Page 85 - Effective healthcare cost containment policies Using the Netherlands as a case study - Niek W. Stadhouders

The marginal benefits of healthcare spending in the Netherlands
(OVB), others have attributed a fixed part of the gains to factors outside the health sector (Cutler and McClellan, 2001; Hall and Jones, 2004). We correct for OVB by using changes in the number of patients as proxy for health trends. The underlying assumption is that when health of a patient group improves, the number of patients that visit the hospital decreases. This assumption is violated if health trends change treatment intensity and outcomes while keeping patient numbers stable, which could happen in the case of waiting lists. However, waiting lists in the Netherlands only exist for a small number of patient groups and are relatively low (Siciliani et al., 2014). When hospitals respond to lower patient numbers by attracting new, healthier patients through supplier-induced demand, OVB may also remain. Other potential sources of endogeneity include time effects. Cost and outcome variables may be correlated to previous years’ values. We corrected for this using a fixed effects model (Wooldridge, 2010). Furthermore, health shocks may affect future spending, which may bias the estimators upwards. As robustness checks, we corrected for health shocks by including time dummies and lagged effects.
We used total QALY loss, corrected spending and the number of patients for each of the 11,000 patient groups as inputs for our empirical estimation strategy. For each patient group we define QALYs ( ) as an unknown function of corrected spending ( ) and number of treated patients (. ): (3)
We assumed diminishing marginal returns: , and assumed that the production function was differentiable at relevant intervals (Boisverf, 1982). We approach the unknown function at the mean by defining the second order Taylor polynomial:
4.2.4 Empirical strategy
(4)
Where α is the group specific productivity parameter, T is the trend in time t, the
coefficients are the cost elasticity parameters and the coefficients are the treatment elasticity parameters. contain fixed effects and is the error term. Evaluated at the mean, the translog function approximates the unknown production function. Using a translog function to estimate the marginal effect of spending is preferred over commonly used explicit specifications –such as the linear or Cobb-Douglas model- if the elasticity is nonlinear and the elasticity of substitution unknown (Boisverf, 1982; Pavelescu, 2011), which are both likely for the heterogeneous patient groups. The elasticity of spending for the mean patient group was obtained by the first derivative of ) with respect to log :
77