Researchers released clinical trial data to persuade the FDA to make IDx-DR the first automated artificial intelligence (AI) diagnostic system to be sold in the United States.
The 900-patient trial found that IDx-DR surpassed pre-specified superior endpoints in sensitivity and specificity, suggesting that it can consistently detect diabetic retinopathy. IDx, the developer of the technology, sees diagnostics as a useful tool that enables primary care physicians to screen for diabetic retinopathy to detect and treat the disease faster.
Proponents of AI-based diagnostics support the technology as a solution to healthcare resource challenges. While some patients in cities have access to trained diagnostic specialists and the latest equipment, the availability of these resources is more complicated in rural communities. Likewise, the number of people who need access to city resources means that demand can outstrip supply, resulting in long wait times for diagnostic tests. Cities also face fragmented availability, with resources concentrated in specialty centers rather than primary care settings that people frequently visit.
One answer to the challenge may be to train, hire and pay more specialists, but given the ageing populations and constrained healthcare budgets in many countries, this may not be a scalable and sustainable solution.
AI-based diagnostics can address resource issues in a way that sustainably expands access to specialized testing. Rather than relying on humans to perform all the analysis, healthcare systems can use AI techniques trained to detect patterns indicative of disease to diagnose or classify patients. This will allow the healthcare system to handle more patients with fewer employees.
The concept took a step forward in April, when the FDA approved IDx-DR, cloud-based software that can detect mild or more severe cases of diabetic retinopathy in pictures taken by retina cameras. When the software makes a diagnosis, the patient is sent to an ophthalmologist for evaluation and treatment.
The Centers for Disease Control and Prevention estimates that between 12,000 and 24,000 people are blinded by diabetic retinopathy each year. However, only about half of people with diabetes visit an eye doctor each year.
IDx has now published approved data. Writing in Nature Digital Medicine, the researchers explain how the trial compared IDx-DR analyses conducted in primary care clinics with findings from photographers certified by the Wisconsin Foundation Photo Reading Center. The trial linked IDx-DR with a specificity of 87.2% and a sensitivity of 90.7%, suggesting that false-positive and negative rates may be low enough to make the test valuable to healthcare systems.
The researchers also reported that images of sufficient quality to be analyzed by AI were available in 96.1% of cases. However, multiple imaging attempts and dilation drops are required to capture high-quality images of some people. The researchers acknowledge that “selective expansion may present challenges to scalable clinical implementation in some cases.” IDx needs to convince primary care sites that the benefits of its system outweigh the costs, while alleviating concerns about the use of novel automated technologies. The success of these efforts (or others) will provide an early indication of physician and patient acceptance of AI testing.