The study has been approved by the Ethics Committee of the Karlsruhe Institute of Technology (Germany) and is registered in the German Register of Clinical Trials (drks.de. Identifier: DRKS00024069). The evaluation of the usability (primary outcome) and trends toward the effectiveness (secondary outcome) of the InCoPE-App is based on a field intervention with a cluster-randomized controlled trial. Blinding in terms of group allocation was not applicable for IWD, nursing assistants, or study personnel. The intervention group (IG) received an 18-week ICT-based intervention (for more details, refer to Section 2.2) along with conventional treatment such as individualized medication, standard care, or therapeutic applications.

The ICT-based intervention was delivered by nursing assistants via the InCoPE-App. Testing of physical performance of IWD was repeated every 3 weeks. Pre- and post-intervention assessments were conducted for global cognition. Participants of the control group (CG) received their conventional treatment and, similar to participants of the IG, underwent tests to assess physical performance every 3 weeks.

Two different study samples were recruited to evaluate the usability as well as the effectiveness of the InCoPE-App. The first sample (for usability purposes) consisted of nursing assistants which are considered the primary users of the InCoPE-App. This sample was recruited with the assistance of the heads of nursing homes in Southwestern Germany who suggested potentially interested nursing assistants. Inclusion criteria were as follows: 1) willingness to participate in the preparatory training with the InCoPE-App, 2) a high level of fluency in German language, and 3) employment at the nursing home for at least 18 weeks or more. Employees who were externally hired (e.g., physical therapists) were excluded from the study. All nursing assistants received detailed information about the study design and confirmed their willingness to participate by signing an informed consent letter.

The second sample (for effectiveness purposes) included individuals with primary dementia living in nursing homes. Participants of this sample were recommended by nursing assistants based on the following inclusion criteria: 1) Alzheimer’s disease, vascular dementia, or other primary types of dementia; 2) mild-to-moderate stage of dementia (Mini-mental State Examination, MMSE: 10–24); 3) aged 65 years or older; and 4) walking ability of at least 10 m with or without walking aid. Exclusion criteria were as follows: 1) secondary dementia, 2) other severe cognitive impairments, 3) other severe neurological conditions, 4) other severe acute diseases, and 5) severe motor impairments. Prior to the study, written informed consent of participants or their legal guardians was obtained. Prior to the intervention, a medical clearance certificate, containing information about the dementia diagnosis and other pertinent information such as medication intake and comorbidities, was obtained from a participant’s general practitioner.

The 18-week intervention was delivered via the tablet-based (Android) InCoPE-App, which was developed by applying a user-centered approach on the base of earlier work of this research group (Barisch-Fritz et al., 2020). Briefly, the main algorithm implemented in the InCoPE-App to individualize cognitive and physical tasks is based on an unsupervised machine learning approach basically calculated through a cluster analysis of 230 IWD (Barisch-Fritz et al., 2019). The sample was drawn out from IWDs who participated in a previous RCT (Trautwein et al., 2017). The results of the cluster analysis identified four homogeneous subgroups: participants with 1) below-average physical and cognitive performance, 2) average cognitive and above-average physical performance, 3) above-average cognitive and below-average physical performance, and 4) above-average cognitive and physical performance. The algorithm of the InCoPE-App enables the allocation of each IWD based on their cognitive and physical performance data (which are also assessed through the InCoPE-App) to one of these four clusters. The approach contains the calculation of probabilities of cluster affiliation through multinomial logistic regression analysis. The individualized composition of exercises is based on about 700 described exercises which are categorized by means of the primary aim and intensity level of each exercise. The training routine for each individual is generated by the InCoPE-App through cluster allocation for the subsequent 3 weeks.

The contents and functions of the InCoPE-App can be summarized as follows: The first function is to assess cognitive and physical performance of IWD. The second is to design, guide, and monitor an individualized program by combining cognitive and physical exercises depending on an IWD’s current physical and cognitive performance level. These automated functions embedded in the InCoPE-App enable nursing assistants who lack advanced knowledge in sports science and/or psychology to test and individually train IWD. The InCoPE-App contains detailed visual materials and annotations to offer visual support during the execution of the proposed tests and exercises. Prior to the start of the ICT-based intervention, nursing assistants were trained in the correct use of the InCoPE-App. This training comprised online modules for self-study (e.g., theoretical basics of adapted physical activity with IWD, objective testing of physical and cognitive performance in IWD, and first steps with the InCoPE-App) and two face-to-face training sessions on the use of the InCoPE-App.

The tests assessing physical and cognitive performance included in the InCoPE-App are based on recommendations from previous studies and were previously administered to IWD (Bossers et al., 2016; Trautwein et al., 2019a) (refer to Section 2.4 for details). The initial testing of physical and cognitive performance is critically important in order to assign the IWD to one of the four training clusters as described above. Physical performance assessments were repeated every 3 weeks during the 18-week intervention period to allow the adjustment of duration, intensity, and contents of physical and cognitive exercises via InCoPE-App. We anticipate that the desired physical and cognitive adaptation to the exercise program will be achieved by gradually increasing exercise difficulty and intensity (e.g., number of repetitions).

One-to-one training sessions between nursing assistant and each IG participant were conducted twice a week. Each 60-min session comprised of the following: 1) a ritualized warm-up, 2) a first set of exercises, 3) a short break with ritualized balance exercises, 4) a second combination of exercises, and 5) a cool down at the end. Actual training duration was about 40 min. The ritualization of training sequences is important to give IWD a sense of security and familiarity (Schick 2015). The warm-up and cool down included walking exercises for the promotion of endurance as well as general mobilization tasks. The main training program took place in the first and second set of exercises. Within these two sets of exercises, strength, coordination, and balance were addressed. Most exercises were combined with a cognitive task (e.g., counting repetitions while doing strength exercises or recall previously learned commands while playing ball). Exercise composition was defined based on the cluster allocation of each individual. For example, an IWD in cluster 1 received exercises focusing on balance, mobility, and strength, as well as cognitive stimulation.

With regard to the examination of the usability of the InCoPE-App, we conducted descriptive analyses. Results from PSSUQ, ISONORM 9241/110-S, and AttrakDiff2 are presented in means, standard deviations (SD), and 95% confidence intervals and compared to published reference values. Logging events for feasibility are also descriptively presented with cumulative numbers and percentages as well as means of adherence to the IG.

The trends toward the effectiveness of the ICT-based individualized intervention were descriptively described using means ± SD. The differences within the groups were tested using paired sample t-test to compare physical and cognitive performance at pre- and post-assessments. To get a deeper insight into the trends toward the effectiveness, a single-subject research design was applied. We conducted a linear regression analysis over the seven assessment times of the four physical performance variables (Johnson and Ottenbacher 1991). The results are reported as regression coefficients (β), the standard error of estimates (SE), and the coefficient of determination (R²). The trends for each individual were coded based on the regression coefficient. A positive trend (coded with +) was set at β ≥ 0.1 for FICSIT-4 and m30CST, as well as β ≤ -0.1 for 6MWT and TUG. No trend (coded with 0) was set at −0.1< β > 0.1. A negative trend (coded with −) was set for β ≤ −0.1 for FICSIT-4 and m30CST, as well as β ≥ 0.1 for 6MWT and TUG.

The first hypothesis addressed the usability and feasibility of the InCoPE-App. The usability was evaluated with three questionnaires focusing on different aspects of user experience. In conclusion, all results were above reference values (which are available for PSSUQ and ISONORM 9241/110-S). Thus, the user experiences of nursing assistants with the InCoPE-App can be regarded as exceptionally satisfactory.

The findings for the PSSUQ can be interpreted by using the means determined by Sauro and Lewis (2016) involving 21 studies and 210 participants (i.e., the lower the score, the better the user experience). The mean values from our study ranged from 1.99 to 2.31 and thus, are better than the means presented by Sauro and Lewis (2016). This instrument has been shown to be effective in small samples and is also highly sensitive in evaluating user experience (Tullis and Stetson 2004). Thus, the InCoPE-App performs well with regard to “information quality,” “interface quality,” and “system usefulness.”

Findings of the ISONORM 9241/110-S questionnaire that operationalizes the seven criteria of the DIN EN ISO 9241-110 (Prümper 1997) were also rated as high (overall mean 1.90 ± 0.88). The mean values ranged from 1.60 to 2.23 and are thus all above the practical benchmark which serves as minimum criterion for a good software. The seven criteria of the DIN EN ISO 9241-110 are principles of interaction regarding the ergonomics of the human–system interaction. Our findings showed that these principles of interaction are well implemented in the InCoPE-App, and no further improvement or adjustment of the App in this regard is needed.

The findings of the perceived hedonic and pragmatic quality of the InCoPE-App were highest rated for the global attractiveness which indicates that the nursing assistants had good overall experiences with the InCoPE-App. Compared to the highest possible rating of three points, the ATT mean of 1.86 ± 1.01 is rather high. The pragmatic quality was also rated as high (mean 1.79 ± 1.03), thus indicating that the usefulness of the InCoPE-App as well as embedded functions were perceived as valuable and applicable in nursing assistants’ daily routines. These findings are in line with the category “system usefulness” of the PSSUQ, which was also highly rated. The values for identity (mean 1.81 ± 0.79) of the perspective hedonic quality were also rated as high. Thus, nursing assistants may have established a closer relationship with the trained IWD through using the InCoPE-App. This might be valuable and can be regarded as a desired working condition, particularly in light of rather precarious working situations in nursing homes (Kliner et al., 2017). The values of the second subdimension, i.e., hedonic quality stimulation, are rated lower (mean 1.37 ± 0.69) compared to the other dimensions. This perspective implies the need for improving own knowledge and proficiency (Hassenzahl et al., 2003). The lower rating on this scale may imply that the nursing assistants felt that they did not experience self-improvement in the field of physical activity promotion or instruction. Further examination of this perceived lack of self-improvement is needed. The rating for the perceived hedonic and pragmatic quality and attractiveness were high compared to other studies using this instrument (Trahms et al., 2017; Mayer et al., 2019; Müller et al., 2021). However, the direct comparison of these findings to previous observations is difficult due to the uniqueness of the InCoPE-App.

In general, the nursing assistants, i.e., users, regarded the InCoPE-App as useful and attractive with a well-designed interaction between human being and technology. Furthermore, they experienced the interaction with the residents guided by the InCoPE-App as valuable. These positive results in user experience did not align with our results with regard to the feasibility. The analyzed logging events focusing on the completed training sessions and physical and cognitive performance assessments of IWD were rather low. Of 30 included IWD, only 50% of the CG and 72% of the IG performed the post-intervention assessments. The overall dropout rate of 36.7% from pre- to post-intervention assessment is very high compared to other studies that conducted physical exercise interventions (Eggermont et al., 2006; Puente-González et al., 2021). Multidomain supervised interventions are generally regarded as feasible and safe with low dropout rates (Park et al., 2019). However, in samples of institutionalized IWD, the dropout rates are often higher (Eggermont et al., 2006). For example, Puente-González et al. (2021) had a dropout rate after 3 months of 19% and after 6 months of 32%. They conducted a multidomain exercise program for IWD delivered in a group setting. In our study, one potential explanation for the relatively high drop-out rate may be the COVID-19 pandemic and related restrictions. Our study was planned and conducted during and between two long periods of lockdowns, in which especially nursing homes were severely affected by contact restrictions. Two nursing homes terminated their participation in our study due to a high number of COVID-19 cases among employees and residents. However, one aim of the ICT-based intervention was to make nursing homes independent from outside exercise and fitness experts; thus, we developed an ICT-based app to enable nursing assistants to test and train residents individually. This approach would also allow them to comply with hygiene and safety measures related to COVID-19. In light of our results, this aim may not have been sufficiently achieved. We can hypothesize that not the possibility to deliver the ICT-based intervention during the lockdowns presented a problem to nursing assistants, but rather the already precarious employment conditions in this setting which were further exacerbated by additional workload and work absences during the pandemic (Wang et al., 2020; Veronese and Barbagallo 2021). However, this needs to be further examined and additional adherence support strategies such as from van der Wardt et al. (2017) need to be considered.

In contrast, the overall adherence of 52.3% in our study may appear lower compared to previously reported mean adherence of 70% in IWD and individuals with mild cognitive impairment (Di Lorito et al., 2020). However, considering the reasons for dropout in our research (e.g., termination of the intervention by the nursing home itself or illness), the remaining sample that participated in more than 25% of the training sessions had an adherence of 85.3%. This means that, on average, the participants fulfilled 30 of 36 possible training sessions.

The findings on the effectiveness of the ICT-based intervention on physical and cognitive performance were not statistically significant when comparing the pre- and post-intervention assessments. The only statistically significant finding was an improvement of balance in the CG. The CG continued their conventional treatment during intervention phase which was not further documented. However, it is possible that this conventional treatment also contained movement-related treatments like physiotherapy or fall prevention, where balance performance may have been addressed. Thus, these findings may be biased. More importantly, the sample size of CG was very small and cannot be adequately interpreted based on inferential statistics. In addition, further studies in this setting should focus on the documentation of the conventional treatment.

With regard to the second hypothesis that postulated positive trends after the 18-week ICT-based intervention, we examined the findings on physical performance by using a single-subject research design over the treatment phase. The fitting of trendlines was used before to objectify courses of changes in single subjects (Johnson and Ottenbacher 1991; Zhan and Ottenbacher 2001). This procedure was chosen as the sample size in our study was too low for applying inferential statistics. The descriptive comparison of the means from pre- to post-intervention assessments revealed that there is a positive shift in the mean cognitive and physical performance (except for 6MWT) in not only the IG but also the CG. Physical performance of the CG was also tested every 3 weeks while receiving conventional treatment. It can be speculated that there was a learning effect that could also have been overlaid by additional conventional treatment.

The analysis of single subjects of the IG over the seven assessment times showed more positive than negative trends. The approach to insert a trendline based on linear regression helps to assess and quantify the trend (Johnson and Ottenbacher 1991). However, the fit of the trendline is not in all cases acceptable regarding the small R² values. This is not surprising considering the variation in the intra-subject measurement results. The variations indicate that there might be several disturbing influences, which is in line with the partly low absolute reliability of the physical performance tests in the sample of IWD (Trautwein et al., 2019b). One potential explanation may be that some assessment tools were not specifically developed for IWD and thus might be impacted by the cognitive disorder (Trautwein et al., 2019b). Another explanation might be that strong daily fluctuations, e.g., with regard to mood or motivation, exist in this target group (Cipriani et al., 2020). Thus, further research and different analysis approaches, for example, on the base of exponential smoothing, are needed to take daily fluctuations in IWD into account.

The main limitation of this study was the small sample size to examine the effectiveness of the ICT-based intervention. The main reason for the small sample is that the study was conducted during the challenging time of the COVID-19 pandemic. Furthermore, the drastically reduced number of IWD in the CG did not allow for a direct comparison of the IG and CG. Thus, our results should be considered preliminary and need to be verified through a randomized controlled trial with a sufficient number of participants as calculated by a power analysis. Furthermore, a following study should integrate information on the amount and content of the conventional treatment for both the IG and the CG.

Another limitation pertains to the assessment tools used for the sample of IWD. This might explain the high intra-subject variation for the results on the physical performance. The partly low absolute reliability of these tools should be further analyzed, and specific and meaningful assessment tools for IWD to assess physical performance should be developed (Trautwein et al., 2019b). Further reasons for the intra-subject variation may be daily fluctuations which should also be examined in future studies. For example, we also collected information about the state of daily form (e.g., with regard to hunger, thirst, sleep, etc.) which was assessed before each training session. The analysis of this information might be a first step to understanding the potential impact of daily fluctuations on physical and cognitive performance in IWD. Other potential mediating or moderating variables that may have affected our study findings are, e.g., frailty, fear of falling, physical fitness readiness, or body composition. In particular, nutrition is a relevant factor when considering cognitive performance (Masoomi et al., 2020; Dominguez et al., 2021). We will document these factors in subsequent studies to analyze these random effects by applying linear mixed models and possibly identify variables that are additionally relevant for the individualization of multidomain exercise interventions.

In general, our individualization approach based on a one-to-one training between nursing assistant and IWD can be regarded critically. On the one hand, the suitability of the daily use of the ICT-based InCoPE-app in nursing homes, which often have a shortage of employees, is imminent and must be critically discussed. Further research in the field of structure evaluation is thus needed. On the other hand, multidomain exercise programs delivered in a group setting seem to have additional benefits for cognitive performance (Kelly et al., 2014). This controversy might be solved by a larger study using the InCoPE-app where more IWD will be tested and trained and thus, small groups for each cluster can be formed. Within these similar clusters, the group setting effects may even be higher as all participants have a similar physical and cognitive performance level. In addition to these considerations, the InCoPE-App might be extended in several ways. One option is to modify the app in a way that will increase the interaction between nursing assistants, i.e., users with IWD. Another option is to implement film material as additional content, which may allow for the InCoPE-App to be used in home care, e.g., by family caregivers. A large number of physical exercises can be carried out with little equipment and limited space at low cost as demonstrated by the vast number of YouTube videos promoting physical exercise in a home setting during COVID-19 (Vancini et al., 2022).