RESEARCH ARTICLE Predicting Impulse Control Disorders in Parkinson Disease through Incentive Biomarkers Juan Marín-Lahoz, MD, PhD ,1,2,3 Saül Martinez-Horta, PhD,3,4,5,6 Javier Pagonabarraga, MD, PhD,3,4,5,6 Andrea Horta-Barba, MS,4,5,6 Ignacio Aracil-Bolaños, MD,3,4,5,6 Helena Bejr-kasem, MD, PhD,3,4,5,6 Frederic Sampedro, MD, PhD ,5,6 Antonia Campolongo, MSN,4,5,6 and Jaime Kulisevsky, MD, PhD3,4,5,6 Objective: This study was undertaken to evaluate whether the feedback-related negativity (FRN)—a neurophysiological marker of incentive processing—can be used to predict the development of impulse control disorders (ICDs) in Parkinson disease (PD). Methods: The longitudinal cohort consisted of consecutive nondemented PD patients with no ICD history. We recorded FRN signals while they performed a gambling task. We calculated the mean amplitude difference between losses and gains (FRNdiff) to be used as a predictor of future ICD development. We performed prospective biannual follow-up assessments for 30 months to detect incident ICDs. Finally, we evaluated how basal FRNdiff was associated with posterior development of ICDs using survival models. Results: Between October 7, 2015 and December 16, 2016, we screened 120 patients. Among them, 94 patients per- formed the gambling and 92 completed the follow-up. Eighteen patients developed ICDs during follow-up, whereas 74 remained free of ICDs. Baseline FRNdiff was greater in patients who developed ICDs than in those who did not (2.33μV vs 0.84μV, p = 0.001). No other significant baseline differences were found. The FRNdiff was significantly asso- ciated with ICD development in the survival models both when not adjusted (hazard ratio [HR] = 0.73, 95% confidence interval [CI] = 0.58–0.91, p = 0.006) and when controlling for dopamine replacement therapy, sex, and age (HR = 0.74, 95% CI = 0.55–0.97, p = 0.035). None of the impulsivity measures evaluated was related to ICD development. Interpretation: Reward-processing differences measured by FRN signals precede ICD development in PD. This neuro- physiological marker permits identification of patients with high risk of ICD development. ANN NEUROL 2022;92:974–984 P arkinson disease (PD) is the second most common neurodegenerative disease. Although the hallmark of PD is motor symptoms, nonmotor symptoms constitute a it. This is the case of impulse control disorders (ICDs), which are associated with treatment with dopamine ago- nists (DAs)3,4 and almost absent in drug-naïve patients.5 major cause of disability and caregiver burden in PD.1 ICDs are characterized by difficulty resisting an Some of them have a poor response to dopamine replace- impulse to perform a typically pleasurable activity that is ment therapy,2 and others are caused or aggravated by finally harmful. In the context of PD, ICDs usually involve View this article online at wileyonlinelibrary.com. DOI: 10.1002/ana.26486 Received Nov 5, 2021, and in revised form Aug 16, 2022. Accepted for publication Aug 17, 2022. Address correspondence to Dr Marín-Lahoz, Neurology Department, Hospital Universitario Miguel Servet; Paseo Isabel la Cat olica 1-3, 50009 Zaragoza, Spain. E-mail: juanmarinlahoz@gmail.comDr Kulisevsky, Neurology Department, Hospital de la Santa Creu I Sant Pau; C/Mas Casanovas 90, 08041 Barcelona, Spain. E-mail: jkulisevsky@santpau.cat From the 1Neurology Department, Miguel Servet University Hospital, Zaragoza, Spain; 2Servet Neuroscience Group, Institute of Health Research of on), Zaragoza, Spain; 3Department of Medicine, Barcelona Autonomous University, Barcelona, Spain; 4Movement Disorders Unit, Aragon (IIS Arag Neurology Department, Sant Pau Hospital, Barcelona, Spain; 5Movement Disorders Group, Biomedical Research Institute–Sant Pau, Barcelona, Spain; and 6 Network Research Center–Neurodegenerative Diseases (CIBERNED), Madrid, Spain 974 © 2022 The Authors. Annals of Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Marín-Lahoz et al: Prediction of ICDs activities that are harmful not on their own, but only because Spain. The protocol, patient information, and consent of the frequency and salience they acquire.6 Nevertheless, form were approved by the institutional review board ICDs cause significant disability and negatively impact the (Hospital of the Holy Cross and Saint Paul Clinical quality of life of PD patients and their caregivers.7 Research Ethics Committee), and written consent was In the general population, these rewarding compul- obtained with full comprehension of the study from each sive actions not directed toward drugs are increasingly patient before any study procedure. The study was referred as behavioral addictions. Pathological gambling is designed and executed according with the Declaration of included as a behavioral addiction in the Diagnostic and Helsinki. We prospectively recruited an ad hoc cohort of Statistical Manual of Mental Disorders, 5th edition.8 Both nondemented PD outpatients between October 7, 2015 in the general population and for PD, the brain effects of and December 16, 2016. Inclusion criteria were PD diag- these activities are comparable to that of drugs.8–10 nosis according to the London Brain Bank diagnostic PD-associated ICDs (PD-ICDs) are potentially pre- criteria,18 Hoehn and Yahr stage up to 3, and full under- ventable. However, DA avoidance may come at the cost of standing of and agreement with the informed consent. worse motor control or higher L-dopa use and should not Exclusion criteria were brain lesions or any neurodegenera- be done systematically but only when benefits outweigh the tive disorder other than PD, dementia,8,19 any condition– risks. This would require ICD risk to be profiled by risk fac- related or not to PD–disenabling the patient from tors or biomarkers. To date, most of the evidence comes performing the proposed evaluations, ICD history according from cross-sectional studies.10,11 A recent review12 consid- to the electronic medical records, positive screening for past ered sex and DAs the only risk factors, because no other or present ICDs according to the short version of Question- factor—not even age and impulsivity—had been confirmed naire for Impulsive-Compulsive Disorders in PD (QUIP), in prospective studies. We later found in a prospective study use of dopamine antagonists, a life expectancy < 1 year, that also depression acts as a risk factor.4 unstable medical conditions, current or recent (4 weeks) par- A strong reward is considered important in the initia- ticipation in clinical trials, and untreated or refractory tion of drug addiction,13 and a highly active reward system depression. Participants were recruited from the same outpa- might be needed to transform normal activities into behav- tient movement disorders clinic. ioral addictions or PD-ICDs.9,14 The feedback-related neg- ativity (FRN) is an event-related potential used to evaluate Assessments reward and incentive processing.15,16 The FRN is consid- At baseline, each patient was evaluated by a neurologist ered to represent reward prediction error, specifically within and a neuropsychologist specializing in movement the dorsal anterior cingulate cortex.16 The amplitude differ- disorders. This clinical assessment included Movement ence between the FRN generated after gains and after losses Disorder Society–Unified Parkinson’s Disease Rating (FRNdiff) has been used to evaluate reward-processing.16 Scale part III (motor score),20 total L-dopa equivalent In a previous study, we showed that apathetic PD patients daily dose (LEDD), DA L-dopa equivalent daily dose have a reduced FRNdiff, whereas PD patients without apa- (DA LEDD),21 and REM Behavior Sleep Disorder Ques- thy have a normal amplitude compared to healthy con- tionnaire.22 Then, a neuropsychological examination was trols.17 These results highlighted the involvement of performed (detailed below). FRN acquisition was per- diminished incentive processing in the downregulation of formed within the next 2 weeks. motivated behavior and illustrated the usefulness of the Follow-up visits were scheduled biannually (+6, FRN to evaluate motivational disorders in PD. +12, +18, +24, and +30 months) to detect ICDs. They In the present study, we aimed to evaluate whether comprised the recording of current medications, and the the differences in reward-processing can be detected prior evaluation of impulse control and related disorders. The to the development of ICDs among PD patients. We presence of ICDs at any time point was considered the hypothesized that an FRNdiff would be associated with main outcome of the study. The biannual evaluations future development of ICDs among PD patients free of were included to allow early detection for a metabolical ICDs. We further hypothesized that the FRNdiff could be study of ICDs,23 not to measure time to development used to predict ICD development. of ICDs. FRN: Stimuli and Procedure. We used a modified version Patients and Methods of Gehring’s gambling task.17 Each trial begins with a fixa- Study Design and Patients tion sign (asterisk). After 500 milliseconds, 2 numbers, We performed a monocentric longitudinal cohort study at 25 and 5, are presented in white against a black back- Hospital of the Holy Cross and Saint Paul, Barcelona, ground. Participants have to bet on 1 of these 2 numbers December 2022 975 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ANNALS of Neurology to increase the starting amount of 1,000 (virtual) Euro using EEGLAB25 on MATLAB R2016a (MathWorks, cents. They are instructed to choose 1 of the 2 numbers Natick, MA). by pressing a button. Immediately after the selection, the numbers change color into red or green. If the selected Neuropsychological and Behavioral Assessment. We number turns green, the participant gains the assessed global cognitive status by means of the corresponding amount of Euro cents (ie, +5 or +25); Parkinson’s Disease Cognitive Rating Scale (PD-CRS).26 conversely, when the selected number turns red, the par- The PD-CRS is a cognitive scale validated and rec- ticipant losses the corresponding amount. This feedback is ommended by the Movement Disorders Society for cogni- shown for 1 second. A new trial is initiated after 3 seconds. tive evaluation in PD.27 We evaluated anxiety and The experimental session comprises a total of 368 trials depression using the Hospital Anxiety and Depression (4 runs of 92 trials). Participants are told that they should Scale28 and apathy symptoms using the Starkstein Apathy adjust their choices based on outcomes in each trial to Scale.29 Impulsivity and risk taking were evaluated by sev- increase their gains. However, the task was programmed eral measures: the Barratt Impulsiveness Scale (BIS11) for to yield wins in 50% of the trials and losses in the impulsivity as a trait, commission errors on the Psychol- other 50%. ogy Experiment Building Language Continuous Perfor- mance Task based on the Conners Continuous Performance Test for motor inhibition, the Kirby Delay FRN: Electrophysiological Recording. We recorded electro- Discounting Task (DDT) for delay discounting, the Iowa encephalogram (EEG) at a sampling rate of 250Hz from Gambling Task (IGT) for implicit risk with potential 19 standard scalp sites (Fp1/2, F3/4, C3/4, T3/4, T5/6, losses, the Balloon Analog Risk Task (BART) for implicit P3/4, O1/2, F7/8, Fz, Cz, Pz) referenced to the 2 mastoid risk without losses, and the Game of Dice Task (GDT) leads using the BrainAmp System (Brain Products, for explicit risk.30 Details on all the impulsivity evalua- Gilching, Germany), Electro-Cap International (Eaton, tions except for the GDT are described elsewhere.31 The OH) electrode caps, and BrainVision Recorder software GDT is a task inspired by the IGT and the Rogers Risk (Brain Products). We registered vertical and horizontal eye Task.32 It was designed to yield explicit risk choices that movements using 2 additional bipolar channels for artifact remain stable and obvious during the task. Participants minimization and rejection. We ensured the impedances have to choose among 4 kinds of dice bets; 2 of those bets of recording sites were <5kΩ. Signals were filtered with a are risky and unprofitable and 2 of them are not. The bandpass of 0.1–35Hz and digitized at a rate of 250Hz. measure derived from this task is the number of safe trials minus the number of risky trials, ranging from 18 to FRN: Processing. We processed EEG signals using SOBI +18. The interpretation is as follows: lower scores repre- (second-order blind identification) to minimize the ocular sent greater impulsivity or risk taking for commission motion artifacts. It is a blind source separation algorithm errors, IGT, and GDT, whereas higher scores represent based on an eigenvalue decomposition of time-delayed higher impulsivity or risk taking for BIS11, DDT, covariance matrices.24 The feedback-locked event-related and BART. potential recording window was set from 200 milliseconds before until 1,000 milliseconds after the feedback stimu- Evaluation of ICDs. We evaluated the short version of lus. We removed any epochs exceeding 300μV in any QUIP at each follow-up and performed a comprehensive channel or 75μV in Fz from further analysis. Then, we interview based on the core components of behavioral averaged and corrected by baseline 50- to 0-millisecond addictions.9,33 The main outcome of the study was the time window the epochs for each condition. Patients with development of a significant ICD according to this inter- <60 epochs for each condition were not analyzed. view. In the cases where an ICD was detected, we also Event-related potentials were quantified measuring administered the QUIP Rating Scale (QUIP-RS).34 The the mean amplitude between 250 and 450 milliseconds evaluator of ICDs was blinded to electrophysiological following feedback presentation for each condition (win studies during all follow-up. and loss) at each midline electrode (Fz, Cz, Pz). The FRN was identified as the negative deflection in the Statistical Analysis.We grouped patients according to the event-related potential during this interval. For each development of ICDs during follow-up and compared participant, we calculated the FRNdiff as the FRN in loss baseline characteristics across groups. Categorical variables condition minus the FRN in win condition at Fz, the were compared by means of χ2 test. For quantitative vari- closer electrode to the FRN generator, in which the ables, we calculated mean and standard deviation potential is greater.15,16 EEG signal processing was done (SD) and compared groups with t test when appropriate. 976 Volume 92, No. 6 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Marín-Lahoz et al: Prediction of ICDs When not, we calculated medians and ranges and highest balanced accuracy. We also calculated the median compared groups using the Wilcoxon signed-rank test. sensitivity and specificity of each cutoff using 2,000 strati- A repeated measures analysis of variance (ANOVA) fied bootstrap replicates. was performed with electrode (Fz, Cz, Pz) and condition For all analyses, statistical significance threshold (win and loss) as within-subject factors to evaluate the (alpha) was 0.05 (2-tailed). We estimated that a sample effects of the condition. Then the outcome (ICD+ and size of 11 in the smaller group (patients developing ICDs), ICD during follow-up) was included as between-groups completing the follow-up, would be required to provide factor (mixed ANOVA). Greenhouse–Geisser correction 0.8 power with the aforestated alpha. As no references of was used for sphericity. We plotted the FRN and the FRNdiff in PD-ICDs had been published, we assumed FRNdiff for each condition and compared the FRNdiff the difference between patients who develop and who between ICD and ICD+ using t test and then adjusting remain free of ICDs to be at least 1.76μV (same difference potential confounders using a general lineal model. we had found between PD with and without apathy).17 We conducted a survival analysis to evaluate the To calculate the size required for the full cohort, we took relationship between the FRNdiff and ICD development into account previously published incidence rate(1 ICD at each time point. For this purpose, we used Cox propor- case per 100 patient-months).37 However, as this rate cor- tional hazard models (time-to-event analysis). We first responds to a cohort in which all the patients were receiv- deployed a model with no covariates and then an adjusted ing DAs (and we planned to include patients regardless of model including known risk factors as covariates. This their use of DA), we counted on a lower incidence rate, adjusted model was considered the main goal of the study. 0.6, for the cohort sample size calculation. Assuming To ensure any association between ICD develop- follow-up losses of 25%, inclusion of 91 participants was ment and FRNdiff was not driven by changes in the med- required. All the statistical analyses were performed using ication after baseline (eg, less DA use during follow-up R version 3.4.3.38 Data used for this study are available among patients with greater FRNdiff at baseline would upon reasonable request. lead to a false association between FRNdiff and the out- come), we deployed another Cox model including DA dose at each time point as a time-dependent covariate. To Results ensure any association between ICD development and We recruited 120 patients, but only 98 performed the FRNdiff was not driven by use of DAs prior to baseline, gambling task. Twenty patients were excluded; 12 patients we also deployed a Cox model including DA accumulated screened positive for an ICD, and 8 had cognitive impair- doses. We explored the inclusion of any other available ment precluding the execution of the study evaluations. variables to improve the model. For this purpose, we used Two patients withdraw their consent for the task. From bidirectional stepwise regression. the 98 patients with EEG registers, 4 were excluded from We explored the classification capabilities of the analysis because they had an insufficient number of valid FRNdiff and the predictability of ICDs using logistic trials per condition (Fig 1). regression and receiver operating characteristic (ROC) cur- Among the 94 patients with valid registers, 18 devel- ves. Although logistic regression cannot deal with dropout oped ICDs within the 30-month follow-up (ICD+), in the way survival models do, we decided to include this exploratory analysis to offer an easily interpretable result and to make the current results comparable to previous studies that have used this methodology.35 We utilized 2 logistic regression models with ICD development as the dependent variable. Both of them included clinical vari- ables, but one of them also featured the FRNdiff. For each model, we generated a ROC curve, and then we compared the area under the curve (AUC) of both ROC curves using DeLong test.36 To show a potential clinical applica- tion of the information yielded by the FRN-based model, we chose 2 cutoff points, the first one to identify patients with low ICD risk (lowest quintile according to the model) and the second one to identify patients at high risk FIGURE 1: Patient flowchart. FRN = feedback-related (highest quintile). We then calculated the cumulative inci- negativity; ICD = impulse control disorders. [Color figure can dence in both quintiles. We also provide the cutoff with be viewed at www.annalsofneurology.org] December 2022 977 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ANNALS of Neurology 74 did not develop an ICD (ICD), and 2 did not per- evaluated variables: age, evolution, medication, impulsivity form follow-up (2.1%). The mean proportion of visit measures, motor status, and FRNdiff. completion was 0.67 (interquartile range = 0.6–0.8). Patients who developed ICDs in follow-up were similar Nevertheless, all the patients included completed several to those who did not, although they had higher cognitive per- follow-up visits, and the median follow-up was 2 years. formance (PD-CRS) and DA LEDD (both not statistically There was no association between dropout rate and the significant). Baseline characteristics are shown in Table 1. TABLE 1. Baseline Characteristics Characteristic ICD, n = 74 ICD+, n = 18 p Sex, male, n (%) 47 (63.5%) 11 (61.1%) 0.85 Baseline age, yr 68.36  9.54 66.31  7.37 0.4 Age at PD diagnosis, yr 62.59  9.26 62.4  13.66 0.94 Time since PD diagnosis, yr 6.77  2.8 5.6  11.78 0.44 Education, yr 11.72  4.83 12.67  4.84 0.46 MDS-UPDRS III 25.49  6.98 24.72  8.88 0.7 Modified H&Y stage, median (range) 2 (1–3) 2 (2–2.5) 0.36 S&E ADL scale 90 (70–100) 90 (70–100) 0.32 Baseline LEDD, mg 552.23  313.14 584.81  326.48 0.7 DA LEDD, mg 138.27  112.74 193.78  103.69 0.07 Accumulated LEDD, g 711.27  1,065.99 490.58  447.77 0.18 Accumulated DA LEDD, g 178.75  222.4 197.23  253.53 0.78 DA use, % 58 (78.4%) 17 (94.4%) 0.22 PD-CRS 87.6  16.37 94.72  15.43 0.11 HADS anxiety 3.29  2.62 3.88  2.65 0.41 HADS depression 2.24  2.75 3.06  2.76 0.28 SAS 4.84  5.9 5.39  6.35 0.73 RBD score 4.76  3.46 6.22  2.76 0.26 BIS11 Total 53.99  8.2 54.28  8.07 0.89 Commission errors 0.39  0.21 0.38  0.27 0.94 DDT low k 0.29  1.15 0.12  0.32 0.30 DDT high k 0.49  1.71 0.67  1.84 0.72 GDT 4.62  8.57 6.22  8.17 0.47 IGT 0.36  21.12 3.53  19.12 0.57 BART 35.41  12.07 34.9  8.28 0.85 BART = Balloon Analogue Risk Task; BIS11 = Barratt Impulsiveness Scale; DA = dopamine agonist; DDT = Delayed Discount Task; GDT = Game of Dice Task; H&Y = Hoehn and Yahr; HADS = Hospital Anxiety and Depression Scale; ICD = impulse control disorder; ICD = PD patients who remain free of ICDs during the 30-month follow-up; ICD+ = PD patients free of ICDs at baseline who develop the disorders during the 30-month follow-up; IGT = Iowa Gambling Task; k = hyperbolic discounting rate; LEDD = L-dopa equivalent daily dose; MDS-UPDRS III = Movement Disorder Society–Unified Parkinson’s Disease Rating Scale part III (motor score); PD = Parkinson disease; PD-CRS = Parkinson’s Disease Cognitive Rating Scale; RBD = rapid eye movement sleep behavior disorder; S&E ADL = Schwab and England Activities of Daily Living; SAS = Starkstein Apathy Scale. 978 Volume 92, No. 6 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Marín-Lahoz et al: Prediction of ICDs conditions (mean  SD); wins ICD were 173.54  22.83, and wins ICD+ were 173.5  25.41 (p = 1); losses ICD were 171.32  24.91, and losses ICD+ were 172.56  20.86 (p = 0.83). The mean FRN for wins was 9  5.91μV and for losses was 7.87  5.37μV. Feedback-locked averages after wins and losses for the entire patient sample are shown in Figure 2. An upward (negative) deflection was observed starting 250 milliseconds after negative feedback (losses). The repeated measures ANOVA on the mean ampli- tude of the FRN with condition (win vs loss) and electrode (Fz, Cz, Pz) as within-subject factors showed the main effects of condition (F1,92 = 46.94, p < 0.001) and electrode (F1.56,143.58 = 116.4, p < 0.001) and the interaction condi- tion by electrode (F1.6,146.79 = 13.504, p < 0.001). The mixed ANOVA with the same within-subject factors and the outcome (ICD+, ICD) as between-group factor showed a significant interaction of condition by outcome (F1,89 = 10.66, p = 0.002). Electrode, condition, and their interaction remained significant, whereas outcome and its interaction with electrode were not significant (p > 0.1). The FRNdiff of the whole sample was FIGURE 2: Event-related potentials associated with monetary 1.19  1.65μV. For those who did not develop an wins and losses. Grand average is feedback-locked at Fz for ICD, the FRNdiff was 0.84  1.51μV, and for those monetary wins and losses. [Color figure can be viewed at www.annalsofneurology.org] who did it was 2.33  1.57μV. The difference between both groups was 1.48μV (95% confidence interval FRN Analysis [CI] = 0.64–2.33, p = 0.001). Figure 3 shows the FRN The number of EEG epochs averaged and included for analysis for wins and losses of each group, and Figure 4 shows did not differ between groups for any of the two main FRNdiff of both groups. The difference between both FIGURE 3: Event-related potentials associated with monetary wins and losses for each group. (A) Average feedback-locked at Fz for monetary wins and losses of patients who did not develop an impulse control disorder (ICD). (B) Corresponding waves of those who did develop ICDs. [Color figure can be viewed at www.annalsofneurology.org] December 2022 979 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ANNALS of Neurology FIGURE 5: Receiver operating characteristic curves (ROCs) of the clinical model with an area of 0.61 and of the clinical and neurophysiological model with an area of 0.8 for the prediction of impulse control disorders in the next 3 years. [Color figure can be viewed at www.annalsofneurology.org] FIGURE 4: Difference between losses and gains (FRNdiff) at Fz for patients who developed an impulse control disorder Classification Models (ICD) within 30 months of the register (ICD+) and those who To better show the utility of FRNdiff as a predictor, we per- remained ICD-free for the same period (ICD). [Color figure can be viewed at www.annalsofneurology.org] formed ROC curves for a logistic model including the same variables (with or without the inclusion of the FRNdiff). These curves are shown in Figure 5. The clinical model groups was also significant after controlling for potential yielded an AUC of 0.61. The model that also included the confounders (DA LEDD, LEDD, and age; 1.40μV, neurophysiological potential had an AUC of 0.8. The AUC 95% CI = 2.22 to 0.58, p = 0.001) and for cumula- of the second model was significantly greater (p = 0.003). tive doses instead of current ones (1.33μV, 95% The cutoff for higher balanced accuracy was 1.97μV CI = 2.13 to 0.53, p = 0.001). (specificity = 80.8%, sensitivity = 72.2%). Table 2 shows the parameters of the full logistic model. The formula shows how to apply the coefficients Survival Model obtained from the logistic model: The FRNdiff was significantly associated with ICD devel- opment, with a hazard ratio (HR) of 0.73 (95% Y ¼ 1:92  0:017*age þ 0:39*ðsex ¼ male Þ CI = 0.58–0.91, p = 0.006). When sex, DA LEDD, þ 4:2*DA LEDD þ 0:01*LEDD  0:53*FRNdiff total LEDD, and age were included as covariates, the FRNdiff remained significantly associated with ICD devel- Age is expressed in years, DA LEDD and LEDD in hun- opment (HR = 0.74, 95% CI = 0.55–0.97, p = 0.035). dreds of milligrams, and the potential in microvolts. None of the other covariates was significant. Patients in the lower quintile (Y < 2.576) had a The replacement of baseline DA dose with current cumulative incidence of 5.26%, or 2.1 cases/100 patient- dose yielded a similar result (HR = 0.73, 95% years. This cutoff yielded a sensitivity of 94.4% and a speci- CI = 0.56–0.96, p = 0.024). The model with accumu- ficity of 37%. Patients in the highest quintile (Y > 0.63) lated DA doses was also similar (HR = 0.69, 95% had a cumulative incidence of 55.6% or 22.2 cases/100 CI = 0.51–0.91, p = 0.01). In both cases, the covariates patient-years. This cutoff yielded a sensitivity of 55.6% and remained nonsignificant. The stepwise selection could not a specificity of 89%. The remaining patients had a cumula- find any variable that could improve the model. tive incidence of 12.9% or 5.2/100 patient-years. 980 Volume 92, No. 6 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Marín-Lahoz et al: Prediction of ICDs but on the Parkinson Progression Markers Initiative TABLE 2. Parameters of the Full Logistic Model (PPMI) cohort.40 It used extensive genotyping to evaluate Parameter Α β p heritability and included a predictive model of ICDs based on genetics. It had nevertheless some limitations that Intercept 1.92 0.48 prevented its use as a clinical predictive tool. First, ICD Age 0.017 0.66 diagnosis was based on a screening scale, which is useful to find an association but might be considered not strin- Sex (male) 0.39 0.54 gent enough for a predictive clinical model. Second, the DA LEDD 0.0042 0.18 model parameters were not reported, making its applica- tion difficult elsewhere. Finally, the polymorphisms associ- Total LEDD 0.0001 0.91 ated with ICD development in the PPMI cohort—found FRNdiff 0.53 0.004 in opioid receptor kappa (OPRK1) and tryptophan DA LEDD = LEDD accounted for by dopamine agonists; hydroxylase (TPH2)—diverge from the polymorphisms FRNdiff = difference between the amplitude feedback-related nega- identified in other populations.41–43 This divergence raises tivity generated after gains and after losses; LEDD = L-dopa equiva- the concern that ICD-related polymorphisms may be dif- lent daily dose. ferent in populations with different genetic backgrounds. In this vein, our method is based on the FRN, which The use of accumulated doses did not improve the reflects incentive processing in the anterior cingulate cor- classification model. tex not only in humans but in other mammals.44,45 The Although the goal of this study was to evaluate the conservation of this event-related potential across species predictive capabilities of the FRN when used in combina- suggests that the genetic background among humans is of tion with known risk factors, we profited from the oppor- little relevance for this application. In the current study, tunity to explore the relationship of multiple modes of in addition to the survival analysis, we report our results impulsivity (and risk taking) with ICD development in a using logistic regression and ROC curves so the current longitudinal cohort, even if none of them was significantly results can be better interpreted in relation to the afore- associated with ICD development on their own. We eval- mentioned PPMI genetic study. uated the predictive power of all these measures combined This is also the first study to identify electrophysio- in an additive model (multivariate logistic regression). logical differences in reward-processing preceding ICD Due to covariance between low and high k, the mean development. This difference was independent of known between them was included instead of including both. risk factors. Previous studies have shown other functional None of them was significantly associated with ICD devel- differences prior to the development of ICDs. One study opment (minimum p = 0.43 for BART). A survival analy- evaluated magnetic resonance imaging (MRI) functional sis did not show a significant association between any connectivity in resting state of 30 drug-naïve PD patients impulsivity or risk measures and ICD development in this who were followed up for 3 years.46 It found that a lower cohort (minimum p = 0.35 for BART). anticorrelation between the default mode network and the right central executive network was independently associ- ated with the development of ICDs. However, the sample Discussion evaluated might not be representative of the PD popula- To the best of our knowledge, this is the first longitudinal tion, as half of the patients developed ICDs within 3 years prospective study of ICD prediction recruiting a PD sam- (or 20.6/100 patient-years), which represents very high ple ad hoc. Among a sample of patients with PD free of incidence compared to other samples.4,37 Furthermore, ICDs, we found that those who were going to develop an each of the patients in the half who remained free of ICDs ICD had greater FRNdiff amplitude than those who were had a QUIP-RS of 0. The relationship between resting not. We also found that this difference was useful to pre- state connectivity and reward-processing has not yet been dict ICD development. Consequently, we report for the studied. This makes the interpretation of the current first time a predictive model for PD-ICDs presented in a results in line with those of resting connectivity too way that can be both replicable and transferable (ie, model speculative. parameters are included). Previous longitudinal prospec- Another study targeted dopamine transporter (DAT) tive studies that recruited an ad hoc cohort did not offer availability (123I-FP-CIT single photon emission com- prediction.37,39 puted tomography [SPECT]) in 71 drug-naïve PD A previous study offered a genetic predictive patients with SPECT performed as part of the diagnostic model.35 This study was not based on an ad hoc cohort process prior to study inclusion.47 They compared the December 2022 981 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ANNALS of Neurology uptake of 6 regions of interest (ROIs) within the basal receive DAs with greater confidence or even at greater ganglia between 11 patients who developed ICDs and doses. 20 who did not within an average follow-up of 2.5 years, This strategy might be considered prognostic with the other 40 patients either excluded or lacking targeting50 and is similar to common practices like the use follow-up. They found 3 of the ROIs to have lower of the CHADS2 (Congestive heart failure, Hypertension, uptake in patients who subsequently developed ICDs. This Age, Diabetes mellitus, Stroke) score to decide which finding has 2 possible interpretations; patients susceptible to patients with atrial fibrillation should receive anti- ICDs may have greater neurodegeneration within the sub- coagulation treatment.51 Prognostic targeting presupposes stantia nigra pars compacta (SNpc), or they may have lower that patients at high risk will benefit the most from an availability of DAT for a similar number of neurons. A low intervention to blunt this risk. Conversely, individual number of DATs or lower affinity to the ligand may explain treatment effect models try to identify the patients who a lower availability. These interpretations are not mutually will benefit the most from an intervention (regardless of exclusive. Whereas fewer neurons in the SNpc would lead the total risk). They require fitting a predictive model to lower dopaminergic activity, fewer DATs in each neur using randomized trial data.50 Nonetheless, the develop- would lead to higher dopaminergic activity. The relation- ment of an individual treatment effect model for PD- ship between dopaminergic activity and event-related poten- ICDs would provide the maximum benefit for every tials has been studied in one PD study.48 It found greater patient regardless of the risk. error-related potentials in L-dopa on state than in L-dopa Regarding transferability of the results, this study has off state, but it did not include FRN evaluation. Taken 2 limitations. First, although FRN recording is not expensive together, these and our results might support the idea of or time-consuming, the patient, the technician, and the greater dopaminergic activity in patients prone to develop equipment need to be at the same place. This means that to ICDs. Nonetheless, this study used a probabilistic learning transfer our findings to the whole population, many register- paradigm and targeted error-related negativity,49 a potential ing sites or patient travel is required. Nonetheless, data that appears 80 to 150 milliseconds after response rep- processing can be centralized and automated. Furthermore, resenting error, not feedback value. Therefore, the relation- FRN recording could take advantage of the EEG equipment ship they found may not hold for the FRN. and technicians currently in use. Nonetheless, models based In contrast to the evaluation of DAT availability, on wet biomarkers would be easier to escalate. Second, as FRN recording is noninvasive and radiation free, and the explained above, pure individualized treatment models are infrastructure required is much simpler. Our methods can desirable and cannot be obtained from longitudinal prospec- be replicated in any quiet environment with 2 conven- tive cohorts like this one, but require randomized trials. It tional laptop computers and an EEG recorder. The time would also be desirable to include patients with depression required is similar to that of an MRI, but with a simpler due to its known relationship with ICD development. Their setting, and without its contraindications and difficulties. exclusion in the present study due to its known relationship Furthermore, FRN recording is currently cheaper than with the FRN precluded us from evaluating its role in FRN both MRI and 123I-FP-CIT SPECT. predictive models. Furthermore, the exclusion of depressed This study may also shed some light on the relation- patients may limit the interpretation of apathy and anxiety ship between impulsivity and ICDs. It is to the best of scores due to the high correlation between these variables. our knowledge the only prospective study evaluating the From the point of view of ICD knowledge, the eval- relationship between several measures of impulsivity and uation of patients who are drug-naïve as well as patients ICD development, and we did not find any association. with active ICDs and patients with remitted ICDs would This may reinforce the results of our previous study in be desirable. For this study, we chose a design that favored which we found that impulsivity was associated with ICD applicability (as the results can be applied to a larger prev- severity but not ICD presence,33 as previous evidence of alent population), but future studies with large cohorts of the relationship between impulsivity and ICD presence drug-naïve subjects are warranted. comes from case–control studies. To conclude, using event-related potentials, we show The model developed in our study discriminated that electrophysiological differences in reward-processing patients with low risk (2% per year) and with high risk precede the development of impulse control disorders in (>20% per year). If validated with another cohort, this PD. We also show that this phenomenon has enough pre- model could change the way clinicians confront the risk of dictive power to deploy preventive strategies in clinical set- PD-ICDs, because many cases could be prevented by tings. We think that the time has come to take full avoiding DAs in patients at high risk, using them only responsibility for what we prescribe in PD and to deploy a when indispensable. Conversely, patients at low risk could coordinated strategy to prevent PD-ICDs. 982 Volume 92, No. 6 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Marín-Lahoz et al: Prediction of ICDs 5. Weintraub D, Papay K, Siderowf A. Parkinson’s Progression Markers Initiative. Screening for impulse control symptoms in patients with de Acknowledgments novo Parkinson disease: a case-control study. Neurology 2013;80: This study was funded by competitive grants awarded by 176–180. Fundacio la Marato de TV3 (grants #2014/U/477 and 6. Maloney EM, Djamshidian A, O’Sullivan SS. Phenomenology and epidemiology of impulsive-compulsive behaviours in Parkinson’s dis- #20142910) and by Fondo de Investigaciones Sanitarias ease, atypical parkinsonian disorders and non-parkinsonian from the Instituto de Salud Carlos III (ISCIII) and Fondo populations. J Neurol Sci 2017;374:47–52. Europeo de Desarrollo Regional grants #PI15/00962 and 7. Leroi I, Harbishettar V, Andrews M, et al. Carer burden in apathy and #PI18/01717. J.P. is the recipient of a grant from Pla impulse control disorders in Parkinson’s disease. Int J Geriatr Psychi- atry 2012;27:160–166. Estratègic de Recerca i Innovacio (SLT008/18), and 8. American Psychiatric Association. Diagnostic and statistical manual J.M.-L. is the recipient of a Juan Rodés contract from of mental disorders. 5th ed. Washington, DC: American Psychiatric ISCIII (JR20/00007). The research group also receives Publishing, 2013. funds from CERCA (CEntres de Recerca de CAtalunya) 9. Rosenberg KP, Feder LC. Behavioral addictions: criteria, evidence, and CIBERNED (Centro de Investigacion Biomédica en and treatment. 1st ed. London, UK; Waltham, MA: Academic Press, 2014. Red de enfermedades NEuroDegenerativas). None of the 10. Aracil-Bolaños I, Strafella AP. Molecular imaging and neural networks funders played any role at any stage of the study. in impulse control disorders in Parkinson’s disease. Parkinsonism We thank patients and caregivers who offered their Relat Disord 2016;22:S101–S105. efforts and made a low attrition rate possible; I. Yarritu 11. Eisinger RS, Ramirez-Zamora A, Carbunaru S, et al. Medications, deep brain stimulation, and other factors influencing impulse control Corrales, M. Cornella, and R. Fernandez de Bobadilla, disorders in Parkinson’s disease. Front Neurol 2019;10:86. who contributed to data acquisition; B. Pascual Sedano, 12. Marinus J, Zhu K, Marras C, et al. Risk factors for non-motor symp- J. Pérez Pérez, and A. Gironell, who contributed to toms in Parkinson’s disease. Lancet Neurol 2018;17:559–568. patient recruitment; and La Marato de TV3 anonymous 13. Volkow ND, Morales M. The brain on drugs: from reward to addic- donors, who made the funding possible. Finally, we want tion. Cell 2015;162:712–725. to specially thank J. Riba Serrano, who contributed to the 14. Kalivas PW, Volkow ND. The neural basis of addiction: a pathology idea and the design of the study and to data acquisition, of motivation and choice. Am J Psychiatry 2005;162:1403–1413. but passed away before he could participate in the inter- 15. Gehring WJ, Willoughby AR. The medial frontal cortex and the rapid processing of monetary gains and losses. Science 2002;295: pretation of data and writing of the manuscript. 2279–2282. 16. Hauser TU, Iannaccone R, Stämpfli P, et al. The feedback-related Author Contributions negativity (FRN) revisited: new insights into the localization, meaning and network organization. Neuroimage 2014;84:159–168. J.M.-L., S.M.-H., and J.K. contributed to conception and 17. Martínez-Horta S, Riba J, de Bobadilla RF, et al. Apathy in design of the study. J.M.-L., S.M.-H., J.P., A.H.-B., I.A.- Parkinson’s disease: neurophysiological evidence of impaired incen- B., H.B.-k., F.S., and A.C. contributed to acquisition and tive processing. J Neurosci 2014;34:5918–5926. analysis of data. J.M.-L., J.P., F.S., S.M.-H., and 18. Hughes AJ, Daniel SE, Kilford L, Lees AJ. Accuracy of clinical diag- J.K. contributed to drafting the text or preparing the nosis of idiopathic Parkinson’s disease: a clinico-pathological study of 100 cases. J Neurol Neurosurg Psychiatry 1992;55:181–184. figures. 19. Emre M, Aarsland D, Brown R, et al. Clinical diagnostic criteria for dementia associated with Parkinson’s disease. Mov Disord 2007;22: Potential Conflicts of Interest 1689–1707. quiz 1837. 20. Goetz CG, Tilley BC, Shaftman SR, et al. Movement Disorder Nothing to report. Society-sponsored revision of the unified Parkinson’s disease rating scale (MDS-UPDRS): scale presentation and clinimetric testing results. Mov Disord 2008;23:2129–2170. 21. Tomlinson CL, Stowe R, Patel S, et al. Systematic review of levodopa References dose equivalency reporting in Parkinson’s disease. Mov Disord 2010; 1. Chaudhuri KR, Sauerbier A, Rojo JM, et al. The burden of non-motor 25:2649–2653. symptoms in Parkinson’s disease using a self-completed non-motor 22. Stiasny-Kolster K, Mayer G, Schäfer S, et al. The REM sleep behavior questionnaire: a simple grading system. Parkinsonism Relat Disord disorder screening questionnaire—a new diagnostic instrument. Mov 2015;21:287–291. Disord 2007;22:2386–2393. 2. Seppi K, Chaudhuri KR, Coelho M, et al. Update on treatments for 23. Marín-Lahoz J, Sampedro F, Horta-Barba A, et al. Preservation of nonmotor symptoms of Parkinson’s disease—an evidence-based brain metabolism in recently diagnosed Parkinson’s impulse control medicine review. Mov Disord 2019;34:180–198. disorders. Eur J Nucl Med Mol Imaging 2020;47:2165–2174. 3. Corvol J-C, Artaud F, Cormier-Dequaire F, et al. Longitudinal analysis 24. Belouchrani A, Abed-Meraim K, Cardoso J-F, Moulines E. A blind of impulse control disorders in Parkinson disease. Neurology 2018; source separation technique using second-order statistics. IEEE Trans 91:e189–e201. Signal Process 1997;45:434–444. 4. Marín-Lahoz J, Sampedro F, Martinez-Horta S, et al. Depression as a 25. Delorme A, Makeig S. EEGLAB: an open source toolbox for analysis risk factor for impulse control disorders in Parkinson disease. Ann of single-trial EEG dynamics including independent component anal- Neurol 2019;86:762–769. ysis. J Neurosci Methods 2004;134:9–21. December 2022 983 15318249, 2022, 6, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/ana.26486 by Readcube (Labtiva Inc.), Wiley Online Library on [26/10/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License ANNALS of Neurology 26. Pagonabarraga J, Kulisevsky J, Llebaria G, et al. Parkinson’s Disease- disorders in Parkinson’s disease. Parkinsonism Relat Disord 2012;18: Cognitive Rating Scale: a new cognitive scale specific for Parkinson’s 1079–1083. disease. Mov Disord 2008;23:998–1005. 40. Marek K, Jennings D, Lasch S, et al. The Parkinson progression 27. Litvan I, Goldman JG, Tröster AI, et al. Diagnostic criteria for mild marker initiative (PPMI). Prog Neurobiol 2011;95:629–635. cognitive impairment in Parkinson’s disease: Movement Disorder Society task force guidelines. Mov Disord 2012;27:349–356. 41. Lee J-Y, Lee EK, Park SS, et al. Association of DRD3 and GRIN2B with impulse control and related behaviors in Parkinson’s disease. 28. Marinus J, Leentjens AF, Visser M, et al. Evaluation of the Hospital Mov Disord 2009;24:1803–1810. Anxiety and Depression Scale in patients with Parkinson’s disease. Clin Neuropharmacol 2002;25:318–324. 42. Lee J-Y, Jeon BS, Kim H-J, Park S-S. Genetic variant of HTR2A asso- ciates with risk of impulse control and repetitive behaviors in 29. Starkstein SE, Mayberg HS, Preziosi TJ, et al. Reliability, validity, and Parkinson’s disease. Parkinsonism Relat Disord 2012;18:76–78. clinical correlates of apathy in Parkinson’s disease. J Neuropsychiatry Clin Neurosci 1992;4:134–139. 43. Castro-Martínez XH, García-Ruiz PJ, Martínez-García C, et al. Behav- ioral addictions in early-onset Parkinson disease are associated with 30. Brand M, Labudda K, Kalbe E, et al. Decision-making impairments in DRD3 variants. 2018. Available at: http://linkinghub.elsevier.com/ patients with Parkinson’s disease. Behav Neurol 2004;15:77–85. retrieve/pii/S1353802018300117. Accessed January 29, 2018. 31. Marín-Lahoz J, Martínez-Horta S, Sampedro F, et al. Measuring impulsivity in Parkinson’s disease: a correlational and structural 44. Vezoli J, Procyk E. Frontal feedback-related potentials in nonhuman neuroimaging study using different tests. Eur J Neurol 2020;27: primates: modulation during learning and under haloperidol. 1478–1486. J Neurosci 2009;29:15675–15683. 32. Rogers RD, Owen AM, Middleton HC, et al. Choosing between 45. Warren CM, Hyman JM, Seamans JK, Holroyd CB. Feedback-related small, likely rewards and large, unlikely rewards activates inferior and negativity observed in rodent anterior cingulate cortex. J Physiol orbital prefrontal cortex. J Neurosci 1999;19:9029–9038. Paris 2015;109:87–94. 33. Marín-Lahoz J, Pagonabarraga J, Martinez-Horta S, et al. Parkinson’s 46. Tessitore A, De Micco R, Giordano A, et al. Intrinsic brain connectiv- disease: impulsivity does not cause impulse control disorders but ity predicts impulse control disorders in patients with Parkinson’s dis- boosts their severity. Front Psychiatry 2018;9:465. ease. Mov Disord 2017;32:1710–1719. 34. Weintraub D, Mamikonyan E, Papay K, et al. Questionnaire for 47. Vriend C, Nordbeck AH, Booij J, et al. Reduced dopamine trans- Impulsive-Compulsive Disorders in Parkinson’s Disease-Rating Scale. porter binding predates impulse control disorders in Parkinson’s dis- Mov Disord 2012;27:242–247. ease. Mov Disord 2014;29:904–911. 35. Kraemmer J, Smith K, Weintraub D, et al. Clinical-genetic model pre- 48. Volpato C, Schiff S, Facchini S, et al. Dopaminergic medication mod- dicts incident impulse control disorders in Parkinson’s disease. ulates learning from feedback and error-related negativity in J Neurol Neurosurg Psychiatry 2016;87:1106–1111. Parkinson’s disease: a pilot study. 2016. Available at: https://www. ncbi.nlm.nih.gov/pmc/articles/PMC5075574/. Accessed September 36. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic cur- 8, 2019. ves: a nonparametric approach. Biometrics 1988;44:837–845. 49. Frank MJ, Woroch BS, Curran T. Error-related negativity predicts 37. Bastiaens J, Dorfman BJ, Christos PJ, Nirenberg MJ. Prospective reinforcement learning and conflict biases. Neuron 2005;47: cohort study of impulse control disorders in Parkinson’s disease. Mov 495–501. Disord 2013;28:327–333. 50. Wilson FP, Parikh CR. Translational methods in nephrology: individ- 38. R Development Core Team. R: A Language and Environment for Sta- ual treatment effect modeling. J Am Soc Nephrol 2018;29: tistical Computing. Vienna, Austria: R Foundation for Statistical Com- 2615–2618. puting, 2008 Available at: http://www.R-project.org. 51. Gage BF, Waterman AD, Shannon W, et al. Validation of clinical clas- 39. Joutsa J, Martikainen K, Vahlberg T, Kaasinen V. Effects of dopamine sification schemes for predicting stroke: results from the National agonist dose and gender on the prognosis of impulse control Registry of Atrial Fibrillation. JAMA 2001;285:2864–2870. 984 Volume 92, No. 6