Abstract
Objective:: Longitudinal trajectories are vital for clinical predictive modeling, yet administrative healthcare datasets exhibit low temporal resolution. This obscures chronological sequence extraction, making traditional timestamp-dependent algorithms ineffective. We propose the TIDE (Topological Inference of Directed Encounters) framework: a decoupled architecture designed to extract, validate, and rank directed clinical pathways from non-temporal data, emphasizing the computational resolution of concurrency and the prognostic utility of “informative absences.” Materials and Methods:: Utilizing a cohort of 404,532 inpatient encounters from the 2024 CMS Limited Data Set mapped via CCSR, the framework first mitigates concurrent documentation artifacts via Jaccard bundling ( J t h r e s h ≥ 0.90 ). After establishing a 5% baseline support floor, candidate trajectories, including statistically generated informative absences, were evaluated using asymmetric probability metrics and an empirical χ 2 gate to prevent long-range collider bias. Sequence non-randomness was verified through a 1000-iteration Monte Carlo permutation test with False Discovery Rate (FDR) control. Finally, we validated the qualifying topological pathways via a 5-stage predictive ablation study using L1-penalized logistic regression, thereby isolating independent prognostic lift without relying on administrative timestamps. Results:: The TIDE sequence engine extracted 463 non-random trajectories. In the ablation study, the Standard Baseline (relying solely on raw, unsequenced features) achieved a PR-AUC of 0.2408. The Fully Engineered Hybrid model, combining negative nodes with the extracted trajectories, achieved the highest ROC-AUC of 0.8093 and positive predictive value (PR-AUC: 0.2675). The L 1 penalty compressed the 207-dimensional hybrid space to 183 robust predictors, isolating 26 significant, multi-step clinical pathways independently predictive of in-hospital mortality. Conclusion:: The TIDE framework successfully extracts and empirically validates clinical trajectories from administrative data characterized by temporal ambiguity or complete absence. With a decoupled architecture, cross-sectional Boolean logic, informative absences, and predictive ablation, the algorithm provides a scalable methodology for clinical sequence discovery without reliance on chronological ground truth.