With the rapid advance in technologies over the past decade, the Internet has become an integral part of the lives of many individuals. Widespread Internet access has benefited people by improving access to online information and providing new opportunities for social communication and entertainment [1]. Although the Internet has offered advantages, excessive usage of the Internet is associated with a psychiatric condition known as Internet addiction (IA) [2, 3]. IA mimics other forms of addiction and is characterized by an inability to inhibit Internet use despite negative impact on physical health and psychosocial functioning [4,5,6,7].

Previous studies have highlighted the importance of studying IA in university students because they are more likely than the general population to use the Internet [1] and vulnerable to develop IA [8]. Additionally, university students have an immediate desire to engage in online activities because the Internet is readily available on laptop computers and smartphone devices [1]. Previous studies report the prevalence of IA among university students to vary between 4% in the USA [9] and 16.8% in Italy [10].

Meta-analysis is a statistical procedure used for combining the results of a number of studies in order to estimate a pooled effect size [11, 12]. Although previous meta-analysis studied the pooled prevalence of depression among medical students in different countries [13, 14], the pooled prevalence of IA remains unknown. In more recent work, IA was found to be associated with atrophy in the prefrontal cortex and cognitive impairment [4]. Such cognitive impairment could significantly affect the academic performance of medical students and fitness to practice after graduation. For this purpose, the current meta-analysis aimed to establish the pooled prevalence of IA among medical students in different countries.

Methods

A comprehensive search was conducted from January 1, 2017, to February 20, 2017. The following databases were searched: PubMed (since 1966), Embase (since 1980), PsychINFO (since 1806), BIOSIS (since 1926), Science Direct (since 2006), and Cochrane CENTRAL (since 1993). The search terminologies included permutations of the keywords for IA (e.g., Internet addiction, problematic internet use) and medical students, medical undergraduates.

The inclusion criteria for this meta-analysis were as follows: (a) undergraduate medical students formed the main cohort of the study population, (b) the severity of IA was measured by a validated questionnaire, and (c) prevalence of IA was the primary outcome of the study population. The exclusion criteria were as follows: (a) a study without an abstract written in English and (b) insufficient information to compute the pooled prevalence of IA.

Articles were de-identified (author(s), year of publication, and journal name) before data extraction [15]. Selection of the relevant publications was conducted independently by two authors, namely that of MWBZ and RL. In the first phase, articles were screened based on their titles and abstracts. The shortlisted articles were then evaluated against the aforementioned inclusion and exclusion criteria. In the event of any disagreement between two authors, it was resolved by discussion with the last author, RCMH. All procedures conformed to the guidelines for meta-analysis of observational studies in epidemiology and preferred reporting items for systematic reviews and meta-analyses where appropriate [16, 17].

The following information was extracted from each of the article, cross-checked by the second author and the last author, and recorded on a standardized electronic data collation form: (a) publication details (names of the authors and year of publication), (b) the number of medical students diagnosed with IA, (c) the total sample size for each study, (d) the mean age of participants, (e) the proportion of male and female participants, and (f) the name of assessment questionnaire used to assess IA, and (g) the mean IA score of all participants.

All statistical analysis was performed using the Comprehensive Meta-Analysis Version 2.0 based on random-effects model and methods established by previous studies [18,19,20]. The random-effect model was used because it assumes varying effect sizes between studies, because of differing study design and study population [21]. The assumption of homogeneity of effect sizes, that is, τ 2 = 0, can be tested by a chi-square statistic which is usually known as the Q statistic [22]. Between-study heterogeneity was assessed with the I 2 statistic, which describes the percentage of variability among effect estimates beyond that expected by chance [23]. As a reference, I 2 values of 25% are considered low, 50% moderate, and 75% high heterogeneity [24].

Meta-regression was performed to identify demographic and disease-related factors that might contribute to the heterogeneity [25] and predict the effective size [26]. The regression coefficients and the associated z values and p values were reported in the meta-regression analysis [27]. Subgroup analysis was undertaken to investigate the effects of categorical variables on the pooled prevalence of IA. We compared the prevalence of IA between subgroups based on the questionnaires used to assess IA.

Egger’s regression test was conducted to determine if publication bias was present [28]. If significant publication bias was present, the classic fail-safe test was performed to determine the number of missing studies that are required in order for the p value of the publication bias among the observed studies to be higher than 0.05.

Results

A cumulative total of 129 citations were identified using the search strategy and screened. One hundred fourteen articles were excluded because these studies recruited non-medical university students or failed to provide sufficient data to calculate effect size. Fifteen articles were further reviewed based on inclusion and exclusion criteria. One article was excluded because it did not provide access to the full text. Three studies were excluded because they included subjects who were studying in basic medical science but not medicine. One study was excluded because it included postgraduate students. The final data set consisted of 10 studies with a total of 3651 participants. The characteristics of studies included are described in Table 1.

Table 1 Baseline characteristics of studies included
Full size table

The pooled prevalence of IA among 3651 medical students is 30.1% (95% confidence interval 28.5–31.8%, Z = −20.66, df = 9, τ 2 = 0.90, I 2 = 98.12). This meta-analysis demonstrates significant heterogeneity across studies (p < 0.0001). Figure 1 shows the forest plot generated for the pooled prevalence of IA among medical students. The meta-regression analyses show that the mean age of the medical students (β = 0.13, 95% CI = −0.94–1.21, Z = 0.24, p = 0.81), gender proportion of male medical students (β = 0.0091, 95% CI = −0,084–0.10, Z = −0.19, p = 0.85), and severity of IA (β = 0.13, 95% CI = −0.036–030, Z = 1.53, p = 0.12) are not significant moderators accounting for heterogeneity. Based on the subgroup analysis, the pooled prevalence of IA diagnosed using the CIAS (5.2, 95% CI 3.4–8.0%) is significantly lower than YIAT (32.2, 95% CI 20.9−45.9%) (p < 0.0001). For publication bias, the Egger’s regression test shows that publication bias is not significant in this meta-analysis (intercept = −3.55, 95% CI−21.4–14.25, t = 0.46, df = 8, p = 0.657).

Fig. 1
figure 1

Forest plot of pooled prevalence of Internet addiction among medical students

Full size image

Discussion

To our knowledge, this is the first meta-analysis that examined the prevalence of IA among medical students in different countries. Our most important result is that the prevalence of IA among 3651 medical students is 30.1%. In one meta-analysis, the global prevalence of IA among the general population was found to be 6% in 31 different countries [4]. In another study, the prevalence of IA among adolescents assessed by the YIAT was found to be 5% in six different Asian countries [38]. This meta-analysis highlights that the prevalence rate of IA among medical students was almost five times higher than that of the general population. The prevalence of IA was higher than global prevalence of depression in medical students which is 28.0% [14]. The high prevalence rate of IA is a major concern because medical students are trained to become health professionals. IA may affect patient safety [10] because it is associated with cognitive impairment [4]. Such impairment can affect medical student’s abilities to concentrate during academic study or clinical placement [39]. Furthermore, IA is associated with poor quality of life [40]. There are several reasons that account for higher prevalence of IA among medical students. First, the accessibility hypothesis suggests that medical students are more vulnerable to IA due to frequent use of the Internet to search for medical information as well as participating in online learning and assessment [8]. Second, medical students may immerse themselves into the virtual reality of the Internet to escape from academic stress [41]. Third, the use of the Internet as a coping strategy for negative psychological states (e.g., stress or depression) may perpetuate IA because students suffering from IA often avoid socialization with other people [1].

Meta-regression analyses demonstrate that continuous variables such as mean age, proportion of gender, and severity of IA are not significant moderators and do not account for the heterogeneity in the pooled prevalence of IA among medical students. Our findings suggest that gender may not play a key role in IA. This can be an emerging trend when both genders have equal access to the Internet [40], especially for medical students. The subgroup analysis demonstrates that there is a significant difference in the prevalence of IA between participants assessed by YIAT and CIAS; therefore, the questionnaires assessing IA is a potential source of heterogeneity. Previous study advised researchers to interpret the prevalence of IA assessed by the CIAS with caution because it was validated in Chinese population but not in other populations [38, 43]. In this meta-analysis, the study which used CIAS was conducted in Iran.

This meta-analysis has several limitations. First, we did not have information on other factors including the purpose of using the Internet (e.g., playing online games, usage of social media), the percentage of individuals who have access to the Internet in each country of study, quality of life of participants, family history of addiction, and academic performance [4, 44]. These factors could influence the pooled prevalence of IA. Second, this meta-analysis is based on six different countries (3 studies from Iran, 2 studies from India, 2 studies from Malaysia, 1 study from Chile, 1 study from Greece and 1 study from Thailand). We could not locate studies conducted in Australia, East Asia, Europe, and North America during our search. This limitation is noteworthy because it limits the generalizability of results. Further studies focusing on the prevalence of IA among medical students are required from these continents. Third, the studies included in this meta-analysis are cross-sectional and this meta-analysis is not able to demonstrate causality or temporal association between IA and undergraduate medical studies. Fourth, we could not perform subgroup analysis to compare the prevalence of IA between medical students from junior and senior years.

Our results have implications for administrators and medical teachers in medical schools. The high prevalence of IA among medical students suggest that medical teachers should be able to recognize signs of IA including inability to reduce internet use, psychological dependency, and withdrawal symptoms [1]. Prompt recognition of IA will facilitate referral for early intervention by mental health professionals. Early intervention can prevent further deterioration of IA and potential negative impact on quality of life [40]. Given the association between IA and other psychiatric disorders, it is of importance to screen for other psychiatric comorbidity including depression and sleep deprivation when assessing medical students suffering from IA. University students found IA as key factor for sleep deprivation [1] because students stay awake during late night hours to surf the Internet [42]. For medical students, the lack of sleep may lead to lack of concentration in lectures, tutorials, and bedside teaching. In addition, sleep deprivation is associated with impairment in executive functioning [45]. Such impairment is detrimental to medical students and junior doctors because it affects decision-making ability and patient safety [10]. Medical educators should increase the awareness of IA among medical students by teaching about IA in undergraduate psychiatry courses. Mental health professionals can offer motivational interviewing, cognitive behaviour therapy, virtual-reality therapy [46] and psychotropic medications to treat psychiatric comorbidity associated with IA.

In conclusion, this meta-analysis reports that the pooled prevalence of IA among medical students from different countries is 30.1%, which is approximately five times than that of the general population. Age, gender, and severity of IA cannot explain high heterogeneity in prevalence, but IA assessment questionnaire are a potential source of heterogeneity. Given the high prevalence of IA, medical teachers and medical school administrators should identify medical students suffering from IA and refer them for intervention. The undergraduate psychiatric curriculum should cover IA and increase the awareness of IA among medical students.