Effects of online group exercises for older adults on physical, psychological and social wellbeing: a randomized pilot trial

Participation measures

Attrition. The attrition rate was used to measure the proportion of participants lost at the end of the study.

Adherence. The adherence was used to measure the conformity of the participants with the exercise program. For each participant, two measures were considered. The first is related to persistence throughout the eight weeks of the exercise program, and it was computed considering the ratio between the number of participations in exercise sessions by a participant and the number of the exercise sessions planned in the program. Participation was measured by logging the attendance to the scheduled training sessions in the virtual classroom (considering also partial participation, where participants skipped exercises). The second measure is related to the level of completeness of the exercise sessions. It was calculated considering, for each session, the percentage of exercises videos that the user actually followed (watched)—excluding the time of preparation and skipped exercises—with respect to the total duration of the exercises planned for the session.

Physical assessment exercises

Specific assessment exercises, developed and validated within the Otago Exercise Program (Campbell et al., 1997; Campbell & Robertson, 2003), were used to measure participants’ leg muscle strength and walking ability at the beginning and at the end of the study, in a face-to-face session with each participant. In particular, the assessment exercises were:

• 30 s Chair Stand test (Jones, Rikli & Beam, 1999): the purpose of this test is to evaluate leg strength and endurance. From seated position, the participant rises to a full standing position and then sit back down again for 30 s. The outcome measure is the number of times the participant comes to a full standing position in 30 s.

• Timed Up & Go test (Podsiadlo & Richardson, 1991; Rossiter-Fornoff et al., 1995): the purpose of this test is to assess older adult mobility. From the seated position, the participant stands up from the chair, walks for 3 m at his/her normal pace, then turns, walks back to the chair and sits back down again. The outcome measure is the number of seconds to complete the test.

Psychological dimensions

We investigated changes in psychological dimensions related to different aspects of physical activity and wellbeing, by collecting participants’ feedback before and after the study on the following measures:

Enjoyment of physical activity. Past literature has shown that intrinsically motivated people tend to engage in physical activity for personal improvement and because they enjoy it (Deci & Ryan, 1985; Pelletier et al., 1995). Enjoyment of physical activity is believed to develop positive attitudes toward exercise, enhance intrinsic motivation, and consequently foster long-lasting adherence to physical activity (Ryan et al., 1997; Wankel, 1993). In order to measure participants’ enjoyment of physical activity at the beginning and at the end of the study, we used the Physical Activity Enjoyment Scale (PACES; Kendzierski & DeCarlo, 1991), which has been validated in several studies, including one with an Italian sample (Carraro, Young & Robazza, 2008). The scale includes 16 items scored on a 5-point Likert scale with the range from 1 (disagree a lot) to 5 (agree a lot). Total enjoyment scores range from 16 to 80 (maximum enjoyment).

Subjective wellbeing. In order to investigate the effectiveness of the application and training in improving subjective wellbeing, we collected participants’ feedback before and after the training period using the Wellbeing scale of the Multidimensional Personality Questionnaire (MPQ; Tellegen & Waller, 2008). This scale was developed to assess positive emotional tendencies as distinct from the absence of a negative emotional disposition. Wellbeing represents individual dispositions to experience positive emotions, and is an important marker of the higher order Positive Emotionality dimension. The scale includes 12 items requiring a true / false response, with the total scoring ranging from 0 to 12. People who score higher in this scale tend to describe themselves as cheerful, optimistic, hopeful, having interesting experiences and engaging in enjoyable activities (Tellegen & Waller, 2008, p. 274).

Social wellbeing

The effects of the technology-based intervention on the social wellbeing was assessed on the basis of participants’ feedback, collected at the beginning and at the end of the study using a measure of loneliness.

Loneliness is an aspect that negatively affects social wellbeing. Shifts in the social environment, and in particular loneliness, are believed to be an important aspect in the life of aging people (see for example Hughes et al., 2004; Liu & Rook, 2013). Loneliness involves individual perception of social isolation and feelings of not belonging and being disconnected, and is a central aspect of a group of socio-emotional states including, among others, self-esteem, optimism, anxiety, anger and social support. To measure loneliness, we used a shorter version of the R-UCLA Loneliness Scale (revised version of the UCLA Loneliness Scale developed by University of California, Los Angeles) (Russell, Peplau & Cutrona, 1980) developed by Hughes et al. (2004). The scale used includes 3 items scored on a 5-point Likert scale, with the total score ranging from 3 to 15, and higher scores indicating higher levels of loneliness.

Participants’ perception

To collect feedback on the participant’s perception of the tool, we used a questionnaire (https://goo.gl/zl7daL) assessing the most stimulating aspect of the overall experience and the usefulness of the various features.

Online social interactions

We briefly summarise the usage of the online social interaction features that were available in the social group: private messages and bulletin board (public messages). The usage was analyzed by looking at the database of messages exchanged among participants.

Private messages were preferred over the public messages, accounting for 75% of all the messages exchanged (544 messages). The most active user was the Coach, who had to contact the trainees to check on their progress on a weekly basis, followed by the Technician. We illustrate the interactions between all the participants in Fig. S1 .

A qualitative content analysis of the online interactions was presented in Baez et al. (2016), where we reported the distinctive use of private and public messages. We developed a coding scheme based on relevant literature about online behaviour and communities, developing a final coding scheme composed of 5 top- and 12 sub-categories. Then two independent researchers coded the messages (Cohen’s kappa: public messages .85 and .84 for top and sub categories; private messages .87 and .85 for top and sub categories correspondingly).

We highlight some observations to put in context the results in this section:

• In private messages the Coach took an active role, dominating the discussions around physical activity and, in particular, by offering support. Trainees instead lead discussions focused on community building, and entertainment (e.g., sharing jokes). When discussing about physical activity, trainees focused on reporting their personal experience with the exercises. Discussions about the application were shared mostly with the technician and were related to issues with the application.

• In the bulletin board, the Coach was much less active, limiting his participation to congratulating users after each week of training. Trainees participated more actively in community building, and in discussions about physical activity where they also provided support and encouragement to their peers. Positive comments about the application were interestingly largely more predominant in public.

Participants of the control group enjoyed the same type of support, although via phone calls, diaries and transcripts of the interactions with the Staff were not available, limiting the comparison of effects at the group level and the detailed analysis of interactions to the Social group.

Attrition

The intervention resulted in a 7.5% attrition rate (corresponding to 3 participants), measured in terms of the proportion of participants lost at the end of the study. Reasons behind the withdrawal of these participants were related to unexpected health and family problems or, in one case, because of Internet connection issues which could not be solved.

Persistence

The overall persistence rate in the two groups was of 76% (SD = 22.6%), when considering the total number of sessions available (Fig. 3A). In the social group the persistence rate was 85%, while in the control group it was 64%. A between-subjects analysis of variance was performed to compare the persistence of both groups while controlling for the initial scores in the measures of gait speed and leg muscle strength (physical measures), and the enjoyment of physical activity. The independent variables were grouped in three equally distributed intervals (Low, Medium, High). The analysis showed a significant interaction between group and the initial measures of leg muscle strength (F(2, 23) = 5.966, p = .008, partial eta squared =.342), but no significant interaction with gait speed (F(1, 23) = 3.42, p = .08, partial eta squared =.13) nor enjoyment of physical activity (F(2, 23) = 1.93, p = .17, partial eta squared =.144). In Fig. 3B we show the relevant interaction plot.

The same analysis, exploring the effect of the initial loneliness score as independent variable (with and without controlling for the other variables) reveal no significant interaction with group.

The interaction between group and leg muscle strength shows a higher persistence in participants of the control group that scored higher in the initial leg muscle strength test. For the social group however, there is no significant difference in the persistence of participants based on their initial leg muscle strength score. This suggests that, when training individually, fitter participants tend to adhere more to the training, but when social elements are in place these differences in fitness no longer determine adherence.

There was also a significant main effect for group (F(1, 23) = 13.151, p = .001, partial eta squared= .364), with the social group showing a higher persistence rate (M = 85.4%, SD = 16.1%) compared to the control group (M = 64.2%, SD = 24.1%). Considering the minimum number of sessions per week as recommended by the Coach (2 sessions per week), it results in a higher level of persistence for both groups: social group (M = 97.5%, SD = 6.8%) and control group (M = 85%, SD = 25.9%).

The lower variability of persistence in the social group can be explained by the social features (normative influence, social facilitation, social learning) that might have motivated users to comply with the community norm. In the control group, the higher variability suggests a stronger effect of individual differences due to the lack of social awareness.

Completeness

The overall completeness rate in the two groups was 90.32% (SD = 17.4%), suggesting that participants tended to complete the working sessions once they started. The completeness rate in the social group (M = 91.75%, SD = 12.46%) was slightly higher compared to that of the control group (M = 88.63%, SD = 22.24%), although not statistically significant. As in the previous measure, a between-subjects analysis of variance was performed to compare the completeness of both groups while controlling for the initial scores in the measures of gait speed and leg muscle strength, enjoyment of physical activity and loneliness. No interaction was found between group and the initial scores, but a main effect for initial leg muscle strength (F(2, 23) = 5.075, p = .015, partial eta squared =.306). We attribute this effect to the duration of the video exercises that were assigned to fitter individuals (e.g., required more repetitions), which were longer for higher levels of intensity.

The higher level of completeness and lower variability in the social group can be explained by the presence of self-monitoring tools (e.g., positive and negative reinforcement) and social facilitation (exercising with others), which were lacking in the training sessions of the control group participants.

Leg muscle strength

A repeated-measures of covariance (ANCOVA) was performed to determine a statistically significant difference in the 30 s Chair Stand scores between control and social group, using pre- and post- measurement points as repeated measures variable, and persistence as covariate (Fig. 4A). The analysis showed a significant main effect for time (F(1, 29) = 37.803, p < .001, partial eta squared =.566) but no interaction between group and time (F(1, 29) = 0.704, p = .404, partial eta squared= .024). This suggests that, while there is an overall improvement, time did not have a substantially different effect on the performance of the two groups nor on participants with different levels of adherence. We attribute this effect to (i) high adherence with the minimal recommendations by the Coach and (ii) the duration of the pilot that might not have been enough to see statistically significant results. The analysis also revealed a main effect of group (F(1, 29) = 6.809, p = .014, partial eta squared =.19) and persistence (F(1, 29) = 10.233, p = .003, partial eta squared =.261).

We should note that despite randomization and the non-significant difference between the self-reported physical activity in the two groups (measured with the Rapid Assessment of Physical Activity Questionnaire by Topolski et al. (2006)), participants in the Social group performed better than participants in the Control group in the 30 s Chair Stand pre-test.

Gait speed

A repeated-measures of covariance (ANCOVA) was performed to determine a statistically significant difference in the Timed Up & Go scores between control and social group, using pre- and post- measurement points as repeated measures variable, and persistence as covariate (Fig. 4B).

The analysis showed a significant main effect for time (F(1, 31) = 11.952, p = .002, partial eta squared= .278) but no interaction between group and time. The results bear similarities with the ones of leg muscle strength test, suggesting overall improvement in gait speed after the training program, but not a significant different effect on the two groups nor on participants with different levels of adherence. The analysis also revealed a main effect of group (F(1, 31) = 11.789, p = .002, partial eta squared =.276) and persistence (F(1, 31) = 5.086, p = .031, partial eta squared =.141). As in the previous measure, we attribute the effect to the overall high adherence and to the duration of the pilot study.

Enjoyment of physical activity

We used the Physical Activity Enjoyment Scale (PACES; Carraro, Young & Robazza, 2008; Kendzierski & DeCarlo, 1991) to measure participants’ enjoyment of physical activity at the beginning and at the end of the study (Fig. 5).

A repeated-measures of covariance (ANCOVA) was performed to determine a statistically significant difference in the PACES scores between control and social group, using pre- and post- measurement points as repeated measures variable, and persistence as covariate. The analysis showed a main effect for time (F(1, 33) = 16.998, p < .001, partial eta squared =.278) but no interaction between group and time. An analysis excluding extreme values resulted in comparable results, with a main effect for time (F(1, 31) = 23.297, p < .001, partial eta squared =.429). These results show that participants enjoyed more engaging in physical activity at the end of the study regardless of the group and their level of adherence.

The effect of the initial level of physical ability on the enjoyment was also explored (muscle strength and gait speed as covariates in the model) but no effects were found.

Subjective wellbeing

Subjective wellbeing was measured by means of MPQ (Tellegen & Waller, 2008) before and after the study (Fig. 6). In the logistic regression performed on the subjective wellbeing score, only time was a significant factor (B =  − 1.050, OR = 0.350, 95% CI [0.129–0.909], p = .034). This suggest that subjective wellbeing improved for participants of both groups, regardless of the version of the app.

Loneliness

In the logistic regression performed on the loneliness score, only the factor time was significant in improving (reducing) loneliness (B = 1.121, OR = 3.068, 95% CI [1.177–8.380], p = .024). To investigate the extent of the improvement according to the initial loneliness score (t1) we also tested the interaction between that initial score (grouped in three equally distributed intervals: Low, Medium, High) and time. Although the interaction is not statistically significant we can see a stronger trend for higher initial levels of loneliness. In Fig. 7 we illustrate the loneliness scores before and after the training.

These results suggest that, overall, the perception of loneliness significantly decreased after the training, regardless of the group and adherence to the training. We attribute this effect to the attention given by the Coach to both groups.

To investigate if the use of social features predicts the improvement in the loneliness score for the social group, we performed a correlation test (with Spearman method) using the number of private and public messages as predictors (Fig. 8). For private messages we took message received, as the exchanges were nearly symmetrical. The results show that number of messages received is significantly correlated with the improvement in the loneliness scores (rho = − .635, p = .003). Public messages, on the other hand, were not correlated (rho = − .244, p = .314).

In virtual group exercising, adherence by persons with low starting levels of physical skills reaches the same level as that of more fit participants

Training adherence outcomes show that the initial level of skill had no significant influence on the adherence of participants of the Social group, while in the Control group we observed that fitter participants tended to adhere more to the training. These results suggest that

1. the online group-exercising could potentially overcome a major issue reported in the literature (De Groot & Fagerström, 2011) in terms of negative effect of group-exercising in the motivation of heterogeneous groups, and

2. it helped in reducing the effect of the initial level of skill in the motivation of participants, with trainees complying to the group norm.

We should notice that despite the small sample we have observed a large effect size (partial eta squared =.327). This pilot thus motivates further studies into the effect of baseline measures on adherence.

In addition, we have seen evidence of the preference of older adults for group exercising. This comes from the debriefing of participants, and after observing the participation of trainees in the Social group, who were able to choose whether to train alone or in company, and largely participated of training sessions in company (after controlling for casual meet-ups) (Far et al., 2015). This is in line with existing literature pointing to the importance of social features in fitness applications (Far et al., 2016).

These results contribute to the literature studying the effects of group-based and home-based training, which on short-term settings have not provided conclusive results (Van Der Bij, Laurant & Wensing, 2002; Freene et al., 2013). However, the hybrid nature of the proposed intervention requires further studies, to compare its effectiveness to that of traditional interventions.

Virtual group-exercising enables tailored home-based intervention with positive physical outcomes and increased persistence

In terms of physical outcomes, at the end of the eight-week program, both Social and Control groups showed significant improvement in gait speed and leg muscle strength, given the high levels of adherence in both groups. It is important to note that by providing the feeling of training together to an otherwise tailored exercise program, participants of the Social Group observed the benefits of performing exercises that were tailored to their individual abilities (traditionally part of an individual training program) while enjoying the extra motivation of the social context (as indicated by the greater persistence of the social group).

These results are in line with previous literature pointing to the equivalent health-related outcomes of traditional group- and home-based training (Freene et al., 2013), though in this setting we have achieved these results with a heterogeneous group. However, further studies are required in order to observe these effects in long-term settings.

Positive effects on enjoyment and subjective wellbeing at the end of the training program (regardless of the control or social condition)

The psychological measures have shown a significant improvement after the eight-week training program. Participants from both groups showed a significant increase in the enjoyment of physical activity, as measured by the Physical Activity Enjoyment Scale (PACES; Carraro, Young & Robazza, 2008; Kendzierski & DeCarlo, 1991), and in subjective wellbeing, as measured by the MPQ (Tellegen & Waller, 2008). This supports the literature associating regular physical activity with positive outcomes in health and wellbeing in later age (Thibaud et al., 2012; Stuart et al., 2008; Landi et al., 2010). However, the difference in the adherence observed between both groups did not account for a statistical difference in psychological outcomes. Further analysis is required to understand whether these measures are not affected by the exercise settings or differences in the software features.

Decrease in loneliness in both groups, attributed to contacts with the Coach (and between participants)

We also observe that, although social features played a part in motivating participation to training sessions in the Social group (Far et al., 2015), there was no significant difference in terms of loneliness scores with respect to the Control group after the eight weeks of training. A plausible explanation for the improvement in the Control group is the frequent contact with participants via telephone. This is common practice in clinical evaluation studies involving older adults, where social visits to the control group are performed to mirror the time and attention provided to the treatment group, and even suggested by several authors (for example, Hogan et al., 2001; Michael et al., 2010; Tinetti et al., 1994). In our study, each participant in the control group was contacted by phone, rather than via the messaging features of the application, and it is possible that the effect of this contact was strong enough to produce a significant decrease in participants’ perception of loneliness. Indeed, previous work relying on the R-UCLA loneliness scale, and with similar intervention periods (6–15 weeks), have achieved significant reductions of loneliness but have also included the physical presence of educators, trainers, and peers during the intervention (Shapira, Barak & Gal 2007; Fukui et al., 2003).

Thus, we found noteworthy the significant decrease in the level of loneliness in the Social group despite the use of remote interactions and in such a short period. What is more, the online interactions in the form of private messages were found to predict the decrease in loneliness.