Prognosis After First-Trimester Threatened Miscarriage: A Systematic Review, Prognostic Accuracy Meta-Analysis, And Prediction Modelling Review

Threatened miscarriage represents one of the most prevalent obstetric emergencies globally. Nevertheless, women experiencing first-trimester bleeding with a viable intrauterine pregnancy are predominantly advised based on empirical experience rather than on precisely quantified risks. Over the past forty years, numerous biochemical, ultrasound, and clinical predictors have been proposed. More recently, multivariable and machine-learning models have also been introduced. However, it remains uncertain which of these diagnostic tests genuinely contribute prognostic value and whether integrated modelling approaches significantly surpass routine assessment methods. We conducted a comprehensive systematic review and a meta-analysis of prognostic accuracy, incorporating an evidence synthesis of multivariable and machine learning prediction models, focusing on women experiencing first-trimester threatened miscarriage with ultrasound-confirmed viable intrauterine pregnancy. Searches were performed across MEDLINE, Embase, CINAHL, CENTRAL, Web of Science, ClinicalTrials.gov, and WHO ICTRP from inception until March 2025, with no restrictions on language. Eligible studies encompassed prospective or nested cohort studies, or high-quality systematic reviews and meta-analyses, which reported first-trimester biochemical, ultrasound, clinical, or combined prediction models alongside subsequent pregnancy outcomes (including miscarriage versus ongoing pregnancy or live birth). Data were systematically extracted into a comprehensive template and synthesized following the principles outlined in PRISMA 2020/PRISMA-DTA, TRIPOD, CHARMS, QUADAS-2, and PROBAST. For thresholds reported in four or more studies, comparable in nature, we employed random-effects bivariate or HSROC models; when fewer such studies were available, the accuracy data were summarized narratively. Ten studies met the inclusion criteria: six primary prospective cohorts and four evidence-synthesis or meta-analytic papers from early pregnancy and emergency settings. Among the primary cohorts, miscarriage risks consistently ranged from 15% to 25% in women with threatened miscarriage and an initially viable intrauterine pregnancy, confirming this as a genuinely high-risk state rather than a benign variation of normal pregnancy. Serum progesterone showed high specificity (>80–90%) but modest pooled sensitivity (∼30%), indicating that very low values strongly “rule in” risk but fail to identify many women who miscarry. In contrast, CA-125 demonstrated near single-test performance in two large meta-analyses (pooled sensitivity and specificity both ≈approximately 90–95%, AUC ≈approximately 0.95), and in one cohort, a cut-off around 31 IU/mL yielded sensitivity over 95% and specificity of 100%. Ultrasound viability parameters, particularly fetal heart rate, crown–rump length, and gestational sac morphology, were the most consistently informative imaging predictors, with specificities often ≥85–90% and good to fair discrimination but limited sensitivity and varied cut-offs. Only two recent cohorts developed formal multivariable models: logistic regression models combining maternal factors, progesterone, and ultrasound features achieved AUCs in the high 8 to about 0.9 range, while a single random forest model reached an apparent AUC of approximately 97, with very high negative predictive value on internal testing validation. All models, however, were from single centers, based on modest event numbers, had moderate-to-high risk of bias, and lacked external validation. Common threats to validity across the literature included non-consecutive recruitment, post-hoc threshold selection, incomplete follow-up, and insufficient reporting to reconstruct 2×2 data. For women experiencing threatened miscarriage in the first trimester with a viable intrauterine pregnancy, substantial prognostic indicators are already available; however, these indicators tend to be fragmented, lack sufficient statistical power, and are seldom translated into practical tools. Ultrasound viability parameters, combined with a limited selection of biochemical markers, particularly CA-125 and progesterone, can effectively stratify risk and facilitate more accurate, probabilistic counseling. Nonetheless, no individual test or model currently satisfies the evidentiary standards necessary for inclusion in clinical guidelines implementation. The priority within the field should now transition from the discovery of increasingly isolated predictors to the cultivation of extensive, prospectively registered, multicentre cohorts characterized by harmonized definitions, prespecified multimodal predictor sets, rigorous contemporary modelling techniques, and comprehensive TRIPOD-compliant reporting alongside external validation. This synthesis delineates the most robust existing indicators, identifies the methodological deficiencies that compromise current models, and offers a concrete roadmap for the development of truly reliable prognostic tools intended for the millions of women worldwide who face threatened miscarriage annually.