Enhancing Progressive Diagnosis Prediction in Healthcare with Continuous Normalizing Flows

Yanchao Tan, Hengyu Zhang, Zihao Zhou, Guofang Ma, Fan Wang, Weiming Liu, Xinting Liao, Vicki Stover Hertzberg, Carl Yang

Abstract

Progressive diagnosis prediction in healthcare is a promising yet challenging task. Existing studies usually assume a pre-defined prior for generating patient distributions (e.g., Gaussian). However, the inferred approximate posterior can deviate from the real-world distribution, which further affects the modeling of continuous disease progression over time. To alleviate such inference bias, we propose an enhanced progressive diagnostic prediction model (i.e., ProCNF), which integrates continuous normalizing flows (CNF) and neural ordinary differential equations (ODEs) to achieve more accurate approximations of patient health trajectories while capturing the continuity underlying disease progression. We first learn patient embeddings with CNF to construct a complex posterior approximation of patient distributions. Then, we devise a CNF-enhanced neural ODE module for progressive diagnostic prediction, which aims to improve the modeling of disease progression for individual patients. Extensive experiments on two real-world longitudinal EHR datasets show significant performance gains brought by our method over state-of-the-art competitors.
BibTeX
@inproceedings{tan2024enhancing,
  author = {Yanchao Tan and Hengyu Zhang and Zihao Zhou and Guofang Ma and Fan Wang and Weiming Liu and Xinting Liao and Vicki Stover Hertzberg and Carl Yang},
  title = {{Enhancing Progressive Diagnosis Prediction in Healthcare with Continuous Normalizing Flows}},
  booktitle = {The Web Conference},
  year = {2024},
  doi = {10.1145/3589335.3651457},
}

Loading related content...

Figure