Serious games could someday have substantial health impact by facilitating patients in learning to cope with chronic pain (CP) or functional somatic syndrome (FSS). A serious game is a kind of computer game that not only aims to provide fun, but also to inform, instruct, and modify [1]. In many patients with CP or FSS, functioning can improve with the moderation of psychosocial consequences or perpetuating factors of their symptoms [2,3]. Increasing evidence suggests that serious gaming can facilitate processes of change in health behaviors and their antecedents, as well as in mental health in patients of any age [1,4]. Randomized controlled trials with stand-alone serious gaming interventions have been reported [5,6]. The findings suggested clinically relevant mental health effects compared to waiting-list conditions, or an equivalent effect compared to traditional face-to-face delivery modes. However, complementary evaluation studies of strong internal and external validity were suggested [1,4]; they are needed to inform and legitimize the implementation of serious games on a large scale, which, assuming that development costs are high and variable costs low, is a plausible requirement to realize value for societies.

This study aims to develop theory with which future developers and implementers of serious games can improve their design and integration in regular health care settings. To achieve this, realist process evaluations were embedded into a quasi-experiment with the serious game LAKA.

CP is pain, with or without a specific organic cause, that persists longer than a usual 3- to 6-months of organic recovery [3]. An FSS is characterized by a persistent pattern of bodily symptoms that cannot be sufficiently pathologically explained after adequate physical examination [2]. Close to 1 in 5 adults has intense pain (ie, rating 5/10 or more, for at least 6 months) [7-9]. FSSs characterized by irritable bowels (11%), chronic fatigue (1%), tension headache (2%), or tinnitus (10%-15%) are prevalent as well [10-12]. The severity and chronicity of lower back pain, which is the most common symptom of both CP and FSS, may be associated with more functional disturbance during people’s lifetime than any other disease [13].

CP and FSS often come with psychological or social burdens. Comorbid mental disorders were found in 35% of patients with CP [14]. Similarly, substantial minorities of patients with FSS also have another FSS or a psychological disorder [15,16]. Compared with individuals without pain, those who experience more than 3 months of disabling pain (7.4%) are about six times more likely to be absent from work and two times more likely to visit a doctor [17]. In addition, economic costs of sick leave and early retirement exceed the total amount of medical expenditures associated with both CP and FSS [18-20].

For managing burdens of CP and FSS, biopsychosocial approaches are considered effective [2,3]. Meta-analyses show positive effects for physical and emotional functioning of various treatment options (ie, medication, complementary medicine, psychological therapy, or multidisciplinary rehabilitation) that are small or medium sized, at most [21-28]. Recommended treatment includes steps of conservative medication, psychotherapy, and physiotherapy, and a multi- or interdisciplinary rehabilitation program when previous treatments do not suffice [2]. Multidisciplinary rehabilitation programs vary in content but commonly include medical, psychological, physical, and occupational interventions [21,26]. New knowledge, with which treatment effectiveness for different CP and FSS patients may be improved, comes from research into the following: (1) the biological mechanisms of specific symptom patterns [29] and (2) how adequately (eg, feasibly or cost-effectively) already understood biopsychosocial mechanisms are targeted by certain treatment approaches, strategies, and modes of delivery [30,31]. Regarding the latter, current interest goes to the potential of facilitating intervention with computer technology [32,33].

CP and FSS conditions are diverse, but a common, partial, explanation is given by the sensitization of the central nervous system [34]. Biopsychosocial approaches target the following: (1) peripheral systems with pharmacotherapy (ie, bottom-up approaches) as well as (2) cortical brain systems with neuroscience education, cognitive behavioral therapy, and exercise (ie, top-down approaches). Changes in coping with stress and pain, avoidance beliefs, rumination, acceptance, or catastrophizing mediate intervention effects on experienced physical and mental functioning [35,36]. An approach for changing coping responses is to promote individual mindfulness (ie, self-awareness, self-regulation, and self-transcendence) in response to adverse experiences (eg, negative emotions or physical sensations) [37]. Another approach is to restructure particular cognitive antecedents of pain-related experiences and behaviors [38-41]. Whether and how these approaches differ with respect to underlying change mechanisms and outcomes has been debated [38,39].

Literature about patient factors for effectiveness provide limited evidence to inform practice. Some studies claimed that neither demographic nor psychological differences between patients predict clinically important variation in treatment effects [42,43]. Other studies, however, stressed that fear-avoidance beliefs, pain acceptance, or depressive symptoms can predict treatment gains to some extent and should therefore be targeted early or additionally in treatment [44-47].

More insight is also needed about the characteristics of interventions and treatment contexts that are responsible for varying intervention effects. Meta-analytic results showed that computer-based intervention (eg, delivery over the Internet) can provide modest chances for symptom reduction, similarly to face-to-face group therapy of similar content [30]. It also showed uncertainty about the transferability from results in self-selecting participants to wider populations and settings. Researchers have been repeatedly suggesting that better explanations of varying effectiveness levels between studies requires contextualization; for instance, consideration of (1) factors such as fidelity of implementation, program compositions, and comparisons and (2) interdependencies and interactions between such factors [31,44,47-49].

For serious games, there is a similar need for theories of what works how, for whom, and when [50,51]. Games or virtual reality have motivational qualities that may support the extinction of phobias, distraction from pain, repetitive physical or cognitive training, or learning about cognitive antecedents (eg, self-efficacy) of health behavior [52-54]. Gaming features, including stories and interactivity, may strengthen behavioral change processes with engagement, intrinsic motivation, positive affect, and sense of presence when processing educational content [52,55]. Various studies showed that effects of serious games can vary with intervention factors (eg, participatory design and duration) and users (eg, gender, age, intelligence, and gaming experience) [1,51,56]. Individual experiences of serious gaming may be difficult to predict and may be context dependent [57,58]. Debriefing is a way of dealing with such variation by leveraging recipient experiences after serious gaming for transferrable individual learning results as intended [59].

To the best of the authors' knowledge, the first outcome evaluation of serious gaming for reducing disease burden in patients with CP or FSS was reported [30,60]. The effect of serious gaming was studied in a quasi-experiment with patients with mostly chronic musculoskeletal pain and psychosocial problems. A comparison was made of symptom changes during a regular multidisciplinary rehabilitation program (100 hours) between (1) an intervention group that received an additional blended intervention with the game LAKA (4 extra hours) and (2) a control group that received no serious gaming. A very small acceleration of physical and emotional symptom reduction was attributed to serious gaming: a standardized regression coefficient for the group × time effect of 0.12 [60]. This effect size estimate corresponds with a comparable estimate from an earlier meta-analytic subgroup analysis (ie, a 0.13 standardized average difference in short-term behavioral or mental health outcomes between groups following multi-component interventions with and without serious gaming) [1]. This suggests that a couple of hours of serious gaming as part of treatment does not provide a general clinically relevant benefit, per se, but can add to the effectiveness of the treatment as a whole. Plausibly, serious games and other computer applications for CP or FSS patients reach stronger, average effects when delivered as stand-alone interventions in other circumstances (eg, when compared with no active treatment) [1,30].

Existing theories on related topics provide a starting point for building transferrable insights into how and why features of a serious gaming could work in certain contexts for patients with CP or FSS (see Textbox 1). Changes of rehabilitation outcomes due to serious gaming are potentially explained with theories about health behavior, including mindfulness [37], relational framing [61], psychological well-being [62], or self-discrepancy theory (SDT) [63]. Why serious gaming features could be a distinctive trigger for learning has been explained with theories about motivational and affective responses to virtual or computer gaming environments [52,64]. Other potentially applicable theories described a comprehensive range of potential context factors for implementing health care innovations [65,66]. More research upon which theory building rested is summarized in Textbox 1 [67-71].

This study primarily aims to provide in-depth information to recipients, developers, evaluators, and implementers on contexts in which certain patients with CP or FSS can use features of serious gaming in ways that result in clinically important health benefits. The overall research question is as follows: (1) How, (2) why, (3) for whom, and (4) when does serious gaming lead to learning and health outcome change? Accordingly, the objectives are enhanced propositions about the following:

The process evaluation uses a realist evaluation approach [72]. Empirical process evaluations are the preferred method to gain a thorough understanding of which intervention features contribute to clinically relevant benefit, how, why, for whom, and under which complex circumstances in real health care contexts [73,74]. Such insights can benefit future choices of development, evaluation, and implementation (eg, selective allocation, tailored design, and rollout in other settings). Realist principles can guide theory-based and contextual, sensitive, process evaluations of complex programs or interventions—terms are used interchangeably [73,75].

A program theory is developed on the basis of empirical evidence that indicates whether initial ideas are supported or must be refuted, extended, or refined. Program theories are built up from configurations (C) of intervention characteristics (I) in context (C) triggering a mechanism (M) that leads to certain outcomes (O) (ICMO configurations or ICMO-Cs). An intervention is something new to, or extracted from, a pre-existing situation, certain aspects of which interact with the intervention to elicit causal effects (ie, context). A mechanism is an underlying (ie, invisible) causal force of empirical events, that is, not a description of successive events under counterfactual situations [76]. Mechanisms have often been framed as the reasoning with which recipients respond to the resources of interventions [77], but ideas about what they are can vary with the topic of interest. The ICMO-C is also applied to conceptualize on a middle range of abstraction for “dealing with different spheres of behaviors and structures to transcend sheer description” [76]. Realist approaches use induction, to discover regularities in empirical phenomena, in alternation with deduction, to formulate expected observable consequences of abstract formal theories, for inferring the best explanation of observable outcome patterns (ie, abduction) [78].

An initial ICMO-C was formulated by taking assumptions from several of the introduced theories on related topics (see Textbox 2 as well as Multimedia Appendix 1 for a comprehensive overview). In addition, it was expected that serious gaming would be more generally accepted and better adhered to after improving the delivery modes of serious gaming in the treatment context, as suggested by previous pilot study results [67].

The process evaluation is designed as an embedded, two-armed, natural, quasi-experimental, mixed methods study [66]. It was carried out by an integrated team of researchers, who are trained in various quantitative and qualitative methods. Priority was given to quantitative methods for investigating patterns in routine clinical patient outcome assessments. Patients were recruited who were following a standardized 16-week multidisciplinary rehabilitation program, of 100 hours on average, in one of four treatment sites of a single Dutch rehabilitation center. At two sites, an additional 4-hour blended intervention with the mindfulness game LAKA was provided during the second half (weeks 9-12) of the rehabilitation program (ie, intervention condition). In the other two sites, gaming was not offered (ie, control condition). Measurements were routinely taken at baseline (t0), at an intermediate time point after 8 weeks of treatment (t1), and posttreatment after 16 weeks (t2). Nonintrusive qualitative data collection took place before, during, and after the experiment. Analyses of qualitative data for formulating hypotheses preceded the inspection and analyses of the quantitative outcome data (see addendum to the registration in the Netherlands Trial Register, NTR6020). All steps from the qualitative and quantitative research are presented in Table 1.

The alternative hypotheses formulated in step 2 specify certain patterns of linear relationships between variables that operationalize outcomes and characteristics of the intervention in context. Such statements were listed, a priori, to strengthen support, or refutations, of tentative ICMO elements. An ICMO-C gains support, or refutation, with quantitative results that should or should not be considered very unlikely under the assumption that the ICMO element is superfluous, as follows:

The Adjudicating Formal Theory and Formulating Hypotheses section explains the ICMO elements to which the hypotheses relate.

The setting and participants were fully described in the outcome evaluation report [59]. Treatment and control sites had similar protocols, sizes, histories, and absence of disruptive activities during the study period. Patients with a physicians' indication of eligibility for multidisciplinary rehabilitation received informed consent from a familiar care provider soon after the second part of rehabilitation treatment started. Participation included additional data collection for the study and permission to process codified, routinely collected, clinical patient data. A total of 275—156 intervention group and 119 control group participants—out of the 329 eligible patients participated (83.6%). Patients in the sample were, on average, 44 years of age (SD 11.3, range 18-67). The proportion of females was 69.8% (192/275). Almost half of the patients (134/275, 48.9%) reported a symptom duration of over 2 years. Patients (N=275) mostly had musculoskeletal pain and concomitant psychosocial problems. The findings of the outcome evaluation suggested no confounding influences by treatment site or patient baseline variables that differed, statistically significantly, in mean levels between the study groups.

Other participants included care providers (ie, three psychologists and one physiotherapist), local managers who were responsible for providing the serious gaming intervention on site, and stakeholders with relevant expertise in information and communication technology (ICT), serious gaming, rehabilitation medicine, health psychology, and spiritual counseling. All were familiar with the setting and provided informed consent on the participation in focus group sessions.

Features of the multidisciplinary treatment and additional serious gaming intervention offered to the intervention group were previously reported in detail [66]. A short description of LAKA is given in Textbox 3. The clinic facilitated tablet computers, suitable rooms with Wi-Fi connections, and the automated planning of four 1-hour sessions, for 1-6 patients simultaneously, in connection to regular therapy hours. Three psychologists and a physiotherapist (two per site) provided support during the first (ie, introduction) and fourth (ie, debriefing) sessions. Topics of debriefing were technology acceptance, play experiences, and learning transfer to daily life. Access to LAKA during the second and third sessions was provided on site by local staff members. In addition, patients could play LAKA at home on a tablet computer with an Internet connection. This connection made it possible to download LAKA via an app store and to safely exchange data with the rehabilitation center for access to the app, storage of progress, and feedback on performance. All patients were expected to attend a debriefing session after completing the game at least once: this took 2.5 hours on average. This intended dose was based on what had been a natural amount to patients during a feasibility study [68].

Interviews and feedback surveys contained open-ended questions, which were intentionally free from theoretical preconceptions. Such open-ended questions were posed first to give space for explanations of initial reactions, including unexpected ones. Topics and additional follow-up questions for interviews were based on the existing theoretical frameworks on related topics (see Introduction section). The prioritization of topics varied with the expected knowledge areas of the participants [79]. Participating patients were considered to be most aware of outcomes and characteristics of the intervention in context, while care providers knew most about rehabilitation mechanisms.

Focus groups were first held with stakeholders who represented various roles and areas of expertise. Participants prepared for the focus groups by playing LAKA and reading pilot study results and adapted intervention specifications. At the beginning of the session, the research questions were presented for collecting local experts' initial expectations. In part two, care providers and implementers built upon shared expectations for specifying local implementation procedures in further detail. The second and third focus group interviews were held with two care providers who worked at the same intervention site. At the beginning of the second focus group, initial expectations about intervention effects on rehabilitation mechanisms were explained. During the natural experiment, care providers also shared feedback informally with MAPV. The third focus group with care providers first generally addressed their postexperimental experiences. These were about how recipients responded to the intervention in the context, including how those responses were shaped by care providers' own reactions. The discussion was enriched by sharing tentative qualitative results (ie, parts of tentative program theory).

The first source of qualitative patient information consisted of the given responses on two open-text requests after finishing playing LAKA: “Please describe in your own words what you experienced when you played LAKA” and “Please explain how the LAKA sessions, according to you, will contribute to your daily life.” Furthermore, semistructured interviews were held with patients. These interviews were held with (1) intervention group patients who were purposively selected by differences in outcome expectations and (2) control group participants with similar baseline characteristics as interviewed intervention group patients. Interviews took place when patients had completed their rehabilitation program. Main topics of each interview were what worked and how, in order to change courses of symptoms and learning in the specific situations of the patient throughout rehabilitation (eg, “Could you explain how your health status developed throughout the rehabilitation program? What aspects of the rehabilitation program really made a difference for you? What was the role of serious gaming, if any, assuming differences occurred? What aspect of the game contributed like that?“). A data saturation point was reached when no new information emerged by asking patients about three good and three bad things about serious gaming. The more initial responses from recipients were already collected and analyzed, the more often MAPV shared an interpretation of a fragment in order to question its accuracy. If potentially relevant topics for the development of program theory were not addressed at the initiative of patients, they were sometimes introduced with questions that referred more directly to specific theory-based expectations (eg, ”Were you concentrating while playing LAKA?“).

The initial program theory informed the collection of quantitative data sources. Table 2 lists the times of assessment and the internal consistency information of all quantitative measurements that were processed for hypotheses testing. Health outcomes were measured with validated questionnaires on depressive mood, pain intensity, and fatigue as part of the routine clinical data collection [68,80-82]. Learning result was measured with a mindfulness questionnaire, of which unpublished psychometric results gave support for being a valid measure (see Multimedia Appendix 2, which contains information in support of a good internal consistency, low social desirability, and patterns of association with other constructs that are similar to those of other validated mindfulness questionnaires) [83,84].

For additional indications about potential mechanisms, a short postgaming feedback questionnaire with complementary scales was added to the routine clinical measurement system. This postgaming feedback questionnaire included Likert-scale items on sense of presence (ie, involvement and realism subscales) [85], positive and negative affect [86], and patient acceptance perceptions with regard to the serious gaming intervention [87,88].

Indicators of patient factors included clinical data about demographic, health status, and coping [89,90] variables as measured at the latest available time point before serious gaming. Classification norms from manuals of the questionnaires were used to divide patients over subgroups. Log data about the circumstances of the intervention in context were disaggregated to the patient level (see Table 2). This includes dummy variables for indicating each intervention site, the care provider who provided debriefing sessions, and the social structure of debriefing sessions: a care provider with a patient alone, a dyad, or a group. Moreover, differences between dates indicated coincidental variation by central planning and general changes of serious gaming sessions in time.

All interview texts (ie, verbatim transcripts) were coded by MAPV in ATLAS.ti, version 7.5.16 (Scientific Software Development GmbH), using sensitizing concepts for intervention, context, mechanism, and outcome unless new codes were needed to cover unexpected statements. The first four patient interviews were also independently coded by MCWJ (two interviews) and AMEEZ (two interviews). Unresolved differences were discussed with HJMV. It was counted how often codes for open-feedback text fragments recurred to get a sense of the relative importance of various themes and coverage of those by the interview data.

In step 1, ICMO relationships were determined per patient interview when explicit statements were found about relationships between positive outcome expectations, mechanisms, and the intervention in context (see Multimedia Appendix 3 for illustrations). In addition, intervention or context factors were coded as barriers when they were explicitly related to suboptimal, weak, or absent activation mechanisms or outcomes. Tentative ICMOs, documented down in theoretical notes, summarized regularities that were found when comparing individual patient results with each other and with the qualitative data that were obtained from other stakeholders. Theoretical notes formed the basis for regular bilateral or team discussions that continued during the next step.

In step 2, decisions were made about which formal theoretical perspectives were most suitable for explaining the qualitative data about intervention mechanisms. This was a starting point for adding, actually resuming, a deductive process to specify indicative patterns of quantitative data (ie, observable effects) of tentative ICMOs. Hypotheses were formulated that were testable with the available (ie, uninspected) quantitative data. Formulations of hypotheses had to take into account the limitations of the sample size, avoiding too many or too complicated hypotheses, and a lack of possibilities to add new quantitative measurement instruments at that time. The research protocol was ethically assessed before data collection began.

The ggplot2 package from R, version 1.1.463 (The R Foundation), was used to visualize levels of change in health outcomes, change of mindfulness, and game acceptance levels between subgroups of patients in context. Provisional classifications of favorable or not favorable were made with combinations of variables about the following: (1) whether a patient received the serious game intervention and how the implementation went, individually, and (2) his or her pre-existing characteristics.

All prediction models were executed in SPSS, version 24.0 (IBM Corp), using applications of the ordinary-least-squares estimation algorithm. All models were based on complete cases. All variables, except for dummy variables and dates, were centered or standardized before inclusion in statistical models. Outcome change indicators are standardized residual scores after regression of postintervention (t2) scores on preintervention (t1) scores. Decisions on the null hypotheses are informed by two-sided P values and 95% CIs. Various sensitivity analyses were performed to indicate the robustness of the methodological choices, including missing data handling. These analyses are discussed among the strengths and limitations.

The first type of procedure applied Hayes’ PROCESS macro, version 2.16, for SPSS to investigate hypotheses 1 and 3 [91]. These procedures calculate CIs using a nonparametric bootstrap approach that, by default, generates 5000 samples. Hypothesis 1 was investigated by fitting models of type 4. These models specified estimations for effects of study group on changes of the three health outcomes as mediated by mindfulness changes. Mediation models were extended to moderated mediation models of type 58 for testing hypothesis 3 (see Figure 1). These models calculated the following: (1) parameters for the interaction between the moderating patient factors and study group for predicting levels of the mediator (ie, mindfulness change), (2) parameters for the interaction between the moderating patient factors and the mediator for predicting levels of the health outcomes, and (3) an index of moderated mediation assessing differences in the total indirect effects between two patient subgroups.

Secondly, multivariable linear regression models were used. A model with mindfulness as the dependent variable, and the subscales of sense of presence and affect as the independent variables, provided relevant test statistics to decide on hypothesis 2. A similar regression model, with game acceptance perceptions as the dependent variable, and patient factors (ie, continuous) as the independent variables, informed, in part, hypothesis 3. Three sets of regression models were used to test whether the variables for intervention or context factors, as the independent variables, affected the variables of (1) game acceptance, (2) mindfulness change, or (3) depression change. To prevent too many factors from being added to a single model, intervention in context variables were added and removed one by one after controlling for patient age and active coping style. Only the planning variables (ie, date differences) were added and removed simultaneously to isolate unique influences, taking interdependency into account.

The final middle-range ICMO-Cs were constructed by assuming elements of the initial ICMO-Cs, refinements, and extensions based on the results from steps 1 and 2 that were not refuted based on the quantitative results from steps 3 and 4. The final conclusions are based on comparisons of the results, with respect to the elements of the ICMO-C, between this study and previous studies.

Postgaming feedback data were collected from 114 out of 156 (73.1%) patients of the intervention group. Nonresponders (mean 40.6 years old, SD 11.7, P=.02) were relatively younger than responders (mean 45.5 years old, SD 11.2) and completed, on average, 1.5 more encounters (15.1 versus 13.6) in the game LAKA (P=.02). No other statistically significant differences were observed between these groups. Two care providers participated in the first focus group, together with a rehabilitation physician and serious gaming expert, a spiritual counselor, an executive and ICT expert, and two managers. The four care providers—three psychologists and one physiotherapist—who regularly facilitated introduction and debriefing sessions participated in the second and third focus groups. The 8 interviewed patients from the intervention group reported differences in the conditions in which they received serious gaming, their demographics, intervention experiences, and outcome levels (see Table 3). All patients reported having at least completed a secondary education. Recruitment of control patients was stopped after two, short, 10-15-minute interviews, because the interviews did not provide useful data.

With regard to why serious gaming can enhance learning outcomes, confrontation with oneself was identified as a necessary form of reasoning. Based on this finding, SDT was chosen as an appropriate general theoretical framework for understanding underlying motivational and affective processes of learning from serious gaming by patients.

Various hypotheses were formulated and tested with available quantitative data for additional indications about the validity of ICMO elements. Hypothesis 1 focused on establishing a quantitative data pattern indicating the initially expected and qualitatively consistent premise that mindfulness is a rehabilitation mechanism, or interim learning outcome, that can be triggered by a serious gaming intervention. Hypothesis 2 was aimed at determining unique covariation between measures of learning results and potential gaming mechanisms, including valence of affect and sense of presence. Although it was initially considered possible to indicate mechanisms with a simple linear relationship between learning and valence of affect, there was no clear basis for this in qualitative results or in the SDT. A positive relationship between the subjective experience of involvement during play and mindfulness was hinted at in the qualitative data, but without explanation of why such a pattern would be generated. Hypothesis 3 focused on determining patient subgroups for whom particularly good or bad outcomes are attributable to serious gaming. Such an empirical pattern was seen as an indication that the self-discrepancies that some of the patients become aware of through serious gaming, that give reason to change behavior, may be absent in other patients. The latter being patients who find themselves free of unsolvable emotional stress, or assume that they can cope with their pain in different ways (ie, no current demand for mindfulness intervention). At the same time, individual inclinations to experience a lack of control could have a negative influence on the use of serious gaming and its results. Hypothesis 4 focused on remarkable covariation between perceived outcomes and factors of serious gaming intervention in context.

Two original ICMO-Cs were constructed by updating initial formulations on the basis of the qualitative and quantitative results (see Textbox 4).

This study illustrates that relevant insights for optimizing biopsychosocial management of CP or FSS (ie, including eHealth) can be developed with an explanatory and context-sensitive evaluative approach. Ideally, such research precedes and goes along with design and assessments of cost-effectiveness and quality improvement with suitable experimental methods (eg, randomized controlled or pragmatic, single- or multicenter, and individual allocation or stepped-wedge methods) [73]. Previously, researchers made similar suggestions for researching effectiveness of treatments for patients with CP, FSS, or mental health problems [31,45,99-101]. After all, consensus building about appropriate allocation and tailoring of treatment to patients has been relying on limited evidence [102]. As realist approaches focus on such issues, there is a large but yet unproven promise for methodological progress to support new insights for intervention quality improvement, treatment allocation decisions, personalization, and setting patient expectations. Specific areas for future theory-oriented research on serious games are implied in Table 4. This table gives an overview of considerations about how the findings of this study relate to those of other studies in various disciplines [1,3,30,31,52,53,59,68,85,87,96,103-116].

Following a realistic approach, this study started with the selection of potentially relevant theories and ended with an integrated fallible model [98]. Serious gaming (ie, a 4-hour intervention during an intensive multidisciplinary rehabilitation program) can offer relevant additional opportunities for enhancing self-awareness and reducing emotional symptoms by triggering increased attention and self-discrepancies. This can be achieved in CP or FSS patients who experience relatively less control over their current stress or pain. This also depends on the fidelity of implementation processes (ie, planning and facilitating) in such a way that gaming can be immediately followed up by exchanging experiences between patients under professional guidance, toward goals compatible with the overall treatment. At the same time, there is no sign as of yet of potential clinically relevant effects of serious game interventions in other treatment circumstances and among patients with CP or FSS.

This already offers some transferable insights for patients, care providers, health care organizations, insurers, policy makers, and businesses who will make future decisions on the future implementation of serious gaming. It may not be wise to offer this form of serious gaming, with similar late timing during rehabilitation, to all patients, including those claiming that their stress and pain are already under control.

Realizing the promise of serious gaming depends on ongoing efforts to develop transferable understandings of how they work, for whom, when, and why. In doing so, focus can be informed by perspectives of important stakeholders. The theoretical scope could be extended to the outcome domains of quality of life of patients, adoption and implementation by organizations, and cost-effectiveness. Understanding about the mechanisms of learning from serious games could be improved by applying SDT when formulating predictions about cognitive and emotional consequences of certain changes in the design and context of use. Finally, future studies could zoom in and out on micro- or macro-level conditions for serious gaming with impact. For example, patients, software developers, and data scientists could collaborate to optimize software (eg, in-game feedback) for adaptation to recipient characteristics for learning and health improvement. Care providers can also be involved in optimizing feedback provision (eg, enabling them to interpret log data) and in optimizing the fit with work routines. Lastly, on an organizational or societal level, policy makers could participate in theorizing on preconditions for implementation, such as financial arrangements and dissemination platforms.