Choice preferences for AI-based medication counselling in patients with chronic obstructive pulmonary disease: a discrete choice experiment

Abstract

Background

Patients with chronic obstructive pulmonary disease (COPD) require long-term and accuracy medication guidance, making it a promising tool for the management of COPD. Nevertheless, patient preferences for AI-based medication counselling in COPD management remain unclear.

Methods

Data from 325 patients were analyzed. We identified 6 attributes of AI-based medication counselling and utilized an orthogonal design to generate a selection set for a discrete choice experiment (DCE). Questionnaires were collected from patients with COPD in Henan Province, comprising a general information questionnaire, DCE questionnaire, five-level EuroQol five-dimensional questionnaire (EQ-5D-5L), medication literacy scale, and e-health scale. We employed a conditional logit regression model to estimate patient-reported preferences for each attribute and calculated their willingness to pay. Additionally, we determined that the 3-class model has the best data fit superiority by calculating the Akaike information criterion (AIC), and performed latent class analysis (LCA) to analyze the heterogeneity of people's preferences for AI-based medication counselling. Based on the results of LCA, we further used multinomial logistic regression analysis to explore the factors influencing patients' preferences for AI-based medication counselling.

Findings

The conditional logit regression model reveals that patients prioritize the attributes of AI-based medication counseling in the following order of importance: accuracy, comprehensibility, consultation channels, presentation format, manual review, and cost. The calculation of WTP revealed that participants are willing to pay an additional 20.81 RMB for a boost in result accuracy from 60% to 100% and 14.33 RMB for an increase from 80% to 100%. Likewise, enhancing comprehensibility to fully understandable from incomprehensible or partially comprehensible garners an extra 13.85 RMB or 4.39 RMB, respectively. Latent class analysis categorized patients into three groups: Class1 (21.2%) favors human-reviewed consultations via websites and audio/dialogue; Class2 (59.7%) prefers app-based consultations without human review and presented with images; and Class3 (19.1%) is less concerned with consultation details but prioritizes unreviewed audio or dialogue formats. Multinomial logistic regression analysis showed that, among Class1 and Class2, gender (p=0.034) age (p=0.017) and e-health literacy scores (p=0.001) significantly influence classification, with females and older, more digitally literate individuals leaning towards Class1. Comparatively, Class1 versus Class3 analysis indicated that higher income (p=0.027) and the presence of other chronic diseases (p=0.005) significantly predict Class1 membership, suggesting that wealthier individuals or those with multiple health conditions are more likely to prefer Class1.

Interpretation

Currently, a significant majority of patients are receptive to AI-based medication counselling. Overall, people prefer AI-based medication counseling that is highly accurate and understandable, accessed through websites, presented in audio or dialogue form, manually reviewed, and low in cost. To effectively guide COPD patients with AI-based medication counseling, it's crucial to enhance the accuracy and clarity of consultation results. However, the study could not include all attributes related to AI-based medication counselling.

Funding

None.