Effects of a 16-week multimodal exercise program on activities of daily living in institutionalized individuals with dementia

The multicenter randomized controlled trial included baseline and post-intervention assessments of cognition function, motor performance and ADL. Participants were allocated to the IG or CG using minimization software (MinimPy0.3 Saghaei & Saghaei, 2011). A power analysis with G*Power 3 (Version 3.1.9.2, two-factor analysis of variance with repeated measurements, two groups, and two measurements, α = 0.05, β = 0.80, η² = 0.01) resulted in a total sample size of 200 participants (Faul, Erdfelder, Lang, & Buchner, 2007). We expected that 35% of participants would drop out of the study or have missing data. Thus, 270 participants for the total sample are required. As attendance and adherence to exercise interventions are not well documented in previous studies (Forbes et al., 2015), and a higher dropout rate was assumed in the IG (e.g., due to motivational problems), the sample of the IG was doubled. Hence, we aimed at a total sample size of 405 participants. Participants were identified by healthcare professionals in the nursing homes. Prior to baseline assessments, informed consent was signed by potential participants themselves or their legal guardian. Inclusion criteria were (1) diagnosis of primary dementia or “suspected dementia” (i.e., without a confirmed diagnosis or awaiting further clinical evaluation) verified by general practitioners and/or based on ICD-10 criteria, (2) Mini Mental State Examination (MMSE) with mild to moderate status (MMSE: 10–24), (3) age > 65 years, (4) being able to walk for approximately 10 meters and (5) clearance by a general practitioner. Participants with secondary dementia, other severe cognitive impairments, neurological or severely acute diseases, severe motor impairments and/or no informed consent were excluded.

The 16-week dementia-specific MEP was implemented, with two sessions per week, each lasting 60 min. Every training session was designed as an imaginary journey with a ritualized arrival and departure element aiming to give participants orientation and familiarization. The mere exercise time included a combination of motor and cognitive tasks and took about 45 min. In detail, strength, balance, endurance and flexibility tasks were performed at various durations and with medium to submaximal intensity. Small training devices, such as dumbbells (1–4.4 Ibs), sandbags (~ 1 Ibs), skipping ropes or pool noodles where integrated into the exercises. Furthermore, different cognitive tasks were embedded in the MEP to stimulate cognitive functions, i.e., memory (e.g., remembering the destination of the last imaginary journey), attention (e.g., remembering a certain sequence of numbers), language (e.g., naming animals), and executive function (e.g., learning to perform according to visual or acoustic signals). Throughout the 16-week MEP, a progression of intensity of motor and cognitive exercises were planned and carried out, e.g., by increasing difficulty level or the number of exercise repetitions. Exercise intensity followed a predefined progression plan throughout the intervention period and was supervised by experienced instructors. Examples of different motor and cognitive tasks and their progression are presented in the appendix (Supplementary Table 1). In order to achieve a high degree of standardization, all instructors received a detailed manual of the MEP and underwent a special training program concerning the content of the MEP as well as the characteristics of the study sample. Furthermore, only instructors with theoretical and practical knowledge in sport science guided the MEP during the intervention study.

ADL (primary outcome measure) were assessed by a proxy-based questionnaire, the Barthel Index (BI; Mahoney & Barthel, 1965) and two performance-based assessments, i.e., the Erlangen Test of Activities of Daily Living (E-ADL; Graessel et al., 2009) and the 7‑item Physical Performance Test (PPT‑7; Reuben & Siu, 1990). The BI is a widely used questionnaire in geriatric care to assess functioning of older individuals in daily life. Usually, it is completed by caregivers and contains ten items: feeding, moving from a wheelchair to bed and return, personal toilet, getting on and off toilet, bathing, walking on level surface, ascend and descend stairs, dressing, controlling bowels and controlling bladder. The BI sum score ranges from 0 (totally dependent) to 100 (independent) (Mahoney & Barthel, 1965). The PPT‑7 contains seven activities of everyday life: 50-feet walk, putting on and removing a jacket, simulated eating, writing a sentence, putting a book into a shelf, a 360° turn, and picking up a penny from the floor. The maximum attainable score of the PPT‑7 is 28, with a higher score indicating a better ADL performance (Reuben & Siu, 1990). The PPT‑7 was found to be feasible and reliable in IWD (Farrell, Rutt, Lusardi, & Williams, 2010). The E‑ADL is the only ADL assessment that was specifically developed for IWD. It is performance-based and comprises five typical routines of daily life: cutting bread, pouring a drink, opening a cupboard, washing hands and tying a bow. The total score of the E‑ADL ranges from 0–30, with a higher score representing better results. The E‑ADL has good validity and test-retest reliability in IWD (Graessel et al., 2009).

For the explorative analysis of responder groups and underlying functions of ADL performance, we defined cognitive function and motor performance as secondary outcomes. We assessed overall cognition with the Mini Mental State Examination (MMSE; Folstein, Folstein, & McHugh, 1975), learning and memory with Digit Span forward and backward (Wilde, Strauss, & Tulsky, 2004), processing speed with the Trail Making Test Part A (TMT‑A; Reitan, 1958), and visual spatial function with the Clock Drawing Test (CDT; Shulman, Shedletsky, & Silver, 1986). To assess motor performance, we used the modified 30-second chair stand test (30s CST) for lower limb strength (Blankevoort, van Heuvelen, & Scherder, 2013), the Short Physical Performance Battery (SPPB; Guralnik et al., 1994) for lower limb function, Frailty and Injuries: Cooperative Studies of Intervention Techniques (FICSIT-4) for balance (Rossiter-Fornoff, Wolf, Wolfson, & Buchner, 1995), and Timed-Up and Go (TUG; Podsiadlo & Richardson, 1991) as well as a 6-meter walk test (6MWT; Graham, Ostir, Kuo, Fisher, & Ottenbacher, 2008) for mobility. The reader is referred to the study protocol (Trautwein et al., 2017) for a more detailed description of all motor and cognitive assessments. Feasibility of the motor assessments were discussed with an international expert panel before the conduct of this study. Furthermore, a standardized testing procedure was developed by the expert panel (Trautwein et al., 2019). Subsequently, the applied test battery has been checked for feasibility in IWD.

Statistical analysis was done with SPSS 27.0 (IBM Corporation, Armonk, NY, USA) for Windows with alpha (α) set a priori at 0.05 for all tests. To investigate the effect of the MEP on ADL performance between IG and CG, we using two different approaches (per protocol, intention to treat). This approach was also addressed but not further specified within the study protocol (Trautwein et al., 2017). First, a per protocol analysis was performed including all participants with sufficient attendance to the MEP (only IG: ≥ 75%, at least 24 of 32 trainings sessions) and a complete assessment (IG and CG: baseline, post) of at least one ADL assessment. In IWD, attendance to PE interventions and possible barriers (e.g., illness, previous exercise experience) vary widely (van der Wardt et al., 2017). A minimum of 75% for the per protocol analysis was defined on the basis of mean attendance rates of previously published studies with similar intervention period, duration, and frequency of training sessions, as well as sample sizes (Telenius, Engedal, & Bergland, 2013; Toots et al., 2016). Second, we imputed missing data of the primary outcomes for an intention-to-treat analysis using multiple imputation (fully conditional specification imputation method, ten imputations, ten iterations). Data of deceased participants was not imputed. ADL performance was considered as primary outcome and motor performance and cognitive function were considered as predictors. Furthermore, to ensure plausibility of data, other constraints like minimum and maximum values in each variable according to observed ranges, rounding according to original data, 100 maximal case draws and ten maximal parameter draws were defined. We considered pooled results as provided by SPSS. If SPSS did not support this pooling procedure for several statistical analyses, we reported ranges observed throughout the imputations.

Before carrying out statistical analyses, required assumptions were tested. Differences in baseline characteristics between IG and CG were calculated using unpaired T‑tests, Chi-square (Χ²) tests and Mann–Whitney U tests. Normally distributed data are presented as means (M) and standard deviations (SD). To investigate group-, time- and interaction effects of the MEP, two-way repeated measures analysis of variance (ANOVA) and paired T‑tests were calculated for primary outcomes. The calculated effect sizes are Cohen’s d for paired and unpaired T‑tests and η_p² for two-way repeated measures ANOVA.

Furthermore, we applied an explorative approach for the secondary aim, i.e., the responder-analysis. Previous studies often focused on main effects and group differences; however, the need to consider individual responses to exercise programs in IWD is currently discussed (Yu, Salisbury, & Mathiason, 2021). We decided against a simple “responder” and “non-responder” division as proposed in a previously published article (Müllers, Taubert, & Müller, 2019), and additionally defined a group with maintaining ADL performance, which may be considered as partial success of physical exercise in IWD. To this end, we divided the IG of the per protocol sample into three groups based on distribution-based methods (i.e., information about the standard measurement error of the assessments within the per protocol analysis): positive‑R (positive change of 10% and more in ADL performance), negative‑R (negative change of 10% and more in ADL performance), and non‑R (change between positive and negative 10% in ADL performance). We compared baseline cognitive function and motor performance between these responder groups using Kruskall–Wallis test, and one-factor ANOVA. Dunn–Bonferroni tests and Tukey–Kramer post hoc tests were used for post hoc analysis. For the analysis of underlying mechanisms of ADL performance, multiple regression analysis with stepwise selection were used with BI, E‑ADL, and PPT‑7 as dependent variables, and the already mentioned motor and cognitive assessments as independent variables. The calculated effect size was f².

In all, 600 IWD were screened for eligibility of study participation between March 2015 and March 2017, and 319 persons were considered as suitable for the study: 201 participants were randomly allocated to the IG and 118 participants to the CG after baseline assessment. Please refer to Fig. 1 for a flow chart of participants. Dropout rate in both IG and CG was 8%.

Table 1 presents the baseline characteristics of the per protocol sample (n = 191). The mean age of the participants was 86 years (SD 6 years) and 86% of the sample was female. A mean MMSE value of 17 (SD 4) indicated moderate cognitive impairment at baseline. More than three quarters of the per protocol sample used walking aids (walker: 70%; walking sticks: 8%). Mean value on the CIRS morbidity index was 9 (SD 5) and the mean number of required medications was 7 (SD 4). In 22% of the per protocol sample, the intake of antidementia medication was confirmed. Statistically significant differences in baseline characteristics of the per protocol sample between IG and CG were observed for the use of walking aids (Χ²(2) = 6.254, p = 0.044), number of medications (t(148) = 2.622, p = 0.010), and body mass index (t(178) = 2.041, p = 0.043). Baseline characteristics of the intention-to-treat sample are provided in the Appendix (Supplementary Table 2).

A responder analysis in the intervention group of the per protocol sample showed that 20.2% (E-ADL), 30.7% (PPT-7), and 31.7% (BI) of participants of the IG responded positively to the MEP (Supplementary Table 4). The majority of participants of the IG showed no changes in the primary outcomes (50.0–68.3%), and between 19.3 and 27.0% had a decrease in their ADL performance that was greater than 10%. With regard to the Barthel Index, no negative-responders were observed (Supplementary Table 4). The comparison of baseline values in cognitive function and motor performance between different groups of ADL responders revealed statistically significant results for BI and E‑ADL performance (Table 4). With regard to the Barthel Index, positive‑R performance at baseline was significantly worse compared to non‑R for SPPB (p = 0.002, d = −0.993), 30-STS (p = 0.008, d = −0.786) and FICSIT‑4 (p = 0.036, r = −0.378). A one-way ANOVA revealed significant differences between E‑ADL responder groups for FICSIT‑4, and CDT. The post hoc analysis showed statistically significantly worse performance of positive-responders in FICSIT-4 at baseline, compared to non-responders (p = 0.012, MD = −1.146, CI₉₅ [−2.084 to 0.208]). For CDT, the post hoc analysis was not significant. We found no statistically significant differences between PPT‑7 responders for baseline motor and cognitive performance.

For baseline Barthel Index, weak to moderate correlations with baseline cognitive function and motor performance were found (|r| = 0.141–0.349, p < 0.05). In a multiple regression analysis model, baseline FICSIT‑4 and TUG were statistically significant regression coefficients (p < 0.05). This model explained 14.3% of variance in baseline BI (F(2, 152) = 13.819, p < 0.005). For baseline E‑ADL, weak correlations with baseline cognitive function and motor performance were found (|r| = 0.118–0.170, p < 0.05) and only baseline MMSE was a statistically significant coefficient in multiple regression analysis (p < 0.05). The model explained 2.4% of the variance in baseline E‑ADL performance (F(1, 197) = 5.843, p = 0.017). For baseline PPT‑7, weak to high correlations with baseline cognitive function and motor performance were found (|r| = 0.198–0.609, p < 0.05). In multiple regression analysis, a model with TUG, gait speed calculated from the 6MWT, FICSIT‑4, 30-STS and DS forward as coefficients explained 51.4% of the variance in PPT‑7 performance (F(5, 196) = 43.581, p < 0.005). Please refer to Table 5 for the results of the multiple regression analyses.

The primary aim of this paper was to investigate the influence of a 16-week multimodal exercise program on ADL performance in IWD. We did not observe any significant time-group effects (η_p²: 0.004–0.019) of the 16-week MEP on ADL performance in IWD. Therefore, we could not confirm our hypothesis that participants of the MEP improved their ADL performance compared to the CG. Previous studies that investigated the effects of PE on ADL performance in IWD living in nursing homes yielded heterogeneous results. While some studies did not find any significant effects (Henskens, Nauta, Drost, & Scherder, 2018; Lamb et al., 2018), others found a slowed deterioration of ADL compared to a control group (Bürge et al., 2017; Littbrand, Lundin-Olsson, Gustafson, & Rosendahl, 2009; Rolland et al., 2007; Toots et al., 2016) or positive effects of PE on ADL performance (Bossers et al., 2016). Nevertheless, all these studies differed in terms of sample characteristics, setting, intervention period and intervention content, hampering a comparison and a critical handling with these findings is thus recommended (Forbes et al., 2015). Studies that yielded positive effects had longer intervention periods, higher training frequency, smaller groups or one-to-one sessions, and/or an adaption of intensity of exercise content during the intervention. We could not reach our aim to increase intensity throughout the intervention, which may be a main reason for our non-significant results. Moreover, even though two training sessions are recommended for nursing home residents (de Souto Barreto et al., 2016), this was feasible for our sample but might not have been sufficient. Furthermore, the baseline differences in walking-aid use were not further addressed within our analysis, but may have influenced the intensity adaption throughout our intervention period. Instructing a group in which some IWD use walking aids and others do not may have complicated the implementation of intensity adaptions, especially with regard to ensuring the safety of the participants. These mentioned challenges underline the difficulty to adapt exercise intensities in a highly heterogeneous sample.

Another reason for our non-significant results may be the high variability of ADL performance that was also reported in a previous study (Bürge et al., 2017). Moreover, even though the BI, E‑ADL and PPT‑7 were used in other studies with IWD before, it must be noted that they may not be specific enough to detect subtle changes in ADL performance in IWD. The sensitivity of the BI and its objectivity in assessing ADL performance of IWD living in nursing homes must be seen critical (Yi et al., 2020). Nevertheless, we decided to use the BI as a proxy-based evaluation tool, and to include an external assessment in addition to the performance-based tests as they may have been influenced by daily form or mood of the participants. However, we are aware that the BI may not have captured the effects of our MEP. Moreover, the E‑ADL was rated as too easy for individuals with mild dementia (Luttenberger et al., 2012). Originally, the PPT‑7 was not developed for IWD and therefore may not be feasible to detect small changes from baseline to post measurement as in our study. Nevertheless, previous longitudinal studies showed that ADL performance of IWD residing in long-term care facilities deteriorates over time (Johansen et al., 2020). Therefore, a small improvement or even maintenance of ADL performance over time could be a valuable outcome.

Indeed, our explorative responder-analysis revealed between 20.2 and 31.7% positive-responders with regard to ADL performance in the IG. Nevertheless, the majority of the participants did not respond to the MEP (i.e., non‑R, between 50 and 68%). We observed worse motor performance (balance, mobility, and lower limb function) in positive‑R compared to non‑R. These results imply that the intensity of our MEP worked well for individuals with weak baseline performance. From a training science point of view, it is crucial to individually modify intensity of interventions in order to achieve an improvement (Bürge et al., 2017; Littbrand et al., 2009; Littbrand, Stenvall, & Rosendahl, 2011). Originally, our MEP was designed to be carried out with increases of intensity of PE to achieve adaptions in participants of the IG. Due to the high heterogeneity of our sample in terms of disease severity, age, and other personal characteristics, we had to intensify our safety efforts and the majority of exercises were carried out while participants were seated. This resulted in lower training stimuli and may be one explanation as to why less performance adaptions might have been achieved. This consideration is supported by the fact that previous research found positive effects of PE on ADL performance when PE was provided in one-to-one guided sessions (Bossers et al., 2016). Therefore, more person-centered approaches considering individual skills and impairments may be warranted (Prizer & Zimmerman, 2018) and an individualization of instructions depending on the degree of cognitive impairment may be also useful. These factors underline the need for individualized MEPs tailored to the specific needs of an IWD in order to impact physical performance and cognitive function. Indeed, the concept of individualized medicine which has already become popular in prevention and treatment of Alzheimer’s disease (Hampel et al., 2017) is also transferable to the design and conduct of individualized exercise programs for IWD (Müllers et al., 2019). Of note, the feasibility of individualized exercise programs for IWD in nursing homes has to be discussed, as time available for PE programs in geriatric care settings is often limited. Organizational and structural aspects from both nursing homes and health care systems could support and facilitate the implementation of PE interventions in nursing homes (e.g., by engaging volunteers) (de Souto Barreto et al., 2016). While previous studies have shown the cost-effectiveness of PE interventions for IWD in community settings (Nickel, Barth, & Kolominsky-Rabas, 2018), the evaluation of costs and personal resources in nursing homes is lacking to date. One may only speculate that individualized PE interventions may require more personal resources at first, but save financial and personal resources in the long-term if they are effective. Mobile health applications may therefore be an effective and efficient possible solution in the implementation of individualized exercise programs in IWD (Barisch-Fritz et al., 2020). In detail, they may represent an easy to administer tool that could support nursing home staff, while conducting PE interventions with information on exercises, training plans, possible risk factors or required training material. Furthermore, mobile health applications offer the opportunity to individually adapt PE programs for example by integrating data-based or artificial intelligence algorithms (Helbostad et al., 2017). Of note, the monitoring of exercise intensity may be a further advantage of mobile health applications. Previous research used and recommended a combination of heart-rate monitoring and rating of perceived exertion (Sanders et al., 2020), but this was mainly done by research assistants and may be too time-consuming for nursing home staff. For a feasible and applicable monitoring of exercise intensity in IWD, a protocol of exercise repetitions or exercise time, as well as externally rated exertion by nursing home staff could be implemented in a mobile health application. Despite all these advantages and possibilities of mobile health interventions, studies examining their usability and feasibility of in nursing homes are sparse (Barisch-Fritz et al., 2020).

Our MEP was originally designed to primarily improve cognitive function and motor performance (Barisch-Fritz et al., 2021; Trautwein et al., 2020), and to address ADL performance indirectly. PE interventions which include ADL-specific tasks and take into account the complex requirements of ADL performance could be more effective. Therefore, we investigated the underlying mechanisms of ADL performance using multiple regression analysis. Explained variance ranged from 2.4% (E-ADL) to 51.4% (PPT-7) with statistically significant coefficients for both cognitive function and motor performance. The small explanation of variance in E‑ADL was expected, as the test contains tasks for upper extremities that were not captured with our assessment battery (Graessel et al., 2009). Nevertheless, our results may contribute to a deeper understanding of different sub-dimensions of ADL performance. If relevant cognitive and motor functions are carefully selected for the conceptualization of PE interventions, potential transfer effects on ADL performance could be more beneficial (Hagovska & Nagyova, 2017). Our results showed that assessments for balance and mobility explained the variance in the BI. For the performance of the PPT‑7, walking speed, lower limb strength, and memory were additionally important. These results are partly in line with previous studies (Garcia-Pinillos, Cozar-Barba, Munoz-Jimenez, Soto-Hermoso, & Latorre-Roman, 2016; Portegijs, Rantakokko, Viljanen, Sipilä, & Rantanen, 2016). However, beyond cognitive function and motor performance, other factors like having depression or feeling pain during specific tasks of daily life (e.g., sit-to-stand transfers) may influence ADL performance in IWD (Mlinac & Feng, 2016) and should be considered in the conceptualization of PE interventions.

The strengths of our study are the high methodological quality and the large sample size. Furthermore, we implemented a MEP that was based on theoretical considerations and proved to be feasible conducted in a sample of IWD living in nursing homes within a pilot study. In addition, we designed the MEP to be highly suitable in everyday life of nursing homes. Our previously assumed dropout rate was lower than expected which may be an indicator for the acceptability of the MEP in the participating nursing homes. Indeed, after the conclusion of our intervention study, many of the 36 participating nursing homes continued the MEP and we received positive feedback from nursing home staff and participants. Furthermore, nursing staff received special training to continue the MEP following the intervention study. This training included information on how to adapt intensity level during exercise sessions when needed. Another strength of the study was the comprehensive acquisition of ADL performance using one proxy-based questionnaire, and two performance-based assessments.

Nevertheless, our study has some limitations that must be considered when interpreting the results. First, we did not reach the intended sample size of 405 IWD because the number of participants that did not fulfill our inclusion criteria was higher than expected. However, a sensitivity analysis (G*Power 3, Version 3.1.9.2) showed that we were still able to detect small to medium effects. Second, our intervention program was initially designed with adjustments of exercise intensity during the 16-week intervention period. However, due to the high heterogeneity of individual characteristics of our study participants such as age or disease severity, we had to increase our safety efforts and could possibly not achieve an adequate training stimulus for all participants. Moreover, the theoretical recommendation of two training sessions per week (de Souto Barreto et al., 2016) was only partially feasible within our study. Hence, we strongly recommend including strategies to support adherence as proposed by van der Wardt et al. (2017). Another factor that may limit our findings is that we did not control for any other interventions that might have taken place during the conduct of our study in the IG no CG. Therefore, we cannot rule out that other social or therapeutic interventions may have biased our results. Another limitation is related to the number of incomplete data sets which reduced the sample size for statistical analysis in the per protocol analysis. Especially the BI as a proxy-based measurement filled in by nursing home staff was less often completed than the other assessments carried out by our study staff. Other reasons for incomplete data sets may be daily form and mood, depressive symptoms, other severe impairments or schedule conflicts. As we did not assess depressive symptoms with an evaluated assessment tool, we were not able to make assumptions about the possible influence of depression on ADL performance. Finally, our study did not include a follow-up assessment or a long-term monitoring of the continuation of the MEP in the participating nursing homes, so we do not have any results about the long-term effects of our intervention.