ABSTRACT
Background: WeChat has become a potent medium for disseminating public health information, especially during the coronavirus disease 2019 (COVID-19) pandemic. WeChat is important for public health organizations when considering users' information needs and preferences to further explore factors that affect user engagement. Methods: We collected data from WeChat official accounts (WOAs) of the Chinese provincial Center for Disease Control and Prevention (CDC) to identify factors affecting and predicting the behavior of user engagement as measured by the level of reading and re-sharing during different phases of the COVID-19 pandemic between January 1, 2019, and December 31, 2020. We used multiple logistic regression analyses to identify features of articles with higher reading and re-sharing levels from 31 Chinese provincial CDCs. We developed a nomogram to predict the effect on user engagement. Results: We collected a total of 26 302 articles. Release position, title type, article content, article type, communication skills, marketing elements, article length, and video length were key determinants of user engagement. Although the feature patterns also varied between different pandemic stages, the article content, release position, and article type were still the most prominent features driving user engagement. Regarding article content, the COVID-19 pandemic report and guidance for public protection were more likely to obtain high-level reading (normalization: odds ratio (OR) = 12.340, 95% confidence interval (CI) = 9.357-16.274) and re-sharing (normalization: OR = 7.254, 95% CI = 5.554-9.473) than other contents throughout the pandemic. When we compared release position with secondary push, users who used main push were more likely to exhibit high-level reading and re-sharing during any period, especially during normalization (OR = 6.169, 95% CI = 5.554-6.851; OR = 4.230, 95% CI = 3.833-4.669). For article type, a combination of text, links and pictures was associated with a higher rate of reading (normalization: OR = 4.262, 95% CI = 3.509-5.176) and re-sharing level (normalization: OR = 4.480, 95% CI = 3.635-5.522) compared to text only. Simultaneously, the prediction model showed good discriminatory power and calibration. Conclusions: Discrepancies exist in article features between different pandemic stages. Public health agencies should make full use of official WOAs and consider the information needs and preferences of users in order to better carry out health education and health communication with the public when public health events occur.