Go back
Title: PREDICTING AND OPTIMIZING IVF SUCCESS USING MACHINE LEARNING- AI MODEL
e-poster Number: EP 439
Category: Miscellaneous
Author Name: Dr. Navin Panchal
Institute: BAPS yogiji maharaj hospital
Co-Author Name:
Abstract :
Objective:
One of the most frequently asked questions by IVF patients is, "What are my chances of success?" Responding to this challenging question often involves considering factors such as the patient's age, AMH value, and the cause of infertility. However, numerous other factors can influence IVF outcomes. We have developed an AI algorithm with data analytics that enables a more precise prediction and also enables us to customise the treatment plan for the likelihood of success in an IVF cycle
Design:
In this study, we applied a Multivariate logistic regression algorithm to analyse the data of 100 parameters tracked for each IVF cycle to establish their association with IVF outcomes. These parameters encompass patient demographics, stimulation protocols, response characteristics, oocyte and embryo metrics, and variables related to embryo transfer. With available data, we tried to calculate independent K and Beta for each variable. To statistically validate our results, we also used Quinlan?s C5.0 decision tree data mining algorithm for comparison, enhancing the robustness of our findings.
Materials and Methods:
The training dataset comprised 300 individual IVF records, each representing a single IVF cycle. We collected data from Four different IVF centres in India, two from GUJARAT and Two from Punjab. We conducted the following experiments: First, we developed a decision tree model using only patient age as the predictor for pregnancy outcomes. This model yielded an accuracy of approximately 56%, reflecting the predictive approach typically used in current IVF clinical practices.
We computed a multiple logistic regression for all the variables. We used P value of <0.05 for statistical significance. In this approach we reduced the number of variables to about 12 statistically significant independent variables. Restricting further analyses to these selected variables produced an association between predicted probabilities and observed responses of about 74% concordant results, an increase in about 18% from a logistic fit with just age as the predicting variable.
we also constructed a comprehensive decision tree model, incorporating around twelve key variables from the available set of 100 potential predictors. This model achieved a predictive accuracy of approximately 75% for pregnancy outcomes, demonstrating improved performance over the age-only model
Result:
Our analysis demonstrated a significant enhancement in predictive accuracy when incorporating additional predictors beyond age for forecasting IVF outcomes. Furthermore, results from traditional statistical methods appear to corroborate the outcomes derived from data mining techniques, both in terms of predictive accuracy and the range of variables considered
