The purpose of this study is to assess the impact of using autonomous artificial intelligence (AI) system for identification of diabetic retinopathy (DR) and diabetic macular edema on productivity of retina specialists in Bangladesh. Globally, the number of people with diabetes mellitus is increasing. Diabetic retinopathy is a chronic, progressive complication of diabetes mellitus that affects the microvasculature of the retina, which if left untreated can potentially result in vision loss. Early detection and treatment of diabetic retinopathy can prevent potential blindness. Study Aim: To assess the impact of using autonomous artificial intelligence (AI) system for detection of diabetic retinopathy (DR) and diabetic macular edema on physician productivity in Bangladesh. Main study question: Will ophthalmologists with clinic days randomized to use autonomous AI DR detection for all persons with diabetes (diagnosed or un-diagnosed) visiting their clinic system have a greater number of examined patients with diabetes (by either AI or clinical exam), and a greater complexity of examined patients on a recognized grading scale, per physician working hour than those randomized not to have autonomous AI screening for their diabetes population? The investigators anticipate that this study will demonstrate an increase in physician productivity, supporting efficiency for both physicians and patients, while also addressing increased access for DR screening; ultimately, preventing vision loss amongst diabetic patients. The study has the potential to contribute to the evidence base on the benefits of AI for physicians and patients. Additionally, the study has the potential to demonstrate the benefits (and/or challenges) of implementing AI in resource-constrained settings, such as Bangladesh.
Bangladesh PRODUCTIVity in Eyecare (B-PRODUCTIVE) Trial Study Aim: To assess the impact of using autonomous artificial intelligence (AI) for identification of diabetic retinopathy (DR) and diabetic macular edema on productivity of retina specialists in Bangladesh. Hypothesis: Autonomous AI increases retina specialist productivity Main Study Question: Will retina specialists complete a greater number of diabetic eye exams per working hour (including persons reviewed by AI whom the retina specialist does not need to see personally) when they use autonomous AI in a randomized clinical trial? Design: Cluster-randomized (by clinic day) controlled trial. Randomization: By clinic day. Each morning the clinic manager will open an opaque envelope, which informs the manager if it is an Intervention (AI) or Control (non-AI) day. Interventions: All patients in both groups go through the eligibility checklist. If approved, they will be evaluated by autonomous AI. This is done to decrease potential bias (neither patients nor physicians know the group assignment of participants) and concealment (so that neither patients nor doctors can arrange visits on a known "Intervention Day"). Intervention Group: On randomly selected "Intervention" clinic days, if patients screen positive or have insufficient image quality, they continue to the ophthalmologist. If not eligible for autonomous AI, they proceed straight to the ophthalmologist without autonomous AI evaluation. If patients receive a negative result, they do not see the retina specialist, and are referred for a visit at the regular eye clinic (not the retina clinic) in 3 months. Control Group: On randomly-selected "Control Days," all patients see the ophthalmologist, irrespective of the results of autonomous AI evaluation. Masking: The retina doctors are masked both patient group assignment (that is, whether autonomous AI was used for pre-screening or not on the particular clinic day) and also masked to the results of the AI on Intervention days. Patients are also masked to group assignment and autonomous AI results.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
DIAGNOSTIC
Masking
DOUBLE
Enrollment
993
If patients receive a negative result they do not see the retina specialist
Deep Eye Care Foundation
Rangpur City, Bangladesh
Number of Completed Care Encounters Among Clinic Patients With Diabetes Per Retina Specialist Clinic Hour
Number of completed care encounters among clinic patients with diabetes per retina specialist clinic hour. Numerator is the number of care encounters among patients with diabetes (including persons evaluated by autonomous AI on Intervention Days who are determined not to need to see the retina specialist). The denominator is retina specialist clinic time in hours.
Time frame: 105 randomized clinic days
Number of Completed Care Encounters Among All Clinic Patients (With and Without Diabetes) Per Retina Specialist Clinic Hour
Number of completed care encounters among all clinic patients (with and without diabetes) per retina specialist clinic hour. Numerator is the number of completed care encounters (including persons evaluated by autonomous AI on Intervention Days who are determined not to need to see the retina specialist). The denominator is retina specialist clinic working time in hours.
Time frame: 105 randomized clinic days
Specialist Productivity Adjusted for Patient Complexity for Patients With Diabetes
Specialist productivity (care encounters / specialist clinic hour) adjusted for patient complexity for patients with diabetes. The complexity score for each patient participant was calculated by a masked United Kingdom National Health Service grader using the International Grading system, adapted from Wilkinson et al. International Clinical Diabetic Retinopathy and Diabetic Macular Edema Severity Scales (no DED = 0 points, mild non-proliferative DED = 0 points, moderate or severe non-proliferative DED = 1 point, proliferative DED = 3 points and diabetic macular edema = 2 points.) The patient participant complexity score was summed across both eyes. The average complexity score for each arm was calculated. Complexity adjusted specialist productivity was calculated for intervention and control arms by multiplying the respective overall productivity (care encounters per specialist clinic hour) calculation by the respective average complexity score.
Time frame: 105 randomized clinic days
Number of Participants Who Were Very Satisfied or Satisfied With Autonomous AI
After the patient participant completed the autonomous AI process, a survey with a 4-point Likert scale ("very satisfied," "satisfied," "dissatisfied," "very dissatisfied") was administered, concerning the participant's satisfaction with interactions with the healthcare team, time to receive examination results, and receiving their diagnosis from the autonomous AI system.
Time frame: 105 randomized clinic days
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.