Individual-level decision-analysis and value-of-information for uncertainty targets

Investigate individual-level decision-analysis frameworks and value-of-information analyses to guide the selection of acceptable widths for uncertainty intervals and acceptable classification instability when using risk thresholds in the development and use of clinical prediction models that estimate individual-level risk for binary outcomes.

Background

The paper proposes an uncertainty-based sample size approach for developing clinical prediction models for binary outcomes, focusing on precision of individual-level risk estimates and classification stability relative to decision thresholds. The authors emphasize involving stakeholders to define acceptable uncertainty in predictions and suggest framing these choices within a decision-theory perspective.

They note that beyond their quantitative framework, additional research is needed to solidify an individual-level decision-analytic foundation and value-of-information considerations to determine when uncertainty is acceptable and when further data collection is warranted.