Factors Predicting Adherence to Artistic-Singing Groups for Older Adults and their Role as Moderators of the Intervention Outcomes

ABSTRACT

Objectives

Group singing (GS), as an art-based intervention, has demonstrated a wide range of biopsychosocial benefits in older adult participants. However, the factors that predict the adherence of older adults to these programs and that moderate the intervention outcomes were not yet studied, which is the aim of this study.

Methods

A randomized controlled trial was developed to test the efficacy of a GS intervention, from which pre-post intervention data was collected and analyzed. Participants: 149 retired older adults (M = 76.66, SD = 8,79 years old) users of a social care institution.

Results

Principal component analysis of responses to a pre-intervention assessment yielded 10 factors. General well-being (GWB), negative mood and loneliness, blood pressure, and the participants’ years of formal education predicted the number of sessions attended by the participants. GWB moderated the intervention’s outcomes on life satisfaction, social identification, and systemic inflammation.

Conclusions

Years of education, well-being, negative mood and loneliness, and blood pressure at baseline predicted participants’ adherence to a singing group artistic intervention.

Clinical Implications

For future artistic interventions with older adults, screening for participants’ characteristics such as formal education, health and well-being before the intervention is important as it allows predicting adherence and tailoring more adjusted and cost-effective interventions.

KEYWORDS:

• Adherence
• attrition
• artistic interventions
• group singing
• older adults
• well-being

Introduction

World population aging is driving much interest in designing and applying community-based interventions as resources available to prevent disease and promote autonomy and quality of life in older adults (Foster & Walker, 2015). Cognitive training, physical activity, and arts-based interventions are among the most common types of intervention. The latter had recently become a subject of systematic research assessing their benefits and the factors responsible for them (Fancourt & Finn, 2019). Low adherence and moderate attrition rates to arts-based interventions in general and group singing (GS) in particular, however, are a threat to the efficacy of these programs (Clifford et al., 2021; Williams et al., 2018). Nonetheless, there is no research on the factors that predict higher adherence and moderate the effects of arts-based interventions, which is essential for designing more effective programs.

Art-based interventions have shown positive effects on participants’ health (Castora-Binkley et al., 2010; Gordon-Nesbitt & Howarth, 2020). For example, GS is the basis for community-based and cost-effective interventions (Coulton et al., 2015) aiming to promote health and autonomy. Such interventions have shown positive effects on general and perceived health (Galinha, Fernandes et al., 2021; Galinha, Pinal et al., 2021). Furthermore, GS interventions may activate the immune system (Beck et al., 2000) and be related to protective effects on systemic inflammation levels (Galinha, Pinal et al., 2021). Also, for GS interventions that include respiratory and breathing techniques training, positive outcomes on respiratory musculature and capacity have been observed (Fu et al., 2018).

Similarly, GS and other artistic interventions have been associated with increased well-being and reduced depressive mood (Dingle et al., 2021; Noice et al., 2014). For instance, in a GS intervention with 34 sessions, which included breaks for social interaction, participants showed significantly higher levels of social well-being and positive affect and a trend toward better self-esteem than an active control group (Galinha, García‐Martín et al., 2021). Likewise, in other GS RCTs, participants showed reduced depressive symptomatology, anxiety, and negative affect (Mathew et al., 2017), reductions in loneliness and increased interest in life (Johnson et al., 2020), and improved quality of life (Johnson et al., 2013).

Similarly, for GS programs in which participants should learn the songs’ lyrics, positive effects were observed in general cognitive function (Pentikäinen et al., 2021), as well as on memory (Fu et al., 2018; Galinha, Pinal et al., 2021) or executive function of older adults (Mansens et al., 2018).

Despite the potential for beneficial effects of artistic interventions, there is a potential bias in the reported results. Most studies observe moderate rates of attrition and drop-outs in these interventions for older adults. Attrition may hinder the efficacy of interventions and the validity of the research results depending on its extension and distribution between experimental and control groups (Clift, 2013; Dingle et al., 2019). Moreover, factors related to attrition or adherence to artistic interventions have not been sufficiently studied, being unknown who are the older adults that will most probably abandon the interventions and why.

Studies on attrition and adherence determinants, however, already exist for older adults’ interventions based on formative actions and physical and/or cognitive training (e.g., Double & Birney, 2016; Viken et al., 2019). Regarding cognitive and formative interventions, confidence in one’s cognitive capacity and positive attitudes toward cognitive activities were positive predictors of adherence (Double & Birney, 2016). In contrast, negative affect (Howells et al., 2016) or depression (Shaw et al., 1994) were negative predictors of adherence to mindfulness training or health education for older adults with osteoarthritis, respectively. Depression has also been reported as a negative predictor of adherence to physical exercise, as well as demographic factors like participants’ low educational level, and socioeconomic and memory status (Collado-Mateo et al., 2021; Rivera-Torres et al., 2019; Viken et al., 2019). However, perceived physical health and social support were positive predictors (Collado-Mateo et al., 2021; Schmidt et al., 2000).

Consequently, given the lack of studies investigating the factors predicting adherence to art-based interventions, this study aimed to (a) identify the factors related to adherence to a GS intervention and (b) explore whether those factors moderate the effects of the program on different outcomes that have been previously shown beneficial effects of participation in GS programs. Research on this topic will help to understand what leads older adults to drop out of these interventions and tailor more adjusted interventions for older adults, improving their cost-effectiveness.

Methods

Intervention & sample

The Sing4Health RCT (trial pre-registration number: NCT03985917) intervention consisted of 34 2-hour sessions, twice a week that comprised: i) relaxation exercises and vocal warm-up; ii) vocal technique training; iii) repertoire rehearsals; iv) group dynamics and 20-minute breaks for socialization; v) creation of a show; and vi) assessment of each participant’s performance (for details, see, Galinha et al., 2020). All RCT protocols followed the Declaration of Helsinki principles and were approved by the Ethics Committee of the Psychology Research Center of the Autónoma University of Lisbon (Approval Nr. 12-09-2018). All participants provided written informed consent.

An a priori power analysis determined a target sample size of 140 individuals to observe large effect size (ηp² = 0.14) differences in repeated measures ANOVA with an alpha level of .05% and 90% statistical power. Participants were over 60 years of age, retired and users of social care institutions (i.e., daycare centers, nursing homes or home support). The exclusion criteria were the presence of severe sensory or motor impairments and having participated in structured GS interventions in the previous 4 months. Participants were randomly assigned to an intervention group (IG, n = 75) that started the intervention immediately or an active waiting list group (WLG, n = 74) that initiated the intervention 6 months later. The WLG was randomly selected from the same social care institution users that accepted the invitation to enroll in the GS program, however they waited 6 months to be able to participate in the program. In the meanwhile, they benefited from the usual leisure and artistic activities offered by the daycare centers, except singing group activities. Statistical comparisons between the IG and WLG at baseline showed that they were homogeneous in the measured variables (see, Galinha, Fernandes et al., 2021; Galinha, García‐Martín et al., 2021; Galinha, Pinal et al., 2021). Note that in the present study, no between groups comparisons are made and that the 149 participants were considered for analyses regardless of their group. Most of these participants were females above 70 years of age, with four or fewer years of formal education and low household monthly income. Most participants (77.9%) reported one or more diagnosed medical conditions and yielded low cognitive function scores measured by the Montreal Cognitive Assessment (Nasreddine et al., 2005).

Materials and procedure

A battery of psycho-emotional, psycho-social, and cognitive tests, along with the collection of biological samples and socio-demographic data, was performed before and after the intervention. Table 1 presents a summary of the collected measures, and more details on the materials can be found in the supplementary material at https://osf.io/2wn39/?view_only=0fea361ce5f944aaac4ccb8daf2912a8.

Table 1. Tests, questionnaires and biological and physical parameters assessed in the Sing4Health intervention protocol at each testing point. Scores obtained for these variables at pretest comprised the data matrix for the principal component analysis.

Satisfaction with life scale	WAIS digit-symbol subtest correct responses	Blood glucose level in mg/dL
PANAS Negative Affect	RAVLT words recalled after List 5	Blood total Cholesterol in mg/dL
PANAS Positive Affect	TMT-A points correctly linked per second	Systolic Blood Pressure in mmHg
WHOQoL-Bref Perceived Quality of Life	Montreal Cognitive Assessment score	Diastolic Blood Pressure in mmHg
WHOQoL-Bref Perceived Health subscale	DASS-21 Depression subscale	C-Reactive Protein levels in mg/L
WHOQoL-Bref Physiological Health Quality of Life	DASS-21 Anxiety subscale	ESR percentage over the expected maximum normal level given sex and age
WHOQoL-Bref Psychological Health Quality of Life	DASS-21 Stress subscale	FVC as a percentage of reference values
WHOQoL-Old Sensory Abilities subscale	UCLA Loneliness Scale	FEV1 as a percentage of reference values
WHOQoL-Old Autonomy	Rosenberg Self-Esteem Scale	FEV6 as a percentage of reference values.
Social Well-Being Composite score*	BPI Maximal Pain subscale	PEF as a percentage of reference values
	BPI Average Pain subscale	One-foot stance seconds in balance
	BPI Pain Intensity subscale
	BPI Pain Interference subscale

Abbreviations: PANAS: Positive and Negative Affect Schedule; WHOQoL: World Health Organization Quality of Life; WAIS: Wechsler Adult Intelligence Scale; RAVLT: Rey Auditory-Verbal Learning Task, TMT: Trail Making Test; DASS: Depression, Anxiety and Stress Scale; BPI: Brief Pain Inventory; ESR: Erythrocyte Sedimentation Rate; FVC: Forced Vital Capacity; FEV1: Forced Expiratory Volume in 1 second; FEV6: Forced Expiratory Volume in 6 seconds; PEF: Peak Expiratory Flow. *The Social Composite score combined the scores from the Social Well-Being scale, WHOQoL-Bref Social Relationships subscale, and WHOQoL-Old Social Participation subscale.

Data preparation

Before statistical analyses, the following data transformations were carried out:

(a) Social well-being scores were created as a composite score of the Scale of Social Well-being (Keyes, 1998), and the subscales of satisfaction with social relationships from the WHOQoL-Bref (The Whoqol Group, 1998), and of social participation from the WHOQoL-Old (Power et al., 2005), showing an internal consistency between .64 and .75;

(b) Trail Making Test (TMT) scores from form A were calculated as the number of points correctly linked per second (i.e., hits/time to complete the test);

(c) C-reactive protein levels were multiplied by 10 and log-transformed to increase their distribution normality;

(d) For any given participant, the observed value for erythrocyte sedimentation rate (ESR) was divided by the value obtained with the formula age/2 for men or (age + 10)/2 for women to get a percentage of ESR over expected maximum average values according to age and biological sex (see, Miller et al., 1983);

(e) The percentage over recommended reference values for forced vital capacity (FVC), forced expiratory volume in 1 and 6 seconds (FEV1, FEV6), and peak expiratory flow (PEF) were calculated following the equations proposed by Direção Geral da Saúde (2016) and the National Health and Nutrition Survey (NHANES III).

Calculation of Mahalanobis distances in a multivariate space including all the psycho-emotional, psycho-social, cognitive, and biological variables assessed at pre- and posttest as well as participants’ number of years of formal education, and an estimation of socioeconomic status showed no multivariate outliers. Additionally, scores exceeding 2.2 times the interquartile range from a given variable mean were considered univariate outliers (Hoaglin & Iglewicz, 1987) and winsorized (Tukey, 1962).

Further, missing values for a single item in a test were replaced by participants’ mean in that test (six participants for WHOQoL-BREF; four for Social Well-Being scale, Rosenberg’s self-esteem scale, and Positive and Negative Affect Schedule (PANAS); three for MoCA; and two for WHOQoL-OLD). Additionally, multiple imputations of remaining missing data were carried out for the intervention and the waiting list groups separately (Sullivan et al., 2018). Since Little’s test showed data were missing completely at random, χ² (2536) = 2579.57, p = .268, predictive mean matching procedures were used. The mean of 50 imputations ran using as predictors all the study variables, as well as participant’s age, sex, marital status, height, weight, and type of social care unit used at pretest and program satisfaction at posttest, was used as the imputed value.

Finally, dichotomous variables were created to run the moderation analyses planned to assess whether the factors associated with adherence to the intervention also moderate the intervention outcomes. These dichotomous variables coded whether the expected intervention effects were achieved or not. Such variables were calculated for those parameters for which previous work on the Sing4health intervention did not find the expected outcomes. Thus, for satisfaction with life scale scores, TMT points per second, social identification with the institution score, and FVC, participants were assigned a code 1 when the difference between post- and pretest scores was equal or larger than 0, and a code 0 otherwise, since increases in these variables were expected from the pre to the posttest. For negative affect subscale of PANAS, UCLA loneliness scale, depression, anxiety, and stress scale scores, and ESR percentage over the expected maximum, code 1 was assigned if the difference between post- and pretest scores was equal or lower than 0, and code 0 otherwise, since a decreased score/level was expected at post-intervention.

Statistical analyses

All statistical tests were performed using SPSS (version 21; IBM) and PROCESS macro (v.3.5; Hayes, 2017). The main goal of this study was to identify those factors at pretest that can predict adherence to the intervention. First, a principal component analysis (PCA) was applied to the data matrix including all the variables at pretest (see Table 1). Adequacy of the data matrix for this procedure was tested using the Kaiser–Meyer–Olkin measure and Bartlett’s test (Field, 2017). Next, PCA with Varimax rotation (Kaiser, 1958) was run on the correlation matrices of the participants’ responses at the pretest assessment. All components with eigenvalues larger than 1 were retained, and only variables with factor loadings greater than an absolute value of .4 for a single component were considered in factor interpretation (Field, 2017).

Multiple linear regression was used to assess the predictive value of the factors retained on the PCA and the sociodemographic variables (age, sex, socioeconomic status, and Education Years) on the number of completed sessions in the intervention. Multicollinearity between predictors was tested by examining correlations (all r < .5), Variance Inflation Factors (all below 2), and tolerance (all over .5). For the sake of parsimony in the model, a backward method was used. To set a limit on the total number of variables included in the final model, at each step, variables were excluded from the model based on p-values with a threshold of 0.1.

The second goal of the study was to explore whether the factors predicting adherence moderated the outcomes of the intervention program in the variables that failed to yield the expected benefits in previous analysis with this sample, namely, life satisfaction, social identification with the institution, TMT, FVC, PANAS negative affect, UCLA loneliness scale, depression, anxiety and stress scale, and ESR. To that end, several moderation models were tested using PROCESS macro model 1 with 5000 bootstrap resamplings. 95% bias-corrected confidence intervals (CI) were used to evaluate indirect effects (Preacher & Hayes, 2008). All the models include the number of completed sessions as predictor (X). Each model has as outcome (Y) one of the dummy variables reflecting the expected change (1 = accomplished; 0 = not accomplished) for the variables mentioned above; and one of the factors predicting adherence to the intervention in the backward multiple regression as moderator (W₁). All the moderation analyses were run controlling for the type of care center used by the participant (coded with 1 – residential care; 2 – daycare center; 3 – home support; or 4 – independent of institution).

Results

Data dimensionality reduction

Bartlett’s test reached significance; approximated χ²₍₅₆₁₎ = 3216.03, p < .001; and the Kaiser–Meyer–Olkin test yielded a score of .76, indicating the adequacy of the pretest data matrix to conduct a PCA on the correlation matrix. Ten factors explaining a total of 73.45% of pretest data variance presented eigenvalues higher than 1 and were selected for extraction. Factor loadings for those variables above the specified threshold for each factor and the name given to each factor can be found in Table 2.

Table 2. Retained factors after the PCA decomposition of the pretest data matrix with the psycho-emotional, psycho-social, cognitive, and biological variables, along with the variable names of those with factor loadings above .4 for each factor.

Variable name (explained variance)	Factor loadings	Variable name (explained variance)	Factor loadings	Variable name (explained variance)	Factor loadings
Factor 1: General Well-Being (21.13%)		Factor 3: Pain (8.95%)		Factor 6: Blood Pressure (4.68%)
Satisfaction with life scale	.697	Current pain intensity	.902	Diastolic Blood Pressure	.777
PANAS positive affect	.584	Maximal pain intensity	.850	Systolic Blood Pressure	.799
Social Well-Being composite score	.781	Average pain intensity	.803
Rosenbergs Self-Esteem Scale	.624	Pain interference	.761	Factor 7: Systemic Inflammation (3.99%)
Sensory Abilities WHOQoL-Old	.477			C-Reactive protein levels	.793
Autonomy WHOQoL-Old	.577	Factor 4: Respiratory Function (7.29%)		ESR maximum normal percentage	.751
Perceived QoL WHOQoL-Bref	.738	Forced Vital Capacity	.922
Perceived Health WHOQoL-Bref	.603	Forced Expiratory volume in 1 sec	.949	Factor 8: Balance (3.43%)
Psych. QoL WHOQoL-Bref	.763	Forced Expiratory volume in 6 sec	.932	One foot stance	.611
Physio. QoL WHOQoL-Bref	.659	Peak Expiratory Flow	.719
				Factor 9: Blood Glucose (3.10%)
Factor 2: Negative Mood & Loneliness (12.01%)		Factor 5: Cognition (5.87%)		Blood Glucose	.874
PANAS negative affect	.811	Digit-symbol subtest WAIS	.879
UCLA Loneliness scale	.680	RAVLT hits for List 5	.638	Factor 10: Blood Cholesterol (2.99%)
Depression scale	.742	Montreal Cognitive Assessment	.868	Blood Cholesterol	.750
Anxiety scale	.785	Points per second in TMT form A	.833
Stress scale	.813

Abbreviations: PANAS: Positive and Negative Affect Schedule; WHOQoL: World Health Organization Quality of Life; WAIS: Wechsler Adult Intelligence Scale; RAVLT: Rey Auditory-Verbal Learning Task, TMT: Trail Making Test; ESR: Erythrocyte Sedimentation Rate.

Factors predicting adherence

The backward multiple regression model produced nine steps, starting with the 10 factors extracted from the PCA and four sociodemographic variables as potential predictors of participants’ adherence to the intervention. Akaike’s information criterion difference between the models produced in steps 7 to 9 was lower than 2 units, and their adjusted R² difference was lower than 0.003, indicating a strong equivalence of these models. For the sake of parsimony, the 9^th step model was selected as the final model, F(6,142) = 5.388, p < .001, adjusted R² = 0.151. This model included one sociodemographic variable, Education Years, and five PCA factors, general well-being (GWB), negative mood and loneliness, respiratory function, blood pressure, and blood cholesterol. Standardized beta values indicated that respiratory function (β = .131, p = .085) and blood cholesterol (β = −.144, p = .061) were not independent predictors of participants’ adherence to the intervention (see Table 3). All other variables in the model were positive predictors of the number of sessions attended: Education Years (β = .180, p = .022, semipartial-correlation = .176), GWB (β = .247, p = .001, semipartial-correlation = .246), negative mood and loneliness (β = .177, p = .022, semipartial-correlation = .175), and blood pressure (β = .180, p = .019, semipartial-correlation = .179). Therefore, higher factor scores and years of education predicted a higher number of sessions attended.

Table 3. Coefficients of the variables included in the final model of the backward multiple regression.

Variable	B	B SE	B CI	β	t	p
General Well-Being (GWB)	3.28	1.01	1.28; 5.28	.25	3.25	.001
Negative Mood & Loneliness	2.35	1.02	.34; 4.36	.18	2.31	.022
Respiratory Function	1.75	1.01	−.24; 3.74	.13	1.73	.085
Blood Pressure	2.39	1.01	.39; 4.38	.18	2.37	.019
Blood Cholesterol	−1.92	1.01	−3.93; 0.09	−.14	−1.89	.061
Education Years	0.71	0.30	0.11, 1.31	.18	2.32	.022
(Constant)	15.35	1.73	11.93, 18.77		8.87	<.001

Predictors of adherence as moderators of the intervention outcomes

Moderation analyses were run using the number of intervention sessions attended as predictor and binomial variables expressing whether the expected change was accomplished for the selected outcome as dependent variables. Only the significant predictors in the final model of the backward regression were used as moderators. The significant moderations are reported below (for more details, see the supplementary material at http://osf.io/raku4/?view_only=f170363c298b48cb82bb81431151e8c6).

Expected change in life satisfaction scores: the interaction between the number of sessions and the GWB factor was significant; B = .03, 95% C.I. [.005, .062], p = .022. For low levels of GWB, B = .02, 95% C.I. [−.009, .064], p = .144, there was no relationship between the number of sessions and the expected change in life satisfaction scores. However, for average, B = .06, 95% C.I. [.029, .093], p < .001, and high levels of GWB, B = .09, 95% C.I. [.047, .143], p < .001, as the number of attended sessions increased, it was more likely that life satisfaction improved (Figure 1A). Similarly, the interaction between the number of sessions and blood pressure factor was significant; B = .033, 95% C.I. [.004, .061], p = .023. For low levels of blood pressure, B = .010, 95% C.I. [−.025, .045], p = .588, there was no relationship between the number of sessions and the expected change in life satisfaction scores. But for average, B = .042, 95% C.I. [.015, .069], p = .002, and high levels of blood pressure, B = .075, 95% C.I. [.032, .118], p = .001, as the number of attended sessions increased, it was more likely that life satisfaction was enhanced (Figure 1A).

Figure 1. Simple slopes analysis of the moderating effect of the factors predicting adherence on the relationship between the number of sessions completed and the probability of observing the expected change at the end of the intervention in selected outcomes. Lines indicate low (dotted light gray), medium (dashed gray) and high (solid black) levels of the moderator variable. a) Moderation of General Well-Being (GWB, left) and Blood Pressure (right) on the probability of observing an increase in satisfaction with life (SWLS) as a function of adherence levels. b) Moderation of General Well-Being (GWB) on the probability of observing an increase in social identification with the institution (FISI) as a function of adherence levels. c) Moderation of General Well-Being (GWB) on the probability of observing a reduction in low-grade systemic inflammation as indexed by the erythrocyte sedimentation rate (ESR) percentage over maximum normal values as a function of adherence levels.

Expected change in identification with the social institution scores: the interaction between the number of sessions and the GWB factor was significant; B = .040, 95% C.I. [.011, .068], p = .006. For low levels of GWB, B = .021, 95% C.I. [−.014, .057], p = .237, there was no relationship between the number of sessions and the expected change in social identification with the institution. However, for average, B = .061, 95% C.I. [.029, .093], p < .001, and high levels of GWB, B = .101, 95% C.I. [.052, .151], p < .001, as the number of sessions increased, the more likely social identification with the institution improved (Figure 1B).

Expected change in ESR percentage over expected maximum normal level: the interaction between the number of sessions and the GWB factor was significant; B = .032, 95% C.I. [.005, .060], p = .021. For low, B = −.023, 95% C.I. [−.061, .014], p = .217, and average levels of GWB, B = .009, 95% C.I. [−.018, .036], p = .529, there was no relationship between the number of sessions and the expected change in ESR percentage over the expected maximum. For high levels of GWB, B = .041, 95% C.I. [.001, .081], p = .043, as the number of attended sessions increased, it was more likely that ESR levels were reduced (Figure 1C).

Discussion

The main goal of the present work was to identify the factors related to adherence of older participants to an artistic GS intervention. A second aim was to explore the moderator role of such factors on the intervention outcomes. Evidence on these factors may help design better art-based interventions and establish a series of monitoring and evaluation processes to lower attrition levels and optimize the benefits of such interventions.

Factors of adherence and attrition

After data reduction of the pretest assessment, 10 factors explaining approximately 73% of the data variance were identified. Only five of these factors remained in the final model of a backward multiple regression analysis. Namely, GWB, negative mood and loneliness, respiratory function, blood pressure, and blood cholesterol. All factors, except respiratory function and blood cholesterol, were predictors of the number of attended sessions. Additionally, of the four socio-demographic variables considered, only years of education was a positive predictor of adherence. Hence, GWB, negative mood and loneliness, blood pressure, and years of formal education predicted adherence to the GS intervention.

Previous reports have shown that participants’ years of formal education and socioeconomic status were positively associated with participant enrollment in GS interventions (Dingle et al., 2019; Johnson et al., 2015) and highlighted the need to test the efficacy of GS interventions on more culturally diverse samples (Johnson et al., 2015). The sample in the present study included a wide range of years of formal education, however the mean education years was low (M = 4.79, SD = 3.33; range = 0–16). Thus, the current results add support and widen the evidence pointing to greater adherence to GS interventions in those older adults with higher education levels. Several reasons may be related with this fact, for example, the perception that GS activities will be easy to perform and that artistic/cultural activities are interesting. Future studies addressing the reasons why these participants may be more motivated to enroll in GS programs than participants with lower levels of formal education are warranted.

GWB was the factor explaining most variance in the pretest assessment (21.13%). Contributions to this factor were made by social well-being, self-esteem, life satisfaction, positive affect, perceived health, sensory abilities, and general as well as autonomy-related, psychological, and physiological quality of life. First, the importance of perceived health, quality of life, and social support on adherence to the GS intervention was highlighted by the participants in the interviews, as the most common reasons reported for abandoning the intervention were health-related issues (personal or familiar), and the enrollment of friends or a partner that were randomized to a different intervention group (anonymized). Second, the associations of well-being with other positive variables and processes of human functioning are well reported in the literature (Diener et al., 2017). Higher GWB may be associated with higher motivation, self-confidence, and energy to participate in activities and social interactions (Ryan & Deci, 2000; Webster, 2008), fostering further engagement in the GS intervention.

Negative mood and loneliness factor, which explains a 12.01% of the data variance at pretest, is composed of depression, anxiety, stress, loneliness, and negative affect scores. This factor was an independent positive predictor of the number of sessions attended by the participants. So, the higher the scores in those variables, the larger the number of sessions completed. This result contrasts with previous literature as loneliness has been previously reported to impose a series of barriers to adherence to group interventions (Stuart et al., 2022). Previous studies on the adherence of older adults to cognitive/educational and physical exercise interventions have found that negative affect and depression are inversely related to adherence to those interventions (Howells et al., 2016; Rivera-Torres et al., 2019). However, Shaw et al. (1994) reported that both very low and very high depression scores were related to attrition in different interventions for osteoarthritis patients. Methodological differences between the previous and present studies may underlie these contrasting results. For instance, differences in the population (users of day care centers, nursing homes or home support versus community-dwelling older adults; healthy versus osteoarthritis patients), or in the intervention program (artistic GS versus cognitive/educational or physical exercise) or the regime of the intervention (presential versus smartphone application) between the studies may explain such discrepancy in the results (i.e., Howells et al., 2016; Shaw et al., 1994). Future studies should further clarify the role of negative emotions and loneliness in the adherence to artistic intervention programs.

The blood pressure factor explained nearly 5% of total data variance at pretest and was a positive predictor of adherence. In contrast, high blood pressure has been previously associated with physical (e.g., obesity, diabetes) and mental illness (e.g., depression, anxiety, and stress) and with more hospital inpatient visits (National High Blood Pressure Education Program, 2004; Rozario & Masho, 2018), which would expectedly lead to lower adherence to the intervention. To the best of our knowledge, however, no previous studies have reported the role of blood pressure as a predictor of adherence to interventions for older adults. Additionally, the blood pressure factor in the present study reflects a composite of diastolic and systolic blood pressure measurements, which present similar factor loadings (i.e., .777 and .799, respectively), precluding interpretations based on each parameter separately. Further, no systematic control of antihypertensive drug usage was carried out. Consequently, future studies should investigate whether there are differences in the predictive power for these two blood pressure parameters on adherence to art-based interventions and whether this relationship is affected by antihypertensive treatments.

Factors that moderate the intervention benefits

Participants showing average or high scores on the blood pressure factor were more likely to increase their life satisfaction scores from pre- to posttest, the higher their adherence to the GS intervention. In contrast, participants with low scores on the blood pressure factor showed moderate probabilities of boosting their life satisfaction scores regardless of their adherence levels. Previous studies have found a significant association between low blood pressure and cognitive decline in older adults (Gabin et al., 2017; Maule et al., 2008), which could partially explain the differential benefits of the program for these groups of participants. Participants with lower blood pressure and cognitive function may have benefited less from the program due to their difficulties following the maestro’s indications and learning the songs’ lyrics, which may have hindered their self-confidence and satisfaction with life. To the best of our knowledge, no previous studies have reported the role of blood pressure as a moderator of the effects of GS interventions. Therefore, the present results should be considered exploratory, indicating the need for further research.

Besides being a factor that explained adherence, the GWB factor demonstrated to be a key moderator of the probability of observing benefits from the intervention in life satisfaction, social identification with the social care institution, and systemic inflammation levels.

In participants with high and moderate GWB, moderation results showed that greater adherence to the intervention was associated with a higher probability of increasing life satisfaction and social identification with the institution after the intervention. Participants with low levels of GWB showed a different trajectory because, regardless of the number of sessions attended, they present a high probability of increasing life satisfaction and social identification scores. Probably, the latter group were more permeable to an initial increase in life satisfaction and social identification, which did not increase further with a higher intervention “dose.” Maybe this was due to a ceiling effect on the first group or because underlying motives prevented further increases in life satisfaction for low GWB participants. Nevertheless, these results expand evidence of increased life satisfaction and well-being as a result of participation in GS interventions (Dingle et al., 2021; Johnson et al., 2020) in that they indicate that participants with different GWB profiles may need a different intervention “dosage” to benefit from those improvements.

Finally, participants high on GWB also showed higher probabilities of reducing their ESR levels (i.e., lower systemic inflammation) with higher adherence to the intervention, which was not the case for participants with low or moderated levels of GWB. Previous studies have shown that systemic inflammation markers and well-being scores are related (Elenkov et al., 2005; Lasselin et al., 2016). Indeed, higher levels of psychological and physiological well-being reached through engagement in active lifestyles are associated with reduced inflammatory markers (Lasselin et al., 2016). Nonetheless, the current results are the first to extend those active lifestyles to include art-based activities. Additionally, they present novel exploratory evidence on a graded effect, where different levels of well-being are associated with opposite trajectories of the relationship between adherence to a GS intervention and inflammation level reductions, which should be further studied.

Conclusion

This study raises novel results for further clinical investigation and application. Years of education, general well-being, negative mood and loneliness, and blood pressure at baseline predicted participants’ adherence to a singing group artistic intervention. Thus, screening participants on these characteristics in future artistic interventions are essential for preventing attrition and maximizing benefits. For example, participants with lower education levels showed lower adherence to the intervention, which may be related with lack of motivation by artistic interventions or with the perception that singing group activities (e.g., reading and memorizing lyrics and songs and harmonizing with the group) as challenging to perform. Therefore, intervention goals should be defined according to participants’ perceived abilities and the selection of activities closely adjusted to participants’ interests and motivations.

Results also showed that general well-being was a key moderator of the program’s effects. Participants with moderate and high levels of general well-being at baseline will probably improve life satisfaction, social identification with the institution, and systemic inflammation levels to a larger extent as the number of intervention sessions increases. This means that the duration or higher doses of the intervention may be important for a greater efficacy of the intervention. Similarly, blood pressure also moderates the probability of observing intervention benefits for participants’ life satisfaction. Low blood pressure, usually associated with cognitive decline in adults with advanced age, predicted a lower likelihood of positive outcomes. For improved results, future interventions with participants with lower global well-being may include strategies providing more support, feedback, and reassurance from the intervention team, where smaller groups of participants are advisable.

Limitations and strengths of the study

One of the limitations of this study is its exploratory nature: following a data-driven approach without hypotheses guidance. Additionally, PCA factors are composed by several variables, where the specific contribution of each of these building blocks cannot be isolated. This fact calls for further studies to shed light on more specific predictors of adherence to art-based interventions. The strengths of the study include being an RCT and using a sample of older adults from a social care institution and low socio-educative background, providing evidence generalizable to this target population for future gerontological research and interventions.

Clinical implications

• Future artistic interventions for older adults will benefit from knowing potential characteristics of attrition of the participants;

• Future interventions will benefit from knowing potential moderators of the effects of artistic interventions on the outcomes;

• Future artistic interventions are advised to screen for these potential factors of attrition and moderation of the intervention effects (e.g., low education and well-being levels), and to adjust the intervention programs accordingly.

Contributions

ICG – Conceptualization, Methodology, Validation, Investigation, Data Curation, Formal Analysis, Review and Editing, Supervision, Project Administration, Funding Acquisition; DP – Writing original Draft, Formal Analysis; MLL – Methodology, Review and Editing, Supervision.

Data availability

Details of the procedure, databases of the study figures and tables are available on the Open Science Framework platform (https://osf.io/2wn39/?view_only=0fea361ce5f944aaac4ccb8daf2912a8).

International review board

The Study Protocol and the informed consent for participants were approved by the Ethics Committee of the Centro de Investigação em Psicologia, from the Universidade Autónoma de Lisboa, CIP UAL – Approval Nr. 12-09-2018.

RCT registration number

Sing4Health; trial registration number: NCT03985917, June 14, 2019.

Acknowledgments

To Anabela Pires (Singer) for the Artistic Direction of the Intervention and Sérgio Fontão (Maestro) and Pedro Baião (Pianist) for the Singing Group Intervention Implementation. To Maria D’Assis Ribamar and Alexandra Antunes from the SCM of Almada; and Etelvina Ferreira, Neusa Freixinho, Maria Teresa Barata from the SCM of Lisboa for the promotion and logistics support of the intervention, and the recruitment, contacts, transportation, and support of participants’ mobility. To the City of Almada and OPART TNSC (National Opera House) for providing theaters and rehearsal rooms.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/07317115.2022.2151390.

Additional information

Funding

The study was supported by a grant from Portugal’s Participatory Budget (Project n.° 626, OPP 2017; Dispatch n.° 11409 C/2017) Coordinated by Direção-Geral das Artes [Projeto n.° 626, OPP 2017].