Transcranial Magnetic Stimulation Pilot Study for Methamphetamine Use Disorder

Overview: This is a single-arm clinical trial, with historical controls as a comparison group for select outcomes. Subjects with MAUD will receive 16 sessions of dual-target theta burst stimulation to the DLPFC and MPFC over 4 weeks. We will follow outcomes for 12 weeks. Outcomes include treatment retention, craving, self-reported MA or stimulant use, urine drug screen results, depressive symptoms, anxiety symptoms, sleep quality, quality of life, response inhibition, and functional connectivity. Magnetic resonance imaging (MRI) to measure functional connectivity and a flanker task to measure response inhibition will be completed at baseline and four weeks. More detail is provided in the outcome measures section. Subjects will also complete the Big Five Inventory at baseline, a measure of personality characteristics, to explore how these relate to outcomes including retention in treatment and the study.

MR Image Acquisition: MRI will be completed at baseline and after the last TMS session. The MRI sessions will be conducted using research dedicated MRI scanners at each site. Anatomical images will include volumetric T1 and T2 weighted images with a 1.0 mm isotropic spatial resolution. Resting state fMRI will be performed to collect 20 minutes worth of data.

Statistical Analysis:

Retention in the Study and Psychosocial Treatment: We will describe the proportion of subjects who complete the 4-week TMS treatment period and complete each subsequent monthly follow-up visit. We will use Kaplan-Meier curves to describe retention in the study and in psychosocial treatment. If non-retention is common enough, we will use Cox regression to explore baseline measures as predictors of retention. We anticipate that multivariate analysis will not be feasible with the sample size.

Symptoms and Impulse Control Measures: Primary analyses for other measures will focus on changes over the 4-week TMS treatment period. Changes in symptoms evaluated weekly or biweekly (e.g. craving, depression, affect, anxiety, sleep) will be assessed using generalized linear mixed models with appropriate distributions. We anticipate a Poisson distribution for days of MA or other stimulant use and will use a binomial distribution with a logit link to evaluate changes in positive urine drug screens. Time will be the primary variable of interest to assess the slopes of change during treatment. Paired t-tests or Wilcoxon signed-rank tests will be used to evaluate changes in measures completed at baseline and after 4 weeks of treatment. We will compare measures at baseline and 4 weeks to those at 8 weeks and 12 weeks similarly, but in separate analyses since decay in effects may occur after TMS ends.

Functional Connectivity Analysis: fMRI functional connectivity analysis will be performed using a standard analysis pipeline. Functional images will undergo pre-processing including brain extraction, motion correction, spatial smoothing (6 mm FWHM), and temporal filtering (.008 Hz \< f \< 0.08 Hz). Following preprocessing, the fMRI signal will be corrected for potential sources of noise using image-based estimates and motion correction parameters. The resulting corrected time-series will be used for all functional connectivity analyses. Functional connectivity will be measured by extracting time-series data from the pre-processed imaging data for the regions of interest (ROIs). Multiple ROIs will be examined and will be defined as spheres (6mm radius) based on coordinate locations previously published by Yeo and colleagues. Specifically, we will focus on connectivity in the cingulo-opercular network involved in cognitive control and salience (DLPFC - anterior insula; DLPFC - anterior cingulate) and reward processing/motivation circuit (MPFC - ventral striatum). Analyses will be averaged across right and left hemispheres but we will also explore differences between right and left hemispheres. The time series from the ROIs will be cross-correlated with the time-series from the other ROIs to determine the strength of functional connectivity between regions. The resulting Pearson's r will be converted to Fisher's z scores to improve normality for the statistical analysis. We will treat each ROI pair connection (DLPFC - anterior insula, DLPFC - anterior cingulate, MPFC - ventral striatum) as a dependent variable. Primary analyses will use paired t-tests or Wilcoxon signed-rank tests to compare connectivity at baseline vs. after TMS treatment. We will explore correlates of connectivity and changes using Pearson or Spearman correlations and linear regression or mixed models.

Exploratory Analysis: Follow-up exploratory voxel-wise analyses will be conducted for functional connectivity, which will provide thousands of individual predictors. This will help confirm findings in large parcel ROI based analysis. We will use the same statistical models as used for the ROI based analysis described above but at the voxel level. Voxel-wise data creates a high-dimensional problem in which the number of predictors far exceeds the number of participants. Machine learning methods, such as random forest will be used to handle the high-dimensional sub-analyses. Random forest requires a minimum of data assumptions, automatically accounts for non-linear and interaction effects, and it has proven useful in identifying useful predictors in high-dimensional contexts.

Comparison with Historical Controls: We will compare retention in psychosocial treatment programs and positive urine drug screens from chart review with a historical control group of patients with MA use disorder matched on age, sex, site, and psychosocial treatment modality. Treatment retention will be compared using a log-rank test. Positive urine drug screens in each week of follow-up will be compared using generalized estimating equation models with a logit link, clustered on subject, with participation in the TMS study as the variable of interest.