Statistical regression using distributed lag non-linear model is fully described in Gasparrini, Armstrong and Kenward, 2010, and Vicedo-Cabrera, Sera and Gasparrini, 2019.
Computation details for Temperature-attributable mortality (and hospital admission) time series, UK (1900-2099) dataset.
Model setup: natural cubic splines in all dimensions, 3 log-spaced knots in lag dimension, 8 degrees of freedom per year in long-term trend, confounding by day of week.

- Mortality: temperature knots at 0.1, 0.75, 0.9 quantiles, 21 lag days.

- Hospital admission: temperature knots at 0.4, 0.9 quantiles, 28 lag days.