Page 84 - Effective healthcare cost containment policies Using the Netherlands as a case study - Niek W. Stadhouders

Chapter 4
Reductions in mortality may lower health spending due to fewer mortality-related costs, while increased spending can reduce mortality. Due to this reverse causality, straightforward estimation would result in underestimation of the true effect of extra spending on outcomes, i.e. an upward biased threshold. Because last year of life costs are known in the Netherlands, we were able to correct for the cost resulting from changes in mortality directly, and isolate the effect of changes in spending on changes in mortality. Although a strong and valid instrument is preferred to correct for endogeneity, direct correction may be a good alternative when no valid IVs are at hand (Moreno-Serra and Smith, 2015). To this aim, we split the bi-directional causality by disaggregating spending
4.2.2 Correcting for reverse causality
into last year of life costs that resulted from mortality, and the costs that do not
result from mortality, which we call corrected spending ( ):
(1)
By construction, exogenous changes in mortality only influence , allowing estimation of the effect of changes in corrected spending ( ) on mortality. In order for equation (1) to
hold, should be independent of changes in mortality.
(2)
If lower mortality changed the average LYoL-costs, for example if predominantly high-cost deaths were averted, the estimate would be biased downward, while if mostly low-cost death were averted, the effect would be biased upward. Moreover, this correction may be incomplete: if LYoL-costs would increase over time, equation (1) insufficiently corrects for
reverse causality, biasing the estimated thresholds upwards.
For the Netherlands, mean LYoL-costs were known for age groups and gender (van
Baal et al., 2011). In step 4.1, we multiplied the LYoL-costs by the number of deaths for each of the 11,000 patient groups to obtain the total amount of spending as a result of mortality . Due to the uncertainty surrounding LYoL-costs, Monte Carlo uncertainty analysis was used. In step 4.2, we subtracted the LYoL-costs from total spending to obtain corrected spending ( ).
Underlying health status may influence both spending and health outcomes: if fewer patients get ill and die from a disease, for example due to healthier lifestyle, costs for the patient group may be lower, and fewer QALYs are lost due to both mortality and morbidity. Straightforward estimation would erroneously attribute the health gains to the hospital sector, causing thresholds to be biased downward. To correct for omitted variable bias
76
4.2.3 Correcting for omitted variable bias