Details of 759 microvascular breast reconstructions in 515 patients were entered into the online calculator. The predicted rates of post-operative complications were recorded and compared to observed complications identified on chart review. The calculator was validated using three statistical measures described in the original development of the ACS NSQIP model.
The calculator predicted that complications would occur in 11.1% of breast reconstructions while the observed rate was 10.5%. Hosmer-Lemeshow test did not find any statistical difference between these rates (p = 0.69) indicating that the calculator accurately measured what is was intended to measure. The area under the receiver operating curve or c-statistic (measure of discrimination) was found to be low at 0.548, indicating the model has random performance in this patient population. The Brier score was higher than that reported in the original ACS calculator development (0.094 vs 0.069) demonstrating poor correlation between predicted probability and actual probability.
This study demonstrates that while the ACS NSQIP Universal risk calculator can predict the proportion of patients that will develop complications it cannot effectively discriminate between patients who are at risk of complications and those who are not.