The good, the bad and the ugly: What do we really do when we identify the best and the worst organisations?

Identifying high and poorly performing organisations is common practice in healthcare. Often this is done within a frequentist inferential framework where statistical techniques are used that acknowledge that observed performance is an imperfect measure of underlying quality. Various methods are employed for this purpose, but the influence of chance on the degree of misclassification is often underappreciated. Using simulations, we show that the distribution of underlying performance of organisations flagged as the worst performers, using current best practices, was highly dependent on the reliability of the performance measure. When reliability was low, flagged organisations were likely to have an underlying performance that was near the population average. Reliability needs to reach at least 0.7 for 50% of flagged organisations to be correctly flagged and 0.9 to nearly eliminate incorrectly flagging organisations close to the overall mean. We conclude that despite their widespread use, techniques for identifying the best and worst performing organisations do not necessarily identify truly good and bad performers and even with the best techniques, reliable data are required.