Statistical Analysis and Sample Size Determination:
A model was constructed to test the predictive value of ELI on
post-procedural outcomes. We calculated the ELI value for each patient
and then examined its accuracy and value in predicting all-cause
mortality/cardiac events. Sample size determination for this prediction
model accepted a minimum area under the curve (AUC) of 0.7 on C
statistics. A global shrinkage factor of 0.9 has been used to estimate
the minimum number of events (E = 46) to fit the newly established model
in predicting the primary endpoint using a time-to-event multivariable
survival (Cox Regression) analysis10. Recently, we
reported 4-year mortality of 32% following the TAVR procedure at our
institution 11. To
include an adequate sample size of 46 events, a total of 144 patients to
achieve enough power for properly fitting the predictive model of ELI.
We increased the global shrinkage factor by including 272 subjects in
our review to 0.95.
A Kaplan Meier curve was constructed to perform univariate, and the Cox
regression model was used for multivariable survival analyses. Since no
similar study has been identified in our literature search, we used a
receiver operating characteristic curve (ROC curve) to determine the
cutoff point for ELI. A ROC curve was also used to calculate the best
discrimination threshold, which balanced sensitivity and specificity.
The plotted points were then used to create a ROC curve to identify the
best entry for distinguishing high-risk from low-risk sub-groups. The
positive and negative predictive values, sensitivity, specificity, and
accuracy of the prediction model were then calculated for this cutoff
value of ELI. The value of the ELI was used for grouping the patients
for comparison. The relative risk determined the probability of
mortality in the high-risk group to mortality in the low-risk group.
This relative risk may be used to predict post-procedural outcomes of
patients based on their ELI value. Furthermore, this ROC curve method
may highlight differences in threshold values between different patient
demographics and types of prosthesis. All analyses were performed using
the Statistical Program for Social Sciences version 26.0 on the Mac OS
platform (SPSS-IBM Inc. Chicago, IL).