ND progress through several stages in several years, and ultimately lead to dementia, a decline in mental ability severe enough to interfere with activities of daily living. Dementia can result from different causes, the most common being AD. It is often preceded by a pre-dementia stage, known as mild cognitive impairment (MCI), characterized by a cognitive decline greater than expected for an individual’s age, which, however, does not interfere notably with activities of daily living (Petersen et al., 1997). Depending on the etiology and the disease’s stage, dementia can be characterized by cognitive, behavioral, motor, and/or functional symptoms. The biological processes involved in ND are very heterogeneous, and include neuroinflammation, gliosis, synaptic loss, neurodegeneration, cerebral atrophy, and alterations of the blood–brain barrier permeability (Raz et al., 2016). These molecular alterations are due, among others, to alterations in the bioenergy metabolism, to hypoperfusion/hypoxia, and to dysfunctions of the cerebrovascular hemodynamic. From a therapeutic point of view, much research aims to modify the course of the disease or to reduce the impact of the clinical symptoms. Social interaction, physical and cognitive activities, and motivation can have a major impact on the disease progression. Hence, non-pharmacological approaches targeting people’s lifestyle are of particular interest.

Boosted by the publication of a Nature letter showing that video game training can enhance cognitive control in older adults (Anguera et al., 2013), there is now a growing interest in developing SG specifically adapted to people with ND. Evidence is accumulating showing that video-games and VR-applications can successfully be employed for early detection and monitoring of physical and cognitive impairment (e.g., Tarnanas et al., 2013; Aalbers et al., 2016; Negut et al., 2016; Zygouris et al., 2017), but also to train physical and cognitive abilities in people with AD, MCI, and related disorders. In the field of training, most of the research work so far has been conducted employing commercial video-games and cognitive games (such as Wii Fit and Wii Sport, Lumosity) designed for an entertaining purpose, and with a ‘typical’ healthy user in mind. In their review on the use of video-games in people with dementia-related disorders, McCallum and Boletsis (2013) showed that: (a) Exergames, i.e., games that promote physical condition and/or aerobic fitness can positively affect several areas of mobility in participants with mild AD and MCI, such as balance and gait (Padala et al., 2012), and voluntary motor control (Legouverneur et al., 2011); (b) cognitive games can improve cognitive functions, such as attention and memory (Stavros et al., 2010; Weybright et al., 2010) and visuo-spatial abilities (Yamaguchi et al., 2011); (c) physical and cognitive games can have a positive impact on social and emotional functions, for instance they can improve the mood and increase positive affect and sociability (Weybright et al., 2010; Boulay et al., 2011) and reduce depression (Férnandez-Calvo et al., 2011). As the field is young, less evidence is available on the efficacy of SG specifically designed for the training of people with ND. However, evidence collected in three studies suggests that SG and Serious Exergames are acceptable and motivating even for people with dementia. Beneviste et al. (2012) designed a SG based on musico-therapy targeting patients with AD and mild to moderate dementia aiming at improving patients’ self-image and to reduce their behavioral symptoms. Players use Wiimotes with WiiPistol to improvise or play predefined songs on a virtual keyboard. Results of a 2-month usability study conducted on seven AD patients confirmed that the SG was usable by AD patients despite their motor and cognitive impairments, and that patients were overall very satisfied with the game and expressed a desire to repeat the experience. Manera et al. (2015) found similar results with ‘Kitchen and cooking,’ a SG designed to train executive functions and praxis in people with MCI and early AD. The results of a 4-week feasibility study conducted on 21 participants (with MCI or early to moderate AD) confirmed that the game was acceptable and usable both at home and in a nursing home setting, and that patients were able to improve their game performance over the training, as confirmed by the fact that they became faster in the game activities. Finally, Ben-Sadoun et al. (2016) showed that X-Torp, a Serious Exergame designed to train physical, cognitive and social functions, was well accepted by people with dementia and MCI (N = 10) and healthy elderly controls (N = 8), and that during a 1-month trial participants experienced mainly positive emotions, improved their cardio-respiratory fitness, and were able to progress in the cognitive games scenarios. A summary of the training features of these three feasibility/pilot studies in reported in Table 1.

Data on the frequency of use employed in the three feasibility studies is converging with the recommendations reported by Robert et al. (2014). A consensus group including expert of ND (health domain) and of VG/SG design (ICT domain) and commercialization (business domain) met in a standalone meeting, and were asked to respond to questions concerning the ideal clinical population and target of SG, their frequency of use, and their context of use (with whom and where). Results (reported in Figure 1) suggested that: (a) SG were considered between ‘adapted’ and ‘very adapted’ to people with MCI, and between ‘not very adapted’ and ‘adapted’ to people with AD and related disorders; (b) SG was rated between ‘adapted’ and ‘very adapted’ for assessment, stimulation and rehabilitation of people with AD and related disorders, to train family caregivers and healthcare professionals, with the best rated target being stimulation; clinical targets rated between ‘adapted’ and ‘very adapted’ included physical, cognitive and social stimulation, as well as apathy (while agitation and improvements in activities of daily living were considered between ‘not very adapted’ and ‘adapted’); (c) SG should be employed regularly (‘everyday’ and ‘once a week’ were both rated between ‘adapted’ and ‘very adapted,’ while ‘on request’ was rated between ‘not very adapted’ and ‘adapted’); (d) SG were rated between ‘adapted’ and ‘very adapted’ to be employed at home, in day centers and nursing homes, and between ‘not very adapted’ and ‘adapted’ to be employed in the hospital; (e) SG were rated between ‘adapted’ and ‘very adapted’ to be employed with someone (being either a therapist, a family caregiver or a professional caregiver), and between ‘not very adapted’ and ‘adapted’ to be employed alone (or with a robot).

In the present paper we aimed to update and refine these initial recommendations thanks to a Delphi expert panel. A number of methodological changes were performed compared to the 2014 consensus panel. First of all, the 2014 recommendations were collected in a single round, without following the classical Delphi methodology. Indeed in the standard Delphi method (Linstone and Turoff, 1975) experts are asked to answer questions in two (or more) rounds. After each round, a facilitator provides a summary of the experts’ responses, and encourages the experts to analyze, comment and (eventually) revise their earlier answers in light of the commentaries of other members of the panel. The recommendations reported in the present paper followed the classical Delphi method (see below). Second, in 2014 we asked participants to rate each question on a 4-point scale (‘not adapted at all,’ ‘not very adapted,’ ‘adapted’ and ‘very adapted’). Here, we selected instead a 5-point scale (‘not adapted at all,’ ‘not very adapted,’ ‘adapted,’ ‘very adapted’ and ‘completely adapted’) to improve the symmetry of this Likert-type scale. Third, in the 2014 study several questions collapsed persons with MCI and more advanced stages of AD (‘people with AD and related disorders’). As recommendations for these two groups may be quite different, in the present study we kept them as separated categories for all the questions. Forth, we collapsed in a single questions the ‘Where’ and ‘With whom’ questions (see below), in order to better explore the exact contributions of (and interactions between) these two factors. Finally, we added a number of response alternatives to several questions, in order to obtain more precise information (e.g., in the ‘When’ question, we employed 6 response alternatives instead of 3).

The expert panel (N = 23) included researchers and health care professionals working on autism and other neurological and developmental disorders (n = 6), neurodegenerative disorders (n = 10), or both neurodevelopmental and neurodegenerative disorders (n = 2); ICT engineers (n = 2); and game developers (n = 3).

Following the DELPHI method, a list of questions was sent to all participants a week before the meeting via web-survey. Who, Where, When, and What questions were used as guidelines to structure the survey. Specifically, participants were asked to respond to the following questions:

Participants were asked to rate each item on a 1–5 scale (1 = not adapted at all; 2 = not very adapted; 3 = adapted; 4 = very adapted; 5 = completely adapted).

Results are reported in Figure 2A. SG were rated between ‘very adapted’ and ‘completely adapted’ for people with MCI, and between ‘adapted’ and ‘very adapted’ to people with dementia. Repeated-measures ANOVA confirmed that SG were rated as more adapted to people with MCI compared to people with dementia [F_(1,21) = 16.87, p = 0.001], suggesting that SG are considered as more adapted to people with initial cognitive decline than to people which are already loosing autonomy in activities of daily living. However, SG are considered to be adaptable also to people with dementia.

Results are reported in Figure 2B. Participants reported that SG are between ‘very adapted’ and ‘completely adapted’ for assessment, to train physical and cognitive functions, improve wellbeing, and teach contents. Improving autonomy and favoring social exchanges were considered from ‘adapted’ to ‘very adapted.’ For people with dementia, SG were rated between ‘adapted’ and ‘very adapted’ for all the specified targets. This is in line with responses to the question ‘SG for whom,’ and it suggests that all these domains do represent useful targets for SG in this population. Repeated-measures ANOVAs (Bonferroni corrected) conducted to compare people with MCI and dementia for each category suggested that SG were considered as more adapted to people with MCI compared to people with dementia to train cognitive functions [F_(1,20) = 17.44, p < 0.001], to improve autonomy [F_(1,20) = 10.80, p = 0.004] and wellbeing [F_(1,21) = 9.32, p = 0.006] and to teach contents [F_(1,20) = 15.42, p = 0.001]. All the other contrasts did not reach statistical significance (p > 0.007).

Results for all these patient populations (Figure 2C) suggest that SG are mostly adapted (between ‘very adapted’ and ‘completely adapted’) to be employed with a trained caregiver, both in a home and in a clinical setting. The use of SG by patients alone was rated between ‘adapted’ and ‘very adapted’ for people with MCI, and between ‘not very adapted’ and ‘adapted’ for people with dementia. Repeated-measures ANOVA with Group (MCI vs. dementia), Where (home vs. clinical facility) and With whom (alone vs. with a trained caregiver) as within-subject factors confirmed a significant effect of With whom factor [F_(1,19) = 54.82, p < 0.001], with SG use with a caregiver as more adapted compared to SG use alone. The Where factor was not statistically significant [F_(1,19) = 1.30, p = 0.269], thus suggesting that SG are considered as equally adapted to be employed at home and in a clinical facility. A significant effect of Group was also found, with SG rated as more adapted to be employed with people with MCI compared to people with dementia in all the settings [F_(1,19) = 22.03, p < 0.001]. Interestingly, a significant interaction between Group and With whom factor was also found [F_(1,19) = 12.67, p = 0.002], suggesting that employing SG with a trained caregiver is especially important for people with dementia. No other 2-way or 3-way interaction reached statistical significance (all ps > 0.360).

Results (Figure 2D) suggest that all the game frequencies were rated between ‘adapted’ and ‘completely adapted’ for all conditions, and the mean ratings of the different game frequencies can be visually described as skewed Gaussian distributions. Repeated-measures ANOVA with Group (MCI vs. dementia) and Frequency (1, 2, 3, 4, 5 and 7 days per week) as within-subject factors revealed no significant main effect of Group [F_(1,18) = 3.45, p = 0.008], and Frequency [F_(5,90) = 1.72, p = 0.139], and no significant interaction between Group and Frequency [F_(5,90) = 0.60, p = 0.704]. Converging with descriptive data, the ANOVA’s contrast tests revealed an almost-significant quadratic contrast [F_(1,18) = 4.27, p = 0.053], suggesting that categories in the middle of the curve were rated as more adapted compared to extreme values (1 and 7 days per week). For participants with MCI and dementia, the highest scores were obtained for game frequencies from 2 to 4 days a week.

Serious Games were rated as more adapted to people with MCI compared to people with dementia. The results are converging with those reported in our previous recommendations (SG were considered between ‘adapted’ and ‘very adapted’ to people with MCI, and between ‘not very adapted’ and ‘adapted’ to people with AD and related disorders; Robert et al., 2014), and suggest that SG may be more useful for people with initial cognitive decline than to people which are already loosing autonomy in activities of daily living. This can be explained, on one hand, by the recognition that the cognitive decline associated to dementia (working memory, attention, etc.) makes it more challenging to design and employ SG in this population; and on the other hand, by the recognition that early interventions targeting initial cognitive decline are supposed to be more effective compared to late interventions (Barnett et al., 2014). Anyway, SG are considered as adaptable also to people with dementia. This is confirmed by recent studies showing that SG are usable in people with dementia both when played alone (e.g., Manera et al., 2015) and with a clinician (Ben-Sadoun et al., 2016). Ongoing studies are also exploring the efficacy of ICT-based devices (e.g., avatars, contextual helps) in supporting older adults with SG use, showing initial promising results.

Assessment, training and promoting wellbeing were all rated as good targets for SG in people with MCI and dementia. Similarly, in the 2014 recommendations, assessment, stimulation and rehabilitation were all rated as good targets for people with AD and related disorders. Favoring social exchanges was not considered as the best target because most of the existing SG for older adults are not social (i.e., they are designed for a single player). However, emerging SG are also targeting the social domain. Some of these SG are showing promising results (Ben-Sadoun et al., unpublished data). Training targeting the cognitive domain and teaching contents may be more suitable to people with MCI compared to people with dementia, as they require some intact learning abilities to be optimized. However, beyond the selection of a clinical target for each patient’s category, an important aspect to take into account is that the selection of a SG should be personalized, and based on extensive clinical assessments aiming at identifying primary and secondary targets in the cognitive, motivational and emotional domains for each patient (Mishra et al., 2016). The assessment can also help to define the main follow up parameters, and the kind of feedback needed by each patient based on his/her challenges. For all conditions, SG should ultimately aim at targeting improvements in daily activities (autonomy). In other words, improvements in game activities should generalize to untrained abilities (Anguera et al., 2013), and demonstrate an impact on real life. SG design principles, such as inclusion of long-term goals embedded in a cohesive narrative instruction, and of specific generalization activities (e.g., instructional supports), may be important for encouraging transfer of knowledge from the computer to in-person settings. However, even improving autonomy in the SG activity alone could have a positive impact on motivation and quality of life: indeed the need of autonomy is one of the main drivers of the intrinsic motivation to play videogames in younger people (Deci and Ryan, 2000; Przybylski et al., 2010). This does not necessarily extend to people with ND. Is the need of autonomy the main motivational driver also for older adults with cognitive impairment, or are there other needs (e.g., social satisfaction/recognition)? To advance the work in this area, more research should be devoted to the design of SG in these target populations (De Schutter and Vanden Abeele, 2015).

The experts suggested that most of the SG are more effective when the patient is accompanied by a trained caregiver (similarly, in the 2014 paper, SG were rated as more adapted for use with a family/professional caregiver, or a healthcare professional). This is consistent with the general recommendations on SG usability drafted by Alvarez et al. (2012) who suggested the importance of assisting the player to improve his/her game understanding (how to play?) and motivation (why to play?). The presence of a caregiver is considered as more important for people with dementia compared to people with MCI, due to their lower degree of autonomy. The presence of a trained person is important for different reasons, including: to motivate people to use the game; to help progressing in the game scenario (in case people get stuck), reminding them about game rules and commands; to make sure the SG are played safely, especially for physical SG; and to embed the SG in a positive social and emotional context. The fact that the presence of a caregiver represents a key element in SG adoption raises potential ecological and economical barriers to the use of SG outside the research and clinical context. Possible solutions include involving a family caregiver, organizing SG group sessions, or providing remote assistance. Avatars and other game assistance solutions may also be useful to promote independent SG use in people with cognitive impairment (Phan Tran et al., unpublished results).

Concerning the Where question, most of the SG are considered as useful to be delivered both at home and in clinical facilities. Clinical facilities have the advantage to allow a complete standardization of the training, and to provide a secure, controlled environment. The main problem, however, is represented by the travel time and costs. In order to improve study adherence, Ben-Sadoun et al. (2016) used a taxi to transport patients to the clinical facility, resulting in a 100% adherence to the training. This strategy was also used by Maillot et al. (2012) for some of elderly subjects which came to the clinical facility during their Exergame training study. Although this is feasible in the context of a clinical study, it would be important, in the long term, to bring SG in the patient’s home for the classical care. At present home-based training is still challenging due to technical issues (e.g., hardware setup, availability of an internet connection for data transfer), and to difficulties to monitor the training remotely, raising possible security problems. As technologies improve every day, well-designed feasibility studies in home-based setting are urgently needed to verify if home-based stimulation is safe and feasible, and for which patient populations. This is particularly relevant because new generations of older adults will be more and more used to employ ICT at home, thus potentially reducing usability problems.

Training frequencies between two and four times a week were rated as the most adapted for people with MCI and dementia. Evidence collected in recent studies on both cognitive and cognitive-physical training converges with this recommendation (e.g., Anguera et al., 2013; Ben-Sadoun et al., 2016). However, these frequencies should be interpreted with caution. The frequency of use for specific SG should vary depending on a number of variables, including the presence of physical activity (and its intensity), the duration of each game session, the time that patient (and eventually the caregiver) can devote to the activity, and the motivation to play. To maximize the training efficacy, it is important to establish the session frequency and duration (as well as the total training duration) based on the goals that the training is willing to achieve, and on the person’s features. For instance, if the target for a patient is to improve the general physical fitness, a short intense Exergame training is probably less adapted compared to a long training with regular sessions, in which physical activity is progressively increased based on performance improvements. But the session duration and goals need to be adapted to the person’s baseline physical fitness level, taking into account eventual physical constraints and concurrent pathologies (for instance, the training designed for an ex-marathon runner should be radically different from the training for a smoker whose previous level of active physical activity was limited).

Another crucial aspect to take into account when designing training is the motivation aspect. In general, the longer and more intense the training, the better. However, ‘forcing’ employing the SG too often may result in lowered motivation, or even in addiction, as there is a fine line between motivational and addictive aspects (Smith et al., 2011). Ideally, the features of the training should be adapted in order to increase the patient’s motivation to play, thus optimizing the training results. Strategies to improve motivation include, for instance, the presence of a clinician motivating the person, helping to set the training pace, and modifying its frequency in a timely manner based on the patient’s changes; the design of an adaptive game challenge, that keeps the player in the ‘flow zone’ (the feeling of complete and energized focus in an activity, with a high level of enjoyment and fulfillment) and improve the feeling of self-efficacy; for instance, the game becomes more difficult as the player progresses, but steps back to an easier level when the player is tired or show a decline in performance; or the presence of a well-designed game reward system.