The complexity of vortex-induced vibration (VIV) instability analysis originates from the variability of the wind environment, the randomness of the analytical model, the ambiguity of the failure criterion, and the high-dimensional nonlinearity and non-explicitness nature of the instability function. This paper addresses the low sampling efficiency and challenges in approximating failure probability in prior research on reliability. By integrating an adaptive sampling strategy with particle swarm optimization (PSO), an initial VIV reliability metamodel is established, which is subsequently resampled to create a more refined metamodel. A combination of multiple reliability algorithms is then employed to calculate the maximum amplitude of VIV and the failure probability of the VIV interval. The validity and accuracy of the proposed method are demonstrated through two representative numerical examples. The results indicate that the proposed radial basis function (RBF)-PSO-importance sampling (IS) analysis method and its adaptive random sampling strategy can significantly reduce the computational cost associated with extensive finite element model operations and the complexity arising from the high-dimensional nonlinearity and non-significance of the limit state function (LSF). This method enables researchers to achieve more accurate and time-saving solutions for VIV in steel–concrete composite bridges.