Observing Music Therapy in Dementia: Repeated Single-case Studies Assessing Well-being and Sociable Interaction

ABSTRACT Objectives This study compared behavioral expressions of momentary well-being and sociable behavior toward significant others during music therapy and regular social interaction. Methods A 10-week active music therapy intervention was provided for people living with dementia and family caregivers. A bi-phasic AB single-case design was replicated for three sessions per dyad and coded using the Observable Well-being in Living with Dementia-Scale (OWLS) and the Verbal and Nonverbal Sociable Interaction Scale-Care Receiver (VNVIS-CR). Effect sizes (Log Response Ratio) were calculated for each session and analyzed with robust cluster meta-analysis. Results Eleven dyads were included, and 32 sessions analyzed (2102 observations). Within sessions we found a 48% increase in well-being, and a 32% increase in sociable interaction during music therapy. Heterogeneity was high. Dementia severity predicted an increase in nonverbal sociable interaction (93% for moderate dementia). Depression and time did not predict any change. Conclusion The potential of music therapy to increase well-being and sociable interactions toward significant others calls for further investigation of heterogeneity and covariates. Single-case designs are demonstrated to be feasible for these investigations. Clinical implications Preference-based music therapy may alleviate some of the individual and relational consequences of living with dementia, facilitating positive emotions and connection to significant others.


Introduction
An increasing attention toward living well with dementia is present in dementia research and care (Dröes et al., 2017), public action plans, and guidelines (i.e. National Institute for Healt and Care Excellence, 2018; World Health Organization [WHO], 2017). Key targets for psychosocial interventions emphasized by home-dwelling people living with dementia are coping with psychological distress following the diagnosis, maintaining normality and identity, participating in meaningful and enjoyable activities, and having good social relationships (Miranda-Castillo, Woods, & Orrell, 2013;Øksnebjerg et al., 2018;Reilly et al., 2020;von Kutzleben, Schmid, Halek, Holle, & Bartholomeyczik, 2012).
Personalized and active music interventions may be beneficial for several of these objectives for multiple reasons (Brancatisano, Baird, & Thompson, 2020). Music is a strong trigger of positive emotions (e. g., Juslin, 2013), and familiar music may trigger pleasant memories and maintain a sense of identity and coherence in the individual living with dementia (Baird & Thompson, 2018;Särkämö, 2018). Additionally, music is an engaging and inherently social activity (Brancatisano et al., 2020). Furthermore, active music interventions may positively affect cognition (Fusar-Poli, Bieleninik, Brondino, Chen, & Gold, 2018), and meet current recommendations of individualizing interventions for this group (Dowson, McDermott, & Schneider, 2019;WHO, 2017).
Family caregivers report that some of the most distressing aspects of caregiving are the disruptive behaviors commonly described as the behavioral and psychological symptoms of dementia or BPSD (Cheng, 2017;Kales, Gitlin, & Lyketsos, 2015). Most people living with dementia experience these symptoms, with apathy as the most frequent (Livingston et al., 2017). While BPSD are driven by a diversity of causes clustered in the person, caregiver and/or environment (Kales et al., 2015), recent metaanalyses conclude that music therapy may decrease BPSD symptoms (Abraha et al., 2017;van der Steen et al., 2018). Music therapy may be beneficial for both caregivers and care receivers. Thus, including dyads in music therapy interventions seems both clinically relevant and feasible.
To our knowledge, only one randomized controlled trial has included family caregivers in music interventions. Särkämö et al. (2014) found that group singing interventions improved mood in home-dwelling people living with dementia, and decreased family caregiver burden. Other small-scale studies have shown musical interventions to positively affect the relationship of the dyads in qualitative (Baker, Grocke, & Pachana, 2012;Camic, Williams, & Meeten, 2013;Clark, Tamplin, & Baker, 2018;Davidson & Fedele, 2011), as well as quantitative measures (Clair, 2002).
Changes in social behavior is common in dementia (Livingston et al., 2017), and it is reasonable to expect interventions targeting social behavior to be of value to the caregiver and care-receiver dyad. Documentation of the effect of music therapy on sociable behavior in dementia is sparse (van der Steen et al., 2018), but single-case studies have shown increased communication behavior (Schall, Haberstroh, & Pantel, 2015), and mutual engagement (Clair, 2002).
Most clinical studies assessing music therapy administer scales before and after the intervention period (van der Steen et al., 2018). However, the degenerative nature of dementia may mask clinically relevant changes over shorter time periods and does not necessarily reflect a lack of treatment effects. To assess clinically and socially meaningful outcomes for people living with dementia, increasing the sensitivity of measurement instruments and research designs are recommended (Cho, 2018;Dowson et al., 2019;Schall et al., 2015). Singlecase designs may be a feasible approach for smaller samples and are underused in dementia research (Steingrimsdottir & Arntzen, 2015). These designs have a high ecological validity and provide explanatory power when the person serves as their control, the intervention is replicated, and threats to internal and external validity are managed (Manolov, Gast, Perdices, & Evans, 2014).

Aims
Thus, the goal of this project was to investigate the effect of individually tailored music therapy for home-dwelling people living with dementia involving close family caregivers as a collateral therapist. The primary focus was the potential effect on well-being in the care receiver, and sociable behavior toward the caregiver. Additionally, we aimed to demonstrate the utility of a single-case design using fine-grained outcome measures.
Our primary hypothesis was that individually tailored music therapy would increase observed within-person expressions of momentary wellbeing and sociable behavior when compared to a baseline of regular social interaction, within and across sessions. Our secondary hypotheses were positive changes in self-rated emotions when comparing pre-and post-session measures, and stable neuropsychiatric symptoms, increased selfreported long-term well-being, and decreased caregiver burden from pre-to post-intervention.

Ethics
The study protocol was pre-registered (www. clinicaltrials.gov, ID: NCT03011723), and approved by the Regional Committees for Medical and Health Research Ethics in Norway (2016/1374). People eligible for inclusion were invited to participate during a regular appointment at the outpatient clinic and provided with written information about the project. Participants were given time to discuss with a person they trusted before actively registering their interest. Next, an appointment was scheduled in the participants' home, where information about the project was given in person. Informed consent was signed immediately after informing them about the project, facilitating the ability to remember the details about the project and actively deciding if they wanted to participate. Thus, all participants could provide informed consent. Both researchers and music therapists were attentive to signs of withdrawn consent during the study, and repetition of information about the research project was provided when needed (cfr. Dewing, 2007).

Intervention
The active music intervention aimed for two weekly sessions over 10 weeks. Each week, a professional music therapist came to the participant's home for the first weekly session with the dyad. This intervention followed a manual of resource-oriented principles for music therapy (Rolvsjord, Gold, & Stige, 2005). The music therapist guided the dyads in choosing musical activities for the second weekly session, initiated by the collateral therapist. The main principles and content of the music therapy is summarized in Table 1.

Procedures
The study design is presented in Figure 1. First, the personal musical history of the person with dementia and the shared musical history of the dyad were mapped. A treatment plan including personal goals was made in collaboration with the dyad.
Before and after the 10-week intervention period, long-term well-being, caregiver burden, and neuropsychiatric symptoms were assessed. For each music therapy and collateral led-session, selfreported emotional state before and after the session was measured. Music therapists and collaterals logged the activities in each session. The collaterals rate their session as "negative," "neutral" or "positive," to detect adverse effects.
The first, fifth, and tenth music therapist-led sessions were video recorded. Music therapists were instructed to ensure at least five minutes of regular social interaction between the dyad and music therapist, constituting a baseline-phase (A). The following 30 minutes of music therapy with the dyad constituted the intervention-phaseFor each 30 second interval, we dichotomously scored the presence or absence of each behavior (described under "Measures"), using the software Noldus Observer XT 12.5© (Noldus Information Technology, 2015). Three coders including the first author (KGM) were trained until reaching at least 80% inter-rater agreement (Ledford & Gast, 2009). Then, the first 20% of video recordings were coded by main-coder (KGM) and research assistants to assess inter-rater reliability. Feedback to prevent observer drift was given. The assessors were independent of the therapists. No blinding procedure was feasible for the participants, therapists, or coders.

Primary measures 1
The Verbal and Nonverbal Interaction Scale -Care Recipient (VNVIS-CR; Williams, Newman, & Hammar, 2017), is a dementia-specific observational instrument consisting of 26 operationalizations of sociable and unsociable verbal and nonverbal behavior toward a close caregiver. A ratio consisting of the number of sociable items divided by the number of unsociable items is calculated per time point (range 0-13). A higher score indicates more sociable behavior. Two subscales exist, sociable-nonverbal and sociable-verbal interaction. VNVIS-CR is evaluated to have good inter-rater reliability, test-retest reliability, and construct validity (Williams et al., 2017). As VNVIS-CR was not available in Norwegian, formal translation/back-translation was approved by C. L. Williams.
The Observable Well-being in Living with Dementia-Scale (OWLS; 2 Madsø, Pachana & Nordhus; manuscript under review) was developed for this study. The development was based on methodological recommendations (Bakeman & Quera, 2012), including pilot-testing, building on the theory of well-being in dementia, and examining existing observational scales assessing well-being (Algar, Woods, & Windle, 2016;Kitwood, 1997). Finally, the content validity of the scale was established via focus-groupdiscussions with relevant experts. Construct validity (Terwee et al., 2007), was supported by moderate to strong correlations with total-score and nonverbal score of VNVIS-CR. Responsiveness (Terwee et al., 2007), was supported by moderate correlations between effect sizes and change scores of the Neuropsychiatric Inventory-Questionnaire (NPI-Q, Kaufer et al., 2000), and The abovementioned activities were often followed by conversations about music and/or memories coming to mind a These principles originate from the manual of Rolvsjord et al. (2005) the self-reported happiness-subscale of the Visual Analog Mood Scale (Stern, Arruda, Hooper, Wolfner, & Morey, 1997). OWLS consists of 10 items ("initiative/response," "attention," "happiness," "enjoyment," "joking," "mastery," "selfconfidence," "express identity," "positive feedback," and "relationship"). Two items may need further elaboration. "Express identity" refers to positive initiatives and responses during activity related to personal history or self-perception. "Relationship" refers to initiating turn-taking interactions to achieve closeness with significant others. All operationalizations include both verbal and nonverbal indicators of well-being (range 0-10). Higher scores indicate higher well-being.
Scores <2 indicate lack of attention and response toward the activity in the current observations. Items of the scales are presented in Table S2.

Single-case analysis for primary measures
For each case, three bi-phasic intra-subject ABreplications were conducted (Tate et al., 2016), to investigate changes in the level of the primary measures. The baseline phase consisted of 10 ratings during regular social interaction, and the intervention phase consisted of 56 ratings during music therapy. The intra-subject replication of the AB-design gave three distinct attempts to investigate the intervention effect, followed by inter-subject replications in 11 cases (Manolov et al., 2014;Tate et al., 2016). No randomization procedure was applied.
The sum of the OWLS-items and the ratio of VNVIS-CR were plotted in R (R Core Team, 2020;Wickham, 2016) and visually inspected (Ledford & Gast, 2009;Tarlow, 2016). As there is no consensus-based method for visual, nonoverlap, or statistical approaches when conducting single-case analyses, we followed the recommendations of Parker, Vannest, and Davis (2011) to visually investigate if there is a) a baseline trend present and b) if there is a strong trendimprovement in the intervention-phase. Different approaches incorporate and control for different parameters, and some single-case effect size estimates are sensitive to study designs such as length of baseline and intervention phase, length of session and type of recording system (Pustejovsky, 2019).
Variability and autocorrelation in time-series are common, and may complicate visual analysis, leading to both over-and underestimation of treatment effects (Parker et al., 2011;Vannest & Ninci, 2015). In these situations, parametric approaches may be helpful. In addition, a baseline trend in the direction of the hypothesized treatment effect may oppose a threat to the internal validity, but correcting for insignificant or random baseline-trends may overcorrect data and mask treatment effects (Tarlow, 2016). To account for the observed characteristics in our data, we chose the parametric approach Log Response Ratio (LRR, Pustejovsky, 2015Pustejovsky, , 2018. LRR is a promising parametric scale-free approach for calculating single-case effect sizes. The magnitude of the LRR is not sensitive to the sample size and length of the observations, as are several other effect sizes (Pustejovsky, 2019).
Single LRR-estimates do not account for autocorrelation, and variance-estimates may be biased (Pustejovsky, 2015). However, meta-analyses of these effect sizes using robust variance estimation will correct for this (Pustejovsky, 2018). Thus, a meta-analysis of the LRR effect sizes for each individual music therapy session was conducted. Due to the dependency between the effect sizes nested within each case, robust meta-regression is the recommended option (Hedges, 2019). Robust cluster variance estimation with small-sample correction was conducted, with accompanying sensitivity analyses and investigation of heterogeneity with meta-regression (Fisher & Tipton, 2017;Tipton & Pustejovsky, 2015).
Interpreting single-case effect sizes should be based on contextual understanding, and benchmarks may vary (Vannest & Ninci, 2015). The LRR ranges from −1 to +1 but may be converted to "percentage of change," making interpretation straightforward. Pustejovsky (2018) recommends a context-specific interpretation of LRR. In this specific context and based on our knowledge of the specific outcome measures, we interpret a change of 20-50% to be small, 50-70% to be moderate, and >70% to be large.
Finally, the relative frequencies of the OWLS items and VNVIS-CR nonverbal and verbal interaction items in the baseline and the interventionphase were explored to investigate the behavioral content of the two phases (total number of occurrences divided by total number of observations).

Statistical analysis for secondary measures
The secondary measures were analyzed with Wilcoxon signed-rank test for paired samples with continuity correction (R Core Team, 2020). Missing data from pre-to post-sessions were omitted, and missing data from pre-to post-intervention were imputed as no change.

Results
A total of 13 participant-dyads were recruited. Two dropped out after one session (withdrawn consent and admission to hospital), and one after 6 sessions (psychotic symptoms). The latter participant's ("Kate") available data was included in the analysis. Of the 11 participants, nine were diagnosed with Alzheimer's dementia and two with Vascular Dementia. They were aged from 71 to 88 years (M =79.82, SD = 5.27), and 63% were women. Clinical dementia stage was very mild for one, mild for five, and moderately severe for five. All participants experienced at least two neuropsychiatric symptoms at inclusion, with symptoms of depression, apathy, and anxiety as the most common. Demographics and clinical characteristics of the 11 participant dyads are presented in Table S1.

Treatment fidelity
Logs showed musical elements were the main ingredient in all the music therapy sessions (range of duration 23-70 minutes, M = 46.2, SD = 8.85). All participants received 10 music therapy sessions except "Clare" (8 sessions), and "Kate" (6 sessions).
The number of sessions led by the collateral therapist ranged from 0 to 8, with six participants reporting ≥5, and five participants reporting ≤4 sessions (range of duration 10-70 minutes, M = 52, SD = 28.3). Adverse effects were rare. Collateral-sessions were rated in the logs as positive for 86%, neutral for 4%, negative for 4% sessions (6% were not rated).
Plots of all observational data for each singlecase are provided in the supplementary material, with an example displayed in Figure 2. Thirtytwo sessions were included (N baseline = 320 observations, N intervention = 1782 observations). Table 2 displays the LRR effect size estimates per session. Phase-specific item-frequency is presented in Table S2.

Investigating validity
An increasing baseline trend was present in one session for well-being ("Ann," Session 1), and three sessions for sociable interaction ("Beth," session 1, "Helen," session 1, and "Kate," session 1). A consistent pattern of returning to baseline between sessions was detected through visual inspection for 30 of 32 observations of well-being (see Supplementary figures. Exceptions were "Greg," Session 10 and "Iris," Session 10). This indicates more experimental control and an increase in explanatory power of these results. For sociable interactions, no such consistent pattern was present.

Meta-analysis
Robust cluster variance estimation meta-analysis with small sample correction showed that the wellbeing (OWLS) increased by 48% in the music therapy compared to regular social interaction (LRR =0.39, 95% CI = [0.28-0.51], SE = 0.05, t (9.88) = 7.49, p = <0.001***, I 2 = 86.03, τ 2 = 0.04). Furthermore, as the different observations were nested within cases and not independent, sensitivity analyses were conducted to investigate the stability of the results. When testing for different rho's (correlations), the results were stable for different dependencies within the clusters.

Heterogeneity and meta-regression
A high degree of heterogeneity was present in the meta-analysis of both measures. τ 2 describes the underlying variance between sessions, and τ is expressed in the same metric as the effect size (LRR). I 2 is a measure of the percentage of variability in the effect sizes across the sessions that is attributed to heterogeneity rather than sampling error (Fisher & Tipton, 2017) and values over 75% are interpreted as large (Higgins, Thompson, Deeks, & Altman, 2003).
Pre-scores of "depression" from the NPI-Q (Kaufer et al., 2000; dichotomized to 0 = "no or mild depression," 1 = "moderate or severe depression") did not predict the effects of music therapy. Conducting meta-regression with time as a factor does not support our hypothesis of accumulated increases in well-being and sociable interaction over time. Table S2 shows the relative frequencies of all the items from the observational scales in the different phases. The largest changes for wellbeing from baseline to intervention on grouplevel were found in the items "enjoyment" (45.9%), "happiness" (23.2%), "express identity" (36.5%), "relationship" (34.3%), and "positive feedback" (14.1%).
The frequency of the remaining items of both scales changed <10%.  Abbreviations and interpretation: LRR = Log Response Ratio calculated with expected increase in values. SE = Standard Error. LRR is converted to % of change, and study-specific benchmarks for small effect is >20-49% change, medium effect is 50-70% change and large effect is >70% change. Sessions where increasing baseline-trend is detected is marked with i Results from secondary measurements are reported in Table 3. In the music therapist-led sessions, positive self-reported emotions significantly increased from pre-to post-music therapy sessions. Negative self-reported emotions significantly decreased, but a floor effect makes the clinical relevance of the lowered negative emotions uncertain. From the pre-to post intervention period, we did not find support for our hypothesized change in self-reported long-term well-being or caregiver burden. However, the hypothesized stable neuropsychiatric symptoms did significantly decrease from before to after the intervention period.

Discussion
This study investigated individually tailored music therapy compared to regular social interaction for home-dwelling people living with dementia. Within sessions, we found close to a moderate (48%) increase in observed wellbeing. A significant increase in the self-rated positive emotions from pre-to post-sessions in our secondary measures supports these observations. A small (32%) increase in sociable behavior toward significant others was found within sessions, where people with moderate dementia severity showed a large (93%) increase in nonverbal sociable interaction during music therapy. We did not see an increase in long-term wellbeing or lowered caregiver burden at the end of treatment. Nevertheless, a significant withinperson decrease in neuropsychiatric symptoms from pre-to post-treatment suggests some positive long-term effects.
The increase in momentary well-being is in line with recent meta-analyses of music therapy (van der Steen et al., 2018;Zhang et al., 2017). Randomized controlled trials of musical interventions have shown inconsistent results, with some studies finding increased well-being (Cho, 2018;Hsu et al., 2015;Särkämö et al., 2014) and others not reaching statistical significance (Raglio et al., 2015;Ridder et al., 2013) or not finding an effect (Cooke et al., 2010). Studies using narrower timeframes with measures focusing on momentary changes (Cho, 2018;Hsu et al., 2015), as well as utilizing single-case designs and repeated finegrained measures (Clair, 2002;Schall et al., 2015) seem to optimize the evaluation of well-being.
When exploring the behavioral content of OWLS in the different phases, we saw a large increase in expressions of enjoyment and happiness during music therapy. This is of clinical relevance, as several participants expressed "experiencing more happiness" as the most important personalized goal in therapy. Furthermore, we observed a large increase in positive expressions of identity, an objective identified as valuable by homedwelling people living with dementia (Reilly et al., 2020;von Kutzleben et al., 2012). While dementia may lead to loss of role-functions and disrupt one's self-concept, personalized music may serve to maintain and support a feeling of identity (Baird & Thompson, 2018;McDermott, Orrell, & Ridder, 2014). Familiar music triggering valued autobiographical memories may contribute to a more positive self-perception (Brancatisano et al., 2020). People living with dementia have expressed engaging in social participation as an important goal in interventions (Øksnebjerg et al., 2018). We found a large increase in joint interaction and turn-taking during music therapy ("relationship"), resembling the presence of mutual engagement found in the single-case study of Clair (2002). This underlines the potential for music therapy to facilitate social interaction and connection (Brancatisano et al., 2020).
In the current approach, we emphasize sociable behavioral expressions. Meta-analyses have identified a small effect of music therapy in people living with dementia on behavioral disturbances, with measures solely focusing on non-sociable behaviors such as agitation (Abraha et al., 2017;van der Steen et al., 2018). When investigating the relative frequencies of the items in VNVIS-CR, we found increases mainly in the sociable as opposed to the non-sociable behaviors. This suggests assessing sociable behavioral expressions is important for detecting behavioral effects from music therapy (Dowson et al., 2019;van der Steen et al., 2018).
Experiencing connections is identified as important for people living with dementia (McDermott et al., 2014;Reilly et al., 2020). At the group level, we found a large decrease in apathy ("appears aloof"). These changes illustrate how using individually tailored music may enable social capacities in people living with dementia (Brancatisano et al., 2020), potentially enhancing dyadic relationships. This is clinically important, as lack of mutuality is associated with caregiver burden (Cheng, 2017). While caregiver burden was stable from pre-to post-intervention, inspecting the logs in the study showed overall positive experiences of collateral-led sessions.
The slight inverse effect (−12%) of verbal sociable interaction may be explained by the low verbal activity in the music intervention compared to the higher verbal activity in the baseline phase, as the difference mainly concerned responding to questions and using coherent and relevant verbal communication.

Heterogeneity
The high degree of unexplained heterogeneity indicates that the effect forwell-being and sociable interactions varied to a large extent from session to session, both within and across the participants. Except for dementia severity, no other significant predictors were found. However, with increasing dementia severity, the potential for change may be greater as the social impairment is more prominent (Livingston et al., 2017). It seems evident that as dementia progresses, more focus on nonverbal positive expressions are important.
Symptoms of BPSD are known to fluctuate in people living with dementia (Kales et al., 2015) and may have indirectly contributed to the varying levels of well-being and sociable interaction.

Strengths and limitations
This single-case design intervention was conducted in a natural setting with optimal ecological validity and a meaningful comparison through using the subjects as their own controls. Offering individualized and preference-based music as opposed to pre-selected musical alternatives should also be considered a strength of this study. The lack of blinding introduces a risk of bias but is difficult to apply in single-case designs (Manolov et al., 2014). Meanwhile, the assessors were independent of the music therapists. Additionally, the intra-subject replication increases internal validity, and the intersubject replication and meta-analytic approach together increase the external validity (Manolov et al., 2014). Replications in other research sites with other researchers will increase generalizability (Kratochwill et al., 2012). The observational measures (Williams et al., 2017;Madsø et al., manuscript under review) need further validation in different studies and contexts.
Although causal inferences are only possible in strictly controlled experimental designs, one may still imply that the intervention is the most plausible explanation of the changes from baseline to intervention in a single-case design (Kratochwill et al., 2012). A systematic return to baseline in the observations of well-being between sessions increases the likelihood of experimental control of these results. Other competing explanations for the change from baseline to intervention are monotonic trends, maturation, or history (Ledford & Gast, 2009;Tarlow, 2016). Future single-case studies should use multiple-baseline or equivalent randomized designs for investigating sociable interaction (Manolov et al., 2014). Continuing the observations while adding a post-intervention phase during the same day (ABA) could provide an opportunity to investigate whether the increase in expressions of well-being and sociable interaction are short-lived. However, continuing the video observation could also exhaust the participants and terminating the observation in a baseline-phase challenge ethical standards (Ledford & Gast, 2009).
The baseline-phase was only of 5 minutes duration, and we recommend that future studies should increase the baseline length to at least 15-20 minutes in order to secure proper baseline stability (Ledford & Gast, 2009). Using a pre-defined length of all baseline-phases, instead of changing phase when stability or trend was established, is a weakness of our design (Ledford & Gast, 2009). The detection of a baseline trend would be a possible threat to internal validity, as stability in the baseline phase is important to validly compare phases (Manolov et al., 2014). Still, there was only an increasing baseline trend in one session when using OWLS and three sessions using VNVIS-CR, and thus, the baseline trend as such seems to be of little concern for the interpretation of the overall results in this meta-analysis.
The variability of the scores of the OWLS and VNVIS-CR within each session resembles the results of Schall et al. (2015) and suggests that the constructs we measured are naturally fluctuating in this population.
Even though we planned for the dyads to use music between sessions, their logs showed varying completion of this part of the intervention. A more structured approach toward the implementation of the caregiver-led music session may have helped, as conducted by Clair (2002) and Baker et al. (2012).

Conclusion
Individually tailored music therapy did positively impact short-term well-being in the care recipients. In addition, the potential of music therapy to increase sociable behavior toward caregivers warrants further investigation. The behaviors observed in the single-case design are evaluated as relevant for people living with dementia (Øksnebjerg et al., 2018;Reilly et al., 2020), following the advice to have a primary focus on the positive effects music therapy may offer (Dowson et al., 2019;van der Steen et al., 2018). The varying effect of music therapy from session to session calls for future studies to investigate sources of heterogeneity more closely. Here, single-case designs with high ecological validity may be a feasible approach. Observing and measuring covariates may hopefully increase the precision of the prediction of effects from music therapy for well-being and sociable interaction. Such covariates may include coding musical elements of the intervention, relevant symptoms, or elements in the communication from the caregiver or therapist.

Notes
1. Two primary measures were changed after pilot testing, and deviates from the pre-registered protocol. CODEM-instrument (Kuemmel, Haberstroh, & Pantel, 2014) was changed to Verbal and Nonverbal Interaction Scale -Care Recipient (VNVIS-CR, Williams et al., 2017). CODEM measures communication behavior, and the underlying construct resembles VNVIS-CR. Observed Emotion Rating Scale (OERS, Lawton, Van Haitsma, Perkinson, & Ruckdeschel, 1999) was replaced with Observable Well-being in Living with Dementia-Scale (OWLS, manuscript in preparation) the latter focusing solely on positive behavioral expressions 2. Contact corresponding author for details about the Observable Well-being in Living with Dementia-Scale (OWLS)

Clinical implications
• In-home music therapy has the potential to increase momentary well-being and sociable interactions for people living with dementia • Given the relational impact of dementia, including family caregivers in music therapy interventions may support the dyadic need for relation reciprocity