O-087 Patient-centric, machine learning (ML)-based personalised prognostics supports fertility specialists to improve access to assisted reproductive technology (ART) and increase overall live birth (LB) outcomes

Abstract

Abstract Study question Does the use of patient-centric, ML-prognostics counselling report (Univfy® PreIVF Report) affect assisted reproductive technology (ART) conversion (first ART cycle usage) and LB rate (LBR)? Summary answer The use of patient-centric, ML-prognostics counselling report (Univfy® PreIVF Report) by fertility specialists is associated with higher ART conversion and LBR among new patients. What is known already ART is a highly effective and safe treatment for clinical infertility. However, ART remains vastly underutilised resulting in missed opportunities to help more people build families. Commonly used age-based trends often do not address patients' perceived risks including their own ART success probability and ART cost burden as related to their personalised LB probabilities. We previously reported the use of artificial intelligence (AI)/ML to generate patient-centric counselling reports based on ART success prediction models developed and validated for each fertility centre to address their local patient populations in ways that are personalised, relevant and actionable. Study design, size, duration Retrospective cohort analysis. Eight fertility centres from 22 locations across 9 states (US) and Ontario, Canada contributed to the research design, compilation of outcomes data, and interpretation of results. Five centres provided ART utilisation and outcomes data for 15,289 new patients seen in each centre's study period when the Univfy® PreIVF Report was available and data were submitted for aggregated research analysis. Each centre provided 4-6 years of data within the period 2016-2022. Participants/materials, setting, methods The effect of Univfy or No-Univfy Group on ART conversion was analyzed by Chi square tests using aggregated data and separately for each centre's data, for 3 timed analyses, 180-Day, 360-Day and “Ever” (no restriction) after new patient visit. Patients who received the Univfy® PreIVF Report prior to IUI or ART conversion, or had no such conversion after receiving it were placed into the Univfy Group. The No-Univfy Group comprises patients who did not receive a report. Main results and the role of chance Univfy report usage was associated with higher conversions to Direct-ART (by 2.6-, 2.4-, 1.9-folds) and Any-ART (by 2.9-, 3.0-, 2.4-folds) in the aggregated data when analyzed for 180-Day, 360-Day and Ever, respectively; p-value < 0.001. Direct-ART is ART conversion without prior IUI(s); Any-ART conversion includes ART conversion with or without prior IUI(s). In the centre-specific analyses, the fold increase in Direct-ART and Any-ART conversions ranged from 1.8 to 4.5 and 2.2 to 4.7, respectively, in the 360-Day period; p-value <0.001. Univfy® PreIVF Report usage was associated with an increase in estimated LBR ranging from 2.1 to 1.3 folds for the Univfy Group compared to No-Univfy Group (360-Day analysis, p < 0.001) based on conservative versus liberal scenarios. Similar ART conversion and LBR results were observed for 180-Day and Ever analyses, p < 0.001. We used conservative to liberal assumptions for IUI-LBR and NC-LBR because IUI and NC outcomes were not readily available. (Conservative: IUI-LBR 15%, natural conception (NC)-LBR 5%; Liberal: IUI-LBR 25%, NC-LBR 20%). Estimated LBR for ART used clinical ongoing pregnancies and documented live births as LBs and the following LBR assumptions: freeze-all with no transfers yet (50%); gestational carrier ART (45%); ART with unknown outcomes (0%). Limitations, reasons for caution This study was not prospective or randomised. The intended report usage was to support physicians when counselling patients. Although we observed comparable report and ART usage across predicted ART-LB probabilities, there is potential unintentional bias towards higher report or ART utilisation among patients with more favorable clinical characteristics. Wider implications of the findings These results represent our retrospective experience in diverse geographies in North America. We endeavor to collaborate with additional centres to test the reproducibility of AI/ML-driven, validated personalized IVF prognostics on improved overall live birth outcomes and ART access when counselling patients about treatment options. Trial registration number not applicable