Abstract: State transitions in complex systems (e.g., cellular carcinogenesis, grassland desertification) exhibit suddenness and irreversibility. Deciphering their underlying mechanisms and enabling early warning constitute a cross-disciplinary common challenge. Existing methods struggle to characterize causal reconstruction during state transitions. To address this limitation, this study proposes a new paradigm of state-driven causal comparison, realizing a perspective shift from causal description of a single state to comparative analysis of causal structural differences between states. A cross-state causal difference analysis framework and a mathematical optimization model are constructed. The core innovation lies in establishing the structural differences of causal networks between states as a measurable direct object, breaking through the limitation of ex-post comparison of traditional static indicators. On this basis, a node contribution index of causal differences is proposed to identify key driving variables. Taking dyslipidemia as an example, by constructing and comparing causal networks of healthy, critical, and disease states, the developed early prediction model demonstrates superior predictive performance and generalization ability over classical methods on external data, verifying the effectiveness of the proposed approach.

Key words: complex systems, state transitions, causal discovery, dyslipidemia, early prediction