[Comparison of quality by means of funnel plots--a plea for a uniform methodology].

Vertical bar charts depicting unit event rates sorted in ascending order enjoy widespread usage in external quality assurance. Unfortunately they suggest a spurious ranking resulting from instability in the percentile distribution chiefly caused by varying denominators. The popular remedy of simply excluding units below a minimum threshold would solve the problem only partially since units with few operations per annum may evade evaluation altogether merely by the grace of their size. Compared with alternative solutions reviewed in this article Spiegelhalter's funnel plots exhibit clear advantages over statistical control charts or Bayesian modelling. A major drawback of control charts at present is that data are still widely transferred on a yearly rather than a quarterly or even monthly basis. The chief disadvantage of Bayesian modelling lies in the difficulty of obtaining consensus on the required prior distributions. Funnel plots on the other hand provide a flexible and sample size dependent uniform approach while at the same time offering an intuitive interpretation of volume effects. The addition of control and warning limits allows for formal assessment of deviations from target values.