Preferences of Patients With Tuberculosis for AI-Assisted Remote Health Management: Discrete Choice Experiment.

Abstract

Background: Tuberculosis remains a major global public health challenge, especially in low-resource settings where long-term treatment adherence and regular follow-up are critical. The integration of artificial intelligence (AI) into remote health management has the potential to improve care delivery and patient outcomes. However, evidence on the preferences of patients with tuberculosis regarding AI-assisted services remains limited.

Objective: This study aimed to examine the preferences of patients with tuberculosis for AI-assisted remote health management services in China, identifying key service characteristics that influence their choices.

Methods: A discrete choice experiment was conducted among 203 patients with tuberculosis in Hubei province, China. Attributes and levels were identified through a systematic literature review, qualitative interviews, and expert panel consultations. The final design included 6 attributes: interaction method, service provider, service frequency, service content, out-of-pocket cost, and service integration. Each participant completed 8 choice tasks comparing hypothetical service options constructed based on these attributes. Preferences were analyzed using a mixed logit model to account for preference heterogeneity. Additional subgroup analyses were performed to explore variations in preferences across sociodemographic characteristics.

Results: All 6 attributes significantly influenced patients' preferences (all P values <.05). Participants strongly favored services involving physician oversight (P<.001), video-based interactions (P<.001), and comprehensive content (P<.001), while higher costs were associated with lower acceptance (P<.001). Subgroup analyses indicated that higher-income patients demonstrated both a greater willingness to pay and a stronger preference for physician involvement. Female participants expressed a lower preference for AI-assisted physician-led services compared to AI-only configurations. Patients with higher educational attainment also reported lower preferences for physician-involved services. Age-related differences were not statistically significant. Across all subgroups, cost remained a critical determinant of service acceptance.

Conclusions: Patients with tuberculosis expressed a clear preference for high-quality, human-integrated remote health management services, emphasizing the importance of physician involvement and personalized, interactive care. These findings suggest that fully AI-driven models may face resistance and that hybrid models combining AI efficiency with professional oversight are more acceptable. Policymakers and service designers should prioritize affordability, provide targeted financial support for populations considered vulnerable, and invest in digital literacy initiatives to enhance equitable access. This study provides critical evidence to support the development of AI-assisted tuberculosis management strategies that align with patient preferences and improve treatment adherence in low-resource settings.