Page 82 - Effective healthcare cost containment policies Using the Netherlands as a case study - Niek W. Stadhouders

Chapter 4
would introduce selection bias. We defined patient groups based on gender, 5-year age category and disease group. A classification matrix was used to categorize DBC codes into 405 disease groups, based on 3-digit codes from the International Classification of Diseases, version 10 (ICD-10). With two gender groups, 21 age groups and 405 disease groups, 17.010 possible patient group combinations were defined. Of these, 11.079 contained claims. We aggregated claims data to patient group level to obtain total spending per patient group and the number of patients per patient group (step 3.1). Patients submitting claims in multiple ICD-groups feature in multiple patient groups. In total, 91% of total hospital spending was attributed to these patient groups. The remainder mainly constituted additional diagnostics and medication that could not be matched to individual DRGs.
Health-related questionnaires were collected annually from a representative sample of the Dutch population (CBS, 2010-2015). Health status of respondents above 50 years was routinely included in the questionnaires, which allowed us to construct EQ5D scores. Respondents could be divided into gender-based 5-year age groups and whether they visited a hospital during the year.
From health questionnaires to morbidity-related QALY loss
Morbidity-related QALY losses on a patient group level were constructed in four steps (1.1 to 1.4 in figure 4.1). In step 1.1, we matched health status questions to a validated QALY-measurement tool (Gheorghe et al., 2015) to obtain individual EQ-5D scores. In step 1.2, using a Dutch EQ-5D algorithm (Lamers et al., 2006), individual EQ-5D scores were transformed into individual QALY scores. In step 1.3, following Edney et al. (2018), we estimated changes in patient QALY scores over time, correcting for demographic trends (see appendix 4.1). Estimations resulted in a time trend in morbidity-related QALY loss per hospitalised patient by age group and gender. A pooled linear regression estimation rendered mean differences between patients and non-patients, which could be interpreted as the potential health gains the hospital sector could still achieve (see appendix 4.1). Combining the two estimates renders per patient group the mean number of QALYs lost due to illness and the mean change in patient QALYs over time. To incorporate uncertainty surrounding the estimates, we included this step in the Monte Carlo uncertainty analysis. In step 1.4, these outcomes were multiplied by the base number of patients in each patient group in 2012, rendering total patient group morbidity-related QALY loss.
We assumed that the average QALY scores of patients before they visit the hospital remained constant over time, and were not affected by exogenous increases in the health of the population (confounding by indication). If the population would get healthier due to factors outside the health sector, the chance of becoming a patient in a given year may
74