Artificial Intelligence-assisted Primary Care for Influenza Vaccination Among Chronic Patients in China (AIPC-Flu)

A randomized controlled trial will be conducted to evaluate the effectiveness of an AI-assisted primary care intervention to improve influenza vaccination uptake among patients with chronic diseases in China, with additional health economic and feasibility evaluations to assess its cost-effectiveness and potential for scale-up.

The sample size calculation is based on the primary outcome measure (influenza vaccination uptake) and the main analysis method (comparison of rates between two groups). Based on the literature review and survey, the current influenza vaccination uptake among patients with chronic diseases is estimated to be approximately 5%. We assume that the uptake rate could increase to 10% following the intervention. Using the formula for comparing two independent proportions, the minimum required sample size is calculated to be 578 participants in each group, with a significance level (α) of 0.05 and a statistical power (1-β) of 0.90. To account for potential loss to follow-up, assuming a 10% loss to follow-up rate, each group will include 650 participants, with a total sample size of 1,300.

A randomized controlled design will be employed across four representative regions in China, selected to reflect different levels of economic development and geography: a municipality, a provincial capital, an ordinary city, and a rural county. In each region, 2-4 community health centers (CHCs) will be selected, and eligible adults aged 18-64 years with one or more chronic diseases will be recruited in CHCs, where all eligible attendees will be invited to participate. Eligible patients with chronic diseases who provide written informed consent will be randomly assigned in a 1:1 ratio to either the intervention group or the control group. The randomization sequence will be generated in advance by a statistician using a computer-generated random number method. At each study site, participants will be allocated sequentially according to their order of enrollment. Due to the nature of the intervention, blinding of participants and on-site intervention implementers is not feasible. Local coordinators, outcome assessors, data managers, and statistical analysts will remain blinded to group allocation throughout the study to minimize assessment and analysis bias.

Building on the evaluation of intervention effectiveness, we will conduct a health economic evaluation to assess the economic value of the influenza AI-assisted primary care intervention. Costs will be estimated using a micro-costing approach from both health-system and societal perspectives, covering the costs of developing, deploying, and maintaining the AI assistant as well as the primary care physicians' labor-time costs; incremental cost-effectiveness ratios (ICERs) will be calculated as the additional cost per additional person vaccinated, per percentage-point increase in vaccination uptake, and per quality-adjusted life-year (QALY) gained, complemented by sensitivity and budget-impact analyses. In addition, a feasibility evaluation combining quantitative and qualitative methods will assess the real-world feasibility and scalability of the influenza AI-assisted primary care intervention. Quantitatively, a questionnaire survey will assess chronic-disease patients' experience with and acceptability of the AI assistant and the physician recommendation. Qualitatively, semi-structured interviews will be conducted with CDC staffs, primary care physicians, and patients with chronic diseases to explore the reasons underlying vaccination decisions, the acceptability of the AI-assisted intervention, and the role of primary care physicians in promoting vaccination.

Statistical Analysis

1. Primary Outcome Analysis An intention-to-treat (ITT) approach will be used as the primary analysis strategy, with all randomly assigned participants analyzed according to their original group allocation. Between-group comparisons will be performed using t-tests for continuous variables and chi-square tests for categorical variables. The primary outcome is influenza vaccination uptake among participants within one and three months after the intervention. A log-binomial regression model will be used to estimate the relative risk (RR) and its 95% confidence interval (95% CI), with CHCs included as a fixed effect and adjustment for prespecified covariates. If the log-binomial model fails to converge, a modified Poisson regression model will be used instead. The adjusted absolute risk difference and its 95% CI will also be estimated from the regression model to evaluate the public health significance of the intervention. Prespecified covariates will include region, sex, age, type of chronic disease, previous influenza vaccination history, education level, occupation, type of health insurance, and income. Prespecified subgroup analyses will be conducted by including subgroup variables and their interaction terms with the intervention group in the primary analysis model to assess heterogeneity of intervention effects across participant characteristics. Prespecified subgroups will include region, sex, age group (18-54 vs. 55-64 years), type of chronic disease, previous influenza vaccination history, education level, occupation, type of health insurance, and income.
2. Secondary Outcomes Analysis Secondary outcomes include time to influenza vaccination uptake among participants within three months following the intervention; influenza risk perception; influenza vaccine literacy; vaccine confidence; and intention to receive influenza vaccination. Time-to-event analyses for time to influenza vaccination uptake will be conducted using Kaplan-Meier methods and compared between groups with the log-rank test. Cox proportional hazards models will be used to estimate hazard ratios (HRs) with 95% confidence intervals (CIs). Influenza risk perception, vaccine confidence, and intention to receive influenza vaccination will be analyzed as binary outcomes using the same log-binomial or modified Poisson regression models as for the primary outcome, accounting for CHC effects. Results will be reported as relative risks (RRs) and 95% CIs. Influenza vaccine literacy will be analyzed as a continuous outcome using linear regression models, accounting for CHC effects. Intervention effects will be estimated after adjustment for prespecified covariates, with results reported as adjusted mean differences (aMDs) and 95% CIs. Prespecified subgroup analyses will be performed by including subgroup variables and their interaction terms with intervention assignment in the analysis models to assess heterogeneity of intervention effects across participant characteristics. Prespecified subgroups will include region, sex, age group (18-54 vs. 55-64 years), type of chronic disease, previous influenza vaccination history, education level, occupation, type of health insurance, and income.
3. Sensitivity Analysis A per-protocol (PP) analysis will be conducted as a sensitivity analysis to assess the robustness of the primary findings. The PP analysis will include intervention-group participants who complete the prescribed AI-assisted primary care intervention, as well as participants in the control group. In addition, within the intervention group, intervention engagement will be assessed on the basis of AI assistant usage and physician recommendation. Participants will be categorized into high- and low-engagement groups. Differences in intervention effects between these engagement levels will then be compared to assess the association between intervention intensity and influenza vaccination uptake.