Qualitative and Quantitative Synthesis of the Results

For a qualitative analysis of the results, study country, study population, eHealth literacy measurement tools, types of health-related behaviors and measurement tools, and the relationship between eHealth literacy and health-related behaviors were presented descriptively. The characteristics of the study population were described by age and morbidity status. The specific contents of health-related behaviors were summarized, and they were also classified into the following 3 categories: health-promoting behavior, health-supporting behavior, and disease management behavior. The relationship between eHealth literacy and health-related behaviors was evaluated by whether the effect of eHealth literacy was positive or negative.

For evaluating the association between eHealth literacy and health-related behaviors by a quantitative method, the pooled correlation coefficient was estimated by Fisher z-transformation and construction of the inverse transformation [29]. We used the correlation coefficients of individual studies and treated each result as a separate study when multiple subgroup results were reported in 1 study. The pooled correlation coefficient presented with a 95% CI was tested by performing hypothesis testing to determine whether the correlation was statistically significant. Interpretation of the pooled correlation coefficient was conducted according to Cohen criteria [30]. Cochran Q-statistics and I²-statistics were used to assess the heterogeneity within the studies included in the meta-analysis, and we applied either the fixed-effects model or random-effects model, depending on the significance of heterogeneity (P<.10 and I²≥50%) [29]. To test the validity of the study results, publication bias was evaluated using a funnel plot and Egger regression, and in case of suspected publication bias, the severity of bias was tested using the trim-and-fill method to estimate the degree to which the publication bias would affect the validity of the study results.

The total effect size (ie, pooled correlation coefficient) was derived from each group of studies divided by the participants’ mean age, morbidity status, and types of health-related behaviors, and from all studies that could be quantitatively synthesized. Through this, we tried to evaluate changes in the effect size according to detailed characteristics. All statistical analyses were performed using Comprehensive Meta-analysis, version 2 software (Biostat).