Cumulative risk adjusted monitoring of 30-day mortality after cardiothoracic transplantation: UK experience.

OBJECTIVE: Guidelines are needed for real-time quality monitoring in heart and lung transplantation. The cumulative sum (CUSUM) methodology with boundary limits derived from the sequential probability ratio test (SPRT) provide a means of monitoring performance without the need for repeated statistical testing. The variable life adjusted display (VLAD) complements the SPRT chart and provides a directly interpretable assessment of performance. We present our experience with these charts in monitoring cardiothoracic transplant outcomes in the UK.
METHODS: Thirty-day-post-transplantation mortality after isolated first time transplantation of the heart (n=1634) or lung (n=1162) in adults, between July 1995 and March 2004 in eight centres were monitored. CUSUM charts, with and without risk-adjustment and risk-adjusted VLAD plots were constructed. Thirty-day mortality rates for the UK as a whole were taken as the reference values for the unadjusted charts and a 50% increase in risk provided the basis for construction of boundary lines for the SPRT. Risk-adjustment was based on multivariate models previously developed from the national database.
RESULTS: For heart transplantation without risk-adjustment, four centres crossed the lower boundary, indicating 30-day mortality was in-line with or better than seen nationally. Two centres were close to signalling an alert, warning of a rise in mortality rate, but neither chart signalled an alarm. After risk-adjustment one centre's graph moved towards the centre of the chart, indicating monitoring should continue and the other signalled an alarm. For lung transplantation the unadjusted mortality rate at one centre was confirmed acceptable and the results remained inconclusive for five. At the other centre, following an alert to a possible increase in mortality half-way through the sequence, results improved. Case-mix adjustment served to pull the charts away from the upper boundary lines; no chart suggested any cause for concern. For most centres the VLAD charts oscillated around the horizontal axis.
CONCLUSIONS: CUSUM charts are useful tools for monitoring performance, and provide a basis for visually comparing results between centres and identifying periods of 'bad runs'. Risk-adjustment, which down-weights higher risk activity, avoids inappropriate reaction to unadjusted breaches of alert and alarm lines.