A recent study supported by Orbis International and published in Nature examined the use of autonomous artificial intelligence (AI) in enhancing the clinical productivity of healthcare professionals (HCPs)—specifically managing diabetic retinopathy (DR).
Let’s start with some background.
Investigators noted that lack of access to essential services is a major cause for health inequity across the U.S. and world, with the following demographics affected:
- Racial and ethnic minorities
- Low socioeconomic status persons
- Rural populations
With this in mind, “access can be improved by increasing overall capacity of the healthcare system,” they wrote.
Go on …
Two solutions were considered to resolve this global issue:
- Expand the healthcare workforce (which would require additional time and resources to train new workers)
- Increase efficiency to improve capacity
As such, researchers hypothesized that the use of AI could improve clinical productivity and increase efficiency
Has there been much research on autonomous AI in this capacity yet?
Not so much. In fact, the researchers noted a lack of real-world evidence surrounding autonomous AI and healthcare productivity.
And this led them to develop a clinical productivity model for a pre-registered, cluster-randomized clinical trial (RCT).
Give me some details.
The RCT was conducted at retina specialist clinics of the Deep Eye Care Foundation (DECF)—a community-based eye health care organization (HCO) in Bangladesh—located in Rangpur, Bangladesh, over a 5-month period in 2022.
A total of 2,109 patients (mean age: 50.9 years; 47% male) diagnosed with diabetes were randomized into one of two groups:
- Intervention (1,189 participants)
- Control (920 participants)
And those eligible for AI?
Out of this total, 993 participants were eligible for testing via an autonomous AI (LumineticsCore, Digital Diagnostics, Inc.) :
- Intervention (49.7%; 494 participants)
- Control (50.3%; 499 participants)
Note: LumineticsCore (formerly IDx-DR) is an AI-based diagnostic system for DR that has been cleared by the FDA to make a preliminary diagnosis* without the need for a clinician to interpret images or results.
*If DR is detected, the patient is referred to an eyecare professional (ECP).
How many clinic days were there?
A total of 105, broken down to 51 (intervention group) and 54 (control group), with an average of 54.5 clinic patients per day.
Break down these clinical visits.
In the control group, participants visited a retinal specialist regardless of autonomous AI results.
In the intervention group, however, participants did not see a specialist during their visit if the autonomous AI did not detect signs of DR—these patients were instructed to return in 12 months;
Overall, patients only saw a specialist if the autonomous AI detected DR and further treatment was required.
And how did the AI detect DR?
Per the study, via prognostic Early Treatment of Diabetic Retinopathy Scale (ETDRS) and Diabetic Retinopathy Clinical Research (DRCR) standards.
Further, the three retinal specialists who participated as investigators in this study were also validated by these standards.
Talk numbers.
In all, the data indicated a 40% increase in patients who completed a high-quality eye exam (per hour) due to the use of autonomous AI.
Additionally, an estimated 75% of participants in the intervention group completed the AI exam only.
What was the feedback from these clinicians?
The retina specialists reported that the autonomous AI enabled them to focus their clinical time on more complex cases requiring treatment or patients needing their expertise.Per the data, when calculating specialists’s productivity (Adjusted for complexity), AI increased overall productivity by a factor of 2.65.
Meaning…
That those patients with more straightforward clinical cases were able to be seen in a timely fashion—all because those patients using an AI exam didn’t need to wait to see a specialist.
And the patients’ feedback?
Overall, high satisfaction was noted among patients who underwent an AI exam.
Another reason this is critical: Long clinic wait times, particularly in geographical areas with limited resources, have historically been a major issue for those patients unable to take the time away from their daily responsibilities including work for an exam.
Any limitations?
The authors noted the following:
- Study conducted in a single health system
- Located in a low-income country
- Use of only three physicians (retina specialists)
- Use of an autonomous AI to diagnose a single disease in patients without symptoms or a history of DR
Lastly… significance?
According to DECF Executive Director Khairul Islam, MD, this study is a significant breakthrough in eyecare for low- and middle-income countries “where scarcity of skilled healthcare professionals and facilities are always a hindrance.”
“These challenges can be overcome through integrating autonomous AI, which increases the efficiency and productivity of our doctors and, thus, helps us reach more patients by providing them with timely eye care,” Dr. Islam stated.