Study participants will receive 7 weekly 60-minute MT or CBT sessions over the first 7 weeks. All sessions will be delivered remotely through Zoom’s Health Insurance Portability and Accountability Act (HIPAA)–compliant, encrypted, and passcode-protected videoconferencing software. Although the duration of these types of treatments may vary in real-world settings, this study will match the MT and CBT groups for patient-therapist contact time to ensure rigorous comparative effectiveness comparison.

MT will be delivered by board-certified music therapists. The MT protocol is based on the neuroscience of music [54,105], the social-cognitive processing model [51,52], the scientific literature on MT [72], and the clinical and research experience of the study team [106]. Healthy psychological adjustment during cancer survivorship requires the processing of feelings and thoughts related to one’s cancer journey in a supportive context [51,52]. Therefore, the intervention protocol is structured to build patient-therapist rapport and allow for greater emotional expressivity and deeper cognitive processing. Guided by the social-cognitive processing model [51,52], the first few sessions will focus on building a trusting relationship with the therapist, whereas the later sessions will focus on the cognitive processing of cancer-related experiences and current fears, worries, and hopes, using playlist creation and songwriting as an outlet for reflection, expression, and meaning making. Research also suggests that patients prefer receptive modes of engagement during their initial exposure to MT [75,107]. As such, the early sessions will focus on receptive techniques (eg, music-guided stress management), and the later sessions will progress to more active MT techniques (eg, songwriting). The protocol includes homework activities in between sessions to promote self-management skills (eg, the use of music-guided deep breathing), strengthen social connections (eg, sharing meaningful songs with loved ones), provide outlets for cognitive processing (eg, writing or reflecting on themes for song lyrics), and serve as transitions to the subsequent sessions.

The CBT protocol is based on the scientific literature [32,33] and the clinical and research experience of the study team [108]. CBT therapists will include licensed social workers because they are the most commonly employed mental health providers in cancer care and, therefore, reflect the real-world practice and implementation of CBT in oncology settings [109]. Informed by the cognitive behavioral model of anxiety, CBT focuses on addressing the thoughts and behaviors that exacerbate or reduce anxiety [35-37]. The protocol consists of psychoeducation on anxiety, relaxation techniques, cognitive restructuring, and strategies for planning activity engagement and managing realistic worries [38,39]. These components target the somatic symptoms of anxiety (eg, muscle tension) as well as the thoughts (eg, “What if my next scan is bad?”) and behaviors (eg, excessive symptom monitoring) that trigger and exacerbate anxiety. Once identified, these problematic thoughts and behaviors are replaced by thoughts and behaviors that prevent and reduce anxiety. Each session will follow a consistent format that includes an overview of the session content, review of the homework exercise from the prior session, information on a new skill for managing anxiety, discussion of the upcoming homework exercise, and a plan for completing the exercise before the next session. Patients will receive a workbook with materials for each session and homework activities that promote practice of the skill learned during the session.

All interventionists will be trained about the specific research protocol and educated on the importance of adherence to the protocol methods. The interventionists will be either English speaking or bilingual in English and Spanish. The training of study interventionists will include didactic information on anxiety in cancer survivors and review of intervention materials. Interventionists will be provided with ongoing supervision over the course of the trial. During the trial, the CBT and MT sessions will be recorded and stored on secure, encrypted MSK servers. To ensure that study therapists adhere to the treatment protocol, document treatments appropriately, and maintain fidelity to the core functions of the intervention, the session recordings for the first 2 patients seen by a therapist will be reviewed. The review of session recordings will be based on treatment fidelity checklists outlining the core intervention components. If a therapist adheres to at least 80% of the treatment fidelity checklist items, then subsequent fidelity monitoring will be reduced to 2 randomly selected session recordings per patient. Deviations from the MT or CBT protocols will be discussed with study therapists during supervision meetings, and strategies will be suggested to minimize the number of deviations. Therapists who fail to adhere to at least 80% of the treatment fidelity checklist items will be retrained. Similar treatment fidelity strategies have been successful in other MT [106] and CBT trials [108,110] of patients with cancer conducted by the study team.

Patient-reported outcomes (Table 1) will be completed on the web using Research Electronic Data Capture (REDCap; Vanderbilt University) [111]. Participants will also have the option to complete these study assessments on paper or over the telephone with research staff. Study assessment materials will be available in English or Spanish. Patient-reported outcome measures for anxiety, comorbid symptoms, and quality of life have been validated in Spanish [112-117]. Where needed, Spanish translation of other assessments took place in standard forward and reverse translation.

The primary outcome is severity of anxiety symptoms, as assessed using the HADS anxiety subscale. The reliability, validity, and factor structure of this 7-item subscale has been established in patients with cancer with a Cronbach α of .83 [118,119]. A score of ≥8 indicates the presence of anxiety symptoms. Research has identified a minimal clinically important difference (MCID) of 1.7 points [120]. The HADS anxiety subscale will be administered at baseline and at weeks 4, 8, 16, and 26.

Given that anxiety is strongly associated with depressed mood [11-13], fatigue [14-17], insomnia [10], pain [121], and cognitive dysfunction [122], these comorbid symptoms will be assessed using validated instruments. Depressive symptoms will be assessed using the HADS depression subscale (Cronbach α=.79) [118,119]. Fatigue will be assessed using the Brief Fatigue Inventory (BFI), a 9-item scale that has been validated in cancer populations with a Cronbach α of .96 [123]. Insomnia symptoms will be assessed using the Insomnia Severity Index [124], a 7-item instrument validated in cancer populations with a Cronbach α of .90 [125]. Pain severity and pain-related interference will be assessed using the Brief Pain Inventory (BPI). The BPI has been demonstrated to be a reliable, valid, and responsive measure with a Cronbach α of .77 to .91 [126]. Cognitive difficulties will be assessed using the Functional Assessment of Cancer Therapy–Cognitive Function (FACT-Cog), version 3, a 37-item questionnaire with 4 subscales (perceived cognitive impairments, impact on quality of life, comments from others, and perceived cognitive abilities). The FACT-Cog is a reliable instrument validated in cancer populations with a Cronbach α of .89 [127]. The FACT-Cog instrument will only be administered to participants who reply “YES” to the following question at baseline: “Are you experiencing difficulties with memory, concentration, or other cognitive abilities?” As anxiety is a key determinant of quality of life in cancer survivors [23-26], the Patient-Reported Outcomes Measurement Information System–Global Health (PROMIS–Global Health) will also be administered. This validated measure contains 2 domains related to quality of life: mental health (Cronbach α=.86) and physical health (Cronbach α=.81) [128]. These measures will be administered at baseline and at weeks 4, 8, 16, and 26.

The Mao Expectancy of Treatment Effects (METE) is a 4-item instrument originally developed as the Acupuncture Expectancy Scale to measure outcome expectancy. It has demonstrated validity and reliability (Cronbach α=.82) and is correlated with patient self-reported efficacy and satisfaction [129-131]. Prior research has also demonstrated that outcome expectancy is associated with treatment response [129,132,133] and can, therefore, help inform personalized treatment decisions. The score ranges from 4 to 20, with higher scores indicating greater expectancy of benefit. This measure has been adapted for the MELODY study to assess patient expectancy of MT and CBT. The METE will be administered at baseline before receiving treatment. It will also be administered at week 8 to explore whether the expectancy changes after the treatment process.

As participants may have different preferences toward mental health treatment based on health beliefs or prior experiences with interventions, they will be asked at baseline whether they prefer MT or CBT or none [134]. Their preferences will also be assessed at week 8 to explore whether preferences change after their experiences with either MT or CBT.

Prior research has demonstrated variability in how people derive reward and pleasure from music, which may affect their experience with MT [135]. Therefore, the Barcelona Music Reward Questionnaire will be administered at baseline to assess the level of reward associated with music. This instrument has been validated with a Cronbach α of .92 [135].

In some groups and cultures, there is stigma associated with mental health needs, which may affect treatment experiences [48,49]. Therefore, the Stigma Scale for Receiving Psychological Help will be administered at baseline. This instrument has been validated (Cronbach α=.72) in different cultures to assess the stigma associated with receiving psychological health from mental health providers [136,137].

Sociodemographics (eg, age, sex, race or ethnicity, and education) and other relevant clinical characteristics (eg, cancer type, stage, treatment history, and time since cancer diagnosis) will be collected at baseline. The Generalized Anxiety Disorder 7-Item Scale (GAD-7) will also be administered as a diagnostic screen for generalized anxiety disorder at baseline [138]. The use of medications prescribed for anxiety (eg, antidepressants, anxiolytics, hypnotics, and sedatives) will be tracked using weekly medication diaries at weeks 0, 8, and 26.

It may be difficult to capture the impact of MT and CBT using quantitative measures alone. Methodological experts recommend mixed methods approaches to understand patient experiences with health interventions [139-141]. Thus, qualitative method specialists will conduct 45-minute, semistructured individual interviews over the phone at week 8 (end of treatment). To reach thematic saturation from interviews, the research team will purposively sample 30 survivors in each treatment arm to be interviewed (N=60), aiming for balance across age, sex, race, ethnicity, and treatment response. Interviews will be conducted shortly after the intervention period to enhance the recall of personal experiences during therapy sessions. Interviews will be conducted using a semistructured guide that covers the following topics: (1) acceptability: treatment satisfaction, usefulness of the intervention, barriers and facilitators to session attendance, and adherence to and engagement with between-session assignments; (2) impact on anxiety symptoms and coping: perceived benefits and harms of the intervention on anxiety and comorbid symptoms, how one copes with stress and anxiety, integration of learned skills into daily life, and plans for maintaining anxiety management skills; (3) digital literacy and experience: views toward technology, comfort level with telehealth, and transfer of acquired digital skills to other aspects of daily living; and (4) unexpected experiences: unanticipated benefits, harms, or insights from engaging with the interventions. The interview guide will include flexibility in asking probing questions to enable participants to elaborate on their experiences. All the interviews will be audio-recorded and transcribed for thematic analysis.

Analyses will be conducted using the intention-to-treat principle (ie, participants will be analyzed according to their randomly assigned treatment group regardless of dropout or treatment adherence status). Given that CBT is widely recognized as a first-line therapy for anxiety, the study team opted to test the noninferiority of MT to CBT, with the goal of understanding whether MT can be an appropriate treatment option for patients who lack CBT access or do not wish to pursue CBT. To provide complementary information for treatment decision-making, the study team will also test which intervention is superior for addressing fatigue and other symptoms co-occurring with anxiety. For all specific aims, the main analytical tool will be a linear mixed-effects model (LMM) because the primary outcome (anxiety) and secondary outcomes (fatigue, depression, insomnia, pain, cognitive dysfunction, and health-related quality of life) are repeated continuous outcomes over time [142]. This statistical procedure takes into account within-subject correlations from repeated measurements in the same subjects and allows the estimation of between-group differences without necessitating the exclusion of participants with missing data. The general template of each LMM will model the outcome as a function of the treatment arm and assessment time (categorical), controlling for the randomization stratification variables (baseline anxiety medication use, preferred language, and study site) and including a subject-specific random intercept. The general LMM template will be tailored to test the specific aim hypotheses by adding interaction terms (eg, time-by-intervention) and additional covariates of interest to the model and by reparametrizing the assessment time variable to focus on specific contrasts.

Outcome measure trajectories will be plotted by randomization arm over time. Each outcome measure will be summarized at each assessment time by treatment arm using descriptive statistics. Comparisons between randomization arms with respect to changes in symptom outcomes will be based on specific coefficients from time-by-arm interactions added to the general LMM template described above in the General Description section. Specifically, the model will include all time points (categorical) at which the outcome was assessed (baseline and weeks 4, 8, 16, and 26), and the time-by-arm interaction will include multiple coefficients corresponding to the time-by-arm interaction at each discrete postbaseline time point. However, statistical inferences will focus on the interaction coefficients at 8 and 26 weeks, which are interpreted as the differences between arms in change from baseline to 8 and 26 weeks, respectively. For hypothesis 1(a), the 2 primary end points are HADS anxiety score changes at 8 weeks and 26 weeks. On the basis of the LMM coefficients for the time-by-arm interaction at weeks 8 and 26, 2 tests will be performed to determine whether MT is noninferior to CBT at reducing the HADS anxiety subscale score within a noninferiority margin of 0.35 SD for both comparisons. The noninferiority of MT to CBT will be tested at a significance threshold of P<.025 for each comparison, controlling the overall type I error at P=.05 for the primary end point comparisons. For hypothesis 1(b), a similar LMM evaluating BFI scores over time will be used to test whether the MT arm had significantly greater improvement in fatigue (BFI score) compared with the CBT arm at weeks 8 and 26. In contrast to the primary end point noninferiority comparisons, these tests will be superiority tests. For other comorbid symptoms (eg, depression and insomnia) and quality of life, the methods described for hypothesis 1(b) will be used to compare the treatment arms. Given that a small number of cancer survivors may experience disease recurrence during the 26-week study, sensitivity analyses will be performed, excluding those individuals who experienced a recurrence during the study period. These sensitivity analyses will not replace the primary analysis.

An essential part of patient-centered care is recognizing that not all patients will respond to treatments the same way. As such, exploratory, hypothesis-generating heterogeneity of treatment effect (HTE) analyses will be conducted to identify patient-level factors associated with treatment response to either MT or CBT by incorporating 6 relevant variables (sex, race, ethnicity, education, outcome expectancy, and time since cancer diagnosis) and variable-by-intervention interaction terms in the mixed-effects model described above in the Aim 1—Hypotheses 1(a) and 1(b) section. These 5 exploratory LMM-based analyses will be guarded against inflated type I error owing to multiple testing by adjusting the variable-by-intervention interaction P values for the false discovery rate [143,144]. This type of subgroup analysis, although patient centered, may need to be interpreted with caution and cannot replace the primary analyses. The evaluation and reporting of HTE analyses will be based on the approach proposed by Kent et al [145]. However, recognizing other factors may also contribute to treatment responses, the study team will also apply Bayesian [146,147] and machine learning [148,149] methods, which can identify HTE and subgroups based on multiple variables simultaneously and are potentially more powerful than traditional univariate methods.

All interviews will be transcribed verbatim and imported into MAXQDA 2022 (VERBI Software) for analysis. The semistructured interviews will be analyzed using theoretical thematic analysis procedures as per the guidelines by Braun and Clarke [150,151]. Such an analysis is aimed at identifying patterns driven by an a priori theoretical framework or research questions. Themes will be identified using a semantic approach in which themes are derived from “the explicit meaning of the data and the analyst is not looking for anything beyond what a participant has said” [150]. Two coders will code the interview transcripts. To ensure the trustworthiness of the analysis, the primary coder will code each transcript, whereas the secondary coder will code 30% of the transcripts. Coding discrepancies will be resolved through discussion until consensus is reached. The codes will then be organized into categories, reviewed, and subsequentially organized into broader themes. Interviews from each treatment arm will be analyzed separately and then compared for convergence and divergence. The themes will be linked to patient-reported outcomes to explore whether specific themes vary according to the magnitude of change in the HADS anxiety score. To augment the HTE analyses, the study team will also explore whether qualitative themes differed by subgroup (eg, sex, race, ethnicity, education, outcome expectancy, and time since cancer diagnosis) within each treatment group.

As the only certain way to avoid biases from missing data is to collect complete data [152], the study team will strive to minimize missing observations by using study procedures that reduce participant burden. Participants who experience difficulties completing assessments on the web will have the option to complete assessments on paper or over the telephone with the research staff. Participants with time constraints or other barriers to completing outcome assessments will be given the option to complete only the primary outcome, that is, the HADS scale. Participants who withdraw from treatments will be encouraged to continue completing the outcome assessments. Reasons for withdrawal will be recorded for those who completely withdraw from the study. Given that missing data are inevitable in a prospective study, sensitivity analyses will be performed to assess the impact on the results of adjusting for patient disease progression or death, and analytical strategies that are as robust as possible to data losses will be applied. The study team will first explore whether missingness is associated with observed variables (eg, randomization arm and baseline outcome measures) by comparing patients with complete and incomplete data. Notably, the LMMs described previously will include patients with incomplete data under the missing at random assumption. However, the exploration of the data may deem the missing at random assumption to be inappropriate. Multiple imputation and pattern mixture models are well-established methods for addressing these issues [153,154]. Sensitivity analyses will be performed to evaluate the robustness of the LMM results by refitting the models after imputing the missing week 8 and week 26 outcomes using multiple imputation.

The sample size is based on the primary hypothesis that MT is noninferior to CBT for anxiety reduction at weeks 8 and 26. Two tests will be performed to determine whether MT is noninferior to CBT at reducing HADS anxiety at weeks 8 and 26, each with a noninferiority margin of Cohen d=0.35 SD for both comparisons. These comparisons will be based on the LMM coefficients for the time-by-arm interaction at weeks 8 and 26; however, the power calculation is based on 2-sample t tests of differences between the arms in their change scores. Therefore, the power estimate is conservative, that is, in the LMM-based analysis, there will be slightly higher power than the power calculation presented below, with all other assumptions held constant. Assuming 150 participants are randomized to each arm, 15% attrition, and a 1-sided significance threshold of P=.025 (controlling the overall type I error at 0.05 for our 2 primary end point comparisons), there will be 80% power to find MT noninferior to CBT with respect to the HADS anxiety subscale scores at weeks 8 and 26 within a noninferiority margin of Cohen d=0.35 SD.

The noninferiority margin was informed by a large study that used multiple rigorous methods to empirically estimate a difference of 1.7 points as the MCID for the HADS scales [120]. In preliminary data from CBT recipients with baseline HADS anxiety subscale scores ≥8, the SD for the HADS anxiety subscale score was 4.2 at week 8 [110]. In addition, in a large acupuncture trial of cancer survivors, among 159 patients with baseline HADS anxiety subscale scores ≥8, the HADS anxiety subscale score SD was 4.0 during follow-up [155,156]. On the basis of these studies, it is assumed that the HADS anxiety subscale score SD will be approximately 4.2. Dividing the raw-score MCID of 1.7 points by 4.2 yields a standardized difference, that is, Cohen d=0.40. The noninferiority margin of Cohen d=0.35 SD is smaller than the standardized MCID (Cohen d=0.40) of the HADS anxiety subscale; as such, the noninferiority margin and statistical approach will find MT to be noninferior to CBT only if MT is not meaningfully worse than CBT.

Regarding the sample size for the qualitative analyses, the number of participants required to draw meaningful conclusions from semistructured interviews is determined by saturation, that is, the point at which existing themes are fully understood and no new themes emerge through further data collection [157]. Prior research has indicated that thematic saturation can typically be obtained with <15 interview participants, although these findings were based on relatively homogeneous samples [158]. Given that this study is enrolling a more heterogeneous population, the plan is to conduct 30 interviews in each treatment arm to better capture the diverse perspectives.