Resource-constrained health systems in low- and middle-income countries (LMICs) are considered a key obstacle to achieving sustainable development goals (SDGs), specifically SDG 3, to “ensure healthy lives and promote well-being for all at all ages,” and SDG 4, to “ensure inclusive and equitable quality education and promote lifelong learning opportunities for all” [1-3]. As LMICs are experiencing an unprecedented increase in the number and use of digital devices such as mobile phones, digital health and education initiatives that capitalize on the widespread use of these devices are emerging [4,5]. A digital device is a physical piece of equipment that uses digital data in some way, such as sending, receiving, storing, or processing data [6]. Despite controversies over various digital initiatives and programs using digital devices, such as “One Laptop Per Child,” policy makers and practitioners are still interested in the potential of digital technology to address issues such as the digital divide, the impact of resource constraints on cognitive development, and developmental disabilities in children [7-11]. However, will digital device exposure and intensity of the exposure affect cognitive development in children in LMICs? This question is difficult to answer because most of the research targets high-income countries, adolescents, and adults [12-14].

In addition to foundational skills such as literacy and numeracy, digital literacy and skills are also key to implementing the SDGs in this digital age [2]. Therefore, many have suggested using digital devices for children’s cognitive development to improve information processing, communication skills, and educational attainment [10,15]. Studies have demonstrated that digital devices provide only short-term improvements in children’s cognitive abilities during interventions [11,16-19]. A recent study on patients with an intellectual disability diagnosis showed significant improvement in cognitive function using digital devices [20]. Moreover, most studies have focused on the experience of high-income countries or different age groups and patient-only analyses [14,20-23]. The evidence provided by these studies may not be relevant to LMICs because of their different socioeconomic environments and cultures. Although socioeconomic characteristics of households, both within and without, are known to significantly affect cognitive development, few studies on digital device–using interventions, including on the impact of various levels of digital device exposure on children’s cognitive development, have been conducted in LMICs [24].

This study addressed the following questions through a survey and cognitive function tests of children in LMICs:

This cross-sectional study was designed to identify and analyze the correlation between cognitive function and exposure to digital devices in children aged 3 to 9 years at 2 schools in Cambodia. The study focused on the following urban and rural regions in Cambodia to consider the effects of various conditions such as average literacy rates, wealth and income distribution, and educational environments: Sisophon in Banteay Meanchey (rural population: 24%-73%) and Sangkat Chaom Chaov in Phnom Penh (rural population: 6%-40%). Sisophon has a relatively small share of the nationally estimated high-wealth quintile (22.9%) and a lower average of schooling years than Sangkat Chaom Chaov (high-wealth quintile: 84.4%). The schools selected were Xavier Jesuit School (rural) and Mirero School (urban). Xavier Jesuit School has both kindergarten and elementary classes. In total, 4 classes—2 kindergarten classes, 1 first grade class, and 1 second grade class—were selected. As Mirero School has only elementary classes, we selected 2 classes each from the first and second grades.

To target children eligible for the cross-sectional study, we selected regions and schools with the advice of the Korea International Cooperation Agency Cambodia Office, which employs experts who are aware of the overall living environment, including education, in Cambodia. After the selection of areas and schools, classrooms and grades of age that met the inclusion criteria were selected.

Study participants were enrolled from 2 elementary schools in Cambodia’s 2 regions: rural and urban. The inclusion criteria for children were as follows: (1) male and female students aged 3 to 9 years, (2) no environmental change in their home or school during the study, and (3) children who provided consent from their legal guardian or themselves. Students were excluded from the test if they could not use the study’s digital device because of physical conditions or when legal guardians did not provide consent for the test. Figure 1 shows details of the participant selection flow.

The participants were divided into 2 groups based on their responses to the questionnaire about digital device experience: group A comprised children with digital device experience, and group B comprised those without digital device experience. The participants in group A were subsequently divided into 4 subgroups according to the duration (in minutes) of digital device use: group A-1: <30 minutes per day, group A-2: 30 to 60 minutes per day, group A-3: 60 to 90 minutes per day, and group A-4: >90 minutes per day. Of the 232 participants, 162 (69.8%) were in group A, with 95 (58.6%) in group A-1, 47 (29%) in group A-2, 10 (6.2%) in group A-3, and 10 (6.2%) in group A-4, whereas group B had 70 (30.2%) participants.

All participants provided written informed consent before enrollment in the study. Consent was granted by their guardians after the study was explained in writing. Ethical clearance was obtained from the Ministry of Health’s National Ethics Committee for Health Research in Cambodia (212).

We gathered data on demographic and socioeconomic characteristics through surveys and data on cognitive function using the Cambridge Neuropsychological Test Automated Battery (CANTAB) test. The survey targeted the basic socioeconomic characteristics and demographics of the participants and their guardians from April 22, 2019, to May 4, 2019. Most of the guardians were either the parents or grandparents. At both schools, the survey was delivered to the guardians living with the participant. The response rate was 82.9%. Most (335/404, 82.9%) of the participants returned the completed form to the school. In cases where the guardians could not read, schoolteachers assisted them in completing the survey at school. The questionnaire had 2 parts. The first elicited demographic information about the children, such as sex, age, and siblings, as well as digital device exposure and use. The second collected information about the family’s socioeconomic status, including residential conditions, guardians’ occupation and education level, household income, and educational expenditure (Multimedia Appendix 1). The participants’ cognitive function, depending on their experience with digital devices, was evaluated through the CANTAB test using a tablet device. The CANTAB test includes highly sensitive, precise, and objective measures of cognitive function correlated with neural networks [25]. It includes tests that evaluate 4 cognitive areas: attention and psychomotor speed, executive function, memory, and social and emotional cognition. The participants were asked to perform 5 tests: motor screening task, reaction time, spatial working memory (SWM), pattern recognition memory, and spatial span. We collected 53 test result variables, including 11 key variables, for measuring the outcome of each test (Table 1 and Multimedia Appendix 2).

To compare the demographic differences between groups A and B, a chi-square test was performed on categorical variables such as sex, and a Mann-Whitney U test was performed on continuous variables such as age. To compare the differences in cognitive function between the 2 groups, normality was tested for the variables using the Shapiro-Wilk test. Variables satisfying normality were examined using a 2-tailed t test, and those that did not satisfy normality were compared between the groups using the Mann-Whitney U test. For each comparison, the effect size was calculated for the 2 groups according to digital device exposure and for the 4 subgroups according to digital device use time, and the results of the normality test were compared for possible type 1 statistical errors. In comparing the 2 main groups, Cohen d was calculated for the t tests, and r was calculated using the Mann-Whitney U test. To compare the effect size among the 4 subgroups, eta squared (η²) was calculated for ANOVA, and epsilon squared (ϵ²) was used for the Kruskal-Wallis test. The threshold of statistical significance was set at P<.05, and an effect size greater than the small size, depending on its type (Cohen d≈±0.20: small, r≈±0.10: small, η²≈0.01: small, and ϵ²≈0.01: small), was considered significant for 2-tailed t tests. To confirm the association among demographic and socioeconomic characteristics, digital device familiarity, and cognitive function in children, univariate regression analysis was performed for 3 cognitive domains and 11 variables flowing from the CANTAB test. In the multivariate linear regression analysis, only variables that were statistically significant (P<.05) through univariate regression analysis were selected, and their effect on the cognitive function variable was evaluated with and without adjusting for age and sex. All statistical analyses were conducted using R (version 3.6.3; R Foundation for Statistical Computing) and Python (version 3.7; Python Software Foundation).

A total of 335 children participated in this study. Of these 335 children, 4 (1.2%) who did not answer the questions about digital device exposure and 6 (1.8%) who answered inconsistently (eg, talking about the purpose of the device, not the experience) were excluded from the data analysis; in addition, 13 (3.9%) children with duplicated test results and 80 (23.9%) without values for the key variables of each test were excluded. Ultimately, of the 335 children, 232 (69.3%) were included in the data analysis (Figure 1).

Of the 232 children, 162 (69.8%) were in group A, and 70 (30.2%) were in group B. The 162 students in group A comprised 110 (67.9%) Xavier Jesuit School students and 52 (32.1%) Mirero School students, a significant difference (P<.05). No significant difference between the groups was observed in terms of sex: in group A, 59.3% (96/162) of the participants were male students, whereas in group B, 56% (39/70) were male students (P=.72). The mean ages of the participants in group A and group B were 7.3 (SD 1.5) years and 7.6 (SD 1.3) years, respectively, without significant differences (P=.15). The proportion of participants with monthly family income of <US $150 was higher in group B (38/70, 54% vs 53/162, 32.7% in group A), with significant differences (P=.004). The proportion of participants with monthly family income of >US $350 was higher in group A (25/162, 15.4% vs 4/70, 6% in group B). The proportion of students whose mothers had secondary education and above was far higher in group A (73/162, 45.1% vs 17/70, 24% in group B), with significant differences (P=.01). The proportion of families who spent >US $30 per month on education per child was also higher in group A (24/162, 14.8% vs 4/70, 6% in group B), without overall differences between the 2 groups (P=.07; Table 2).

When comparing the results of the CANTAB test between the 2 groups, the SWM between errors (4 boxes; SWMBE4) variable of group A had a median of 2.0 (IQR 1.0-3.0), whereas that of group B had a median of 2.0 (2.0-3.0); there were significant differences (P=.01). However, the effect size was small (P=.02). The SWM strategy (SWMS) variable of group A had a median of 12.0 (IQR 10.0-58.3), whereas that of group B had a median of 10.0 (IQR 9.0-12.0); there were significant differences (P.002). The effect size was small (P=.04). There were no significant differences between groups A and B for the other variables. The smaller values for the motor screening task mean latency (MOTML) and reaction time median 5-choice reaction time (RTIFMDRT) variables were positive, but the median values of group A were 15.5 and 21.2 points higher, respectively, than those in group B. Conversely, the reaction time median 5-choice movement time variable for the exposure group was 13.5 points lower, without a significant difference in median values among the other variables (Table 3).

When comparing the differences in cognitive levels based on the duration of digital device use, the results showed no significant differences among the 4 subgroups, with all effect sizes being small. Although there were no significant differences, the median values of the 2 variables measuring attention and psychomotor speed in group A-4 were higher than those in group A-1. Regarding the visual memory variables, there were no significant differences among the 4 subgroups, and the distribution did not show a linear relationship. Regarding the executive function test, as use time increased, the number of errors decreased, leading to positive results. Regarding the SWM test, the distributions of the 4 subgroups were not different (Table 4).

All results of univariate linear regression analysis to determine the relationship between demographic characteristics and socioeconomic status and CANTAB test score have been presented in Multimedia Appendix 2.

The measures showed that the 3 variables MOTML, RTIFMDMT, and RTIFMDRT from the Attention and psychomotor speed cognitive function were significantly associated with Age (all P<.001), School (all P<.001), and Education expense per child (all P<.001) in the univariate model and multivariate model 1. However, in multivariate model 2, adjusted for age and sex, MOTML was significantly associated with School, and RTIFMDRT was significantly associated with Family type (Table 5).

The 3 Memory cognitive function variables pattern recognition memory percent correct immediate, pattern recognition memory percent correct delayed, and spatial span forward span length were associated with Age, Education expense per child, Family income, Family type, and School. In the univariate model and multivariate model 1, pattern recognition memory percent correct immediate was significantly associated with Age (all P<.001), Family type (all P=.07), and School (all P<.05). However, in multivariate model 2, it was significantly related only to Age (P<.001) and Family type (P=.01). In the univariate model and multivariate model 1, pattern recognition memory percent correct delayed was significantly related to Age (all P<.001), Education expense per child (all P<.05), and Family income (all P<.05). However, in multivariate model 2, it was significantly associated only with Age (P<.001) and Family income (all P=.08). In all 3 regression models, spatial span forward span length was strongly related to Age (P<.001; Table 5).

The variables SWM between errors (4, 6, and 8 boxes), SWMBE4, SWM between errors (8 boxes), and SWMS corresponding to Executive function were related to Age, Digital device exposure, Education expense per child, Family income per month, Family type, Sex, and School. In the univariate model, SWM between errors (4, 6, and 8 boxes) was significantly related to Education expense per child (P=.03) and Sex (P=.05). However, in multivariate model 1, it was not associated with any variable. SWMBE4 was significantly associated with School (all P<.05) in the 3 regression models. Both Age (all P<.05) and Sex (all P<.05) had a significant association in the 3 regression analyses of SWM between errors (8 boxes). The SWMS variable showed a significant relationship with the 3 variables Age (all P<.05), Digital device exposure (all P<.05), and School (all P<.001) in all regression analyses (Table 5).

Since 2015, when the 2030 Agenda for Sustainable Development was adopted at the United Nations General Assembly, many initiatives for the application of digital technologies to policies and programs for development, particularly those associated with SDGs 3 and 4, emerged in various countries, particularly LMICs [5,29]. However, there is insufficient evidence on how digital device use affects children’s cognitive improvement in LMICs [8,30,31]. This study focused on urban and rural Cambodian schools to provide a comprehensive perspective on the adoption and use of digital devices among children. We used a cross-sectional study to examine the cognitive level of 3 domains using the CANTAB, focusing on how exposure to digital devices affects cognitive development in elementary school–age children.

Our study found significant differences in demographic and socioeconomic characteristics such as school location, family type, and family income according to digital device exposure. We found that children demonstrated neither superiority nor inferiority in cognitive scores on the CANTAB in 3 cognitive domains depending on digital device use. Given the lack of empirical research on the impact of digital device exposure on cognitive development, we offer two key findings that make a significant contribution to policies and programs for the application of digital devices to health care and education: (1) there is no significant association between exposure to digital devices and cognitive development; (2) however, socioeconomic conditions such as school location, family income, family type, and education expenditure are significantly related to cognitive function [31].

Several policy implications can be drawn from this study’s findings. The first is the importance of an enabling environment that maximizes the impact of digital devices on cognitive function and development. The evidence demonstrates that socioeconomic factors significantly affect cognitive function and development in infancy, including memory and enforcement functions [24,27]. For example, US studies found that children from higher socioeconomic backgrounds had higher achievements in visuospatial, memory, and executive functions than those from middle or low socioeconomic backgrounds. Similarly, students from higher socioeconomic backgrounds in Brazil were found to have better memory and executive function results [24]. To what extent do socioeconomic conditions affect children’s cognitive function and development in LMICs? The findings of our study demonstrate that these conditions, specifically education expenditure, enhanced cognitive function and development to a greater extent than did children’s exposure to digital devices. Comments such as “we have higher priorities than laptops” expressed by many delegates to the 2005 World Summit on the Information Society, during which the “One Laptop Per Child” devices were introduced, were supported by our findings. Proper investment in education is a much more significant task than purchases of, and exposure to, digital devices, particularly in LMICs.

The second implication concerns the design and setting of digital devices. Many initiatives on the use of digital devices for children’s cognitive function and development are based on a theoretical framework called the “brain’s rearrangement capacity,” which posits that children learn to associate what they see, hear, and know with symbolic characters [31]. Thus, we collected data indicating the symbolic characters that children see and hear on digital devices (Multimedia Appendix 1). Most children with digital devices experience a desire to play games and watch videos, such as those available on YouTube. By playing games and watching videos, children may not learn to associate what they see, hear, or know with symbolic characters. Our finding on the insignificant impacts of exposure to, or use of, digital devices on cognitive development suggests that the design and setting of the devices that children are exposed to should be effective enough to stimulate and accelerate the “brain’s rearrangement capacity.” Giving children access to digital devices with suitable designs and settings is more important than ever amid a pandemic such as the COVID-19 pandemic, during which almost all schools had closed for long periods worldwide [28].

Our study had several limitations. First, the levels of children’s cognitive function and development were measured only during the survey; the long-term impact of digital device use on cognitive function and development was not assessed. This is a typical limitation of cross-sectional studies with a defined time frame. A longitudinal panel study based on groups of children with different socioeconomic backgrounds and levels of exposure to digital devices is needed to measure the long-term effects of digital devices on cognitive function and development. Second, our study was conducted in 2 areas of Cambodia. We focused on both rural and urban areas to examine the impact of differences in socioeconomic levels. However, this did not cover a sufficient area. Nevertheless, considering the lack of studies on the impact of digital device exposure on children’s cognitive function and development in LMICs, particularly those focusing on both digital device exposure and socioeconomic conditions, our study expands the understanding of the enabling environment required for digital device–use initiatives aimed at children’s cognitive function and development.