Adaptive techniques could improve diagnostic effectiveness in five key disease areas, a Government Accountability Office report states, but only if the data is high quality.
Low-quality data is hampering artificial intelligence (AI) and machine learning (ML) from making more inroads in healthcare diagnostics, according to a new report from the US Government Accountability Office (GAO).
In addition, the report found, these technologies are yet to fully demonstrate real-world performance in diverse clinical settings.
"Our policy options--like improving data access and collaboration--may help address the challenges," the report stated.
Potential benefits of machine learning in the diagnostic process include earlier detection of diseases, more consistent analysis of medical data, and increased access to care, particularly among underserved populations, the report said.
The GAO identified a variety of ML-based technologies for five selected diseases: certain cancers, diabetic retinopathy, Alzheimer's disease, heart disease, and COVID-19. Most rely on imaging data such as x-rays or magnetic resonance imaging (MRI), but the report noted that these technologies have yet to be widely adopted.
Three broader approaches could assist these diagnoses: autonomous, adaptive, and consumer-oriented ML diagnostics.
According to the GAO, relying upon information supplied by the US Food and Drug Administration (FDA), which oversees use of these algorithms in diagnoses, incorporating additional data during the machine learning process (the adaptive approach) may improve accuracy, but only if the data being automatically updated is of high quality. Barring that, these processes could cause algorithms to perform poorly or inconsistently.
Diagnostic errors affect more than 12 million Americans each year, with aggregate costs likely in excess of $100 billion, the GAO said, citing a report by the Society to Improve Diagnosis in Medicine.
The report recommends that policymakers promote collaboration among technology developers, providers, and regulators when developing or adopting machine learning diagnostic technologies. This collaboration could expedite the creation of ML-ready data, according to officials at the National Institutes of Health interviewed by the GAO.
Providers should consider setting aside time for their employees to engage in these innovation activities, the GAO said.
Scott Mace is a contributing writer for HealthLeaders.