Sample sizes ranged from 14 [14] to 1470 [15] participants. Almost half the included studies had sample sizes below 50.

Fourteen studies were conducted in Europe [16-30], and 11 studies were conducted in North America [14,15,31-40], one of which recruited participants from 63 countries via the internet [15]. One study was conducted in Singapore [41]. Four studies were conducted in middle-income countries, three of which were conducted in Brazil [42-44]. One study was conducted by authors based in China and Taiwan, but it was unclear where the study itself was carried out [45]. None of the included studies were conducted in low-income countries. Details of study designs and population for each trial are compared in Tables 1-3 and a summary is given below.

Eleven studies included only medical students [14,16,17,19,20,24-26,39,40,42,45]. Five studies included only nursing students [22,27,29,37,41] and four included only surgical residents [18,30,31,38]. The remaining studies included primary care doctors (n=2) [36,43]; dental students (n=2) [21,32]; anesthesiology residents (n=1) [35]; urologists (n=1) [15]; speech and language science students (n=1) [44]; participants of the Major Incident Medical Management and Support course, which typically includes doctors, nurses, and paramedics with an interest in prehospital care (n=1) [23]; nursing and medical students (n=1) [28]; and medical students, residents, and specialists in Obstetrics and Gynecology (n=1, reported in one article and one conference abstract) [33,34].

Characteristics of the interventions included are compared in Tables 4-6 and a summary is given below.

Two of the included studies evaluated “gamification” interventions [15,36]. The remainder evaluated serious gaming interventions. Two of these were group interventions, in which a serious gaming intervention was projected to a traditional classroom of learners who played together [39,44].

A total of 22 interventions had original design purposes other than entertainment, of which 11 interventions were classified as “Message Broadcasting - Educative” [15,17,20,28,36,37, 39-41,43,44]; four were classified as “Training - Mental” [19,22,23,29], three were classified as “Training - Physical” [18,21,32], and four were classified as both “Training - Mental” and “Training - Physical” [16,27,35,45]. All of the interventions with design purposes other than entertainment were classified under “Education” for “Scope.”

The remaining eight interventions were commercial off-the-shelf games designed only for the purpose of entertainment but used for training of motor skills as part of “Serious Diverting” interventions. These were all classified as “Training - Physical” for “Purpose” and as “Entertainment” and “General Public” for “Scope” [14,24-26,30,31,33,34,38,42].

Serious gaming/gamification was compared with traditional learning in 19 studies [18-23,27-31,35,37,38,41-45], with digital education in 7 studies [14,16,17,32,36,39,40], and with other serious gaming/gamification interventions in 6 studies [15,24-26,31,33,42].

One study addressed patient outcomes [36]. Fourteen studies assessed knowledge [15,17,19,20,28,32,36,37,39-41,43,44]. Twenty-three studies addressed outcomes relating to skills [14,16-19,21-35,37-39,41,42,45]. Four studies assessed outcomes related to attitudes [17,24,25,28,43]. Sixteen studies addressed participant satisfaction [15-17,19-22,26,27,29,32, 37,39,42,43,45].

Figures 2 and 3 summarize the risk-of-bias assessments for the included studies. A total of 25 of the included studies were considered to be at high risk of bias [14,16-23,26-32, 35,36,38,40-45] according to Cochrane guidelines, because they had a high or unclear risk of bias for either the sequence generation or allocation concealment domains [12]. All three cRCTs were rated high for incorrect analysis, as none accounted for clustering in the analysis.

Effects of the interventions are compared in Multimedia Appendices 2-4 and a summary is given below.

One study measured patient-related outcomes [36]. This study compared serious gaming/gamification with an online posting intervention for primary care physicians and reported significantly shorter time to control blood pressure in the intervention group for only a subgroup of participants whose patients were already on antihypertensive medication at the start of the study (P=.02), although this may not be clinically significant (117 vs 125 days). Data were insufficient for calculation of standardized mean differences (SMD). The quality rating assessed using Grading of Recommendations, Assessment, Development and Evaluations (GRADE) was low.

Figure 4 summarizes the results of studies reporting knowledge outcomes.

Four RCTs reported higher posttest scores in a serious gaming/gamification group than in a traditional learning group, with a mostly moderate magnitude of effect [20,28,37,43]. Interventions included a videogame to prepare students to enter the operating room [28] (SMD 1.05, 95% CI: 0.68-1.41), a urology educational adventure game [20] (SMD: 0.69, 95% CI: 0.35-1.03), pediatric respiratory disease-assessment game for nurses [37] (SMD: 0.65, 95% CI 0.23-1.07), and the InsuOnline serious game [43] (SMD: 0.40, 95% CI: 0.06-0.73). Comparisons were made between usual learning, written script, and traditional lectures.

One RCT of a group intervention, where speech and language science students played a serious game together in a classroom via a projector, found no difference in anatomy and physiology knowledge compared to a self-study control [44] (SMD: 0.05, 95% CI: –0.74 to 0.83).

Two cRCTs were also included [19,41]. One showed evidence of a large magnitude of effect for a blood transfusion serious game as compared to usual education, although the effect may not have been statistically significant (SMD: 1.95, 95% CI: –0.20 to 4.11) [41]. The second study showed no evidence of effect for a geriatric game compared with standard education, although this study measured perceived knowledge rather than an objective measure (SMD: 0.01, 95% CI: –1.50 to 1.61) [19].

All the individually played games with an objective assessment of knowledge suggested that serious gaming/gamification was superior to traditional learning. The quality rating assessed using GRADE was low for this outcome and comparison (Table 7).

Five studies found no evidence of a difference. Studies included comparison of serious gaming on dentin bonding and an online lecture control [32] and serious gaming with digital education on patient safety [17] and Web-based vascular anatomy study aids with and without game elements [40]. One study of a group serious gaming intervention found no difference in pediatric knowledge between groups who played a projected board game in teams in a conference room with Web-based pediatric flashcards [39]. One study found that compared with an online educational posting, serious gaming (an online spaced-education game) may improve knowledge (large magnitude of effect) [36] (SMD: 0.82, 95% CI: 0.42-1.22). The quality rating assessed using GRADE was low (Table 8).

One study of a spaced-education game found that interventions with greater question spacing (four questions every 4 days rather than two questions every 2 days) resulted in higher posttest scores, with a moderate magnitude of effect [15] (SMD: 0.50, 95% CI: 0.38-0.64). The quality rating assessed using GRADE was moderate, as the one included study had a low risk of bias in all but one domain.

A total of 24 studies addressed skill outcomes. Figure 5 summarizes the results of studies reporting skill outcomes.

Twelve RCTs [21,22,27-31,35,37,38,42,45] and three cRCTs [19,23,41] compared serious gaming/gamification to traditional learning in this outcome category. The results were inconsistent, and studies were generally of low quality, making it difficult to draw conclusions about the efficacy of these interventions.

Six studies reported significant differences between groups for overall skill assessments in favor of serious gaming [27,28,35,37,42,45], with a magnitude of effect ranging from small to large. However, SMDs for two of these studies could not be calculated due to missing data [27,45]. Interventions included games with scenarios simulating clinical environments [27,28,35,37] and serious diverting interventions for improving practical skills [42].

Three studies comparing serious gaming/gamification with traditional learning used multiple measures for assessing skill outcomes; differences in favor of serious gaming/gamification were observed for some, but not all, of these skill measures, and the studies did not present an overall estimate of the effect [22,30,38]. Effect sizes could not be estimated, as SDs were not reported and attempts to contact the authors for further data were unsuccessful.

Two studies reported no significant difference in skill outcomes when comparing serious gaming/gamification and traditional learning and another reported no differences in pre- and posttest scores in either group [31].

Three cRCTs were also included [19,23,41]. One showed evidence of an effect of small magnitude, favoring a blood transfusion game group [41] (SMD: 0.33, 95% CI: –1.19 to 1.86); the second study found evidence of a moderate magnitude of effect, favoring the standard educational activity group, although skill measures were self-perceived as opposed to objective [19] (SMD: –0.77, 95% CI: –2.53 to 1.00); and the third showed no evidence of effect for a triage trainer game [23] (SMD: –0.18, 95% CI: –2.37 to 2.02). Each of these results may not be statistically significant.

There is some evidence that serious gaming/gamification interventions are more effective for improving skills than traditional learning. The quality rating assessed using GRADE was low, as the risk of bias was unclear for multiple domains and all the included studies had fewer than 400 participants.

Five studies comparing skill outcomes for serious gaming/gamification and other modalities of digital education found no evidence of a difference in outcomes between groups [16,17,22,31,32]. In these studies, serious gaming was compared with an online video on dentin bonding [32] and with an electronic module (e-module) on patient safety [17] and management of an acutely unwell patient [16]. Another study reported higher postintervention skill score in a virtual reality control group than a commercial off-the-shelf intervention, with a large magnitude of effect for the time taken to complete surgical skill tasks (peg transfer and bimanual carrying; SMD: –1.56, 95% CI: –0.31 to –2.81), but reported no difference for distance travelled with surgical instruments when completing these tasks [14]. The quality rating assessed using GRADE was low.

We are uncertain whether any particular type of serious gaming/gamification is more effective than the other for improving skills. In three of the five studies comparing two serious gaming/gamification interventions, games involving motor skills, visuospatial skills, and manual dexterity may be more effective than interventions involving cognitive skills for improving laparoscopic surgical skills [24-26,33,34], but the quality of available evidence is very low.

Figure 6 summarizes the results of studies including professional attitudes outcomes.

Two RCTs compared a serious gaming/gamification intervention with traditional learning and measured outcomes related to professional attitudes. There was some evidence of a small magnitude of effect for a serious game, preparing students to go into the operating theatre for the first time, compared with traditional learning (SMD: 0.49, 95% CI: 0.14-0.84) [28]. A study comparing an insulin-prescribing game with an onsite learning activity for primary care reported insufficient data for comparisons between groups [43].

One cRCT was also included [41]. When reanalyzed with the number of clusters as the sample size to account for clustering in the analysis, there was evidence of intervention effectiveness, but this may not have been statistically significant and the analysis was likely underpowered (SMD: 1.23, 95% CI: –0.55 to 3.02). The quality of evidence for this outcome and comparison was rated very low according to the GRADE assessment.

One study compared a serious game and an e-module on patient safety and reported no difference between groups in perceived patient safety behavior or reported stress [17].

The quality of evidence was rated low according to the GRADE assessment.

One study (reported in two papers) compared two serious diverting interventions, one was a first-person shooter (FPS) and one was a non-FPS, and reported no significant differences in self-efficacy or positive engagement modes [24,25]. Data were insufficient for calculation of effect sizes. The quality of the evidence was very low according to GRADE assessment.

Eleven RCTs [15,16,20-22,26,27,32,33,39,45] and two cRCTs measured outcomes relating to satisfaction [19,41]. Seven studies did not measure satisfaction in a comparison group [19,21,22,27,32,35,45], Diehl et al measured satisfaction with the intervention and comparison group using different scales [43], and Kerfoot et al reported results for participants in both groups combined [15]; therefore, these studies did not meet the inclusion criteria for this review. The remaining studies showed mixed evidence and are compared below.

One study reported significantly better attitudes toward learning among a serious gaming group (a urology adventure game group) compared with a written script [20]. The quality rating assessed using GRADE was low.

Three studies reported higher satisfaction scores for serious gaming/gamification on managing acutely unwell patients [16], patient safety [17], and training during a pediatric clerkship [39] compared with an e-module [16,17] or Web-based flashcards [39] covering the same topics. The quality rating assessed using GRADE was low.

Results of a participant survey [26] suggested that more participants in the FPS gaming group than in the non-FPS gaming group found the intervention beneficial for their performance on a surgical simulator. No significance test was reported. The quality rating assessed using GRADE was very low.