The local Ethics Committee at the Brugmann Hospital approved this study (“2022/118”). After receiving detailed information regarding the study, all the participants provided written informed consent.

Conflict of Interest Statement

The authors have no competing financial interest, potential conflict of interest, or financial relationship with commercial entities to report.

Funding Sources

The authors were funded by the Belgian Fund for Scientific Research (F.N.R.S., Belgium), the Brugmann Foundation (CHU Brugmann, Belgium), the seed money fund from Université Libre de Bruxelles (ULB, Belgium), and the IRIS Foundation (King Baudouin Foundation, Belgium), although these funds did not exert any editorial direction or censorship over any part of this article.

Author Contributions

Anais Ingels: conceptualization, methodology, validation, investigation, formal analysis, writing – original draft, supervision, project administration, and funding acquisition; Christie Chenut, Alix Pendry, and Felix Hever: patient recruitment; Clémence Dousset: investigation; Hendrick Kajosch and Mathieu Hein: coordination with clinical units for patient access; Salvatore Campanella: conceptualization, methodology, validation, writing – review and editing, resources, supervision, and project administration.

Funding Statement

The authors were funded by the Belgian Fund for Scientific Research (F.N.R.S., Belgium), the Brugmann Foundation (CHU Brugmann, Belgium), the seed money fund from Université Libre de Bruxelles (ULB, Belgium), and the IRIS Foundation (King Baudouin Foundation, Belgium), although these funds did not exert any editorial direction or censorship over any part of this article.

Data Availability Statement

The data that support the findings of this study are not publicly available but are available from Ingels A.