The association of wearable sensor measures of time use with cardiovascular disease

Keywords: Physical activity- Accelerometer- Cardiovascular disease- Statistical methods- Big Data- Genomics
Background
Cardiovascular disease (CVD) is the leading cause of mortality, globally and in the UK1. CVD prevention is therefore a public health priority. Physical inactivity is associated with adverse health outcomes (e.g. higher all-cause mortality, CVD rates and Type 2 diabetes rates)2; moderate-to-vigorous intensity physical activity seems to reduce CVD risk3,4. However, evidence is limited by the use of self-reported questionnaire data which is crude and prone to measurement error. Therefore, more research is needed to understand quantitatively the change in risk conferred by different physical activity (PA) profiles, including the role of light intensity PA.
Measurement of PA using accelerometers, which are devices worn on the wrist, produces vast amounts of data that can be challenging to analyse as it is compositional. Statistical methods developed in different contexts may be helpful e.g. compositional data analysis and isotemporal substitution10-12. Accelerometer data also comes as a time series: functional data analysis and ARIMA modelling are candidate methods to capture its particular structure13. There is limited literature applying these methods to PA data, so research and development is needed to understand which methods are most useful in creating PA profiles for epidemiology.
Until recently, there was little understanding of genetic influences on PA behaviours8. Understanding genetic influences on PA behaviours allows better understanding of its underlying biology and causal inference of its potential association with CVD outcomes.
Aims
This project aims to improve understanding of PA and its association with CVD, using objectively measured PA data and genetic epidemiology. The objectives are:
1. To identify and develop methods to characterise activity profiles from time-series device data.
2. To perform epidemiological investigations into the association between PA patterns and risk of CVD outcomes.
3. To perform genome-wide association studies (GWAS) to identify genetic variants associated with PA profiles and to apply Mendelian Randomization to assess causal relevance of PA profiles with CVD outcomes.
Data and Data Preparation
UK Biobank is a prospective study of 500,000 individuals, aged 40-69 at recruitment5. Participants have given various health-related measurements and have been genotyped6. Linking with health registries allows participants' health outcomes to be tracked. Over 100,000 participants wore wrist-worn accelerometers for a week7,9.
Methods
A review of statistical methods potentially relevant to PA data analysis will be performed. Methods developed for analysis of compositional and/or time series data in different fields will be assessed and developed for application in this epidemiological context.
Prospective epidemiological analysis of PA profiles with CVD outcomes will be carried out. Cox regression analyses will be conducted in order to assess the association of the newly derived PA metrics with ischaemic heart disease and stroke with adjustment for age, sex, income, education, alcohol, smoking and sedentary behaviour in the first instance.
A genome-wide association study (GWAS) will be performed on the newly derived PA metrics. The goal is to apply genomic understanding in analysing the potential causal role of PA in disease aetiology via Mendelian Randomisation.
References
1. Murray CJL et al. Lancet. 2012;380(9859):2197-2223.
2. Lee I-Min et al. Lancet. 2012;380(9838):219-229.
3. Bennett DA et al. JAMA Cardiol. 2017:1-10.
4. Stewart J et al. JRSM Cardiovasc Dis. 2017;6:204800401668721.
5. Sudlow C et al. PLoS Medicine. 2015: 12(3), p.e1001779.
6. Bycroft C et al. Nature. 2018: 562(7726), p.203.
7. Doherty A et al. PloS one. 2017: 12(2), p.e0169649.
8. Bauman AE et al. Lancet. 2012;380(12):258-271.
9. Willetts M et al. bioRxiv. 2017:187625.
10. Ch