Objective: Different statistical methods are used to evaluate the performance of risk prediction models. While Brier score and R² methods are used to evaluate the overall performance of the model, ROC curve and discrimination curves are used to evaluate discrimination ability. Additionally, calibration curves and the Hosmer-Lemeshow test statistic are used to evaluate the calibration.

It is aimed to evaluate the performance of SNAP-II and SNAPPE-II risk prediction models with the statistical methods used to evaluate model performance.

Materials and Methods: SNAP-II and SNAPPE-II risk prediction models were used to estimate the mortality probabilities of 864 infants hospitalized in the Neonatal Intensive Care Unit of Çukurova University Medical Faculty Balcalı Hospital between July 2013 and 2015.

Results: The Brier Score of the SNAP-II risk prediction model was found to be 0.076 and the Nagelkerke R² to be 0.479. The ROC-AUC value was 0.885 (CI: 0.851-0.919). The cutoff point was determined as 14.50. The p value of the Hosmer-Lemeshow test was found to be 0.619 (χ²=6.249). While the Brier score of the SNAPPE-II risk prediction model was 0.056, the Nagelkerke R² value was 0.619. The ROC-AUC value was 0.921 (CI: 0.890-0.953). The cutoff point was determined as 26.50 The p value of the Hosmer-Lemeshow test was found to be 0.968 (χ²=2.338).

Conclusion: Both SNAP-II and SNAPPE-II risk prediction models have high model performance. In addition, it was found that the SNAPPE-II risk prediction model was superior to SNAP-II in terms of overall performance, discrimination ability and calibration.