4.7.3. Issues specific to the Netherlands

Since the 2006 healthcare reform, Dutch citizens are covered by mandatory private insurance while public insurance no longer exists. The government continues to play a regulatory role and subsidises premiums among the low-income population.

The cost of illness (COI) analysis was performed on 2013 reimbursement cost database of the main private insurer named Vektis who is responsible for collecting discharge data from every Dutch insurer. This equates to 98% of cost coverage and 95% of the population. Remaining costs are sourced from OOP deductible: this cost component was corrected at the individual level according to age and gender (see [Thi{\'{e}}baut, 2017 [59]]). This sample contained 13,4 million individuals aged 18 and over. People in nursing homes are not included in this sample.

Diseases were identified at an individual level according to 4 digit Anatomical Therapeutic Chemical (ATC) codes recorded in 2013 for each insured Dutch citizen who had been prescribed medicines over the year. If medicines were delivered at hospitals and nursing homes then the ATC code was not recorded. This leads to disease identification difficulties for some illnesses, namely Chronic Kidney Disease (CKD), Cirrhosis and Cancer. On the other hand stroke, heart disease, COPD, depression and neurologic disorders were identified by computing a likelihood score based on a weighted sum of ATC codes. The weights were estimated via a random forest analysis and provided by our partners at National Institute for Public Health and the Environment (RIVM) (see forthcoming working paper by Lany Slobbe and Marc Koopmanschap for more details).

Statistical analysis performed for COI estimates for the Netherlands follows the methods described in reports by [Thi{\'{e}}baut, 2017 [59]] and [Cortaredona and Ventelous, 2017 [13]]. However some adjustments were made according to the nature of the data. For more information, please refer to the report [Thi{\'{e}}baut, 2017 [59]].

Because of the data limitation, it was impossible to reliably estimate the probability of using healthcare (1st stage model) for age categories older than 80 years. For this reason, we extrapolated these probability values from categories 75-79 to older age groups. In addition, as mentioned above, it was not possible to reliably define the diagnosis for Cirrhosis, Cancer and CKD. Therefore, such costs were used from France. In addition, because of the nature of the microsimulation model, the extra costs of comorbidity from France were used. While the Dutch insurance system is mostly private-based and the French system is not, in both countries the population coverage is comprehensive, OOP payment share of total health expenditures is relatively small, the governments play an important regulatory role, and the organisation of care provision is quite similar. Therefore, this extrapolation was deemed appropriate.