AI Model Uses Routine IVF Data to Forecast Miscarriage Risk
Researchers in China developed an AI model using five standard clinical measurements—age, BMI, FSH, TSH, and platelet count—to predict early pregnancy loss after IVF. The XGBoost algorithm achieved high accuracy in distinguishing live births from miscarriages in a cohort of over 1,200 patients. While not diagnostic of implantation failure, the tool offers potential for improved patient counselling and risk stratification.
In October 2026, researchers at Shanghai First Maternity and Infant Hospital in China developed an AI model using five routinely collected clinical markers—age, BMI, follicle-stimulating hormone (FSH), thyroid-stimulating hormone (TSH), and platelet count—to predict early pregnancy loss following IVF or ICSI. The study analyzed data from 1,201 couples who achieved a pregnancy confirmed by ultrasound, of whom 222 (18.5%) experienced miscarriage within the first 12 weeks. The XGBoost machine learning algorithm emerged as the most accurate predictor, with FSH levels proving the strongest individual factor. The model was validated on a separate patient cohort from the same hospital and demonstrated strong discriminatory ability. An online calculator based on the model has been developed to allow clinicians to input patient data and generate personalized risk estimates. While not diagnostic of implantation failure—since it only applies after ultrasound confirmation—the tool aims to support clinical counselling and risk stratification. The study, published in *BioData Mining* in September 2026, acknowledges limitations: it relies solely on single-center data, excludes embryo quality, genetic factors, and immune status, and requires external validation across diverse populations before clinical adoption.
The broader context includes growing interest in AI applications within reproductive medicine, particularly around improving outcomes and patient support after fertility treatment.
The potential impact of this AI tool could be significant for patients undergoing IVF, particularly in reducing uncertainty during early pregnancy. By offering a personalized risk estimate based on standard clinical data, it may help clinicians tailor monitoring and support strategies, potentially improving psychological outcomes for individuals facing the anxiety of early pregnancy loss. However, such tools may also amplify emotional distress if risk predictions are misinterpreted or communicated poorly. The broader fertility industry—clinics, insurers, and patient advocacy groups—may adopt similar models if validated, influencing treatment pathways and insurance coverage decisions. A comparable precedent is the use of AI in predicting preterm birth using routine maternal data, which similarly raised hopes for early intervention but also concerns about overmedicalization. What remains uncertain is how regulatory bodies will assess such tools before clinical deployment, or whether they