Coarse-grained materials are widely used as railway subgrade fill, and their resilient modulus is essential for evaluating recoverable deformation under repeated train loading. In railway subgrades, the initial stress state varies with depth and location, but its effect on resilient modulus has not been fully quantified in existing models, which generally do not distinguish between initial static axial stress and cyclic stress. To address this issue, cyclic triaxial tests were conducted on coarse-grained subgrade fill under different confining pressures, initial axial stresses, and cyclic stress amplitudes. The results show that, under the same cyclic stress amplitude and confining pressure, specimens with higher initial axial stress exhibit higher resilient modulus. Based on these results, a resilient modulus prediction model incorporating initial axial stress was proposed and validated using both the present test data and published datasets. The proposed model performs well under isotropic or fixed initial stress conditions and provides higher accuracy than the conventional three-parameter model when predicting responses under varying initial axial stresses at the same total axial stress. The proposed model improves resilient modulus prediction for coarse-grained railway subgrade materials under different initial stress states.