Purpose: To develop an innovative machine learning (ML) model that predicts personalized risk of primary ovarian insufficiency (POI) after chemotherapy for reproductive-aged women. Currently, individualized prediction of a patient's risk of POI is challenging.
Methods: Authors of published studies examining POI after gonadotoxic therapy were contacted, and six authors shared their de-identified data (N = 435). A composite outcome for POI was determined for each patient and validated by 3 authors. The primary dataset was partitioned into training and test sets; random forest binary classifiers were trained, and mean prediction scores were computed. Institutional data collected from a cross-sectional survey of cancer survivors (N = 117) was used as another independent validation set.
Results: Our model predicted individualized risk of POI with an accuracy of 88% (area under the ROC 0.87, 95% CI: 0.77-0.96; p < 0.001). Mean prediction scores for patients who developed POI and who did not were 0.60 and 0.38 (t-test p < 0.001), respectively. Highly weighted variables included age, chemotherapy dose, prior treatment, smoking, and baseline diminished ovarian reserve.
Conclusion: We developed an ML-based model to estimate personalized risk of POI after chemotherapy. Our web-based calculator will be a user-friendly decision aid for individualizing risk prediction in oncofertility consultations.
Keywords: Chemotherapy; Fertility preservation; Oncofertility; Primary ovarian insufficiency; Risk calculator.
© 2021. This is a U.S. government work and not under copyright protection in the U.S.; foreign copyright protection may apply.