To study the relation between the participant’s subjective appraisal of the gaming experience, we administered an exit survey at the end of each session, which was loosely based on the intrinsic motivation inventory and tapped into the question of enjoyment (The experience was enjoyable, The experience was interesting), competence (The game was difficult), tension (The experience was stressful; I found this to be a frustrating experience. These games are visually intense), choice (I liked to play this game again; I will play this game again. I did not like this experiment), and value (This game help improve my mental wellness, I think this game is useless; These games are cognitively stimulating). Instead of a 7-choice scale to measure degrees of satisfaction, we decided to implement a 4-choice response (definitely disagree, somewhat disagree, somewhat agree, and definitely agree), ranging from –2 to 2, but we allowed participants to add comments if they did not know the answer. We coded these responses to 0 or I don't know.

Saliva cortisol is often measured as a biomarker of a latent psychological modulation to the stress system [73]. Cortisol production displays a typical diurnal rhythm with a peak in the morning hours (8-10 AM), followed by a periodic pattern consisting of ultradian oscillations (20-120 min periods) [74,75]. We therefore ensured to schedule the visits at the same time for each individual, and we timed the experiment to avoid measuring the sharp rise in the awakening cortisol. Free circulating cortisol was assessed from saliva, collected using the Salivette device (Figure 2). The samples were frozen at −80°C, centrifuged at 4°C for 10 min, and analyzed using luminescence-immunoassay (IBL International) using enzyme-linked immunosorbent assay (at PERFORM Centre). The intra- and interassay coefficient variability was less than 12% and 5%, respectively.

Since 1970s, the electrodermal activity (EDA) and the heart rate variability (HRV) have been used as proxy markers of sympathetic autonomic response to arousing or stressful stimuli, finding successful application in laboratory experiments involving human-computer interface and games [76-82]. We used the E4 device (Empatica, Inc), a light wrist-wearable device for continuous monitoring of the physiological signals. The E4 band (Figure 2) was properly fit on the wrist of the nondominant arm. This device is equipped with 4 sensors: (1) photoplethysmography sensor to measure cardiac activity, (2) EDA sensor to measure skin conductance as a marker of arousal and stress, (3) 3-axis accelerometer (ACC) to measure the amount of movement along the x-, y-, and z-axes, and (4) optical thermometer to measure skin temperature.

To facilitate the replicability of our study, we downloaded preprocessed data from the Empatica data portal. We used our in-house MATLAB software on these data to extract data for each activity (by averaging signals’ amplitudes normalized to the duration of different activities). An activity denotes a specific interval measured from the onset of each experimental step. Activities were delimited using a tag button on the E4, which was pressed every time we administered a new task. A research assistant also kept track of detailed timing of the experiment. Because EDA is sensitive to the activity of sweat glands, the room temperature was maintained at 23°C. However, we also collected data from the temperature sensor and ACC on the E4 device to rule out confounding effects related to thermal or movement noise.

The triggering of physiological signals to stress serves the metabolic modulation of reward processing [83] and attention [84] systems that help individuals initiate an appropriate adaptive response to stress; thus, it can interfere with cognitive functions performed during or immediately after stress. We included a set of simple cognitive tasks to assess whether, despite the potential learning effect due to repeated nature of the design, increasing game difficulty would affect performance in cognitive tasks. The first set of cognitive tests used a picture encode/delayed recall memory test, which we have previously shown to be sensitive to stress modulations of the hippocampal function both under tasks [71] and under stress-inducing drug conditions [85]. In addition, we administered a simple Stroop word color test to assess an individual’s cognitive processing speed and their selective attention accuracy. The Stroop test has been shown to be sensitive to physical exertion and cognitive challenge [86,87]. All cognitive tests were presented at a minimal level of difficulty (with 3-second delay between stimuli) to ensure these tasks did not introduce added stress.

The encode/recall test consisted of two parts. In the first part (the encode condition), the participant was presented with 50 black and white images of random scenes, places, people, and objects. Each image stayed on screen for 3 seconds; thus, the participant had time to commit it to memory. This set of 50 pictures included 14 unique pictures and 36 nonunique pictures that were shown more than once. The participant was asked to press m as soon as they saw a familiar picture. They were told to remember these pictures for a recall test at the end of the session. In the second part (delayed recall), 45 black and white images were shown. This set included 15 pictures from the encode session and 30 new pictures. Participants were asked to press m as soon as they detected a previously seen picture. In both sessions, each picture remained on the screen for 3 seconds. If no answer was provided, the screen would proceed to the next picture. This test was implemented in E-Prime, and we recorded the number of hits, misses, and false alarms, both for the encode and the recall conditions. We had designed 3 sets of unique encode/recall pairs. In other words, none of the pictures seen in session 1 were repeated in the pictures presented in sessions 2 or 3.

We also wanted to measure the reaction time (RT) to the Stroop effect, that is the time it takes a player to process the incongruency between the color of a word read, and the color represented by the word. At each trial of this test, a brief training session was held to ensure the players understood the test objectives. Following the training session (1 min), a randomly selected color word (red, yellow, blue, or green) was presented to the participants. The color of the word on the screen may or may not have been congruent with the word itself. The stimuli (50) were randomly selected from a stack of 320 cards (80 of which were incongruent). The participants were asked to press a colored key on the keyboard, which was associated with the color of letters of the word, as fast as they could. In our implementation, we provided immediate feedback on whether the participants provided the correct answer or not. The participants were told that their scores would depend on how fast they responded to each card, but the time interval between cards was set to 5 seconds. The test was implemented in E-Prime, and we recorded the number of correct responses and the RT to correct response.

Figure 4 illustrates the effects of games and activities on different signals measured during the experiment (mean, SEM). Session by activity interactions had a significant effect on EDA (χ²₁₀=26; P=.004), HR (χ²₁₀=24.5; P=.006), HRV (χ²₁₀=96; P<.001) but not on Cortisol (χ²₆=6.5; P=.37). In all sessions, we observed the highest levels of cortisol at the baseline, followed by an exponential decline in cortisol levels. Detrending the cortisol levels to remove this decline did not change the outcome of the statistical inference.

The main effect of Session was significant for EDA (χ²₂=20.6; P<.001), HR (χ²₂=23.4; P<.001), and HRV (χ²₂=33.3; P<.001). The effects of control variables (acceleration and duration) were not significant on any of the outcome measures (albeit a trend was observed for a positive association between acceleration and HR (P=.08). There were no significant differences between S₁ and S₂, but EDA in S₃ was significantly higher than that in S₁ (mean difference 4.3 muS; P<.001; 95% CI 2.77 to 5.9) and S₂ (mean difference 3.97 muS; P<.001; 95% CI 2.4 to 5.5), expectedly as a result of physical activity. The HR in S₃ was higher than that in S₁ (mean difference 10.02 bpm; P<.001; 95% CI 7.5 to 12.2) and S₂ (mean difference 9.6 bpm; P<.001; 95% CI 7.2 to 12 bpm) and the previous two sessions, but the difference between S₁ and S₂ was not significant. The main effect of Activity was significant only for HR (χ²₅=28.8; P<.001) and HRV (χ²₅=79.3; P<.001).

We had hypothesized that if a game was stressful, then it would interfere with cognitive processing. Indeed, we found that different game sessions affected the rate of correct immediate recall in the encode (χ²₂=10.2; P=.006) and delayed recall (χ²₂=24.6; P<.001), as well as on Stroop’s RT to correct (χ²₂=6.5; P<.05) but not on Stroop rate of correct response (χ²₂=2.5; P=.29). Post hoc tests showed significant reduction of performance in the exergame session (S₃) compared with baseline (brain training, S₁) during encode (mean difference −12.6%; P=.02; 95% CI −23.5 to −1.68) and recall (mean difference −22.96; P<.001; 95% CI −35.6 to −10.2). A similar trend was observed in comparison of Car Race (S₂) vs S₁ (for encode, mean difference −10.9; P=.09; 95% CI −23.3 to 1.45; and for recall, mean difference −11.48; P=.06; 95% CI −23.4 to 0.47). No difference in encode rates was observed between S₃ and S₂, but there was a trend of reduced correct recalls (mean difference −11.48%; P=.08; 95% CI −24.0 to 1.06). Stroop RT was also slower in S₃ vs S₁ (mean difference 440 ms; P=.006; 95% CI 130 to 759) and S₂ vs S₁ (mean difference 269 ms; P=.03; 95% CI 20 to 510). Because encode and recall rates were correlated (r=.631; P<.001), we controlled for the effect of encode hit rates in the GEE model that estimated the effect of the session on recall hits. Regardless, the effects of Session on recall performance remained significant (χ²₂=16.1; P<.001). These results suggest that even though the participants had become familiar with the tasks, their performance did not improve by the third session; it declined instead, suggesting that the exergame was the most challenging and potentially the most stressful game of the three. Figure 5 illustrates how this stress effect on performance was exacerbated in the group with lower MoCA scores.

To examine whether cognitive reserve would predict the stressfulness of the gaming experience, the sample was split based on MoCA scores to low-MoCA (MoCA<27; n=8; mean 24.62, SD 1.41) and high-MoCA (MoCA ≥27; n=11; mean 28.27, SD 1.19) groups.

Independent t tests did not show significant differences in age (95% CI −3.8 to 2.3; P=.53), Perceived Stress Scale (95% CI −3.36 to 4.36; P=.73), General Self-Efficacy (95% CI −3.6 to 2; P=.53) or Loneliness scale (95% CI −19 to 2; P=.13)

Figure 7 illustrates the patterns of game appraisal in the MoCA groups. In the screening questionnaires, the high-MoCA groups had a generally more positive attitude toward playful experience. The Kruskal-Wallis test demonstrated that in the pre-enrollment appraisal questionnaire, the low-MoCA group considered digital games too hard to learn (χ²₁=4.05; P<.05), but other scores were not significantly different. In the postgame appraisal questionnaires, the low-MoCA group disliked the experience more (χ²₁=5.1; P<.03) and found the game more useless (χ²₁=4.92; P<.03), less interesting (χ²₁=4.90; P<.03), and visually more intense (χ²₁=5.6; P<.02); no other differences were statistically significant.

We expected individuals with lower MoCA scores to find the games more challenging and therefore be more stressed by it. To compare the groups, we first normalized all the physiological variables to the within-session and within-subject baseline. Figure 8 illustrates group differences in the pattern of physiological signals over time. Because the physiological profile was significantly different between sessions, we compared group differences separately for each session. A GEE model with independent variable MoCA was then tested for each physiological dependent variable. The results are summarized in Multimedia Appendix 1. The only significant differences in reflexive responses to the experiment were observed in the cortisol levels in the first session (S₁) and in the HR during the CareRace (S₂).

In S₁, cortisol in the low-MoCA group had a significantly larger drop from the baseline (mean difference −0.218; P=.046; 95% CI 0.004 to 0.432). This effect can reflect the fact that the low-MoCA group was more stressed about taking part in the study from the onset. Although not statistically significant, this group showed a steeper decline in cortisol level in the exergame session as well.

In S₂, the low-MoCA group increased their HR with respect to baseline and compared with the high-MoCA group. A trend for increased HR during cognitive tests during S₁ (Encode and Stroop) was also observed in the first session. These effects might suggest that irrespective of games, the low-MoCA group found the cognitive challenges in the experiment difficult and stressful.

We expected that the performance of the low-MoCA group in cognitive tasks would be significantly reduced compared with that of the high-MoCA group. As Figure 5 illustrates, the interactions between MoCA Group and Session were mainly driven by session effects, which affected both groups similarly. The GEE test (Session_i×MoCA) revealed significant interaction effects on all cognitive variables (all P<.005), except Stroop Correct (χ²₅=7.4; P=.19). Post hoc analyses show that for both groups, performance in memory tasks was significantly lowered in S₃ vs S₁ (12% less hits in immediate recall, P<.001; and 22% less hits in delayed recall, P<.001). The Stroop RT was significantly increased in S₃ vs S₁ (mean difference 461 ms; P=.001). In the low-MoCA group, a small reduction in Stroop accuracy was observed between S₃ and S₁ (mean difference −2.1%; P<.03) and between S₃ and S₂ (mean difference −2.33%; P=.008). The Stroop RT to correct in S₃ was also the highest and significantly different from that of S₁ (mean difference 535 ms; P<.001). The only significant difference between the two groups was in percentage of correct Stroop in S₃ (mean difference 2.03%; P<.03). Although the game difficulty affected the Stroop performance in the low-MoCA group, it did not have any impact on the performance of the high-MoCA group.

Our snowball sampling survey was important in informing the results about the characteristics of those who found the topic of serious games interesting. Interindividual variations in game perception and motivation are important factors that engender the gameplaying experience [58,59]. The AGPHA framework [60] builds on the appraisal theory of stress and coping [90]. When confronted with a new challenging encounter, the primary appraisal process is to categorize it as irrelevant, benign-positive, or stressful, depending on what implications it would have for the individual’s well-being. If the person has no investment in the outcome of the challenge, then they will have no need for it and will not commit to engaging with it. On the contrary, if they perceive immediate or potential benefits, they are willing to try it and enter the secondary appraisal stage. It is in the second stage that the individual focuses on evaluating the challenge: Is it feasible and within my physical cognitive abilities?

Our recruitment advertisement targeted a large mailing list of people interested in preventive health care and attracted 42 responses to the survey. Therefore, the number of individuals who entered the primary appraisal stage was not large. Of those, only 19 agreed to enter the secondary appraisal process. Those who did not participate were more likely to agree with the statement Digital games are too hard to learn, suggesting that from the onset, those with less confidence about their cognitive abilities excluded themselves from the experimental phase. Overall, the patterns of response (Figure 3) illustrated that the participating group had a generally more positive attitude toward gaming activities. The participating and nonparticipating groups did not differ in any other characteristics (eg, age, education, self-efficacy)

It must be emphasized that the participating sample had high self-efficacy and was well educated and socially and cognitively active. Thus, the following discussion hinges on the fact that the participants in the study were not representative of the aging population who are in need of cognitive or physical assistance. To address their needs by gamified digital interventions, we must repeat this experiment in a group with pronounced cognitive or physical disabilities (eg, in a nursing home or a rehabilitation center). Theoretically, any factors (such as accessibility or executive difficulty) that influence game appraisal will theoretically change the subsequent reflective and reflexive outcomes as well.

The second question we asked was whether playing a new digital game would cause stress. If this were the case, then we would expect to observe a significant increase in cortisol and ambulatory signals after the introduction of each game. As the patterns of signal in Figure 4 illustrate, this was not the case.

Cortisol is the hallmark of a latent psychosocial response to stress, but we did not see any significant cortisol effect after any of the gaming activities in the experiment. One explanation is that the highest levels of cortisol were measured at arrival time. This could have resulted from the participant’s anticipation of the session, or from exertion of walking to our center. Because the hypothalamic pituitary adrenal stress system is regulated through a negative feedback mechanism, it is plausible that while the body was downregulating the high levels of basal cortisol, the system was not as responsive to the stress that may have been introduced during game play [91]. Detrending the cortisol signal to remove the sharp decay showed only a mild (and statistically insignificant) increase in cortisol post exergame.

While the cortisol response results from complex and relatively slow neuroendocrine processes, the EDA and HRV mark sympathetic autonomic response and have widely been used in human-computer interface studies [76-82]. The exergame was physiologically more demanding and higher physiological response was expected. Indeed, we observed significant differences in EDA, HR, and HRV, which significantly changed during the exergame (Figure 4) independent of the effect of movement and body temperature (measured on E4).

Interestingly, reflexive and reflective responses were not corresponding. Despite being more difficult, the exergame was not subjectively rated as more stressful (Figure 6). Nevertheless, the lowering of performance in cognitive tasks, which became statistically significant in the stress-sensitive recall task [71], indicates that it had a more pronounced interference with cognitive processing. This interpretation is consistent with previous reports that have shown a link between cognition and training with exergames [92-95]. In other words, the fact that exergame affects the cognitive performance immediately proves that it exercises the brain. Given that effects of stress on cognition are mediated through reward processing [83] and attention [84] systems, the challenge for game designers as well as gerontologists is to evaluate to what extent, for whom, and which exercises can become beneficial to the neurological health of aging individuals.

Our third question was whether differences in cognitive ability (based on MoCA) predicted differences in reflective (Figure 7) and reflexive (Figure 8) response to games.

In terms of primary appraisal, the low-MoCA group found digital games too hard to learn. In the secondary appraisal (ie, after playing), the low-MoCA group was less likely to enjoy the game or find it interesting, less likely to find it useful, and more likely to find the game visually more intense. However, the groups did not differ in finding the games difficult or stressful. According to the appraisal theory, these anticipatory and perceptual differences should predict an increased stress response.

Indeed, physiological reactions to the games were stronger in the low-MoCA groups, although they were not statistically significant. The exception was in cortisol levels. The low-MoCA group had 21% more decrease in cortisol levels with respect to baseline during the first session (brain training). This effect may have resulted from the anticipatory stress of the first session, or from the MoCA test itself, which was administered before the experiment began, and possibly caused a social evaluative stress.

Despite differences in perception and cortisol dynamics, cognitive performance in the first session was not significantly different between the two groups, and they both had high rates of correct responses to cognitive tests (>80%). However, in both groups, the performance in those tasks was lowered by increasing game difficulty and more pronounced in the low-MoCA group (Figure 5).

Van Reekum et al [80] have demonstrated the correlation between autonomic responses (cardiac and electrodermal signals) and performance in an action video game. It is plausible that the low-MoCA group experienced higher states of arousal as a result of adaptive coping to compensate for their initial discomfort with the game. This interpretation is consistent with the observation of Birks et al [58], who have argued that adding extrinsic motivation increases the efforts of those who do not have high levels of affinity with their game avatar. Thus, an increased stress response in the low-MoCA group could indicate the player’s increased engagement with the game. Whether this engagement will lead to a desire to repeat play needs to be tested.

To the best of our knowledge, this is the first quantitative crossover study to have compared the psychophysiological responses of older adults to different genres of marketplace games, with different levels of complexity in aesthetics, dynamics, and mechanics of games. Intergenerational differences in such studies must address differences in needs and values, as well as differences in physiology.

Poels et al [79] used a similar study design, but their participants were 19 young players. In the study by Poels et al, participants were monitored for hedonic and physiological reactions to four action games with different aesthetic features (two first-person shooters and two race games, with varying degrees of visual complexity). Their study indicated that initial physiological reactions predicted the likelihood of repeat playing in the long term.

Mandryk and Atkins [81] found that galvanic skin response, facial electromyograms, and cardiovascular responses can be used in machine learning algorithms to dynamically compute the degrees of valence and arousal during a gameplay session in 24 young gamers. Their study showed a high convergence between the subjective ratings of games and the machine-predicted levels of emotion and arousal.

In our study, postgame appraisal suggests that the desire to play the game again was highest in the brain training game (14 of 18 players indicated that they will play those games again), which was associated with lower levels of physiological response. In other words, it appears that younger players in the study by Poels et al [79] played games to induce a physiological response, whereas older players in our study preferred a game that did not induce strong reflexive responses.

Our findings are consistent with previous reports that in general, more seniors prefer the ease and pleasure of casual games (preferably with an intellectual component) over more cognitively demanding action games [3,56,57]. Our observations raise the following important questions that cannot be addressed in this study. What are the intergenerational differences in game appraisal? How are these differences in game appraisal correlated to age-related differences in physiological and cognitive reserves? Are physiological measures appropriate for evaluation of seniors desire to engage in the game challenge?

In a meta-analysis of 48 studies, involving uniplayer games (excluding exergames), Van der Vijgh et al [96] demonstrated and argued that although games can explain up to 57% of stress-related physiological variations (HR and blood pressure), it is not the game alone but the context of the experimental design and the study characteristics that moderate the stressful response to games.

In this study, we wanted to make the participants comfortable with games (eg, by sequencing the games from simple in the first session to hard in the last session). However, these games were presented in a serious context by the title of the study An Objective Assessment of Interactions Between Appraisal, Arousal and Cognitive Benefits of Electronic Playing. If the study characteristics (which we modified by altering the games in the different sessions) are important, then arguably, our complex experimental setting, and the inclusion of the memory and Stroop tests (even though simplified) could have altered the phenomenology of the experience and thus have hindered the physiological responses to the gameplay alone. Nevertheless, our complex experimental setup allows to illustrate the complexity arising from contextual factors that interact with psychophysiological adaptation. In real life, games are not experienced as isolated laboratory events, but in relation to other activities. The complexity of Group×Session×Activity interactions with physiological signals (Figure 8) provides evidence to concur with Van der Vijgh et al [96] that it is the context in which games are experienced that counts.

Acute stress is not necessarily a maladaptive response. Stress interacts with learning and cognitive function through myriad physiological and behavioral cascades [64,84]. The evolutionary function of an acute physiological stress response is to focus and consolidate the lessons to be learned from surviving new challenges [65,84,97,98].

Van Reekum et al [80] have demonstrated a correlation between autonomic responses (cardiac and electrodermal signals) and performance in an action video game. In our design, neither could we reliably record nor did we want to measure the player’s performance in the commercial games that we presented. Instead, we used the simple version of a Stroop test in which RT is expected to be affected by acute stress [71,86], and which seems to acutely improve after casual exergaming in a younger sample [87]. We also used an encoding/recall task, which we have previously shown to be sensitive to acute stress [71,85]. Although these measures were not chosen to be difficult (eg, the number of Stroop trials was very small) or to measure specific cognitive domains, they served as control variables to help us assess whether game difficulty could have an impact that exceeded the learning effect.

We found that the RT to a very simple Stroop (4 color, 75% incongruent stimuli) increased with game difficulty and that delayed recall accuracy decreased with game difficulty, especially in those with lower MoCA scores. Although we cannot infer about how games interacted with cognition, we demonstrate that such cognitive tests may offer a lower cost measure to trace effects of repeated play on neuropsychological factors that underlie cognition. As we have argued in introduction of the AGPHA framework [60], to document such data can potentially guide the decisions made in the design cycle, in terms of how to activate specific cognitive targets by modifying game aesthetics and dynamics.

This study is limited primarily by the sample size, which was estimated with the expectation of 80% of power to detect a cortisol response to an explicit stress challenge [71]. In older adults, stress responses to the same task may be blunter than those in a younger sample [99]. Age-related variations in stress sensitivity may be connected to a myriad of endocrinological or psychological factors, all of which need to be carefully modeled using larger samples [100]. Adding multifactorial quantitative measurements (growing in availability and reducing in cost) to larger crowdsourced experiments such as the one conducted by Birks et al [101] could provide a finer grained picture of the physiological embodied qualities of a game experience. This would be of considerable importance for the clinical industries that are emerging around the utilization of game-based interventions in mental health care [32].

In addition, the complexity of our setup makes our ambulatory measurements more susceptible to instrumentation noise. Despite setting up the procedure to be identical and obtaining repeated measures of cortisol and ambulatory signals at baseline, we did not achieve interclass correlations above 70%. Endocrine measurements (eg, cortisol) are sensitive to factors such as time of the day (circadian rhythms), activity before sampling, and states of health and medications. Although we controlled and measured all of these factors, our small sample size does not allow including all of them in our statistical models. These factors must be controlled for in larger studies.

The second limitation of this study is inherent to its experimental design. We wanted these experiments to be participatory and gave primacy to the participant’s appraisal of games in relation to their cognitive benefits. We therefore avoided stressful elements such as keeping scores or time pressure. The main experimental variants were gradual increase in game complexity and MoCA categorization. Our repeated measures experimental design is advantageous in terms of statistical power but is not immune to habituation effects, which is a very important and confounding factor in stress studies.

The third limitation of our study is to have split the sample by MoCA scores. MoCA includes 30 questions, and a score of 1/0 per each question, to account for differences in short-term memory, orientation, visuospatial and executive functions, language abilities, abstraction, and attention. This categorization, especially in the context of evaluating a complex and interactive medium such as a game, is too reductionist. To use MoCA as a discriminating factor provided methodological simplicity to offer a proof-of-concept example of the application of AGPHA framework, but it did not account for the variations in behavioral strategies that compensate for decline in those specific cognitive domains. These issues must be further investigated.

Finally, our test studio (blue-floored and mirrored-walls) and the various assessment tools we used were quite burdensome and far from ecological. To repeat this experiment in a familiar setting and with a simpler design, such as in a living laboratory, a nursing home, or a senior’s community center, will likely produce different results.