Estimating cost-effectiveness from claims and registry data with measured and unmeasured confounders