Multivariate model predicts immune imbalance in recurrent pregnancy loss and recurrent implantation failure

Recurrent pregnancy loss (RPL) and recurrent implantation failure (RIF) are thought to arise from distinct yet partially overlapping causes, with a substantial number of cases associated with immune system alterations. We hypothesized that a peripheral blood signature integrating natural killer (NK) cell receptor status, monocyte activation, myeloid-derived suppressor cell (MDSC) abundance, and regulatory T cell (T Reg ) levels would more accurately distinguish each disorder from non-pregnant healthy controls than any single biomarker. We enrolled 194 women and performed deep immunophenotyping of NK cells, monocytes, MDSC, and T Reg . Variable selection was performed with the Boruta algorithm, followed by multivariate logistic regression modelling. For RPL, the final model included five biomarkers, achieving an area under the curve of 0.95 and an accuracy of 90.7%. For RIF, the model retained four biomarkers, yielding an area under the curve of 0.85 and an accuracy of 79.5%. Logistic regression was deliberately chosen to prioritize clinical interpretability and facilitate future translation into a point-based diagnostic score.