GenAI-Assisted Rheumatoid Arthritis Bone Erosion Risk Assessment and DMARD Selection (RABER)

This arm will include at least 100 consecutive adult inpatients with rheumatoid arthritis (RA) receiving routine clinical care. For each patient, the treating physician will first independently assess bone erosion risk and select a disease-modifying antirheumatic drug (DMARD) treatment strategy based on available clinical information (pre-AI assessment). The physician will then review a GenAI-generated clinical decision support report containing predicted bone erosion risk levels and DMARD recommendations and record an updated risk assessment and treatment decision (post-AI assessment). Pre-AI and post-AI physician decisions will be compared against independent expert reference standards to evaluate the impact of GenAI-assisted decision support. Treating physicians retain full responsibility for final clinical decisions.

The intervention consists of a GenAI-generated clinical decision support report provided to treating physicians during routine RA care. Using patient clinical information available at the time of assessment, the GenAI system generates predicted bone erosion risk probabilities (0-100%), risk levels (0-3), and DMARD treatment recommendations. Physicians review the report after completing their independent pre-AI assessments and may revise their risk assessments and treatment decisions based on the additional information. The GenAI system provides decision support only and does not autonomously make or implement treatment decisions.

The percentage of participants for whom the treating physician changes the rheumatoid arthritis bone erosion risk level after reviewing the GenAI-generated clinical decision support report, compared with the initial assessment before reviewing the report. Risk levels are categorized from 0 to 3. A change is defined as any difference between the pre-AI and post-AI risk levels. The percentage is calculated using participants with complete paired assessments.

Agreement between physician-assessed rheumatoid arthritis bone erosion risk levels (0-3) and the reference standard will be evaluated separately before and after review of the GenAI report using Cohen's kappa coefficient. The pre-AI and post-AI kappa coefficients will be compared to assess changes in agreement following GenAI-assisted decision support. Cohen's kappa ranges from -1 to 1, with higher values indicating greater agreement beyond chance. The difference between post-AI and pre-AI kappa coefficients will be reported, with positive values indicating improved agreement. The reference classification will be established during the data analysis stage by three independent expert physicians who are not involved in patient treatment, with disagreements resolved through consensus.

Agreement between physician-selected disease-modifying antirheumatic drug (DMARD) treatment categories (0-4) and the reference standard will be assessed after physicians review the GenAI-generated clinical decision support report. Agreement will be quantified using Cohen's kappa coefficient, with higher values indicating greater agreement beyond chance. The analysis will include participants with complete post-AI physician treatment selections and corresponding reference standard classifications. The reference classification will be established during the data analysis stage by three independent expert physicians who are not involved in patient treatment, with disagreements resolved through consensus.