Proportion ratios and 95% CIs of variables associated with place of death (home/hospice versus hospital) in England 1993–2010.

PRs were estimated from the log-binomial regression models. The clustering effect within the LSOA geographical units was adjusted using the general estimating equation (GEE) method. In additional to variables listed in the table, models were also adjusted for the calendar year of death. A PR greater than 1 indicates higher probability of death at home/hospice than the reference category. The p-value for overall association of individual factors with PoD was smaller than 0.001 in all models.