Assessing digital health inequalities in Indonesia: an extended SEM–ANN model of socioeconomic status, mHealth usage, perceived access to healthcare services, and sustained physical activity

Background Enhanced access to health services and lowered health inequalities can possibly be associated with the usage of digital technologies. Unfortunately, differences in socioeconomic status (SES) and technological skill could be associated with disparities in their usage. Indonesia remains a nation with rural–urban disparity. Hence, to create strategies for equitable health service, it is crucial to comprehend the extent to which mobile health usage (MHL) and SES are associated with perceived sustained physical activity (PPA) and access to healthcare. Purposes This research extended the 2026 Sumarliah et al.’s framework by incorporating perceived access to healthcare services (PAHS) into the model and assessing its association with MHL, PPA, and SES among a broader sample of Indonesian rural–urban respondents. To deliver predictive and explanatory evaluations, it further integrated artificial neural network (ANN) and structural equation modeling (SEM) assessments. Methods This research adopted a stratified regional sampling to support representation of six major Indonesian regions: Papua-Maluku, Nusa Tenggara-Bali, Sumatra, Java, Sulawesi, and Kalimantan. A sum of 1,566 valid responses (78.3%) were retained, which comprised 871 urban (55.6%) and 695 rural (44.4%) respondents. To evaluate the gathered data, a three-stage quantitative approach was utilized, which involved SEM, multi-group SEM and ANN examinations. Results There was a positive association between PAHS and SES, as well as between MHL and SES. The PAHS was also positively linked to MHL. Conversely, PAHS was not significantly associated with PPA. As such, the indirect link between MHL and PPA via PAHS was insignificant. Multi-group SEM verified the much stronger positive correlations among PAHS, MHL, and SES within urban samples instead of rural samples. Supportive evidence to predict PPA was acquired through ANN exploration, where maximum normalized predictive importance was achieved by the MHL input construct. Conclusion From the present research it is evident that PAHS had statistically insignificant association with PPA, which is different from earlier works’ findings. The likely reason is that although PAHS can be associated with the usage of health services, it may be inadequate to account for differences in respondents’ PPA.