Health inequities are significantly exacerbated by a lack of access to essential healthcare services. While expanding the healthcare workforce is one solution, increasing efficiency offers a more immediate and resource-effective approach. This study addresses the concerning decline in clinic productivity observed in the United States and other countries. The researchers hypothesize that autonomous artificial intelligence (AI), capable of making independent medical decisions, can significantly improve clinic productivity. Recent FDA authorization and Medicare/Medicaid reimbursement of such AI systems provide a foundation for this investigation. This study aims to rigorously test this hypothesis using a preregistered, randomized controlled trial.

The introduction cites several studies highlighting the global challenge of healthcare access and the growing gap between overall economic productivity and the declining productivity within the healthcare sector. The authors refer to research showing declines in clinic productivity in the US and other countries, suggesting that this productivity gap contributes to rising healthcare costs. They posit that autonomous AI, now approved by the FDA for certain medical applications and reimbursable by major insurance providers, may offer a solution to this problem, but there has been a lack of real-world evidence.

The B-PRODUCTIVE study was a preregistered, prospective, double-masked, cluster-randomized clinical trial conducted in Bangladesh. The study utilized a mathematical model based on rational queueing theory to guide its design and ensure unbiased estimation of healthcare productivity. Clinic days served as clusters, and randomization was concealed until the day of the clinic. The study used the LumineticsCore (formerly IDx-DR) autonomous AI system, which had previously demonstrated safety, efficacy, and lack of bias in clinical trials. The intervention group utilized the AI system for diagnosis, while the control group did not. Both physician and patient participants provided informed consent. The primary outcome measure was clinic productivity for diabetic patients, calculated as completed care encounters per hour per specialist. Secondary outcomes included overall clinic productivity, complexity-adjusted specialist productivity, and patient and provider satisfaction. Statistical analyses included Student's t-tests and linear regression modeling with generalized estimating equations to account for clustering.

The study confirmed the primary hypothesis: the use of autonomous AI significantly increased clinic productivity. For diabetic patients, the intervention group showed a 40% increase in productivity (1.59 encounters/hour vs. 1.14 encounters/hour in the control group, *p* < 0.001). The secondary outcome, overall productivity, also showed a significant increase (4.05 encounters/hour in the intervention group vs. 3.36 encounters/hour in the control group, *p* = 0.004). When adjusted for patient complexity, autonomous AI increased specialist productivity by a factor of 2.65. Patient and provider satisfaction were high in both groups. The number of DED treatments scheduled per day did not differ significantly between groups. However, the average complexity of patients seen by specialists was significantly higher in the intervention group, indicating that the AI was effectively identifying and triaging patients, allowing specialists to focus on more complex cases. The autonomous AI system demonstrated high sensitivity (93.9%) and specificity (84.0%) compared to human graders.

The B-PRODUCTIVE trial provides robust real-world evidence demonstrating the significant potential of autonomous AI to improve healthcare system productivity. The findings directly address the underappreciated issue of the healthcare productivity gap and its contribution to health inequities. The success of this intervention highlights a novel approach to increasing healthcare capacity, particularly relevant in resource-constrained settings. The results contrast with other cost-saving measures that may compromise quality of care. The high patient and provider satisfaction further supports the feasibility and acceptability of autonomous AI integration. The study's limitations, such as the single health system and focus on a specific disease, highlight the need for further research to validate these findings across diverse settings and conditions.

The B-PRODUCTIVE trial conclusively demonstrated that autonomous AI significantly improves clinic productivity, offering a viable strategy to address global health disparities. The 40% increase in productivity underscores the potential of AI to expand access to high-quality care, particularly in low- and middle-income countries. Future research should explore the generalizability of these findings to other healthcare systems, disease contexts, and AI technologies.

The study was conducted in a single health system in a low-income country with a limited number of specialist physicians and focused on a single disease (diabetic eye disease). The AI system was validated for patients without symptoms or a history of the disease, which might limit the generalizability of the results to other patient populations. The study design, while robust, might not be directly transferable to healthcare settings with scheduled appointments. The use of a single AI system also limits generalizability to other AI technologies.