Exenatide once weekly for alcohol use disorder investigated in a randomized, placebo-controlled clinical trial

Background Alcohol use disorder (AUD) is a chronic, relapsing brain disorder that accounts for 5% of deaths annually, and there is an urgent need to develop new targets for therapeutic intervention. The glucagon-like peptide-1 (GLP-1) receptor agonist exenatide reduces alcohol consumption in rodents and nonhuman primates, but its efficacy in patients with AUD is unknown. Methods In a randomized, double-blinded, placebo-controlled clinical trial, treatment-seeking AUD patients were assigned to receive exenatide (2 mg subcutaneously) or placebo once weekly for 26 weeks, in addition to standard cognitive-behavioral therapy. The primary outcome was reduction in number of heavy drinking days. A subgroup also completed functional MRI (fMRI) and single-photon emission CT (SPECT) brain scans. Results A total of 127 patients were enrolled. Our data revealed that although exenatide did not significantly reduce the number of heavy drinking days compared with placebo, it significantly attenuated fMRI alcohol cue reactivity in the ventral striatum and septal area, which are crucial brain areas for drug reward and addiction. In addition, dopamine transporter availability was lower in the exenatide group compared with the placebo group. Exploratory analyses revealed that exenatide significantly reduced heavy drinking days and total alcohol intake in a subgroup of obese patients (BMI > 30 kg/m2). Adverse events were mainly gastrointestinal. Conclusion This randomized controlled trial on the effects of a GLP-1 receptor agonist in AUD patients provides new important knowledge on the effects of GLP-1 receptor agonists as a novel treatment target in addiction. Trial registration EudraCT: 2016-003343-11. ClinicalTrials.gov (NCT03232112). Funding Novavi Foundation; Research Foundation, Mental Health Services, Capital Region of Denmark; Research Foundation, Capital Region of Denmark; Ivan Nielsen Foundation; A.P. Moeller Foundation; Augustinus Foundation; Woerzner Foundation; Grosserer L.F. Foghts Foundation; Hartmann Foundation; Aase and Ejnar Danielsen Foundation; P.A. Messerschmidt and Wife Foundation; and Lundbeck Foundation.


Introduction
Alcohol use disorder (AUD) is an essential contributor to the burden of global disease (1). In Denmark, the cumulative all-cause ten-year mortality risk is almost 30% after a first-time hospital contact due to an alcohol problem (2). Only three medications are approved by the Food and Drug Administration (FDA) to treat AUD; disulfiram, naltrexone, and acamprosate (3). About 40% of patients treated for AUD relapse within the first three years (4), and new targets for therapeutic interventions are urgently needed for this devastating chronic disease (1,3).
The endogenous glucagon-like peptide 1 (GLP-1) is a 30 amino acid peptide hormone produced in the intestinal L-cells in response to food intake (5), as well as in the nucleus tractus solitarius of the medulla oblongata (6). GLP-1 stimulates insulin secretion, inhibits glucagon secretion, and notably, dampens appetite and food intake (5). GLP-1 receptor agonists are approved by the European Medicines Agency (EMA) and FDA to treat type 2 diabetes and obesity (7). Since drugs of abuse and alcohol activate the same reward system that underlies food reward (8), it is conceivable that appetite-regulating peptides such as GLP-1 target areas associated with reward and addiction. In support for this hypothesis, several studies have reported expression of GLP-1 receptors in brain areas associated with reward and addiction (6,(9)(10)(11)(12)(13)(14)(15)(16). Furthermore, treatment with GLP-1 receptor agonists reduce alcohol intake and decrease relapse-like alcohol drinking in nonhuman primates (17) and rodents (18). In humans, a recent study reported that the GLP-1 receptor 168Ser allele variant was associated with increased alcohol intake in humans (19).
However, the effects of a GLP-1 receptor agonist on alcohol consumption in humans remain unknown. To this end, we performed a randomized, placebo-controlled clinical trial lasting 26 weeks plus a long-term six-month follow-up to evaluate the efficacy of the once-weekly GLP-1 receptor agonist exenatide (Bydueron ® ) at a dose of 2 mg in patients diagnosed with AUD according to  In total, 127 treatment-seeking AUD patients, who had a minimum of five heavy drinking days, i.e., 60/48 grams of alcohol or more per day (men/women) in the past 30 days, were included. Since the pharmacodynamic and pharmacokinetic of a GLP-1 receptor agonist in patients with AUD has not been investigated, we chose a dosing regimen consistent with established tolerability and efficacy in treatment of type 2 diabetes, i.e. exenatide, 2mg subcutaneously once weekly. Importantly, exenatide crosses the blood brain barrier (20), and a similar dosing regiment, i.e. 2 mg subcutaneously once weekly, has recently shown efficacy in other neuropsychiatric disorders including nicotine dependence (21) and Parkinson's disease (20) suggesting a central engagement, possibly mediated, at least in part, by dopamine signaling (22).
The primary endpoint was reduction in heavy drinking days, recorded with the time-line followback method (23). A subgroup of the patients had a functional Magnetic Resonance Imaging (fMRI) scan and a single-photon emission computerized tomography (SPECT) scan performed at baseline and at week 26. Using the fMRI technique, we investigated whether exenatide onceweekly would reduce alcohol cue reactivity in brain areas involved in drug reward and addiction, and in top-down regulation of impulsivity (24), as pre-clinical-and clinical evidence suggests that GLP-1 receptor stimulation may be associated with improved cognitive impairment (25). By use of the SPECT scan, we measured the availability of the striatal dopamine transporter (DAT), a key modulator of extracellular dopamine. Dopamine plays a pivotal role in the neurobiological underpinnings of reward (26), and a large body of evidence suggests that brain dopamine homeostasis changes following chronic alcohol intake (27).

Characteristics of the patients
From 7 th of August 2017 to 1 st of October 2019, 152 patients were screened for eligibility, and 127 patients were enrolled; 62 were randomly assigned to the exenatide group, and 65 were assigned to the placebo group ( Figure 1). Overall, the two treatment groups were balanced with respect to baseline characteristics ( Table 1). All patients were Caucasians, with a mean age of 52 years. The majority of the patients were men (60%). On average, they had 17 heavy drinking days and an overall alcohol intake of 2400g of pure alcohol over the last month, and 80% fulfilled the criteria for severe AUD, i.e., more than five symptoms, according to DSM-5 (see baseline characteristicsand flowchart for the patients included in the brain-imaging sub-study in Appendix 1 and 2). Of the 127 patients included a total of 58 patients completed the trial, i.e., participated in the last follow-up after 26 weeks of treatment; 25 patients finished prematurely, i.e., participated in a final examination before 26 weeks of treatment. Fifty-five patients participated in the long-term sixmonth follow-up visit (Supplemental Figure 1), with the last visit held on the 10 th of October 2020.
The mean (SD) number of injections in the exenatide group was 22.6 (2.2) and in the placebo group 22.1 (2.8) (Supplemental Table 1). There was no difference (p=0.46) between the two groups in time to trial discontinuation ( Figure 2). In addition, 25 healthy controls matched with gender, age, and educational status of the patients included were recruited for the fMRI substudy.

Efficacy
For both groups, the number of heavy drinking days (Table 2, Figure 3) and total alcohol intake ( Table 2) were strongly reduced, but there were no significant differences between the two groups.

Exploratory analyses
Exenatide once weekly did not reduce the number of heavy drinking days in the prespecified subgroup analyses (baseline heavy drinking days, severity of DSM-5 criteria, and geography) (Supplemental Table 6). However, an exploratory subgroup analysis (Supplemental Table 7) including BMI-subgroups, revealed that in obese patients with a BMI >30 kg/m 2 (n=30), exenatide reduced heavy drinking days by 23.6 percentage points (95% Cl -44.4 to -2.7, p=0.034) ( Figure 4) and reduced total alcohol intake per 30 days by 1205 grams (95% Cl -2206 to -204, p=0.026) relative to placebo ( Figure 5). In patients with a BMI <25 kg/m 2 (n=52), treatment with exenatide increased number of heavy drinking days by 27.5 percentage points (95% Cl 4.7 to 50.2, p=0.024) relative to the placebo group. However, in this subgroup (BMI <25 kg/m 2 ) the total alcohol intake did not differ between treatment groups. Other exploratory posthoc subgroup analyses were performed to investigate if there were sub-groups responding differently than others on the intervention. However, no significant differences were observed with respect to gender, baseline craving (PACS score), baseline AUDIT score, baseline number of days without alcohol, baseline total alcohol consumption, fMRI-subgroup (n=22), and SPECT-subgroup (n=16).
Besides the exploratory subgroup analyses, we also looked at the reduction in WHO Risk Drinking Levels (28). Both groups reduced their risk drinking levels, but there was no significant difference between the two groups (Supplementary Table 6).
To explore if there were a correlation between change in HbA1c and change in Heavy drinking days, the Pearson correlation coefficient was computed in the imputed dataset (n=127) to assess linear relationship. Here, we found a weak negative correlation between the two variables, r(12755) = -.27, p = .001. We also found a weak negative correlation between changes HbA1c and changes in total alcohol intake, r(12755) = -.36, p = .001.

Six months long-term follow-up
There was no difference between the two groups at the six-month follow-up after exenatide or placebo discontinuation ( Table 2, Supplemental Table 8), except for a higher AUDIT-score (5.1 points, 95% Cl 0.9 to 9.3, p=0.02) in the original exenatide group (adjusted from the end of treatment) compared to the placebo group.

fMRI Alcohol cue-reactivity
The predefined region of interest (ROI) masks was acquired from the WFU PickAtlas. The At baseline, the exploratory whole-brain analysis showed no significant difference in cue reactivity between the placebo group and the exenatide group. When comparing cue reactivity in all patients with healthy controls, significant differences were found in the left superior-, and middle frontal gyrus, caudate, and insula (p=0.001). However, at the 26 weeks rescan, these differences were no longer significant. At the week 26 assessment, cue-induced activation was significantly reduced in the exenatide group compared to the placebo group in the following brain areas (Appendix 1: Supplemental Table 13); left caudate nucleus and septal area ( Figure 7A), and right middle frontal gyrus ( Figure 7B). There was no significant change in cue reactivity in the placebo group at the rescan, but the exenatide group showed a significant reduction in cue-induced activation in the temporal lobe, hippocampus, and parahippocampus (rescans per-protocol; Appendix 1: Supplemental Table 14, Supplemental Figure 5; rescans per-protocol including premature rescans: Supplemental Table 15, Supplemental Figure 6).

fMRI Spatial working memory
The voxel-wise analysis showed a significant reduction in the exenatide group at the week 26 rescan compared to placebo in response to the 2-back>1-back task in two clusters in the right frontal pole and right superior frontal gyrus, within the dorsolateral prefrontal cortex ROI ( Figure   8 & Appendix 1: Supplemental Table 16). The additional right dlPFC ROI analysis showed no significant change in the exenatide group at week 26 compared to placebo in task-related activations (F(1,31), p=0.122, partial η2 = 0.076). The reduction in task-related neuronal activations in the exenatide group occurred in the absence of change in cognitive performance on the SCIP (p=0.93).

SPECT Dopamine transporter availability
After adjustment for age, there were no significant differences comparing baseline DAT availability of the patients with AUD and healthy controls in striatum F (1,62) Figure 9B).

Safety
Gastrointestinal symptoms, bodyweight loss, fatigue, and injection site reactions were the most common adverse events reported, and the incidence was higher in the exenatide compared to the placebo group (nausea, 37.1% vs. 15.4%; decreased appetite, 24.2% vs. 9.2%; vomiting, 22.6% vs. 7.7%; overall weight loss, 67.7% vs. 40.0%; fatigue, 12.9% vs. 4.6%, and injection site reaction, 41.0% vs. 0.0%) ( Table 3). The gastrointestinal side-effects reported lasted up until the first five weeks of treatment, and the weight loss continued throughout the trial. The injection site reactions were typically small nodules of 1-2 centimeters, hard, mobile, skin-colored, and were reabsorbed within six weeks, leaving no scar. Serious adverse events were reported almost equally between the two groups (exenatide 24.2% vs. placebo 18.5%), and there were no cases of acute pancreatitis or elevation of pancreas enzymes above upper limits. One patient in the exenatide treatment group committed suicide seven weeks after withdrawal from the trial. One patient in the placebo group was found dead after being hospitalized three times in one week for alcohol withdrawal symptoms.

Discussion
This is the first RCT investigating the effects of exenatide in AUD patients. Treatment with exenatide once weekly was not superior to placebo in reducing the number of heavy drinking days in the prespecified analysis. The negative results could reflect the characteristics of the AUD patients included in our RCT. Data from preclinical trials showed that high alcohol-consuming animals decreased their alcohol intake significantly more than low alcohol-consuming animals when treated with a GLP-1 receptor agonist (29,30). In the present trial, 80% of the patients fulfilled the DSM-5 criteria for severe AUD. However, their severity profile, based on baseline alcohol intake and heavy drinking days (Table 1), was less severe than observed in other AUD pharmacotherapy trials (31,32). Another explanation could be that the potent placebo response could have masked a possible beneficial effect of exenatide ( Figure 3-4). The observed potent placebo response could be due to the standardized cognitive behaviour therapy (CBT) against AUD (38) offered to all participants in the study, but it could also be due to the less severity profile of the AUD patients included, which is typically linked to a higher placebo response (33). Large placebo responses are also reported in other clinical AUD trials and shown to be negatively correlated with the treatment intervention effect sizes (34).
Predefined fMRI brain-ROI analysis found a reduced alcohol cue-reactivity in the exenatide group compared to the placebo group in the ventral striatum, a region that plays a pivotal role in addiction and relapse ( Figure 6). This finding is important because it implies that AUD subjects treated with exenatide lose the incentive salience of alcohol-associated cues. The exenatide-induced reduction in cue-reactivity in the septal area (35) observed in the whole-brain analysis ( Figure 7A) is particularly intriguing as this is an area connected to reward (15), and a brain area where GLP-1 receptors are highly expressed (6). These findings are in accordance with a central effect of exenatide as mentioned in the introduction. Future fMRI studies investigating the effects of GLP-1 receptor agonists on alcohol-cue induced activation should include the septal area as a region of interest (ROI). Impairments in cognitive processes related to executive function in AUD patients (36), may negatively influence clinical outcomes due to deficits in self-regulation (37). In the fMRI spatial working memory test, we found reduced cue-reactivity in the dorsal prefrontal cortex in the exenatide group compared to the placebo group, possibly indicating a reduced effort to maintain cognitive performance (38).
The SPECT sub-study revealed no significant differences in DAT availability at baseline between the AUD patients and healthy controls, which is in accordance with some earlier findings (39), but in discordance with others (40). After 26 weeks of treatment, the analysis revealed a significant reduction of DAT in the striatum, caudate, and putamen in the exenatide group, compared to placebo, which might compensate for the decreased dopamine activity previously reported in AUD patients (41). Notably, this effect is most likely not acutely induced since no change in DAT availability was observed after acute treatment with exenatide in healthy volunteers (42).
Even though the results from the exploratory post hoc BMI subgroup analysis are preliminary, we think they are of substantial interest because overlapping dysfunctional brain circuits are observed in individuals who suffer from obesity or addiction (8), and deranged GLP-1 signaling is also reported in obese individuals (43). In addition, an fMRI study in obese vs. lean individuals showed that exenatide infusions "normalized" the brain response to a food-paradigm in obese patients with a BMI > 30 kg/m 2 compared to lean individuals (44). Moreover, several GLP-1 receptor agonists have recently been approved to treat obesity (BMI>30 kg/m 2 ), and other compounds are under development (7). The reason why the number of heavy drinking days was increased in the subgroup of exenatide-treated patients with a BMI < 25 kg/m 2 compared to placebo-treated patients could be that those lean individuals treated with exenatide experienced a larger decrease in blood sugar (45), and this might be associated with increased alcohol craving (46). The significant increase in urinary oxidative stress markers in the exenatide group was previously reported in type 2 diabetes patients treated with exenatide (47), but the clinical significance of rising levels of urinary stress parameters 8-oxoGuo and 8-oxodG is currently unknown (48).
Notably, increased urinary oxidative stress parameters in patients with type 2 diabetes is associated with increased mortality risk (49), and the clinical impact of these biomarkers should be further investigated.
GLP-1 receptor agonists have shown beneficial skeletal effects in rodents (50). However, in the present trial, no differences in bone turnover markers were observed between groups, indicating that bone-related adverse effects are not of concern in this patient population.
Both the exenatide group and the placebo group exhibit an overall reduction in DUDIT-score after 26 weeks of treatment. However, the exenatide group had a significantly higher DUDIT-score compared to placebo after 26 weeks of treatment (Table 2). An exclusion criterion was a diagnosis of any active substance use disorder (SUD) except for nicotine. Individuals who had a DUDIT score > 6 (men), >2 (women) were screened according to ICD-10 SUD criteria and, if diagnosed with SUD, excluded from the trial. Only four of the 25 included participants with a positive DUDIT score (range between 1-22 points) finished per protocol. This is essential information for a follow-up study, where it may be relevant to exclude all individuals with a positive baseline DUDIT-score to increase study compliance.
The previously reported safety profile of exenatide once weekly is consistent with the present safety data. Our most significant safety concern was the risk of pancreatitis in patients with AUD (3) combined with the associated risk of exenatide treatment (51,52). Importantly, none of the patients experienced a rise in blood amylase above upper limits or developed pancreatitis.
Surprisingly, the injection site reactions to exenatide were a bigger problem for the patients due to unexpected concerns from their relatives, who might have been unaware of their AUD diagnosis.
This led to a 6.5 % withdrawal rate specifically due to injection site reactions in our AUD trial compared with only 0.5% in exenatide treated patients with type 2 diabetes (53). The gastrointestinal (GI) side effects, which are well-recognized but typically transient (54) was in the exenatide group (44.1%) higher than reported in diabetes trials (55,56). Also, 23.6% of placebotreated patients experienced GI side-effects, indicating that this group of patients may have a GI vulnerability (57). Only a single RCT has investigated the effects of pre-treatment with antiemetics, reporting a significant reduction in nausea and vomiting in exenatide-treated healthy subjects (58).
Large drop-out rates are often observed in AUD intervention trials (59) and the present study iswith a drop-out of 54.3% -no exception. Although our sensitivity analysis (Supplemental Table   9) confirmed the robustness of the results even with imputations of missing data, the present dropout rate (69 out of 127) remains a concern when evaluating the reproducibility and reliability of the findings. Weekly visits for 26 weeks might have been a contributory factor. However, in accordance with the EMA guidelines (60), we chose a study duration time of 26 weeks to see if there was a sustained treatment effect, lasting longer than the 12 weeks often reported for alcohol RCTs (61).
The approved 2 mg dosing regimen for diabetes patients is reported as the maximally efficacious dose for glucose control, reduction in body weight, and tolerable side effects (62). Our data also shows that AUD patients obtain the same incretin response as diabetes patients with respect to improved glycemic control, weight loss, and side effects. Also preclinically, the standard exenatide dose used in preclinical food reward trials (63) has shown effects in preclinical alcohol self administration experiments (64,65). We did report a central effect in the brain imaging sub-studies, but of course, we cannot rule out that the standard dose given, was too low to elicit a reduction in number of heavy drinking days. However, the mean plasma exenatide level in this study were four times as high as reported as the minimal effective concentration in humans ≈ 50 pg/ml (66). Also, due to safety concerns in this vulnerable group of patients, we did not raise the dose above the registered dose for treatment of type 2 diabetes.
Previous studies in diabetes patients have reported that while 45% of individuals receiving exenatide generate low-titer anti-exenatide antibodies (67), there is no apparent correlation between antibody titers and the effect of exenatide on mean HbA1c (55,67). To the best of our knowledge, there is also no evidence of altered exenatide clearance in AUD patients. The renal elimination of exenatide (5) is an advantage in this group of patients, who typically have a heightened risk of hepatic injury (1).
One would expect a correlation between reduced brain alcohol cue reactivity and alcohol consumption. However, this was not the case in the present study, neither for the whole group of patients (n=127) nor for the subgroup of patients that were fMRI-scanned (n=22) or SPECT- 19 scanned (n=16). The sample size of the fMRI BMI-subgroups with BMI<25 (n=7) or BMI>30, (n=5), was too small to further explore, if the overall fMRI striatal responses were correlated with heavy drinking days in the overweight or obese subgroups. Only a few RCTs on AUD patients, including fMRI measurements at baseline and follow-up, have been performed (68), and most studies have been underpowered or have to varying study populations to report significant clinical treatment effects (69). The study protocol, statistical analysis plan and de-identified individual participant data, except raw fMRI-, MRS-data, and alcohol dictionaries, will be available at the Mendeley database (70) .
Data will be available at publication, and access criteria are a methodologically sound proposal with an approved aim directed to the corresponding author, and requestors will have to sign a data access agreement. Data will be available for five years.

Patients
All potential participants received oral and written information about the project. Before signing the written consent form, the alcohol breath concentration had to be below 0.5‰, which is the same limit as driving a motor vehicle in Denmark (71). Eligible patients were 18-70 years of age, diagnosed with AUD according to DSM-5 and alcohol dependence according to ICD-10 and treatment-seeking. Inclusion criteria required a minimum of five heavy drinking days, i.e., 60/48 grams of alcohol or more per day (men/women), in the past 30 days, measured by use of the Time-Line Follow Back (TLFB) method (72). Key exclusion criteria included severe mental disorder, other drug use disorder, a history of diabetes, pancreatitis, alcohol withdrawal seizures, and current treatment with drugs against alcohol dependence (disulfiram, acamprosate, naltrexone, and nalmefene). Full in-and exclusion criteria are listed in Supplemental Table 10. The healthy controls included in the fMRI-sub study (n=25) were matched by sex, age, and educational level.
All patients were recruited from outpatient alcohol treatment facilities in the suburbs of Copenhagen or through our project webpage, and healthy controls, via the project webpage. No patients were involved in setting the research question, planning the study, interpreting-, or writing up the results. The results of the trial and the assigned intervention will be disseminated to all patients and healthy participants.

Procedures
The randomization was stratified in terms of sex, age (+/-40 years of age), and number of heavy drinking days at baseline (four strata), and the patients were randomly assigned 1:1 by REDCap (73), to receive exenatide once weekly (Bydureon®), 2 mg or placebo subcutaneously. The weekly injections were administered by an unblinded project nurse who did not participate in any assessments or behavioral treatment sessions. No randomization was performed in the imaging subgroup, as all eligible patients were invited to participate.
Patients who participated in the brain imaging sub-study were scanned before receiving the first injection and again after 26 weeks of treatment. Throughout the trial, patients received the assigned treatment blindfolded by an unblinded nurse at the outpatient clinic to whom they also delivered their weekly alcohol diary. Patients were assessed by blinded project staff at the time of screening, at week 4, 12, 20, 26 (end of the main trial), and at the long-term six-month follow-up visit (Supplemental Table 11, Supplemental Figure 2). At every assessment, weight, somatic symptoms, or diseases since the last visit were recorded, and safety blood samples were collected. In case medical assistance was needed, a 24-hour phone line was available. As a safety precaution due to earlier associations of pancreatitis caused by GLP-1 receptor agonist treatment (74), blood pancreas amylase was measured at all assessments. Participants with initial severe GI side-effects received injections every second week for the first six weeks to reduce GI-symptoms. All harms were recorded up until ten weeks after termination of the intervention e.g week 26.
Throughout the trial, all patients received the assigned treatment as an add-on to standard AUD behavioral treatment, which included therapy sessions every second week, with a combination of motivational interviewing (MI), cognitive therapy, and family therapy with a blinded therapist.
Patients discontinuing the trial after a minimum of eight weeks were encouraged to participate in a premature final visit and rescan. Only patients completing the week 26 visit (premature + per protocol) were invited for the long-term six-month follow-up visit.
The healthy fMRI control group was assessed for eligibility before brain imaging at the Neurobiology Research Unit at Rigshospitalet, Copenhagen, Denmark. See Appendix 1 for full details of the fMRI sub-study and Appendix 2 for the SPECT sub-study.

Outcomes
The primary endpoint was change in heavy drinking days, from baseline to week 26, as recorded by the TLFB-method. Secondary endpoints included changes in total alcohol consumption;

Statistical analysis
The study was designed to have 90% power to detect a 28-percentage point treatment difference between the two groups with an estimated drop-out of 40%. We planned to include 114 patients, but due to a 60% drop-out, we extended enrolment until the 1 st of October 2019 or until 144 patients were included, whichever came first. All continuous outcomes were analyzed with an ANOVA adjusted for baseline until the last observational endpoint, and missing data were imputed with the use of multiple imputations in the mice package (75) in R software version 3.6.0 (76), method = "pmm" (predictive mean matching), and the number of imputed datasets = 100.
No adjustment for covariates was performed. SCIP data were analyzed with a linear mixed model, adjusted for benzodiazepine intake at the time of the assessment. DUDIT data were analyzed with a censored regression model due to zero-inflated values. An exploratory subgroup analysis based on the World Health Organization (WHO) body mass index (BMI) categories (77) was performed to see if the effect of the treatment was related to baseline BMI. The statistical analysis plan was uploaded to ClinicalTrials homepage (78) and the dataset was locked before any analysis were performed. All statistical analyses, except the post hoc analysis regarding exenatide plasma levels, were performed blinded. The hypothesis-test was two-sided, the level of statistical significance was 5%, and a confidence interval of 95%. All efficacy and safety analyses were performed after 24 the intention-to-treat principle. Analyses were performed with the R software version 3.6.0 (76).
See appendix 1 and 2 for the complete statistical method for the fMRI-and SPECT analyses.      Table 7 is included. Data were analyzed with an ANOVA adjusted for baseline, and missing data is imputed with the use of multiple imputations as described in the text. Data represent mean ± SEM.    The exenatide group showed a reduction at follow-up in the response to 2-back>1-back task compared to the placebo group (2-way mixed effect ANOVA; placebo n=16, exenatide n=17, control n=25) in two prefrontal clusters (frontal pole x,y,z=34, 54, 20, corrected p<0.002; superior frontal gyrus x,y,z=4, 46, 46, corrected p<0.001). Boxes represents upper and lower quartiles, the line represents the median, and the X represents the mean.