Subscribe to RSS
DOI: 10.1055/s-2006-925889
Data mart based risk stratification in heart surgery: comparison to a prospective risk score validation
Aim: A data mart system was developed which integrates all relevant data from various hospital information systems (HIS) to support easy comprehensive research. The consolidated data set has enabled us to apply six commonly used risk scores in a large patient population. The present study evaluates the validity of this approach.
Methods: Current and historical data from three HIS (surgical, anesthesiological, and laboratory) were extracted and integrated in a data mart system. By reconstructing the semantic meaning of all score variables we could retrospectively apply six cardiac risk schemes for postoperative mortality in 13882 of 15364 patients (90.4%). Using the area under ROC curve (c-index), the scores were evaluated with regard to their discriminative power and compared with a prospectively scored collective of 504 patients from Geissler et. al. [2000].
Results: The performances of the data mart based risk scores were equal or even better than the prospective controls: Euro 0.786 vs. 0.786, Initial Parsonnet 0.784 vs. 0.755, Cleveland Clinic 0.782 vs. 0.731, Pons 0.761 vs. 0.745, Ontario Province Risk 0.752 vs. 0.752, and French 0.750 vs. 0.719. In both populations the Euro- and Initial-Parsonnet-Score have the best discriminative power while the difference between the two scores is smaller in the data mart set.
Conclusion: The data mart approach enables individual hospitals in a very flexible way to apply and validate a variety of risk models by using data from daily clinical practice. The resulting risk scores show a high validity and are valuable tools for risk-adjusted outcome analysis.