Association of changes in cardiorespiratory fitness with health-related quality of life in young adults with mobility disability: secondary analysis of a randomized controlled trial of mobile app versus supervised training

For the current study, the initial randomized controlled trial design was retained in order to compare the effects of the two interventions on HRQoL, which was among the pre-specified secondary outcomes of the trial (http://​www.​isrctn.​com/​ISRCTN22387524) [33], using an equivalence design. In addition, a longitudinal design with multiple cross-sectional waves was used merging all participants together, in order to study the association between changes in cardiorespiratory fitness and changes in HRQoL over 1 year. The decision to analyze randomized controlled trial data for fulfilling this aim is based on the previous observation that cardiorespiratory fitness levels improved in the study participants and on the availability of repeated measurements of both cardiorespiratory fitness and HRQoL. Since participation to the intervention could affect the association under study, analysis was also performed separately for each intervention group. The present study adheres to the CONSORT guidelines.

This study included young adults (18–45 years) that had acquired mobility disability over the past 3 years and had participated in the randomized controlled trial [33]. Individuals were considered as having mobility disability if they had self-reported mobility limitations in performing essential daily activities, such as getting dressed, or doing usual household and work tasks. Individuals with severe mobility limitations requiring mobility assistive devices or being unable to walk at a moderate to low intensity, were excluded [33]. Further inclusion criteria were having access to a smartphone and being able to understand and speak Swedish [33]. Recruitment took place in March 2018 and was performed by rehabilitation coordinators, at occupational, rehabilitation, and primary health care centers in the Stockholm region, Sweden [33].

During recruitment at the rehabilitation centers and before the first examination, staff from the TWITCH Health Capital in Stockholm randomly assigned participants to one of the two intervention groups. Participants were randomized to receive either a mobile app program (n = 55) or a supervised health program (n = 55) for 12 weeks [33]. In order to achieve an equal number of participants in the two intervention groups, a block randomization procedure was performed using a block size of two [33]. Randomization was performed using the SAS Proc Plan (SAS Institute Inc., Cary, NC, USA).

Due to the nature of the interventions, blinding of the participants was not possible. However, participants were not aware of their allocation group before the first examination. The investigators were blinded to the randomization and to the group allocation during the follow up assessments [33].

The sample size calculation of this study was based on moderate to vigorous physical activity (MVPA), which was the primary outcome of the randomized controlled trial [33]. For this population, a difference of 10 min/day in MVPA between the two intervention groups was desirable, and therefore this threshold was chosen for the sample size calculation [33]. The total number of participants needed to have this effect with 80% power, 5% of significance level with an expected 20% loss to follow up was 100 participants [33]. In order to prevent further loss in power from unexpected reasons, the decided sample size was 110 participants [33].

Both intervention groups had multiple components designed based on behavior change techniques, including the use of intrinsic motivation, self-monitoring, and goal setting, which have deemed to be effective in increasing physical activity levels in adults [35].

Statistical analyses were performed using Stata/IC 15. A significance level of < 0.05 was set for all statistical tests. Baseline characteristics of all participants and within each intervention group are presented in Table 1 as mean ± standard deviation (SD) for the continuous variables and as frequencies and percentages for the categorical variables. In order to test for mean changes in V˙O₂max and the SF-36 domain scores between baseline and each follow up examination, two tailed t-tests were applied. The results are presented as mean changes ± SD between baseline and each follow up examination.

The effects of the mobile app program compared to the supervised health program on the SF-36 scores were examined using intention-to-treat analysis. Between group differences in the mean changes of SF-36 scores from baseline to midpoint, endpoint, and 1-year were estimated using simple linear regression models, with the supervised group chosen as referent.

Mixed-effects linear regression models were fitted for estimating the association of changes in V˙O₂max at midpoint, endpoint, and 1 year from baseline, with changes in the SF-36 scores. Mixed-effects linear regression is an extension of the generalized linear model and was suitable for this analysis, as it accounts for correlations between repeated measurements within participants and can model both mixed and random effects [42]. The intraclass correlation coefficient (ICC) was calculated for each model. All models were adjusted for age, sex, BMI, chronic conditions, and smoking, since they are predictors of both cardiorespiratory fitness [26] and HRQoL [43]. In addition, separate models were fitted for each intervention group, in order to examine possible effect modification of the intervention group on the relationship between changes in V˙O₂max and changes in the SF-36 scores.

Sensitivity analysis was also performed for comparing baseline characteristics between those who attended the 1-year follow up and those who did not. In addition, changes in V˙O₂max, MVPA and SF-36 scores from the midpoint to the endpoint of the intervention, were compared between those who attended the 1-year follow up and those who did not. Two-tailed t-tests and chi-squared tests were used for these comparisons with level of significance α = 0.05.

In total, 40/55 participants from the mobile app and 49/55 from the supervised health program were included in the intention-to-treat analysis, as they had completed the SF-36 questionnaire in at least one follow-up examination. Between the two intervention groups there were not significant differences in the mean changes of SF-36 scores from baseline to each follow up. (see Table 2). Comparing the mobile app group to the supervised group at the endpoint, the difference in the mean change of the physical and mental component summary was − 3.06 (95% CI: − 7.81 to 1.68) and − 3.62 (95% CI: − 8.96 to 1.72) respectively. At the 1-year follow up these differences were 0.09 (95% CI: − 6.27 to 6.46) for the physical component summary and 0.57 (95% CI: − 5.58 to 6.72) for the mental component summary. Figure 2 illustrates the mean scores ± SD of the physical and mental component summary of SF-36 at each examination point for each intervention group.

Mean changes in V˙O₂max levels and the SF-36 domain scores between baseline and each follow up examination are presented in Table 3. V˙O₂max was significantly higher at the follow up examinations compared to baseline. In addition, the scores of physical functioning, bodily pain, mental health, and mental component summary were improved at 1 year from baseline. Physical role functioning, general health, and vitality scores were improved during the intervention, but did not remain at the same level in the 1-year follow up. In addition, physical functioning, social functioning, and physical component summary scores decreased during the intervention but reached the baseline levels at the 1-year follow up.

The mixed-effects linear regression models indicated that changes in V˙O₂max at midpoint, endpoint, and 1 year from baseline, were positively associated with changes in the physical role functioning (β = 1.06, 95% CI: 0.11 to 2.01, ICC = 0.86), general health (β = 1.37, 95% CI: 0.56 to 2.19, ICC = 0.86), vitality (β = 0.83, 95% CI: 0.02 to 1.64, ICC = 0.80), and emotional role functioning (β = 1.34, 95% CI: 0.27 to 2.42, ICC = 0.79) domains of SF-36 (see Table 4). After adjusting all models for possible confounders (age, sex, BMI, chronic conditions, and smoking), only the associations of changes in V˙O₂max with changes in general health (β = 1.30, 95% CI: 0.48 to 2.13, ICC = 0.86) and emotional role functioning (β = 1.18, 95% CI: 0.11 to 2.25, ICC = 0.78) remained statistically significant. When measuring these associations within each intervention group, changes in V˙O₂max were significantly associated with changes in the SF-36 scores only in the supervised health program. More specifically, both crude and adjusted models in the supervised group indicated associations between changes in V˙O₂max and changes in the general health (adjusted model β = 1.88, 95% CI: 0.87 to 2.90, ICC = 0.87) and emotional role functioning (adjusted model β = 1.37, 95% CI: 0.18 to 2.57, ICC = 0.85) domains.

Sensitivity analyses indicated that participants who did not attend the 1-year follow up did not differ from the participants who attended the 1-year follow up regarding the baseline characteristics. Among the participants that did not attend the 1-year follow-up, 65% were from the mobile app group. In the mobile app group, 7 participants were dissatisfied with the allocation and dropped out before receiving the intervention. During the follow up, the attrition rates were similar in both groups. In addition, those who did not attend the 1-year follow up but had attended the endpoint follow up, had smaller increases in their V˙O₂max from midpoint to endpoint, compared to those who attended the last follow up (0.9 vs 2.4, p = 0.03).

A small number of previous studies focused on clinical populations has evaluated the effects of mobile app training on HRQoL [44‐47]. Among those studies, two found that receiving mobile app training had higher effects on HRQoL compared to receiving usual care which included unsupervised self-rehabilitation [44] or one home visit by a physician who provided nutrition and exercise advice [45]. Another study found similar effects between a mobile app program and the provision of an exercise brochure [46]. However, the post-intervention scores were higher compared to baseline in both groups [46]. Similarly, another study without control group found higher post-intervention scores at follow up compared to baseline [47]. Overall, the available evidence indicates that mobile app interventions might improve HRQoL in clinical populations. In the current study, the effects of the mobile app program were compared to a supervised health program which according to literature was expected to improve HRQoL [15, 48, 49]. Furthermore, the attrition rate within the mobile app group in our study, especially after the allocation, is higher compared to other studies of mobile app training [50]. This could be explained by the fact that previous studies did not have supervised training as a comparator, but instead provision of advice or no intervention, which might had been seen less attractive than using a mobile app [50]. However, the total attrition rates of this study are similar to previous studies of mobile apps [50].

Associations between cardiorespiratory fitness and specific domains of HRQoL have been previously observed in cross-sectional studies, however showing partially inconsistent results. More specifically, a study of healthy young males that looked into the different domains of HRQoL found positive associations only with the general health domain [27], whereas two other studies in a similar population that assessed the component summary scores of HRQoL found positive associations with both physical and mental health [28, 29]. A study of postmenopausal women in which all domain and component summary scores of SF-36 were assessed, found that cardiorespiratory fitness was associated only with the mental component summary [30]. In contrast, a study of two national representative samples in Sweden that assessed self-perceived physical and general health with two questions, found that cardiorespiratory fitness was a predictor of physical health [51]. The discrepancies in specific domains between these studies could be explained by the different components of HRQoL that were assessed and the different sexes of the participants. Nevertheless, the results of the current study might differ from previous studies for several reasons. First, in contrast to previous studies, this study looked at the associations between changes in cardiorespiratory fitness and HRQoL. In addition, the focus was on people with mobility limitations who had received physical activity interventions, a factor that seemed to modify the association of cardiorespiratory fitness changes with HRQoL.

This study also supports previous evidence regarding the possible benefits of supervised training on HRQoL. Two studies of older adults with mobility limitations reported higher increases in general health [48] and emotional role functioning [49] respectively, in those receiving supervised training compared to unsupervised training at home. Moreover, in a meta-analysis studying the effects of physical activity interventions on HRQoL, it was shown that interventions including supervised training were more effective compared to educational-motivational interventions [15].

This study has several strengths. Most importantly, to the best of our knowledge this is the first study to compare the effects of mobile app training to supervised training on HRQoL as well as to examine the association of changes in cardiorespiratory fitness with HRQoL, in people with mobility disability. In addition, the methods chosen had several benefits. Most importantly, the randomized controlled trial study design using intention to treat analysis, was allowing to compare the effects of the two interventions on HRQoL with a very low risk for confounding as all the characteristics of the participants were randomized in each group. Moreover, studying the population longitudinally allowed us to assess whether changes in cardiorespiratory fitness were associated with changes in HRQoL on an individual level over 1-year period, accounting for possible correlations between the repeated measurements with the use of mixed effects models. Finally, cardiorespiratory fitness was estimated with a submaximal cycle ergometer test, minimizing the risk of exposure misclassification, while HRQoL was measured with a widely used questionnaire [38], which allowed comparability with previous studies.

Nevertheless, there are also some limitations that need to be considered. A threat for the internal validity of this study is the high attrition rates. Especially when comparing the effects of the two interventions on HRQoL, it is likely that the estimates have been affected by selection bias. As the sensitivity analysis showed, higher proportion of the dropouts was from the mobile app group. Although there were no differences regarding changes in HRQoL from midpoint to the endpoint between those who attended the 1-year follow up and those who did not, it is possible that participants who did not attend the midpoint or endpoint follow up had small or negative changes. This might have led to the underestimation of the differences in the mean change of HRQoL scores between the two groups, having the supervised group as reference. Regarding the estimates of the association of changes in cardiorespiratory fitness with HRQoL, mixed effects regression models use the maximum likelihood method to handle missing data by estimating the mean values from the available data at each time point [52]. However, if the missing values are related to both HRQoL and cardiorespiratory fitness, these estimates might have been biased [52]. Since it is likely that those with lower HRQoL and cardiorespiratory fitness were lost to follow up, the estimated associations might have been diluted.

The generalizability of this study might have also been limited due to the high attrition rates and the exclusion criteria. Certain characteristics of the study population such as higher proportion of females than males, were representative of young adults with mobility disability living in Stockholm [3, 4]. However, the participants that were lost to follow up might have been different than those who remained to the study. These differences might have been related to the severity of mobility disability during the follow-up, possibly leading to higher retention rates among those with less severe symptoms. In addition, exclusion of people without access to a smartphone or that were not speaking Swedish during the enrollment, further restricts the generalizability of this study. Moreover, people enrolled in this study were recruited from rehabilitation centers, excluding individuals that did not seek care. Therefore, the results of this study could possibly be generalized to young adults with moderate mobility disability living in Stockholm or in other cities in Sweden, who would seek care in rehabilitation centers.

Further limitations of this study lay on the interpretation of the results. More specifically the sample size of the study was probably not adequate to detect differences in the effects of the mobile app compared to the supervised program on HRQoL, since it was calculated based on the primary outcome. However, equivalence of these effects cannot be implied, as such inference would require a larger sample size [53]. Furthermore, the nature of the data does not allow the inference of causal effects of cardiorespiratory fitness on HRQoL. Although according to theory this relationship could be very likely [54], the possibility of inverse causality cannot be excluded. Therefore, it could also be possible that HRQoL affects cardiorespiratory fitness.

More evidence is needed regarding the effects of mobile app training on HRQoL in adults with mobility disability. Future research could include larger samples and compare these effects not only with supervised training but also with usual treatment. In addition, further studies in young adults with mobility disability are needed in order to replicate the associations between changes in cardiorespiratory fitness and HRQoL, to investigate the existence of a causal relationship, and to better understand the role of supervised training in this association.