Author: Alan Pan, MS (2025)
Primary advisor: Xiaoqian Jiang, PhD
Committee members: Yejin Kim, PhD and Farhaan Vahidy, PhD, MBBS
PhD thesis, McWilliams School of Biomedical Informatics at UTHealth Houston.
ABSTRACT
The post-stroke journey is highly variable. Upon hospitalization for a primary acute ischemic stroke (AIS), patients encounter a multitude of clinical decisions that ultimately influence the trajectory of their health recovery and outcomes. The overarching scope of our work is centered around leveraging routinely captured sources of health information – such as electronic medical records (EMR) and administrative medical claims – to assess how patterns in clinical workflows and care pathways can be extracted and transformed to generate insights from observational data repositories. Within this framework, we utilized a suite of computational methods to evaluate predictors of adverse health outcomes among the stroke population from a tertiary healthcare system.
Our first 2 research aims capitalize on a linkage between clinical and claims data to capture details on post-acute patterns-of-care for a subset of older, Medicare-insured AIS survivors. Although EMRs are expansive, these sources are often limited with respect to information on post-discharge care utilization and states of recovery. Furthermore, a minority of post-stroke follow-up studies have included evaluation of more patient-centered outcomes, such as home time (HT) – defined as the number of days spent at home or in non-institutionalized care. We hypothesized that variability in 1-year HT could, in part, be explained by patterns in early (30-day) post-acute care (PAC) utilization. Our corresponding research objectives were multi-fold. First, we identified high-risk patient clusters and defined individual and health systems factors that contribute to reduced 1-year HT. Second, we utilized sequential pattern mining (SPM) techniques to derive probabilistic representations of 30-day PAC pathways and their associated risks of undergoing unfavorable trajectories with respect to HT. We anticipate these findings can help coordinate early PAC pathways and improve long-term health trajectories.
Our third aim incorporated a re-examination of the acute care setting and was underscored by growing recognition that optimizing hospital environments and systems-of-care are crucial to attaining patient-centered goals. Specifically, we sought to derive intervenable targets for minimizing the onset of adverse in-hospital complications known to prognosticate poor health outcomes among AIS survivors. To address this objective, we extracted data on clinical workflows during 2 distinct phases of the stroke care continuum. First, among older adults treated for an AIS in the emergency department (ED), we evaluated the impacts of prolonged ED boarding durations (i.e., holding admitted patients in the ED while awaiting an inpatient bed) on incident delirium risks. Second, we analyzed whether nighttime interruptions and sleep disruptions similarly led to increased delirium onset. Our findings demonstrate that efforts to minimize delays in sub-acute care transitions as well as maximize hospital restfulness may represent implementable strategies for enhancing stroke systems-of-care.
In light of expanded big data applications in stroke outcomes research, gaps in knowledge translation persist as to the extent to which real-world health data can be harnessed to effectively coordinate care. Our multi-faceted evaluation of EMR and medical claims highlights an underutilized application of time series data and reinforces the value of clinical workflows and care pathways as an elucidative modality of health information.