Confounding and missing data in cost-effectiveness analysis: comparing different methods

Health Economics Review

Table 3 The observed and model-based estimates

The observed and model-based estimates of cumulative costs and effectiveness measured by the area under the curve (AUC).
^aThe observed group means, and their standard deviations (SD) and group differences are not adjusted for confounders. 95% confidence intervals (CI) and tests were based on the t-test.
^bThe bootstrapped group means and differences, and their standard errors (SE) are based on multiply imputed and bootstrapped data but unadjusted for confounding.
^cThe adjusted group means and their differences are based on regression modelling. The rows marked adjusted mean and SE are based on predictive margins, frequentist inference, bootstrap and multiple imputation.
^dThe rows denoted by predictive means and SD are based on the hierarchical Bayesian model, posterior predictive group means and their differences, and the corresponding SDs and 95% credible intervals (CrI).
^eNA corresponds to not applicable results.
^fP-values correspond to the null hypothesis “no difference between groups” or “ICER equals zero”.
^gP-values are generally not sensible in Bayesian inference, thus they are not reported.

ISSN: 2191-1991