FDA outlines considerations for digitally derived measures in clinical investigations

The US Food and Drug Administration has published a white paper on the development and use of digitally derived measures in clinical investigations, covering patient relevance, evidence requirements, validation, usability, and potential sources of measurement error.

The Baseline

• A new white paper from the US FDA outlines the considerations for developing and using digitally derived measures in clinical investigations.
• The paper emphasizes patient relevance, context of use, and appropriate verification and validation when determining whether digitally derived measures are fit-for-purpose.
• It also addresses risk-based evidence requirements, usability, and potential sources of measurement error, including considerations for measures incorporating AI.

The US Food and Drug Administration (FDA) has published a new white paper bringing together considerations from existing agency guidance on digitally derived measures (DDMs) used as outcomes in clinical investigations. Jointly developed by the agency’s drug, biologics and device centers and its Oncology Center of Excellence, the paper outlines considerations for determining whether a DDM is appropriate for its intended context of use.

DDMs are measures derived from data collected using digital health technologies (DHTs), including those enabled by AI. They may be used as clinical outcome assessments, biomarkers, or components of multicomponent endpoints derived from multimodal data. According to the FDA, continuous data collection outside healthcare settings may help detect changes in health that are not evident during periodic clinical visits, and support earlier detection of treatment effects or safety signals. DDMs may also support the implementation of decentralized trials.

Commenting on the publication, Anindita Saha of the FDA’s CDRH and CDER emphasized the importance of the intended measurement and its supporting evidence:

“The value of a DDM isn't about the sophistication of the technology producing it. It comes from whether the measure is clinically relevant, meaningful to patients, and supported by evidence appropriate to its context of use.”

The white paper recommends beginning by identifying the meaningful aspect of health to be measured and defining the concept of interest, target population, and context of use. The FDA “encourages the use of clinical outcomes that are both clinically relevant and capture what is meaningful to patients,” defining a meaningful aspect of health as “any specific aspect of feeling or functioning in daily life that is important to patients.” Alongside clinical expertise, qualitative data from patients and caregivers “play an essential role” in identifying and describing these aspects. Developers should also consider the potential value and burden of using a DHT in the target population.

Once a DDM has been selected, the FDA recommends establishing a justification linking it to the relevant aspect of health and identifying the evidence needed to support its relevance and validity. A DHT used to generate the measure should undergo verification and validation to establish fitness-for-purpose. Verification assesses whether underlying parameters are measured accurately and precisely; analytical validation determines whether the DHT appropriately measures the relevant clinical event or characteristic; and clinical validation establishes whether the DDM measures its intended concept, reflects a meaningful aspect of health, and tracks changes in disease or clinical status.

The FDA applies a risk-based approach, with supporting evidence tailored to the DDM’s intended use. A DDM used as a primary endpoint in a pivotal study is expected to undergo prospective validation with prespecified performance thresholds in the intended population, whereas an exploratory endpoint may rely on retrospective or bridging evidence. For DDMs incorporating AI algorithms, developers may also consider a credibility assessment proportionate to model risk and context of use.

The paper also addresses usability and potential sources of error, including sensor noise, incorrect wearable placement, environmental conditions, missing or variable-quality data, activity misclassification, and software or algorithm updates. For DHTs intended for prolonged or repeated use, the FDA recommends evaluating adherence, user burden, sustained usability, and potential user fatigue.