The tablet-based story game application was developed in Unity 3D [62] and programmed entirely using the C# scripting language. For an overview of the rationale, developmental design-related decisions, stories, and programming details of the game, refer to the technical story game development guide (Multimedia Appendix 2 [5,47-53,58,62-79]). The story game provided 72 gaming sessions, each of which mainly comprised listening to 1 long (eg, approximately 10 minutes) or 2 short (eg, twice approximately 5 minutes) stories accompanied by story illustrations. Game settings prevented children from playing ≥1 session every 24 hours. The 72 sessions were categorized into 18 play phases of 4 grouped sessions containing 4 (longer) to 8 (shorter) stories from the same book series and author. In total, 87 stories from 14 different book series were implemented in the game. Generally, the game contained three main game modes: (1) the main intervention task environment, which involved actual story listening, story rating, and responding to content-related questions; (2) the virtual hub world, in which players consulted their game progress on a map; and (3) an avatar customization system, where players could buy accessories for their game avatars or buy new avatars. Most of the artwork in the game, such as 3D models, textures, and animations that were used to create the virtual hub world, the avatars, and their accessories, were custom-made; acquired from the Unity Asset Store [80]; or acquired from child-friendly projects of the Dyslexia Research Collaboration team at KU Leuven, such as Diesel-X [58]. The resulting art style was characterized by flat primary colors, avatars with exaggerated cartoon-like proportions, and simplified basic models and animations suitable for the chosen target age group [63,81].

The main intervention task presented recorded story audio, all in Dutch, with accompanying illustrations (Figure 1A), followed by a story appreciation rating (Figure 1B) and audio questions with a simple multiple-choice system (Figure 1C). Questions were formulated by the research group, and listening comprehension was examined by recalling information explicitly mentioned in the story text. Although not directly assessed, the language level of the questions was intended to match the language and cognitive capacities of children aged 5 years. The question part put a strong emphasis on feedback by awarding 1 coin per correct response, with added flair in the form of animations and sound effects (Figure 1D). Each session contained 3 content-related questions. Hence, given the total amount of 72 game sessions, players could maximally earn 216 coins throughout the game.

Similar to classic digital games such as Super Mario Bros, the different game sessions were tied together in a so-called virtual hub world. The virtual hub world provided an overview of each game session in the form of a cylindrical stage. Every time a player successfully completed a game session, his or her avatar jumped from one cylindrical stage to the next, leaving a star on the stage that was just abandoned (Figure 2). As such, players gained information on their progress throughout the game and could earn a maximum of 72 stars (ie, the total number of game sessions).

To provide added incentives and rewards to keep the player engaged, an avatar customization system was implemented, which featured a virtual store where players could spend their coins (Figure 3). These coins were earned through correct responses to the content-related story questions in the main intervention task. The store offered new avatars and accessories for current avatars, which were displayed in the virtual hub world. The items put an emphasis on quantity to make sure that the player could acquire roughly one new item per story session without running out long before reaching the end of the story game.

In a single game session, first, players watched a short cut-scene of their avatar positioned on one of the 72 possible stages in the virtual hub world, representing the 72 game sessions and, thus, their progress throughout the game. Second, the main intervention task occurred, in which the player was required to actively listen to 1 or 2 stories, depending on the particular story session. After listening, the player rated the story based on a green or red thumb (ie, like or dislike, respectively) and received 3 content-related multiple-choice questions for which one coin could be earned per correct response. Following the short quiz, the player automatically ended up in the avatar customization system, where the earned coins could be spent on accessories for the avatar or to buy extra avatars. Finally, the player was sent back to the hub world in which the new or newly decorated avatar jumped to the subsequent cylindrical stage while a dancing star appeared on the one that was just abandoned (for a demonstration movie of the story game, see Multimedia Appendix 3). Useful player data for individual player accounts (eg, game progress, question response accuracy [QRA], play dates and hours, and story rating information) were automatically logged on our university server and sent daily to the research group.

The subtest Understanding Spoken Paragraphs was adapted from the Clinical Evaluation of Language Fundamentals (CELF)-fourth edition, Dutch version [82] and was used to evaluate baseline listening comprehension. The experimenter read 2 short stories aloud. Each story was followed by 5 content-related questions. The child received 1 point per correct response. The maximum score on the test was 10.

Receptive vocabulary was measured using the Peabody Picture Vocabulary Test–III, Dutch version [83], which contains 17 subtests with 12 test trials each. In each trial, the child heard a target word and was asked to select the corresponding picture from the 4 alternatives. The test was interrupted when the child made ≥9 errors in 1 subtest. The raw receptive vocabulary score was computed by subtracting the sum of the errors across all administered subtests from the number of trials that was last assessed. This raw score was then converted into a standard score and represented the final receptive vocabulary score (mean 100, SD 15).

The Word Structure subtest from the CELF Preschool–second edition, Dutch version [84] was used to evaluate morphological awareness. In this task, the ability to correctly apply Dutch morphological rules (eg, conjugation, derivation, flexion, pronouns, and degrees of comparisons) was measured. The test contained 23 items, each containing 2 pictures. The first picture was fully described by the experimenter (eg, “Dit meisje eet” meaning “This girl eats”) followed by an incomplete description of the second picture (eg, “en dit meisje” meaning “and this girl…”), which had to be completed by applying a certain morphological rule of Dutch (eg, “and this girl sleeps” or “en dit meisje slaapt”; morphological rule: conjugation of a regular verb in third singular form). On the basis of the instructions in the test manual, the test was interrupted when the child made 7 consecutive mistakes. The maximum score was 23.

After the intervention period, parents independently filled out a short questionnaire that included three questions related to motivation, encouragement, and sustained attention during story gameplay. Accompanied by a member of the research group at school, intervention enjoyment was also measured in all children using 2 components of the Fun Toolkit survey instrument [85], which was developed to gather children’s opinions on technology. On the one hand, children were asked to assign story game enjoyment on a 5-point Likert scale–based smiley-o-meter, in which the 5 scales were represented by smileys (ranging from a very unhappy smiley meaning I did not like the game at all to a very happy smiley meaning I liked the game very much). On the other hand, participants were asked whether they were willing to redo the intervention (see Multimedia Appendix 4 for an overview of the 5 child and parental categorical questions and their response possibilities).

Of the 91 children who were asked to play the story game, 7 (8%) were excluded from the data set because of technical game problems during the intervention period (bugs; explained in (Multimedia Appendix 2 [5,47-53,58,62-79]). Approximately 2% (2/91) of children were excluded as they never started the general digital gaming intervention properly (ie, they played <10% of the GG-FL or AC and the story game). Hence, useful story game data were available for 90% (82/91) of the children. Data visualization and statistical analyses were conducted using RStudio [86,87].

Story game enjoyment was addressed in 3 ways. First, the postintervention child and parental questionnaire outcomes were analyzed by investigating the relative frequencies per response category of each question. Second, 2 members of the research group independently evaluated the GameFlow criteria by assigning a state of fulfilled, partly fulfilled, or not fulfilled. Then, after reaching a consensus, all criteria received a final, single state. Third, in-game enjoyment–related data were evaluated, such as general story appreciation, which was measured as the proportion of given likes and dislikes per story.

To gain insights into the feasibility of the intervention, we first defined the general QRA, which indicated the difficulty level and thus the feasibility of the story-listening part of the game. This was computed as the proportion of correct and incorrect responses per question. Second, we visualized the attrition rate, which determined the number of children who completed the intervention and at what point eventual dropouts occurred. Third, we calculated the individuals’ final game exposure and training duration. Final game exposure was computed as the ratio (percentage) of the number of actually played sessions to the total number of available sessions (ie, n=72). The individuals’ corresponding final training duration represented the number of days the child played, starting from the day of the first story session until the day on which the participant played for the last time, irrespective of game completion. The measure of the final training duration was particularly informative about the feasibility of the training intensity. By analyzing the players’ final game exposure and training duration, we categorized players into different gaming profiles based on intervention completion (complete or incomplete) and schedule compliance (compliant or noncompliant). Categorizing participants based on these criteria provided information on the possible need to adjust the training schedule and intervention duration to increase the feasibility of future studies. A total of 4 additional nonparametric median-based Theil–Sen regressions were performed using the mblm package in R [88] to predict individuals’ final game exposure or final training duration, either based on initial individual story appreciation or QRA. These analyses provided information on how to keep players engaged from the start and prevent them from dropping out or slowing down. The Theil–Sen technique allows for a robust line fitting as the estimation is obtained by calculating the median of the slopes of all possible pairs of data points [89]. This nonparametric regression technique was opted for, given the relatively small sample size and violated assumptions to perform ordinary least square regressions. Initial individual story appreciation and individual response accuracy at the start of the intervention were defined as a player’s mean story appreciation and mean QRA in the first two game phases, respectively (ie, mean ratings of the first 12 stories and mean response accuracy for the first 24 questions). As for predicting game exposure based on initial story appreciation or initial individual QRA, the analyses were restricted to a sample of players who did not completely finish the story game (32/82, 39%), as the inclusion of participants who completed the full intervention would render a ceiling effect in the results. As for predicting the training duration based on initial story appreciation or individual response accuracy, we excluded 10% (8/82) of participants who played <80% of the total game content, as their training duration in days did not represent a reliable intervention trajectory. As such, the training duration was predicted based on the data of 90% (74/82) of participants.

The third and last part of the Results section tackles a possible growth in language comprehension. We considered the mean QRA (percentage) per game phase as an indicator of language comprehension. However, we acknowledge that stories and questions do not belong to a validated language instrument, and we did not directly test the stability of the difficulty of the questions and stories. However, we were obliged to select measures beyond the existing language instruments, given the long intervention period. Each game phase contained 12 content-related questions. Using the lmerTest package in R [90], we performed a multilevel linear growth model in the sample that finished at least 80% of the game (74/82, 90%), with game phase as a within-subjects variable and baseline listening comprehension, morphological awareness, and receptive vocabulary as between-subjects covariates. The game phase was coded such that the fixed intercept of the model represented the mean QRA of the first game phase if a score of zero was obtained for all 3 baseline language covariates. We also modeled a randomly varying intercept and slope across individuals and added 3 additional phase×baseline language interaction terms (1 per baseline language covariate). As such, we could test the hypothesis that the mean QRA, which was our indication of language comprehension, improved over the game phase when controlling for each baseline language skill, as well as whether children with lower baseline language skills showed larger growing potentials concerning mean QRA.

Postintervention child questionnaire data revealed that 74% (61/82) of participants liked the game very much. Moreover, 66% (54/82) of participants confirmed that they were willing to redo the training. Postintervention parental questionnaires revealed (1) that most parents observed a high (39/82, 47%) or relatively high play motivation (32/82, 39%) in their child, (2) that most of the players never (43/82, 52%) or only sometimes (25/82, 30%) needed encouragement to play, and (3) that more than three-quarters of the parents (65/82, 79%) observed a state of sustained attention during gameplay.

Although the GameFlow model was applied in the design process, Tables 1 to 8 clarify the extent to which the game criteria were properly implemented and whether our game did, only partly, or did not fulfill each GameFlow criterion [5]. As shown in Tables 1 to 8, the GameFlow model–based evaluation revealed that only the criteria belonging to the concentration principle were completely fulfilled. As for the other elements, except for the elements of social interaction (as it was intentionally not implemented, given the young age of our participants) and challenge, we concluded that most of their corresponding criteria were completely or partially fulfilled. To avoid plagiarizing from Sweetser and Wyeth [5], the wording of the descriptions of the GameFlow criteria in Tables 1 to 8 differs from the criteria wording used in their article [5], although the meaning of the criteria did not change. However, we want to emphasize that we considered the original wordings when developing and evaluating the story game, as we wanted to take into account the importance of applying evaluation tools in the exact same way as they are validated.

Figure 4 visualizes the general story appreciation, calculated as the proportion of available likes and dislikes per story. Each story was liked by >75% (ie, >62/82 listeners) of the listeners for whom data were available, indicating that the stories were enjoyable.

Figure 5 visualizes the general QRA, calculated as the proportion of correct and incorrect responses per question. All but 11 questions were answered correctly by at least 75% (62/82) of the listeners for whom data were available, suggesting a rather stable difficulty level for the questions.

Figure 6 shows the attrition rate and the proportion of active participants and dropouts throughout the entire story game progress. Of the 82 players, 74 (90%) listened to 80% of the stories (approximately 57th session of the total 72 story sessions), and 50 (61%) managed to finish the game completely. Hence, 39% (32/82) of players dropped out at some point during the 12-week intervention. Figure 6 also shows that the first player dropped out after the 12th story, corresponding to the start of the third game phase. From that point onward, the dropout proportion stayed relatively stable until the 80% completion point, after which it gradually increased toward the end of the game.

Figure 7 demonstrates individual story game trajectories, individuals’ final game exposure and corresponding training duration, and their overall distributions. Apart from the visible finding that a few participants had a low training duration as they dropped out early, Figure 7 also shows a large variation in the training duration in children who nearly or completely finished the story game. On the basis of the categorization criteria (complete or incomplete and compliant or noncompliant), we established three different gaming profiles: (1) complete and compliant players (18/82, 22%; Figure 8A), (2) complete but noncompliant players (32/82, 39%; Figure 8B), and (3) incomplete and noncompliant players (32/82, 39%; Figure 8C). Note that we did not observe any players who perfectly followed the training schedule but dropped out at some point.

The Theil–Sen regression analysis revealed that the mean story appreciation of the first 2 game phases was not significantly predictive of final game exposure among the 39% (32/82) of children who dropped out at some point (β=.15; P=.08). Figure 9A shows that most children who quit at some point still assigned positive ratings to all stories of the first 2 phases. In contrast, the Theil–Sen regression analysis revealed that the mean QRA of the first 2 game phases significantly predicted final game exposure, such that a lower accuracy at the start accelerated the dropout point (β=.35; P<.001; Figure 9B).

Story appreciation in the first 2 game phases significantly predicted the final training duration in a negative way among the 90% (74/82) of children who finished 80% of the game (β=−0.16; P=.003; Figure 10A). These results indicate that players who liked the stories at the start of the game were more likely to play the game at a higher rate. The mean QRA at the start did not significantly predict the final training duration (β=.00; P=.68) among children who finished or almost completely finished the story game (Figure 10B).

The game enjoyment–related results give rise to 2 important discussion points. First, caution is required when interpreting the positive questionnaire outcomes. On the one hand, the phenomenon of social desirability (ie, the tendency to provide answers that correspond to widely accepted social norms) could have biased the responses of both children and parents [91]. On the other hand, the short attention span and underdeveloped cognitive capacities of young children could have resulted in extremely positive but less reliable responses on the smiley-o-meter. Indeed, in line with our study, Zaman et al [92] established an overrepresentation of positive responses when applying the smiley-o-meter in young children aged <7 years. Accordingly, in another study, the authors stated that children aged ≤7 years do not yet grasp the principles of a visual analog scale, of which the smiley-o-meter is an example [93]. As put forward by Zaman et al [92], the search for how to optimally measure the enjoyment of technology in preschoolers is still ongoing. The second discussion point related to enjoyment concerns the GameFlow-based evaluation. More specifically, we acknowledge that the agreement of the developed game with the GameFlow model contains a form of subjectivity, as the evaluation was performed by 2 members of the research group who were involved in the design process itself. Irrespective of this limitation, the fully and partly implemented GameFlow criteria embedded in the current game were presumably satisfactory enough to attain a certain level of enjoyment. This assumed enjoyment level was also supported by the cautiously interpreted questionnaire and in-game enjoyment–related outcomes, which indicated game appreciation in most players. Optimizing the partly or nonfulfilled criteria (eg, fixing the remaining bugs and clarifying the end goals) could increase the level of enjoyment even more. To deal with subjectivity, a future study could let adults who were not involved in the design process (eg, parents of playing children) rate the game.

This paragraph discusses 3 relevant points related to feasibility outcomes. First, although most of the participants (50/82, 61%) completed the intervention, we still acknowledge that 39% (32/82) participants dropped out at some point, mostly in the last game phases. This slightly corresponds to another home-based preventive intervention study in Dutch children at risk for dyslexia, which reported an attrition rate of 34% [94]. Unfortunately, the attrition rate of this paper still exceeded the desirable benchmark of 30% [95], and further research on why participants tended to drop out at some point is necessary. In a study by Justice et al [96], who investigated the feasibility, efficacy, and social validity of a 12-week home-based shared storybook reading intervention in preschoolers with language difficulties, the authors compared completers with noncompleters on a variety of child- and parent-specific characteristics and established risk factors of dropout, such as lower maternal age and lower parental educational levels. Distinguishing completers from noncompleters based on intrinsic and extrinsic features did not fall within the scope of this paper, albeit the research methods applied in the study by Justice et al [96] offer possibilities for future research that could be of added value to optimize the feasibility of the current intervention program. Irrespective of the rather high attrition rate, we still established a completion rate of 80% (approximately 57th session of the total 72 story sessions) in 90% (74/82) of the participants, suggesting that shortening the intervention by approximately 15 sessions, which almost equals 4 weeks in this paper, would result in lower dropout occurrence. However, an important sidenote relates to the fact that the participants in this paper combined the intervention with another tablet game (GG-FL or AC games). This could also have affected general motivation and perseverance. The second discussion point related to feasibility bears upon the variation in training intensity among the participants. As treatment fidelity in terms of quality and quantity has been found to predict learning outcomes in previous studies [94,97], the variation in training intensity cannot be neglected when investigating the educational impact of interest in future studies. However, in their reading intervention study, Katzir et al [98] found larger gains in sight word efficiency in a group of grade 1 to 3 children at risk for reading difficulties who received a 44-hour fluency-based therapy over a 9-month period compared with children who followed the exact same intervention within a more intensive period of 2 months. Hence, albeit the need for statistical confirmation in a future study, spacing out the story game intervention over time in this paper might have even benefited the expected educational learning outcomes. The last feasibility-related discussion point is linked to the prediction of the final game exposure and training duration based on story appreciation and QRA at the start of the intervention. The outcomes of these predictive analyses point to limitations in the current game design but give rise to suggestions for optimizing intervention feasibility in future studies. For example, given the importance of initial story appreciation to maintain the advised training intensity, a possible suggestion involves offering the storybook series in the preferred order of the participant. More specifically, providing a catalog menu in which players could choose which story series occur first might increase the story appreciation at the beginning and, as a result, the engagement to follow the advised schedule. Moreover, given the role of initial QRA in dropout occurrence, offering player-adapted questions based on an individual’s language knowledge and cognitive capacities, which then increase in difficulty, instead of fixed, predefined questions for all participants, might prevent participants from withdrawing from the study. A more interactive approach in which players could request and receive explanations of possible difficult words might also increase the chance of successfully responding to the questions from the start, lowering the chances of early dropout. However, a disadvantage of these proposed adjustments to prevent attrition relates to the fact that the growth of in-game QRA data then becomes less easily interpretable at the group level, as participants would follow an individually adapted trajectory. However, large-scale adaptability has been widely considered as one of the strengths of serious digital gaming interventions [4]. Furthermore, according to our GameFlow-based evaluation in its current form, the criterion of so-called player-adjusted challenges belonging to the overall challenge principle is lacking in the current story game. Although this principle was beyond the scope of the story game, its implementation could increase player enjoyment and decrease the attrition rate [5].

The results related to growth in language comprehension must be interpreted with extreme caution, and several important limitations should be elucidated. The first and most important limitation relates to the research design. In fact, the actual impact of the game on language comprehension can only be solidly established by conducting a randomized controlled trial (RCT), which was not performed in this study. Hence, although we found a larger growth in mean QRA in children with lower listening comprehension, the current research design does not allow us to draw univocal conclusions on the potential of the game to train language comprehension. Thus, a future RCT study, preferably including (1) a group that is purely playing the story game without combining it with other interventions such as GG-FL or AC games, (2) a no intervention control group, and (3) a control group receiving an alternative placebo treatment that does not specifically train language skills [99], is of crucial relevance to further disentangle the gaming effects on language comprehension. The second important limitation is related to the concept of content validity. We considered the in-game questions as a measure of language comprehension. However, despite the selection of age-appropriate stories based on library visits and the construction of the questions based on the vocabulary of the story content, the stories and questions did not belong to a validated comprehension test instrument, casting doubt on the certainty that we truly measured language comprehension in our participants [100]. In addition, we are unsure whether all questions and stories implemented in the game were of equal difficulty, although we found relatively stable response distributions for most of the questions. Hence, changes in performance might not be because of the intervention effects but simply because of the varying difficulty of the questions and stories. Conducting the pre- and postintervention measurements of independent language comprehension tests within an RCT design with appropriate control groups (as mentioned previously) would eliminate this concern. The third limitation discusses the distribution of the mean QRA in the first game phase, which already showed a tendency toward the ceiling. Hence, children with high accuracy scores from the start, which were related to higher baseline vocabulary and listening comprehension, did not have the potential to increase any further during the intervention, influencing the interpretation of the game phase–listening comprehension interaction. The high accuracy scores at the start of the study are presumably attributed to the fact that none of the participants in this paper experienced language difficulties. Indeed, originally, the story game was not developed to train language comprehension within the framework of the current research project but rather to investigate the potential of envelope-enhanced speech in children with a pure cognitive risk for dyslexia. Another factor that may influence the high initial accuracy scores relates to the type of posed questions. The questions intentionally only involved identifying or recalling items that were explicitly mentioned in the story text and thus examined language comprehension only at the literal language level [101]. Hence, questions at the inferential level, which, for example, require implying emotions, predictions, and connections between information within the text or with the child’s own experiences [102], were not part of the story game in its current form. For preschoolers, inferential questions are certainly more difficult to answer than literal ones [101]. However, given the scope of the story game within the current research project (ie, exposing children at risk for dyslexia, in a joyful way, to either envelope-enhanced or nonenhanced speech), questions were not supposed to be or become difficult but only served as a way of (1) ensuring active listening and (2) ascertaining that players earned as many coins as possible, increasing the chance of maintaining the intervention for 12 weeks. On the basis of these aforementioned limitations, the preliminary and cautiously interpreted results concerning a possible growth in language comprehension give rise to 3 important ideas for story game improvement if it would be used to train language comprehension at some point in populations with specific problems in this domain (eg, children with low socioeconomic status; bilingual children; and children with DLD, hearing impairment, or dyslexia). First, there is a clear need to include inferential questions, as Tarvainen et al [18] emphasized the importance of inferential language training in fostering language comprehension in preschoolers. Second, apart from optimizing the research design by conducting RCTs with adequate intervention and control groups, it is also of crucial relevance to include test batteries that assess comprehension at both the inferential and literal levels. The currently used CELF-4th edition, Dutch version [82] and Peabody Picture Vocabulary Test–III, Dutch version tasks [83] mainly focus on language comprehension at the literal level [101], emphasizing the need to add inferential language comprehension test batteries to the research protocol. Finally, researchers must take into account and tackle the possibility that children with lower language levels might show less motivation in storybook interventions than their typically developing peers [103].