4.4. Extra cost at death

The estimation approach for cost of death differed in the three countries. In the French case, extra cost of death was estimated using a similar two-part modelling approach, as the mean marginal difference of the predicted outcome (total health expenditure in 2014) with a death dummy (corresponding to dying in the first semester of 2015) switched on or off. As data on costs accumulated throughout 2014 was available, and the information on whether a person died was provided for 2015 as well, the difference was estimated for people who died in the first semester of 2015, compared to the people who stayed alive in the same period. Including people who died in the second semester of 2015 in the analysis was ruled out because the date of death for these individuals was too far from 2014, and therefore likely that the extra cost of death would be underestimated.

We conducted this analysis separately for two samples: with at least one model-defined NCD, and without any NCDs. These costs were then added on top of the other healthcare expenditures. To avoid potential double-counting of death-related costs, the main cost analysis was restricted to the people who were still alive on 31 December 2014, and therefore the proportion of those who died in the first semester of 2015 was small in relation to the total. In Estonia, and the Netherlands, it was impossible to estimate extra deaths costs using this approach due to data limitations, therefore for all countries the extra costs of death were extrapolated from the French data using the approach described above.