Table 1.
Comparison of treatments reflected by number of mature follicles at trigger.
Fig 1.
Decision tree analysis (Evtree) for blastocyst optimization.
The predictive modeling analysis determined that [E2] at trigger <1584 pg/ml resulted in 80% probability of >40% blastocyst development success (n = 7, node 2). Female age, male age and number of oocytes retrieved were not significant variables (Node 0). [E2] at trigger significantly impacted outcome when [E2] was >1584 pg/ml, based on the number of mature follicles at trigger (<12 vs. >12) (node 2). The branch originating from the left of node 3 (nodes 4–10) demonstrates outcomes of treatments with < 12 mature follicles at trigger. When <3025 IU gonadotropin was administered in this branch, a 90% blastocyst development success occurred (n = 10, node 5). Total IU medication >3025 IU for with an [E2] at trigger of <2748 pg/ml and minimum [AMH] <1.783 ng/ml for 80% blastocyst development success (n = 11, node 8), otherwise success was poor (nodes 9, 10). The branch on the right of node 2 (nodes 11–21) depicts outcomes for treatments with >12 mature follicles at trigger. When [E2] at trigger >1584 pg/ml, success was significantly improved by administering <3025 total IU medication with and trigger by day 10 (node 13). [E2] <3056 pg/ml resulted in 70% (n = 7, node 14). [E2] >3056 pg/ml when maximum [FSH] was >6.5 mIU/m1 resulted in 85% success (n = 12, node 17). Success was > 90% success seen with [E2] >1564pg/ml, >12 mature follicles at trigger, >3325 IU medication and minimum [AMH] of <3.156 ng/ml (n 9, node 20).
Fig 2.
Receiver operating characteristic curve of the blastocyst optimization model.
The accuracy of the decision tree was calculated at 86.5% with an error rate of 13.5%. The area under the ROC curve (Robin et al. 2011), which is a plot of the true positive rate (sensitivity) against the false positive rate (specificity), is approximately 0.87 with confidence intervals of 0.7996–0.9312. This indicates that any final node on the model will yield 29% greater success than chance alone at the lowest predictive accuracy and 43% greater success than chance alone at its highest predictive accuracy. This is added validation of the predictive model in the current study.