Background and aims Preterm babies have higher mortality than terms. Risk-adjusted mortality (RAM) is useful for making comparisons among different NICUs. GA, BBW, sex, singleton birth and antenatal steroid have been used to estimate mortality (M) of preterm. The aims of this study are 1. To compare the performance of GA, BW, and Logistic Regression (LR) in predicting M of VLBW infants. 2. To compare the RAM in different areas and periods.
Methods Cohort data from 2000 to 2011 were used. M is defined as death prior to discharge. Exclusion criteria included 1) Transferred after 24 h of age; 2) Death within 24 h of admission and 3) Lethal malformation. We developed a LR model to predict M [expected probability (Pro)]. ROC curves were used for assessing performance of predicting M. To compare the RAM, we calculated (O-E) Pro (observed Pro – expected Pro) values in each patient and used these values for comparisons.
Results 9207 VLBWs were enrolled. The calculated probability of death by LR model was: P = 1/(1+e-Z), where e= natural logarithms and z = (-0.62•[prenatal steroid]) -(0.219•GA)- (0.004•BBW)- (0.327•[singleton]) + (0.286•[male]) + 8.438. Area under ROC were 0.858 for LR (95% CI: 0.847–0.869), 0.841 for BBW (95% CI: 0.829–0.853) and 0.827 for GA (95% CI: 0.815–0.839).
There were significant differences of RAM in different locations and years (Figure 1).
Conclusions The use of LR is better than GA and BW in predicting M of VLBWs. RAM can be used as a tool for quality improvement.