Spotting the pantomime villain: do the usual approaches correctly indicate when waiting times got shorter?

To assess whether changes in length of wait are correlated (directly) with changes in size of list, i.e. whether census-, event- and enrolment-based estimates of the percentage that waited 0-2 months are valid measures. EyeNet Sweden supplied dates of extraction and enrolment for 1,061,246 cataracts extracted between 1992 and 2009 inclusive. Changes in size of list were calculated as enrolments minus extractions. Fifty-nine times out of 62 the enrolment-based measure reported an increase in length of wait when size of list increased or a decrease in length when size decreased (L(B) = 90%, 95% confidence interval = 80-100). But 47 times out of 62 the event-based measure reported a decrease in length of wait when size of list increased, or an increase in length when size decreased (L(B) = 40%, 95% confidence interval = 6-74). The census-based measure did likewise 34 times out of 63 (L(B) = 6%). The three approaches gave different results despite using the same combination of cohorts and categories, the same point on the distribution of waits, the same rules about which waits (and which parts of these) count and the same length of follow-up. Census- and event-based measures often concealed the true direction of change in length of wait.