Colakoglu Selahattin, Durmus Mustafa, Polat Zeynep Pelin, Yildiz Asli, Sezgin Emre
Department of Clinical Development, Albert Health, Istanbul, Turkey.
School of Medicine, Biruni University, Istanbul, Turkey.
J Med Syst. 2025 Jun 9;49(1):76. doi: 10.1007/s10916-025-02202-2.
Understanding user engagement with conversational agents is key to their sustainable use in mobile health and improving patient outcomes. This retrospective study analyzed interactions with a multimodal conversational agent in the Albert Health app to identify usage patterns and barriers to long-term engagement in self-care and chronic disease management.
We retrospectively analyzed interactions from 24,537 users of a Turkish-language mobile health app (between January 1, 2022, and December 31, 2023). Interactions with the app's multimodal conversational agent (voice and text) were categorized by demographics, interaction type, and engagement mode. Descriptive statistics summarized patterns, while Mann-Whitney U, Chi-square, and logistic regression identified group differences and predictors of sustainable engagement.
Most users were female (56%) and aged 30-45 (44%). The majority (92%) used general health programs, with only 8% in disease-specific ones. Common interaction types included health information (32%), small talk (20%), and clinical parameter logging (16%; e.g., blood pressure). Voice use was frequent in fallback (80%; unclear/ out-of-scope input), small talk (64%), and medication tasks (53%), while screen input was more common for clinical logging (61%) and health queries (59%). Engagement peaked in the first week and declined after 10 days. Sustainable engagement was associated with disease-specific program use (OR = 0.67, 95%CI: 0.60-0.74, p < 0.001), greater voice interaction (OR = 1.005, 95%CI: 1.004-1.006, p < 0.001), and a balanced mix of clinical and non-clinical use (OR = 1.56, 95%CI: 1.43-1.70, p < 0.05).
This study highlights user preferences for voice interaction and health information access when using a multimodal conversational agent. The high rate of single-session users (58%) points to barriers to sustainable engagement, emphasizing the need for better user experience strategies.
了解用户与对话代理的互动情况是其在移动健康领域可持续应用以及改善患者治疗效果的关键。这项回顾性研究分析了与阿尔伯特健康应用程序中多模态对话代理的互动，以确定自我护理和慢性病管理中的使用模式及长期参与的障碍。
我们回顾性分析了一款土耳其语移动健康应用程序的24537名用户（在2022年1月1日至2023年12月31日之间）的互动情况。与该应用程序多模态对话代理（语音和文本）的互动按人口统计学、互动类型和参与模式进行分类。描述性统计总结了模式，而曼-惠特尼U检验、卡方检验和逻辑回归确定了组间差异和可持续参与的预测因素。
大多数用户为女性（56%），年龄在30至45岁之间（44%）。大多数（92%）使用一般健康程序，只有8%使用特定疾病程序。常见的互动类型包括健康信息（32%）、闲聊（20%）和临床参数记录（16%；如血压）。在回退情况（80%；输入不清晰/超出范围）、闲聊（64%）和用药任务（53%）中语音使用频繁，而在临床记录（61%）和健康查询（59%）中屏幕输入更为常见。参与度在第一周达到峰值，10天后下降。可持续参与与特定疾病程序的使用（OR = 0.67，95%CI：0.60 - 0.74，p < 0.001）、更多的语音互动（OR = 1.005，95%CI：1.004 - 1.006，p < 0.001）以及临床和非临床使用的平衡组合（OR = 1.56，95%CI：1.43 - 1.70，p < 0.05）相关。
本研究突出了用户在使用多模态对话代理时对语音互动和获取健康信息的偏好。单会话用户的高比例（58%）表明存在可持续参与的障碍，强调了制定更好用户体验策略的必要性。
