This project investigates whether associations between accelerometer-measured behaviours and health outcomes are underestimated due to within-person variability and measurement error, and whether repeated measurements can be used to recover individuals’ habitual behavioural exposures and their true health effects.
The main objectives are to:
1.Quantify within-person variability and reliability of key accelerometer-derived behaviours using repeated measurements in UK Biobank, including estimation of intraclass correlation coefficients (ICC).
2.Derive estimates of habitual (long-term) behavioural exposure and correct for regression dilution bias using established calibration approaches.
3.Compare uncorrected and corrected associations between behavioural exposures and health outcomes to assess the impact of regression dilution on effect size estimation.
4.Examine associations between corrected behavioural exposures and major health domains, including cardiometabolic, mental and cognitive, sleep-related, aging-related and cancer-related outcomes.
5.Explore behavioural variability itself as a potential exposure associated with adverse health outcomes.
Accelerometer-based measures provide objective assessments of daily behaviours but are subject to substantial within-person variability when based on short-term recordings. This variability can attenuate exposure-outcome associations in epidemiological analyses. UK Biobank offers a unique opportunity to address this limitation, as a subset of participants has repeated accelerometer assessments over time. Leveraging these data allows correction for regression dilution bias and more accurate estimation of habitual behavioural exposures. Improving the validity of behavioural exposure assessment is essential for correctly quantifying health risks and informing public health relevance.
UK Biobank data will be accessed only by the named applicant Yuying Huang and supervising researcher Tong Xia.
