Positive predictive value metrics for multicancer detection tests

Area of Science:

• Oncology
• Medical Diagnostics
• Biostatistics

Background:

• Multicancer detection (MCD) tests screen for multiple cancer types, often providing a predicted cancer signal origin (CSO).
• Positive predictive value (PPV) is a standard metric for single-cancer screening but requires adaptation for MCD tests.

Purpose of the Study:

• To define and evaluate various PPV metrics for MCD tests.
• To create a prototypical MCD test to assess PPV properties.
• To analyze the impact of nontargeted site sensitivity on PPV.

Main Methods:

• Defined PPV metrics including PPV_ALL, PPV_T, and PPV_CSO based on different numerator/denominator populations.
• Utilized case-control study data and population incidence rates to model MCD test performance (sensitivity, specificity, CSO accuracy).
• Calculated PPVs for a prototypical MCD test across a range of sensitivities for nontargeted cancer sites.

Main Results:

• PPV_ALL increased, while PPV_T decreased with higher sensitivity for nontargeted cancer sites.
• Site-specific PPVs were more dependent on CSO accuracy than on cancer site prevalence.
• The choice of PPV definition significantly influences the interpretation of MCD test performance.

Conclusions:

• Multiple PPV definitions are applicable to MCD tests, each with unique implications.
• The clinical utility of different PPVs depends on the specific clinical scenario and intended use.
• Clear communication between MCD test developers and clinicians is essential for selecting and reporting appropriate PPV metrics.