Why context matters: clinical validation of digital measures for use in Parkinson's disease

The clinical validation of digital measures for Parkinsons disease (PD) must be tailored to their specific purpose, or context of use (CoU). However, validation work often assumes that the same measures will work across contexts. We analyzed data from the ObjectivePD substudy of the mPower study, involving 24 people with PD (PwPD) and 22 healthy controls (HCs). We evaluated 31 digital measures from five smartphone-based assessments (gait, tapping, and three types of tremor) for test-retest reliability, criterion validity against the MDS-UPDRS, and performance across three CoUs: differentiating PwPD from HCs, detecting effect of symptomatic treatments, and tracking disease progression. Of the 31 measures, 22 were found to be reliable (Intraclass Correlation Coefficient >0.5), and many correlated significantly with MDS-UPDRS items, such as postural tremor root mean squared (RMS) acceleration ({tau} = 0.51) and finger tapping speed ({tau} = -0.51). Performance was context-specific, however: certain measures excelled at differentiating PwPD from HCs, others were highly sensitive to medication effects, and a different set tracked long-term changes. The validity of a digital measure is not universal but depends on its intended application. This context-driven approach is essential for establishing that a measure is fit-for-purpose and for qualifying its use in future PD clinical trials.