Analysis Method of Real-World Digital Biomarkers for Clinical Impact in Cancer Patients

Area of Science:

• Digital Health
• Wearable Technology
• Clinical Trials

Background:

• Consumer wearables offer promise for monitoring patient wellness in decentralized clinical trials.
• Handling and interpreting data from wearables present challenges, necessitating statistical and data science expertise.
• Gaps in wearable data require robust methods for accurate health outcome prediction.

Purpose of the Study:

• To develop and validate methods for estimating daily steps from Apple Watch data with data gaps.
• To analyze the association between estimated daily step counts and clinical events in cancer patients.
• To identify digital biomarkers of physical activity for predicting clinical outcomes.

Main Methods:

• Utilized data from 50 cancer patients using the DigiBioMarC app and Apple Watch over 28 days.
• Developed three methods to estimate daily steps, addressing various data gap types based on duration and context.
• Employed Cox proportional hazards regression, decision tree modeling, and participant clustering to analyze associations.

Main Results:

• A method accounting for data gap duration and context provided the most reliable daily step estimates.
• Daily step counts on sufficient measurement days predicted time to first clinical event (p=0.068).
• Higher daily step counts correlated with lower hazard probabilities for mortality and clinical events, with 83.3% accuracy in prediction.

Conclusions:

• An effective method was developed to estimate daily steps from consumer wearable data with unknown gaps.
• Daily step counts from sufficient sampling days are strong predictors of clinical event timing and occurrence.
• Increased daily step counts in cancer patients are associated with a reduced hazard of death or clinical events.