Risk prediction models for ovarian hyperstimulation syndrome: a systematic review and meta-analysis

Abstract Background Ovarian hyperstimulation syndrome (OHSS) is a serious complication of controlled ovarian stimulation (COS). The main clinical manifestation of OHSS is increased ovarian volume. OHSS can cause local and systemic tissue oedema, electrolyte disturbances, cardiorespiratory dysfunction, coagulation dysfunction, and other symptoms. These symptoms greatly affect patients’ quality of life. As infertility rates rise and assisted reproductive technology (ART) becomes more common, the risk of OHSS increases. Therefore, early identification of high-risk patients and timely intervention are crucial. Methods The PubMed, Embase, Cochrane Library, Web of Science, CINAHL, China National Knowledge Internet (CNKI), Wanfang, China Science and Technology Journal Database (VIP), and China Biology Medicine (CBM) databases were systematically searched from inception to March 30, 2025. Two researchers independently screened the literature, extracted data, and evaluated the quality of included studies using the updated prediction model risk of bias assessment tool (PROBAST + AI). We conducted a meta-analysis of predictors from the developed models using Stata 15.0 software. Results A total of 16 studies were included, comprising 29 OHSS risk prediction models. The area under the curve (AUC) ranged from 0.628 to 0.998, with 23 models demonstrating AUC > 0.700. Model calibration was performed in 10 studies, internal validation in 14 studies, and 2 studies conducted both internal and external validation. The PROBAST + AI assessment identified a high risk of bias across the included studies, primarily in the research design and statistical analysis domains. The most common predictors identified across the models included: antral follicle count (AFC), estrogen (E2) levels on the day of human chorionic gonadotrophin (hCG) injection, number of oocytes retrieved, polycystic ovary syndrome (PCOS), age, anti-mullerian hormone (AMH), gonadotropin (Gn) days, initial dose of Gn, and body mass index (BMI). Conclusions Our findings indicate substantial variation in OHSS incidence. Interpretation of the results should be with caution due to the limitations of the current evidence. Current OHSS risk prediction models remain under development and require further refinement. Future efforts to build and improve these models should focus on key areas, including research design, sample size, handling of missing data, model calibration and validation, and detailed reporting. Trial registration: PROSPERO CRD420251025876.