INTRODUCTION: The Parsonnet scoring system has been widely adopted in cardiac surgery but is recognised to systematically over-estimate risk. The aim of this study was to develop firstly an alternative scoring system, which better reflects risk based on contemporary data and which can provide estimates of in-hospital mortality rates; and secondly, graphical methods that allow mortality figures, adjusted for risk mix, to be displayed and regularly updated. SCORING SYSTEM: We analysed the records of 4,318 consecutive cardiac surgery patients treated between 1992 and 1995. The analysis identified risk factors that were associated with in-hospital mortality. Logistic regression was used on a random sample of 80% of the data to derive an additive 'risk score' for expected mortality. Initially this was done using only patients who had undergone isolated coronary artery bypass grafting as they represented the majority of cases (2,980/4,318). The same technique was then applied to the second most common procedure, isolated valve surgery (654/4,318), before being applied to all remaining procedures. The simple formula derived expresses risk in terms of significantly associated factors, such as LV function and type and urgency of procedure. This 'risk score' was then validated against the remaining 20% of the data. GRAPHICAL DISPLAY: The Cusum method plots cumulative in-hospital deaths against the number of operations and can be used to monitor results in a run of cases (de Leval JTCVS 1994; 107:914). However it takes no account of variable risk in a mixed practice. Our 'Variable Life Adjusted Display' or VLAD incorporates expected mortality as derived by the above 'risk score'. A surgeon is regarded as being notionally in credit or debit according to how his actual in-hospital death rate compares with the death rate that would have been expected based on the scores for his caseload. This credit/debit is plotted against a count of the surgeon's consecutive cases. We have developed a system for displaying these plots on a computer which allows trends, learning curves, and changing performance, to be monitored in real time, corrected for case mix.