The POPI-Plus tool: prediction model of outcome of pregnancy in in vitro fertilization from a large retrospective cohort

Objective: To create a tool that accurately predicts live birth chances after a positive pregnancy test after elective single embryo transfer (ET).

Design: Retrospective cohort.

Setting: CHUM hospital and Ovo clinic in Montreal, Canada.

Patient(s): Patients with a positive pregnancy test result who underwent their first single ET after in vitro fertilization (IVF) at the CHUM hospital and Ovo clinic in Montreal, Canada, from 2012 to 2016 were selected. A total of 1,995 patients were included in this study.

Intervention(s): The data from both centers were combined and divided into training (70%, n = 1,398) and validation (30%, n = 597) sets. The predictive model was developed using backward selection method for the following variables: age of patient at egg retrieval; log β-human chorionic gonadotropin (β-hCG) (β-hCG) 1; log β-hCG 2; and IVF treatment type. Moreover, the classification tree, random forest, and neural network models were generated.

Main outcome measure(s): The measured outcomes were live birth (live fetus ≥24 weeks of gestation) and nonviable pregnancies. The performance of all models was evaluated by area under the receiver operating characteristic curve (AUC).

Result(s): Advancing age was negatively correlated with live birth. The odds ratio (OR) of age of patient at the time of egg retrieval was 0.95 (95% confidence interval [CI], 0.91-0.99). The log β-hCG 1 and log β-hCG 2 were positively correlated with live birth in the univariate analysis (OR, 4.15 [95% CI, 3.19-5.39], and OR, 3.84 [95% CI, 2.99-4.93], respectively). The β-hCG 1 level needed for a successful pregnancy was lower in frozen ET and modified natural IVF than in simulated IVF (OR, 0.55 [95% CI, 0.34-0.91], and OR, 0.49 [95% CI, 0.26-0.95], respectively). The best performance in terms of the AUC was the updated logistic model: POPI-Plus. The AUC values were 0.76 (95% CI, 0.73-0.79) and 0.78 (95% CI, 0.74-0.82) for the training and validation data, respectively. The other models (classification tree, random forest, and neural network) also performed adequately, with an AUC of ≥0.7, but remained below POPI-Plus. An open-access calculator was generated and can be found on the website of the University of Montreal on the following link: https://deptobsgyn.umontreal.ca/departement/divisions/medecine-et-biologie-de-la-reproduction/the-popi-plus-tool/.

Conclusion(s): The POPI-Plus tool offers individualized counseling for patients after an initial positive β-hCG test result. Future studies will assess its impact on patient anxiety while awaiting viability ultrasound and perform prospective validation on new patients.