Target trial emulation of physical activity and cardiovascular disease risk: What is impact of the exposure assessment method?
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Target trial emulation (TTE) designs provide a framework for strengthening causal inference in observational research, but it is unknown how vulnerable they are to substantial error when the emulated intervention (exposure) measurement is imprecise. In physical activity epidemiology specifically, correcting for confounding via TTE designs but not addressing the large measurement errors arising from self-reports (e.g. questionnaires, which typically capture partial behavioural accounts with low or very low precision) creates uncertainty about possible dominance of type 2 error biases arising from such novel designs. High resolution wearables-based methods capture most movement, providing an assessment of physical activity behaviour with substantially less measurement error. No study has examined how measurement method influences causal inference in physical activity research.
Objectives
We applied TTE methodology to sub-samples of the UK Biobank cohort with repeat exposure measurements, to compare the effects of an emulated physical activity intervention on incident CVD risk, when the physical activity exposure was quantified using self-report vs. wearable devices.
Methods
We emulated a Randomised Controlled Trial, by identifying physically inactive adults (<150 moderate-to-vigorous physical activity (MVPA) mins/week) who had repeat assessments for both wearable and self-reported physical activity. At re-examination, participants were categorized into intervention (adopted ≥150 MVPA mins/week) or control (remained physically inactive) groups. Participants in each group were propensity score-matched to balance lifestyle behaviours, demographic, and health factors. Cumulative risk for CVD incidence was assessed through survival curves, hazard ratios, risk ratios, using Fine-Gray subdistribution and Poisson regression models.
Results
The wearables analytic sample included 490 participants (245 per arm; mean incident CVD follow-up 4.4 years), and the self-report sample included 11,302 participants (5,651 per arm; mean follow-up 6.3 years). In wearables assessments, guideline-adherent participants had markedly lower cumulative CVD risk (8.0% vs. 17.0%; hazard ratio [95%CI] = 0.59 [0.36, 0.98]; relative risk = 0.45 [0.28, 0.72]). In contrast, self-report assessments showed near-identical risk trajectories for intervention and control groups (21.6% vs. 21.2%; hazard ratio = 0.98 [0.89, 1.08]; relative risk = 0.92 [0.84, 1.00]). A series of sensitivity analyses confirmed these findings.
Conclusion
Reliance on self-reports of physical activity in TTE studies may obscure true emulated intervention effects due to non-differential misclassification, increasing considerably risk of Type II error. Exposure assessment using wearable devices may be essential for valid causal inference in TTE studies of physical activity and CVD risk. Future TTE studies should prioritise objective measurements to accurately inform public health policy and guidelines.
