1 Introduction

By 2060, the European population over the age of sixty-five years old is expected to outnumber those under the age of sixteen. Therefore, the proportion of people prone to cognitive decline will increase, because age affects cognition (Deary et al. 2009). Memory, executive functions, speed of mental processing, or reasoning are some of the altered brain functionalities. All them necessary for an autonomous life. Moreover, the economic burden associated with the cognitive impairment are huge for heath-care system as well as for the families that assumes the highest indirect cost.

Various approaches have been proposed to address cognitive decline. An early and accurate diagnosis is crucial for prescribing pharmacological and/or non-pharmacological treatments. However, as reported in Chen et al. (2022), the adherence to the treatment is lower than expected. To overcome this limitation, more holistic approaches suggest combining cognitive stimulation, physical activity, social engagement, and distress reduction (Hertzog et al. 2008). According to Mortenson et al. (2017), non-digital forms of games are natural and more common in the senior population. Although usually used for enjoyment, non-digital games may also aid cognition and improve social interactions and quality of life. Another approach came from serious games based on virtual reality technology, that have emerged as a powerful tool to address the consequences of cognitive decline, by enhancing engagement and producing synergistic effects on cognitive functions (Palumbo and Paternò 2020). Although it may be commonly believed that older adults are less likely to engage with computer-based games, this is not necessarily true for all individuals, and there is likely to be considerable variation in their willingness and ability to use these technologies. A study presented in Ballesteros et al. (2015) suggests the potential of behavioral, cognitive, and emotional effects on the senior adults.

Adlakha et al. (2020) found that serious games based on virtual reality (VR) have been in increasing demand over the past decade as complements to traditional cognitive rehabilitation. This is because serious games improve engagement, socialization, and adherence to therapy. Nevertheless, Maier et al. (2020) state that the effectiveness of cognitive rehabilitation is weak, as long as patients have deficits in multiple cognitive domains, and therapies focus on one cognitive domain or, at best, address one cognitive domain at a time. Another argument in favor of the simultaneous stimulation of multiple cognitive domains is that they mimic activities of daily living. As a complement, Ballesteros et al. (2015) review interventions to delay cognitive decline in healthy adults through: (i) physical activity, (ii) computerized cognitive training, and (iii) social enhancement. The authors highlight that most studies apply one of the three types of interventions previously mentioned and posit the need for therapies that combine them within daily activities.

Sustained engagement in a game is achieved when the challenge is adapted to the evolving skill level of the user (Nakamura and Csikszentmihalyi 2014). The underlying principle of this feature, commonly referred to as Dynamic Difficulty Adjustment (DDA) (Zohaib 2018), is to maintain an optimal balance between challenge and skill, ensuring that the gameplay remains neither overly difficult, leading to frustration, nor excessively easy, resulting in boredom. In the field of serious games for cognitive rehabilitation, DDA emerges as a highly desirable attribute due to its potential impact on therapy adherence.

DDA in the context of cognitive rehabilitation games serves a dual purpose. Firstly, it addresses the diverse cognitive abilities of older adults, recognizing that individuals may exhibit varying levels of proficiency across different cognitive domains. Through real-time assessment and adaptation, DDA tailors the game’s complexity to align with the user’s cognitive capabilities, providing an individually and engaging experience (Sweller 1988; VanLehn 2006; Tekinbas and Zimmerman 2003). Secondly, by preventing extreme levels of difficulty or monotony, DDA promotes sustained user motivation throughout the rehabilitation process (Ryan et al. 2006; Brown and Cairns 2004), which is crucial for achieving the desired cognitive outcomes, because consistent and prolonged interaction with the game enhances the effectiveness of cognitive rehabilitation interventions (Ijsselsteijn et al. 2007).

This review addresses the following research questions: (RQ1) How VR games have been used for cognitive rehabilitation?  (RQ2) What cognitive domains have been addressed by VR games and in which populations have these games been used?  (RQ3 ) Which features have been considered to improve engagement in VR games for cognitive rehabilitation?, and (RQ4) How is the difficulty adjustment of exercises carried out in VR games for cognitive rehabilitation?

1.1 Related reviews

A search of specialized databases yielded reviews focused on VR games designed for cognitive rehabilitation in the elderly. In this section, we review and synthesize the results of eight studies, each of which offers different perspectives on serious games for cognitive rehabilitation, such as design considerations, cognitive domains that have been addressed, and the hardware most frequently used to support the implementation of these games.

Brassel et al. (2021) emphasize that the design and development of VR games for cognitive rehabilitation of the elderly requires a holistic approach, highlighting the involvement of end users in the design process. This aspect implies the recognition and incorporation of individual characteristics such as age, gender, and physical limitations to adapt the VR experience to the needs of the target population. Brassel et al. (2021) further stress the importance of assessing potential adverse effects and ensuring safety throughout the study. Collaboration between game designers and clinicians is encouraged, as the latter bring valuable information to the design process. In addition, the authors highlight the importance of identifying barriers and facilitators to VR use, paving the way for a more user-friendly and effective rehabilitation tool.

Contreras-Somoza et al. (2021) present a series of recommendations on usability evaluation, in which they advocate for usability measurements in terms of efficiency and effectiveness, complemented by the evaluation of subjective usability through questionnaires and scales. The results of this dual approach seek to adapt the VR experience to each user and, consequently, improve adherence to treatment.

In their review on maintaining the functional integrity of the aging brain, Ballesteros et al. (2015) recommend a holistic approach that integrates physical exercise, computerized cognitive training, and social enhancement. Their research suggests that this approach may produce more effective outcomes than individual interventions, which leads to a more substantial impact on the cognitive well-being of older people.

The review by Maggio et al. (2019) explores the application of VR training for individuals with multiple sclerosis, emphasizing its effectiveness in both motor and cognitive rehabilitation. Their findings indicate that VR interventions can impact users’ executive functions, visual-spatial skills, attention, and memory. Notably, the authors underscore the positive influence of VR on gait and balance, attributing these enhancements to the motivational elements inherent in VR experiences.

The review by Francillette et al. (2021) extends the scope of VR applications for cognitive training from neurodegenerative conditions and mild cognitive impairment to mental disorders such as schizophrenia, autism, attention deficit hyperactivity disorder, and post-traumatic stress. The authors argue that VR games can be effectively integrated into therapeutic approaches for people with mental disorders. In particular, Francillette et al. advocate the incorporation of multiplayer capabilities in these VR games, recognizing the importance of social interaction for motivation.

The review by Mansor et al. (2020) evaluates studies focusing on the impact of VR games on cognitive functions such as processing speed, attention, and memory. The authors highlight a notable lack of consensus regarding the efficacy of VR games in preserving cognition. The authors attribute this to the influence of experimental design, to improve the reliability and reproducibility of findings in the field of VR-based cognitive rehabilitation.

Abd-Alrazaq et al. (2022) provide a critical perspective in their review, where they investigate the impact of VR games on improving executive functions in cognitively impaired older adults. Contrary to expectations, their findings suggest that the efficacy of VR games does not differ significantly from that of conventional exercises. Indeed, Abd-Alrazaq goes so far as to state that, even with the incorporation of dynamic difficulty adjustment, no discernible difference is observed in the improvement of executive functions through the use of VR games.

The review of Palumbo and Paternò (2020) presents an overview of virtual reality games for cognitive stimulation covering three main aspects: the technologies that have been used for their implementation, the cognitive domains that have been addressed, and the learning methods and the assessment tools that has been employed. Table 1 shows that tablets are the most widely used technology for the implementation of serious games for cognitive stimulation.

Table 1 Technologies used in the implementation of VR serious games for cognitive rehabilitation (Palumbo and Paternò 2020). AR, VR, and MR refer to virtual, augmented, and mixed reality, respectively

In our exploration, the reviewed studies delve into several dimensions, such as design considerations, the ongoing debate surrounding the impact of VR games on cognitive domains, and the diversity of hardware employed. However, the current state of knowledge, derived from these reviews, shows that there is a knowledge gap regarding the essential properties or characteristics that make a VR game effective for cognitive rehabilitation. Additionally, no guidelines for the design, implementation, and validation of DDA features were found.

2 Methodology

This review was conducted following the methodology described in Ramdhani et al. (2014), which comprises three main phases: search, selection and synthesis of the literature. For the search we want to answer the questions (i) How VR games have been used for cognitive rehabilitation?(ii) What cognitive domains have been addressed by VR games and in which populations have these games been used?(iii) Which features have been considered to improve engagement in VR games for cognitive rehabilitation?(iv) How is the difficulty adjustment of exercises carried out in VR games for cognitive rehabilitation?

Given the proposed questions, this study will focus on applications that run on computers, mobile devices, VR headsets, among others. We are also interested in all types of games, as long as they are proposed for cognitive stimulation or rehabilitation, regardless of game genre. Additionally, only works aimed at the senior population were considered. The scientific articles included and analyzed in this review were obtained from the following database: Scopus, Science Direct, PUBMED, Lilacs, EBSCO, Taylor and Francis, ProQuest and Springer. To this end, the keywords used were: “serious games” AND adapt AND (cognitive OR senior adults OR rehabilitation); and without time limit specified.

Two critical concepts in the context of this work are VR and adaptation. As these terms may have different meanings depending on the context, it is important to clarify the definitions that we have chosen to consider when selecting works.

Although VR has historically been associated with video games, its evolution has led it to find practical applications in fields such as education and healthcare. By allowing the creation of entirely fictional worlds or adaptations of real-life environments, VR technology has expanded beyond its initial gaming-centric context. According to Fuchs et al. (2011), there are many definitions of VR, ranging from a human–machine interface to an advanced interactive simulator or a technology that enables users to become part of an artificial world. Fuchs et al. (2011) present a definition that’s particularly relevant to our context: through VR technology, users are no longer limited to being mere spectators but can become active participants in an environment that provides an additional dimension to their experiences. Given the above, this work will focus on applications that allow users to interact with a virtual environment, regardless if it is immersive or not.

Streicher and Smeddinck (2016) provide insights into the distinctions between the concepts of personalization, adaptability, and adaptation in the context of serious games. Personalization involves offering players a personalized experience based on their individual characteristics, while adaptability refers to the ability of the system to adapt its content and interfaces to meet the specific needs of the user. Adaptation, on the other hand, refers to the automatic adjustment of the game’s difficulty level to the user’s skill level. This aspect is closely related to the state of flow, which is an important consideration in game design.

The inclusion criteria for this review were: (1) articles written in English and available for download, (2) papers published in 2013 and later. Exclusion criteria were: (1) studies that did not describe whether the VR game was used for stimulation, assessment, or both, (2) studies that did not describe the mechanics of the VR game. Later, looking to expand the scope of the search, we sought to include works in Spanish. The Fig. 1 summarizes the study selection process.

Fig. 1
figure 1

PRISMA flow diagram of the study selection process

3 Results

The synthesis of the studies selected for the present review are summarized in Table 2.

Table 2 Papers referenced in this review

3.1 RQ1: How VR games have been used for cognitive rehabilitation?

Table 3 classifies the selected studies according to their purpose (stimulation, evaluation, or both).

Table 3 Purpose of the serious game

3.1.1 Games for both, cognitive evaluation and stimulation

According to Table 3, there are just three articles that used serious games for evaluation and stimulation simultaneously. An example of this approach is the trivia question game presented in Caixinha and Alexandre (2014), in which the user has to remember information about himself. The game follows the progression of Alzheimer’s disease by collecting information about the patient’s life-story and generating a set of questionnaires to track the evolution of the disease, while providing cognitive stimulation.

Farber et al. (2016) proposed two tablet-based games to evaluate and stimulate cognition in senior adults prone to develop Alzheimer disease. In the first game, users combine railway track segments to create paths for trains while avoiding collisions between them. In the second game, players drive a car to a store to purchase the items presented on a list. The authors posited that participants play rounds of increasing difficulty in a predetermined order to ensure a repeatable and consistent gaming experience. For this reason, both games do not include DDA features.

Neto et al. (2018) proposed three serious games to stimulate and measure cognition. These games consist of: (1) sorting a herd of colored sheep into specific areas, (2) identifying the number of sheep and wolves appearing in an enclosure, and (3) remembering the order of a group of cows that appeared on the screen. The three versions offer the possibility to adapt the difficulty of the game to the user’s performance.

3.1.2 Games for cognitive evaluation

Tarnanas et al. (2015) states that patients with mild amnesic impairment have high risk to develop Alzheimer disease. To keep the brain healthy as much as possible, the authors proposed a game that allows the user to take a tour in a virtual museum consisting of six rooms equipped with artifacts from ancient Macedonia. At the end of the tour, visitors are asked about the location and characteristics of the artifacts presented at the beginning of the tour.

Paliokas et al. (2020) propose a virtual supermarket, where users are able to buy the items given in a shopping list. The objective of this game is to compare frail (Markle-Reid and Browne 2003) versus non-frail adults with respect to their performance in spatial orientation, short-term memory, selective attention, and cognitive speed. The game allows users to select one of three difficulty levels based on their preferences.

Chen et al. (2021) used whack-a-mole and hit-the-ball games to compare reaction time and accuracy in three populations: a control group, a group diagnosed with mild cognitive impairment, and a group diagnosed with moderate-severe dementia. The comparison was performed using three levels of difficulty, and for both games there were significant differences in the performance of the populations.

3.1.3 Games for cognitive stimulation

Eichhorn et al. (2018) state that serious games for cognitive rehabilitation could be classified according to the cognitive domain addressed. For example, any game whose purpose is to recall a list of items, objects, or places can be used to stimulate short-term memory; trivia games encourage long-term memory; object matching games improve selective attention, and hitting and catching objects promote inhibition. Like Paliokas et al. (2020), the game described in Eichhorn et al. (2018) comprises three difficulty levels, which can be freely selected by the user.

Knobel et al. (2021) propose a VR game that consists of catching birds in a 3D virtual world. The game was tested on young and healthy people, as well as in patients with impaired visual exploration. All the three groups reported an average usability of 4.5 over 5.0. The authors attribute this result to the immersion of the game and the use of DDA.

Liu et al. (2018) propose to stimulate cognition through a set of five games, each of them addressing one cognitive domain. In the first game, oriented to stimulate observation, the player should find a previously given number on a screen and then throws a ball at it. The second game trains short-term memory by finding on a screen a copy of a number displayed and then hidden. The third game, which trains judgment, presents to the user an arrow that sometimes points left and sometimes points right. In response, the user looks at a colored indicator, and when it is red, the user must select an arrow button that points in the same direction as the arrow on the screen. When the indicator is blue, the arrows should be in opposite directions. The fourth game is rock - paper - scissors, which stimulates reaction times. The fifth game trains math skills by asking the user to select the balloons that are multiples of the number three.

Brasil et al. (2013) implemented seven games to stimulate memory, logical reasoning, and problem solving in people 60 years and older with cognitive impairment. The first game is Crossword, which trains memory and language. The second is Tic-Tac-Toe, which focuses on executive functions. Its purpose is to form vertical, horizontal, or diagonal sequences of three elements of the same type. The third game is Hangman whose purpose is to recognize objects and associate images. The fourth game, named Walking Route, fosters visual and spatial memory by showing a sequence of places of interest, which the player must memorize and reproduce in the same order they were presented. The fifth game is Finding at Sea, which presents numbers and letters that the user should find in the midst of many others. This game requires executive functions, visual examination, memory and concentration. The sixth game, named Right Sequence, shows frames in disorder and the user should organize them in a logic sequence. This game trains memory, information processing, perception and involves inference and deduction. The seventh game is Set of Relationships, which shows images organized on a pyramid, at the top there is a reference figure and the user should choose related images for the pyramid base. This game up-skills memory and executive functions.

3.2 RQ2: What cognitive domains have been addressed by VR games and in which populations have these games been used?

Table 4 presents the rows of Table 2 where the value in “Cognitive Domain” column is “One Game per domain” or “One Game Multidomain”. For example, Jha et al. (2020) proposed an application that uses one game for orientation, a second game for memory and a third for attention, this is a “One Game per domain” approach. On the other hand, Paliokas et al. (2020) proposed a virtual supermarket as an unique game, that assesses spatial orientation, memory, attention, and cognition speed, this will be classified as “One Game Multidomain”.

Table 4 Detailed cognitive domain on each work

Sachdev et al. (2014) proposed a framework that classifies cognitive domains in: motor-perception, language, learning and memory, social cognition, complex attention, and executive functions. Sachdev et al. (2014) also provide a list of diseases related to cognitive decline which comprises Alzheimer, Parkinson, substance abuse, brain injury, vascular disease, and cognitive mild impairment (MCI). Table 5 summarizes the literature review according to criteria given by Sachdev et al. (2014). Just nine articles address a single cognitive domain, other fourteen use a battery of games, where each game tests a single cognitive domain. The remained two studies propose games that trains multiple domains simultaneously.

Paliokas et al. (2020) use the Virtual Supermarket game, where the user must locate and buy items from a list. This game trains spatial orientation, short-term memory, selective attention, and reasoning speed simultaneously. Vallejo et al. (2017) propose another game that trains multiple domains simultaneously. In this game, the authors implement a functional and safe environment where the user trains executive functions, memory, and attention through a cooking task.

Table 5 Cognitive domain addressed by the serious game

Cuevas-Lara et al. (2020) propose an arcade game that trains twelve cognitive domains, using one game per each of the following cognitive domains: (i) temporal attention, (ii) spatial orientation, (iii) visual perception, (iv) spatial perception, (v) selective attention, (vi) sustained attention, (vii) short-term memory, (vii) semantic memory, (ix) linguistic expression and understanding, (x) denomination, (xi) categorization and semantic awareness, and (xii) sequence of functional scopes. The authors recommend that patients walk as much as possible on the day of training.

Table 6 uses the taxonomy suggested in Sachdev et al. (2014) and shows that most studies involve senior adults with mild cognitive impairment or early stage dementia, followed by people with Alzheimer’s. Table 7 presents the clinical tests used in each work either as inclusion criteria or to monitor the cognitive condition of the users during the tests.

Table 6 Population addressed by the serious game
Table 7 Cognitive assessment tools used in each study

3.3 RQ3: Which features have been considered to improve engagement in VR games for cognitive rehabilitation?

Jha et al. (2020) argue that immersion with VR glasses should be considered in games that train orientation, memory, and attention. The game proposed in Jha et al. (2020) also includes a DDA system that makes the game easier or harder depending on the performance of the user. Borrego et al. (2020) propose a game for cognitive stimulation and motor rehabilitation of senior adults that comprises remote interaction between two participants, remote session configuration, and monitoring of the performance. The experimental results reported in Borrego et al. (2020) showed that users perceive the game as stimulating, useful, easy to use, and even fun. These authors, like those of Jha et al. (2020), consider that the DDA is essential for enjoyment. Goumopoulos and Igoumenakis (2020) show a game based on an ontology that uses words, images, and sounds to create personalized exercises. Likewise in the two previous games, the one proposed in Goumopoulos and Igoumenakis (2020) adapts the difficulty to the user’s performance. The authors state that customization and adaptation are necessary to avoid boredom when users play for long periods.

Another trend in literature is to combine physical activity with cognitive training. Ben-Sadoun et al. (2016) evaluate the effect of the physical activity on usability and conclude that physical intensity must be adapted to user performance to make the game more stimulating. Pedraza-Hueso et al. (2015) propose a Kinect-based application for physical and cognitive rehabilitation. The game trains strength, aerobic capacity, and cognition, while adapting the exercise to the user’s skills. Liu et al. (2018) posit that cognitive rehabilitation is more effective when combined with physical activity, and develop a game that projects images on a wall and follows user’s movements using laser scanners. The game aims to correlate physical and cognitive status. The users of Liu et al. (2018) expressed that the game is attractive, even though the customization needs to be improved. Paliokas et al. (2020) highlight that VR games for cognitive rehabilitation should train multiple cognitive domains simultaneously to mimic daily life activities and then improve functionality.

In summary, serious games for cognitive rehabilitation must be oriented to daily life activities and consider: challenges according to players skills, immersion, customization, cooperation, and physical activity. The Table 8 shows the functionalities found in each work.

Table 8 Characteristic on each work

3.4 RQ4: How is the difficulty adjustment of exercises carried out in VR games for cognitive rehabilitation?

Streicher and Smeddinck (2016) provides a detailed explanation of how to personalize and adapt serious games, as they explain how these adaptations can promote motivation and increase the acceptance of serious games, due to the heterogeneity of users and the need to respond to each individual’s traits. The adaptation process can be done with simple heuristics based on thresholds, or with complex evolutionary models that learn from the user. These adjustments are necessary whenever a game has repetitive mechanics or there is a difference between the player’s skill and the proposed challenge, with the goal of achieving positive results in specific objectives.

As mentioned above, a DDA algorithm varies the difficulty of a game based on the challenge perceived by the players. Without this functionality, users get bored when the game is too easy or frustrated when it exceeds their skills (Zohaib 2018). A successful DDA engages users by keeping them in a zone called “flow” by Nakamura and Csikszentmihalyi (2014), where people are fully immersed in the present moment. To maintain the state of flow, a DDA must periodically measure success rate, number of victories, life points, time taken to complete a task, or any other performance metric. For example, Morán et al. (2015) present a game for cognitive stimulation that measures performance as the time required to clean a dirty spot on a virtual surface. Chen et al. (2021) use two games, hit the mole and hit the ball, whose performance indices are reaction time and accuracy. Vallejo et al. (2017) measure the performance in a virtual kitchen as the number of cooking tasks performed successfully. This game, unlike the previous ones, does not include DDA. In contrast, Chen (2007) suggest that each player has his own flow zone, and, therefore, the function defining the challenge as a function of the player performance should be different between a novice and a hardcore player. This statement implies that games should consider manual adjustment of the parameters governing the DDA algorithm. In the case of a game with a linear relationship between challenge and performance, the user could select, for example, the slope of that relationship.

Another scenario involves games where difficulty is tailored based on the user’s physiological measures. Although no examples were found in the context of cognitive training, Muñoz et al. (2018) presented a case where a physical activity-promoting game was adjusted in real-time to maintain desired levels of heart rate. The study compared the effects of conventional exercise and cardiovascular game with adaptation, highlighting the importance of having games that achieve desired levels of physical effort to encourage senior adults to increase the time they dedicate to physical activity. In the physical context, adapting the game increases safety and effectiveness, reducing the risk of injuries, and making the experience more enjoyable for the user.

Other DDAs combine user performance and physiological data to increase or decrease the difficulty of a game. For example, Jha et al. (2020) propose a maze game that activates an assistance system when users remain static in the maze, move away from the target, or one emotion (frustration, excitement, engagement, meditation, and valence) prevails over the others. The emotions of the player are estimated every ten seconds through electroencephalographic (EEG) signals processed by the Emotiv software.

It’s worth mentioning the work of Darzi et al. (2021) where multiple DDA methods were compared using data obtained from a previous study on user performance and physiological measurements. Participants were asked to fill out two questionnaires: the intrinsic motivation inventory and the flow experience measurement. Although their results did not show significant differences in these tests, possibly due to the subjective nature of a survey, they report a correlation between difficulty changes and enjoyment/pressure.

Table 9 summarizes the type of measurements used to assess user’s performance. Twenty five articles were included, while three do not assess the difficulty perceived by the users, nineteen use classical performance metrics (i.e. success rate, number of victories, life points, time taken to complete a task), and the remaining three combine performance metrics with physiological data.

Table 9 Approaches used to assess user-experienced difficulty. The hybrid category refers to studies that combine performance metrics with physiological data

Zohaib (2018) classifies DDA approaches as: heuristic, probabilistic, single and multi-layered perceptrons, dynamic scripting, hierarchical agent-based machine learning systems, and reinforcement learning. In the our review, DDA algorithms are classified into four categories, the first two are the ones presented in Zohaib (2018), and the other two are static and dynamic artificial intelligence (AI). A DDA approach based on empirical rules is called heuristic (Zohaib 2018), while a DDA approach that estimates the probability that the player lost the match, is denoted as probabilistic. A DDA trained once on physiological or performance data is classified as a static AI, while a DDA that continuously learns from such data is classified as a dynamic AI.

Heuristics is a decision-making method that aims to streamline the decision process by defining thresholds, thereby avoiding the need for complex models (Gigerenzer and Gaissmaier 2011). In the context of difficulty adjustment in VR games, an illustrative example is Meteorix (Correa et al. 2014), which was specifically designed for amblyopia rehabilitation. In this game, players control a spaceship and shoot asteroids in a top-down 2D view. If a player successfully destroys the asteroids, the game employs Procedural Content Generation, an automated method for creating digital content such as articles, missions, or level designs. This technique is used to increase the game’s difficulty by adding more targets and introducing deeper levels of challenge. On the other hand, if a player fails, a helpful system is activated to provide assistance. To define the heuristic used in Meteorix, the researchers gathered data from 300 healthy children. They measured factors such as the required depth perception and reaction time to successfully shoot the asteroids. Based on their findings, they established that if a target moves out of the shooting range or if a player fails to hit a target seven times consecutively, the assistance system is activated to aid the player. Caixinha and Alexandre (2014) propose an application that stimulates the memory of Alzheimer’s patients by asking them for information about their relatives. When the user fails 80% or more of the questions about a person five times in a row, those questions are removed from the game. Tarnanas et al. (2015) also use a heuristic approach for an immersive game designed for users with impaired visual exploration, which consists of catching birds flying in random directions. The birds, a total of 30, are presented one at a time. The game consists of 15 levels whose difficulty is defined by four parameters that are the life time and the speed of the birds, and two thresholds (in number of captured birds) that are used to decide when the user goes to a higher or a lower level. The user remains at the same level when the score is between the two thresholds. Rodríguez-Isasi et al. (2014) created a serious game to stimulate spatial vision, memory, and attention in older people. The game consists of a puzzle and a bingo, both with increasing difficulty, which means that next level is more difficulty than the previous one.

When it comes to probabilistic algorithms, they can be defined as optimization problems that seek to maximize engagement (Zohaib 2018). In a study by Segundo et al. (2016), they introduce the game Space Shooter, where players take control of a spaceship equipped with 100 health points and an inventory of 50 bullets. The objective is to increase their score by destroying enemies and overcoming obstacles. Throughout the game, players can come across additional ammunition packs or health points, while their spaceship’s health deteriorates upon contact with enemies. The DDA assesses a challenge function to gauge the player’s experienced level of difficulty. Based on this assessment, DDA applies a policy that adjusts various game parameters, such as the distribution of item offerings and enemy behavior, aiming to strike a balance and create a state of flow-an immersive and optimal experience for the player. DDA utilizes statistical measures, such as the mean and variance of damage suffered within a time interval, to estimate the probability of future defeat. Additionally, a set of empirical rules considers factors like the probability of losing and the player’s health percentage to determine which response policy DDA should apply in order to guide the player towards achieving a state of flow.

In the context of static AI, performance mapping involves mapping the user’s performance and physiological responses, or a combination of both, to a set of emotions. This mapping is done to enable the Affective Driver Distraction system to make decisions (Zohaib 2018). In the realm of rehabilitation, Bian et al. (2016) present a project aimed at addressing Autism Spectrum Disorder, a condition that poses challenges for individuals to function independently in tasks requiring adaptive skills, such as driving a vehicle. Their work utilizes a driving simulator in a VR environment, which includes vehicles, pedestrians, and traffic lights. The task difficulty can be adjusted through intrinsic vehicle parameters, such as brake pedal response, accelerator sensitivity, and steering wheel feedback, or through external variables like lighting conditions and the speed of other vehicles. To evaluate the user’s affective state, the authors propose a DDA system based on physiological signals, including photoplethysmogram, galvanic skin response, and respiratory rhythm. The DDA system was developed offline using physiological data and expert input from therapists in a previous study (Bian et al. 2015). They employed a machine learning algorithm called Random Forest to train a commitment detection model. In the DDA described in Bian et al. (2016), the commitment detection module is activated every 3 min, acquiring data that is then sent to the Random Forest model to generate a binary label indicating the detected level of commitment. Finally, by combining the physiological variables with the performance data obtained from the task and using a predefined table, the DDA determines whether to increase or decrease the game difficulty.

An instance of dynamic AI is presented by Shakhova and Zagarskikh (2019) in a real-time, first-person fighting game where two avatars engage in combat using various weapons. One avatar is controlled by AI, while the other is under the command of a human player. Each player possesses unique parameters, including health points, physical endurance, strength, and agility. At the outset, both health points and physical endurance are set at 100, and the game concludes when one avatar’s health points reach zero, with the other declared as the winner. Players can execute attacks in three directions (right, left, or upward) with two types of attacks (fast or strong) and have the ability to block enemy strikes. Moreover, prior to each battle, players can select a weapon from their inventory, such as a sword, axe, mace, or spear, each with its own advantages and disadvantages. The AI component consists of an agent Footnote 1 that uses a DDA to respond to the player’s actions. This agent utilizes a neural network that models the player’s behavior and employs a genetic algorithm to update its weights. The agent’s behavior includes actions like weapon switching, attacking, and blocking. The AI’s weapon selection is influenced by the opponent’s most frequently used weapon, while the direction and type of attack are influenced by the potential maximum damage that can be inflicted.

The input features considered are the remaining health points, the selected weapon, the presence of recent injuries or blocks, the duration of the fight, and the player’s skill level, which encompasses the number of successful blocks and the maximum damage inflicted. All values are normalized between 0 and 1, and the neural network produces a set of 13 values representing the possible movements of the agent. By employing selection, crossover, and mutation procedures, the weights of the neural network are adjusted to minimize the skill gap between the user and the agent. The approach proposed in Shakhova and Zagarskikh (2019) exhibits the advantage of rapidly adapting to the user’s initial skill level while also monitoring their progress or regression. Furthermore, the utilization of neural networks and genetic algorithms contributes to the creation of agents with believable behaviors, as the agent learns alongside the player.

According to the literature included in this review, five papers use a manual approach, five use a heuristic approach, two use static AI, and 13 do not include a DDA algorithm, meaning that for these games the difficulty remains constant or increment regardless of user performance (see Table 10).

Table 10 DDA approaches

4 Discussion and conclusion

In this study, we conducted a bibliographic research analysis resulting in 317 articles. After applying the selection criteria outlined in Sect. 2, we narrowed down the articles to twenty-five that met our inclusion criteria. These twenty-five articles were then utilized to address the four questions that served as the basis for this review.

Tables 2 and 3 indicate that VR games have primarily been utilized for cognitive training, with a lesser emphasis on assessing cognitive strengths and weaknesses. This trend can be attributed to the fact that clinicians often rely on widely used tools such as MoCA and MMSE for cognitive assessment. Contrarily, the VR assessment tools are diverse and focus on different cognitive domains, which instead of encouraging the standardization of assessment tools cause the opposite. For instance, the five reviewed VR applications for evaluation involved activities like touring a virtual museum, whacking a mole and hitting a ball, shopping in a virtual supermarket, cooking in a virtual kitchen (Tarnanas et al. 2015; Chen et al. 2021; Paliokas et al. 2020; Vallejo et al. 2017, respectively).

Tables 4 and 5 reveal that the majority of VR serious games focus on addressing one cognitive domain at a time, such as attention or memory. This approach allows for targeted and intensive training within specific cognitive areas. However, while there is experimental evidence supporting the effectiveness of this approach, there are concerns regarding its ecological validity (Reijnders et al. 2013), which refers to the extent to which these interventions translate into improvements in real-life performance. As a result, there has been a growing interest in approaches that simultaneously stimulate multiple cognitive domains (Vallejo et al. 2017; Paliokas et al. 2020). The rationale behind this is that cognitive functions are interconnected (Kurashige et al. 2020), and addressing deficits across various domains may lead to broader improvements in overall cognitive functioning.

To the best of our knowledge, no clinical studies specifically address the incidence of DDA on therapy outcomes. Nevertheless, its importance has been recognized by several authors (Zohaib 2018; Morán et al. 2015; Chen 2007; Vallejo et al. 2017; Ben-Sadoun et al. 2018). For example, Ben-Sadoun et al. (2018) outlines nine ergonomic criteria for the design of serious games, which includes “adaptability”. This criterion refers to the integration of various levels of difficulty adapted to the cognitive and physical skills of the user. The objective of adaptability is to generate progressive stimulation, thereby facilitating the creation of different experiences that cater to diverse player profiles. In addition, Seyderhelm et al. (2019), drawing from the previous experiences, such as Hunicke (2005) and Xue et al. (2017), affirm that incorporating DDA in 3D games not only enhances enjoyment but also fosters commitment among users. These components are considered essential for achieving effective learning outcomes. The recognition of DDA’s potential impact on user engagement in the context of serious games aligns with the broader goal of optimizing cognitive rehabilitation experiences for individuals with varying needs and profiles.

Future studies could address key questions, such as the relative effectiveness of each approach in terms of cognitive improvement, user acceptance, and long-term benefit retention. It is also essential to establish clear guidelines for the selection between single-domain and multi-domain approaches, which is required to optimize the effectiveness of interventions and adapt them to the individual needs of older adults.

Based on Table 6, the majority of serious games designed for cognitive rehabilitation are targeted towards individuals with MCI and Alzheimer’s disease. In the context of Parkinson’s disease (PD), serious games primarily focus on rehabilitating fine motor skills, as evidenced by Foletto et al. (2017), de Oliveira et al. (2021). However, these games do not address the cognitive aspects affected by PD, which can be categorized into two main groups according to Fang et al. (2020). The first group includes working memory and executive functions, while the second group encompasses semantic verbal fluency and visual spatial ability. Interestingly, these authors highlight the lack of therapies specifically addressing cognitive impairments associated with PD-MCI and PD-dementia.

Rehabilitation experts concur that everyday activities involve multiple cognitive domains (Paliokas et al. 2020), making game topics such as cooking, dancing, or shopping popular for serious games. However, it is essential to consider user preferences, as what may be appealing to one person could be monotonous for another. To strike a balance between implementation feasibility and customization, serious games can incorporate a single theme with multiple customization options. For example, users can select their preferred cooking recipe, choose the music for dancing, or decide the type of store for their virtual purchases. This approach allows for customization, while maintaining the practicality of game design.

The inclusion of exergame features in serious games for cognitive rehabilitation is based on the premise that physical activity can enhance cognitive function through the increased production of two specific neurotransmitters (Bherer et al. 2021). One of these neurotransmitters is brain-derived neurotrophic factor (BDNF), which supports neuron growth, survival, and function. The other neurotransmitter is glutamate, which plays a role in optimizing learning and memory processes. Supporting this premise, a systematic review conducted by Amorós-Aguilar et al. (2021) identified five randomized controlled trials that concluded that combining aerobic exercise with cognitive training yielded better results than interventions involving either component alone, particularly in stroke and traumatic brain injury cases. A similar conclusion was reached by Vanderbeken and Kerckhofs (2017) in their systematic review. However, both studies highlighted the need for additional experimental evidence to further understand the synergistic effects of physical exercise and cognitive training.

Researchers who advocate for multiplayer games argue that playing with others fosters motivation, enjoyment, and social engagement. These factors, in turn, generate positive emotions that have been shown to correlate with improved cognitive performance (Wang et al. 2017; Alexander et al. 2021). Additionally, a study by Felix et al. (2021) revealed a positive correlation between social engagement and the integrity of gray matter in brain regions associated with dementia. However, to the best of our knowledge, no randomized controlled study has yet provided experimental evidence to support the specific benefits of multiplayer features in serious games for cognitive rehabilitation.

Table 9 shows that most DDA approaches estimate the difficulty perceived by the user from use performance measurements. However, Seyderhelm et al. (2019) criticize this approach arguing that users can maintain a sustained performance while experiencing increasing levels of cognitive load, which prevents them from obtaining the maximum benefit from the therapy. These authors mention that cognitive load can be reported by the user at the end of the task or estimated through physiological measurements such as electroencephalography, hear rate, or pupil dilatation, among others. As far as we know, no serious game for cognitive rehabilitation integrates cognitive load into the DDA.

DDA is justified on the Beck (1992) theory, which postulates that people are in the state of flow when they perceive that an activity poses a challenge according to their abilities. Chen (2007), and Nakamura and Nakamura and Csikszentmihalyi (2014) describe the state of flow for games, and Ritterfeld et al. (2009) and Streicher and Smeddinck (2016) specifically for serious games. People in this balanced state spend more time on a task than when they feel anxious or bored. Therefore, DDA plays a central role in serious games because it must bring users to the state of flow, which is necessary to improve adherence to therapy. However, according to Cruz and Uresti (2017) when people play games, some prefer excitement while others prefer relaxation; hence, there is no consensus on how a DDA might combine physiological and performance measurements to determine if a player is in a state of flow. Furthermore, which physiological variables are most suitable for this purpose remains an open question. Our view is that the gap is caused by the absence of a conceptual framework that integrates relevant theories such as state of flow (Nakamura and Csikszentmihalyi 2014), cognitive load (Mayer and Moreno 2003), cognitive rehabilitation (Seelye et al. 2012), serious games (Carvalho et al. 2015), and feedback systems (Parnandi et al. 2013).

Tong et al. (2017) classify game mechanics according to cognitive domains, outlining their effects as follows: for short-term memory enhancement, games emphasizing memorization are deemed essential; for long-term memory, the recommendation is for trivia-based multiple-selection questions; the ability to switch between tasks is addressed through mechanics involving image matching; inhibitory control is targeted through mechanics involving hitting or selecting objects; object recognition is stimulated by classifying images by name or by functionality; and social cognition though the recognition of personal photos, videos, or belongings, which reinforce self-awareness and communication.

The use of a wide range of assessment tools in cognitive rehabilitation involving serious games (Table 7) creates challenges in comparing success rates across studies. Additionally, even if the MoCA or MMSE were commonly selected for cognitive assessment in these studies, their dependence on the expertise of the test administrator would be a significant source for bias. To address this issue, several computerized tests have been proposed (de Oliveira et al. 2014); however, their integration into the clinical context remains limited. Therefore, the lack of standardized and reliable cognitive assessment tools continues to be the primary barrier to drawing conclusions about which game features (Table 8) are the most relevant in achieving therapy outcomes, and which of them to prioritize when implementing serious games for cognitive training.

One of the key practical challenges in implementing VR games for cognitive rehabilitation in older adults is ensuring accessibility (Contreras-Somoza et al. 2021). It is essential to design these games to be inclusive for a diverse range of users, regardless of their physical or cognitive abilities. Achieving this goal requires evaluating how older people accept technologies like VR headsets, wearable motion capture systems, or hand tracking systems (Ben-Sadoun et al. 2018). Additionally, assessing their overall experience during interactions with virtual environments is also important. These steps will ensure that the VR games are not only effective for rehabilitation but also accessible and engaging for all older adults.

In summary, the development of serious games for cognitive rehabilitation faces two significant challenges. The first challenge involves effectively combining performance and physiological measurements to assess the state of flow in individual players, considering the inherent variability of this state across individuals. The second challenge arises from the lack of comparability among studies, primarily due to the diverse array of methods used to evaluate cognitive therapy outcomes. Consequently, the current state of art falls short in ensuring both the attainment of a state of flow, which is vital for therapy engagement, and reproducibility, which is crucial in the medical domain.

4.1 Limitations

While this study provides valuable information on serious games for cognitive rehabilitation in older adults, certain limitations should be considered. The literature review, although extensive, may not encompass the entirety of the research available in this dynamic field, and future studies could broaden the scope for a more comprehensive understanding. Additionally, concerns arise about the generalizability of findings due to the diversity of older adult populations, cultural contexts, and healthcare settings. Finally, the study highlights the need to adopt diverse perspectives beyond cognitive science, and future research should actively seek collaboration across disciplines to ensure a holistic approach to serious games for cognitive rehabilitation.