Tsui, Alex; (2023) Quantifying cognitive and mortality outcomes in older patients following acute illness using epidemiological and machine learning approaches. Doctoral thesis (Ph.D), UCL (University College London).

Preview	Text Tsui_10174957_Thesis_sigs_removed.pdf Download (8MB) | Preview
Preview	Text Supp fig 1 - TSNE by imputation.pdf - Supplemental Material Download (142kB) | Preview
Preview	Text Supp fig 2 - TSNE by feature.pdf - Supplemental Material Download (1MB) | Preview
Preview	Text Supp fig 3 - Cluster interrogation (hierarchical)_17_02_22.pdf - Supplemental Material Download (87kB) | Preview

Abstract

Introduction: Cognitive and functional decompensation during acute illness in older people are poorly understood. It remains unclear how delirium, an acute confusional state reflective of cognitive decompensation, is contextualised by baseline premorbid cognition and relates to long-term adverse outcomes. High-dimensional machine learning offers a novel, feasible and enticing approach for stratifying acute illness in older people, improving treatment consistency while optimising future research design. Methods: Longitudinal associations were analysed from the Delirium and Population Health Informatics Cohort (DELPHIC) study, a prospective cohort ≥70 years resident in Camden, with cognitive and functional ascertainment at baseline and 2-year follow-up, and daily assessments during incident hospitalisation. Second, using routine clinical data from UCLH, I constructed an extreme gradient-boosted trees predicting 600-day mortality for unselected acute admissions of oldest-old patients with mechanistic inferences. Third, hierarchical agglomerative clustering was performed to demonstrate structure within DELPHIC participants, with predictive implications for survival and length of stay. Results: i. Delirium is associated with increased rates of cognitive decline and mortality risk, in a dose-dependent manner, with an interaction between baseline cognition and delirium exposure. Those with highest delirium exposure but also best premorbid cognition have the “most to lose”. ii. High-dimensional multimodal machine learning models can predict mortality in oldest-old populations with 0.874 accuracy. The anterior cingulate and angular gyri, and extracranial soft tissue, are the highest contributory intracranial and extracranial features respectively. iii. Clinically useful acute illness subtypes in older people can be described using longitudinal clinical, functional, and biochemical features. Conclusions: Interactions between baseline cognition and delirium exposure during acute illness in older patients result in divergent long-term adverse outcomes. Supervised machine learning can robustly predict mortality in in oldest-old patients, producing a valuable prognostication tool using routinely collected data, ready for clinical deployment. Preliminary findings suggest possible discernible subtypes within acute illness in older people.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item