ADLM urges federal regulators to maintain lab oversight of AI tools under existing CLIA standards, citing the need for continued validation.
Addressing Unique Failure Modes in Laboratory AI
While traditional software has historically supported functions such as verifying, interpreting, and reporting results, emerging AI-based applications introduce distinct safety challenges that are not adequately addressed by current Clinical Laboratory Improvement Amendments (CLIA) standards. Conventional software errors typically affect all patient cases matching specific programmed conditions, allowing for straightforward assessment using known outputs. In contrast, AI models can generate case-specific errors that are considerably more difficult to identify and resolve. Furthermore, generative AI tools have the potential to produce inaccurate or unsupported information, omit clinically important details, or exhibit altered behavior following updates to their underlying models, prompts, or knowledge bases.
Focusing on the Total Testing Process
Instead of establishing a separate regulatory framework for artificial intelligence as standalone software, the Association for Diagnostics & Laboratory Medicine (ADLM) emphasized that these technologies should be regulated within the existing total testing process framework of CLIA. The association advocates for targeted updates to CLIA that would reflect the unique challenges presented by machine learning, while also ensuring that clinical laboratorians maintain responsibility for the validation, implementation, and routine monitoring of all AI tools used in patient care.
