<p>Identifying and detecting causal relationships in complex systems is a challenging task. In this paper, we propose a new method to quantify causality in complex systems. The basic idea of the method is that if there is a causal relationship between two variables, then historical information about one variable can be a good estimate of the other. The method establishes an attractor with the help of phase space reconstruction, fully considers the nearest neighbor points at each moment and their change patterns, and constructs a weighted change pattern matrix to portray the global properties of the system. Further, we predict the change by cross-prediction and use cosine similarity entropy to measure the similarity between the predicted and true values and quantify the causal strength in this way. In addition, we extend the method to the multivariate case, which helps us to quantify the direct coupling between the two variables, eliminating the effect of indirect coupling on the results of the process. Finally, we apply the method to the analysis of two physiological datasets, which further reveals the interactions in different physiological systems, and also provides new perspectives for studying causality in complex systems and nonlinear time series.</p>