Disease Continuity Index (DCI): Quantitative Evidence Supporting the "Disease as a Continuous State Space" Hypothesis in Lung Adenocarcinoma

Abstract

Background: Traditional medicine categorizes diseases into discrete labels based on organs or pathology (e.g., lung adenocarcinoma, coronary heart disease, diabetes). However, significant prognostic heterogeneity exists within the same diagnosis, and comorbidity is extremely common, suggesting that these discrete labels may obscure a deeper biological reality. We propose the "disease as a continuous state space" hypothesis: not only is a single disease continuous, but different diseases are also continuous with one another. So‐called different diseases are essentially different regions or dynamical bifurcations within the same multi‐dimensional continuous space, and comorbidity is a direct manifestation of this continuity. Using lung adenocarcinoma as an example, this study constructs a Disease Continuity Index (DCI) to provide quantitative evidence for this hypothesis and demonstrates a methodological shift from "applying AI to solve established problems" to "using AI to redefine the problems themselves." Methods: Transcriptomic data from The Cancer Genome Atlas (TCGA) lung adenocarcinoma cohort (n=539) were used. A variational autoencoder (VAE) was first employed for unsupervised extraction of a three‐dimensional latent space. Univariate Cox regression identified the dimension most associated with overall survival (z1), and its percentile rank was inverted to define DCI (0–1, with higher values indicating worse prognosis). The overall prognostic value of DCI and its risk stratification ability within Stage I patients were evaluated. The latent space was visualized using t‐SNE, and the correlation between DCI and the principal axes was calculated to quantify its continuous gradient. Results: DCI significantly stratified overall prognosis (log‐rank p=0.0008) and identified a high‐risk subgroup within Stage I patients (n=290, p=0.0176). Latent space visualization revealed a remarkably strong ordered gradient of DCI along the t‐SNE2 axis (Spearman ρ = -0.923, p < 0.001), whereas TNM staging was completely intermingled, showing no natural separation. In univariate Cox analysis, DCI was significant (HR=1.67, p=0.045); after adjusting for stage in multivariate analysis, DCI showed an independent trend (HR=1.67, p=0.06). Conclusion: DCI exhibits a remarkably strong continuous gradient in the latent space, orthogonal to the staging system, providing quantitative evidence for the "disease as a continuous state space" hypothesis. This index can identify high‐risk Stage I patients missed by traditional staging. More importantly, this study offers a novel perspective for understanding comorbidity—comorbidity is not the coincidental coexistence of multiple independent diseases, but an inevitable manifestation of multiple bifurcation points within the same continuous space. This framework represents a paradigm shift from "applying AI to solve established problems" to "using AI to redefine the problems themselves," laying the foundation for a quantitative, individualized language of disease description that transcends organ‐ and disease‐name‐based distinctions.