[Methodological requirements for center comparisons (profiling) in healthcare]

Quantitative comparisons of health institutions (e.g. hospitals, rehabilitation centers, or disease management programs) have become increasingly common in Germany, focused at assuring and improving the quality of the healthcare system. For evaluation, frequently only raw center means of one specific outcome are calculated and ranked. This evaluation method is insufficient because it does not reflect the multidimensional profiles of the institutions or their development in the time domain. Also, the observed means are not corrected for random effects and thus suggest larger differences among institutions than actually exist. Additionally, raw means do not allow fair comparisons since they do not take into account baseline differences among institutions. Instead of presenting raw means, multilevel statistical models with institutions as random effects and carefully selected potential confounders as fixed effects should be applied and used for the estimation of center effects. For presentation of the results, profiles plots and control charts should be preferred to the widely used rankings and league tables.