An analysis pipeline to provide personalised time-use suggestions, via group LASSO regression on the UK Biobank data, for optimising various cognitive outcomes.

Please see the below papers for more details and context:

• "'Small Steps' towards improving 24-hour time-use behaviours to decrease the risk of dementia: protocol for a personalised, web-based randomised controlled trial in community- dwelling older adults". Mellow et al. (2025). Submitted/under review (link to preprint v1 at medRxiv).
• "An interactive tool to personalise 24-hour activity, sitting and sleep prescription for optimal health outcomes". Mellow et al. (2025). Under review (link is to preprint v1 available at Research Square).

This repository provides the analysis code for

• preparing, cleaning anbd processing the raw UK Biobank data
• fitting group LASSO models for each cognitive outcome with optimisation of the regularisation (lambda) parameter via 10-fold cross-validation (mean square prediction error)
• group LASSO diagnostics
• 4-simplex grid creation and ellipsoid fence constraint calculations (with interactive plots of 4-simplex time-use compositional data [see below])
• machinery to create predictions from new data inputs (time-use and covariates/other person specific predictors)
• optimisation calculations of LASSO model cognitive outcome predictions over ellipsoid fence constrained time-use grids and personal predictors

The process is summarised in the below flowchart:

Please note the user input elements in the above flowchart are facilitated via a Shiny app available as another repository: tystan/ideal-day.

(Screenshots of the plotly/html interactive figures)