PRISM | Institutional Repository
Communities in PRISM
Select a community to browse its collections.
Recent Submissions
Health Economic Evaluation of Antimicrobial Stewardship, Procalcitonin Testing, and Rapid Blood Culture Identification in Sepsis Care: A 90-Day Model-Based, Cost-Utility Analysis
(2024-11-19) Sligl, Wendy I.; Yan, Charles; Round, Jeff; Wang, Xiaoming; Chen, Justin Z.; Boehm, Cheyanne; Fong, Karen; Crick, Katelynn; Clua, Míriam G.; Codan, Cassidy; Dingle, Tanis C.; Prosser, Connie; Chen, Guanmin; Tse-Chang, Alena; Garros, Daniel; Zygun, David; Opgenorth, Dawn; Conly, John M.; Doig, Christopher J.; Lau, Vincent I.; Bagshaw, Sean M.
Abstract Objective We evaluated the cost-effectiveness of a bundled intervention including an antimicrobial stewardship program (ASP), procalcitonin (PCT) testing, and rapid blood culture identification (BCID), compared with pre-implementation standard care in critically ill adult patients with sepsis. Methods We conducted a decision tree model-based cost-effectiveness analysis alongside a previously published pre- and post-implementation quality improvement study. We adopted a public Canadian healthcare payer’s perspective. Two intensive care units in Alberta with 727 adult critically ill patients were included. Our bundled intervention was compared with pre-implementation standard care. We collected healthcare resource use and estimated unit costs in 2022 Canadian dollars (CAD) over a time horizon from study entry to hospital discharge or death. We calculated the incremental net monetary benefit (iNMB) of the intervention group compared with the pre-intervention group. The primary outcome was cost per sepsis case. Secondary outcomes included readmission rates, Clostridioides difficile infections, mortality, and lengths of stay. Uncertainty was investigated using cost-effectiveness acceptability curves, cost-effectiveness plane scatterplots, and sensitivity analyses. Results Mean (standard deviation [SD]) cost per index hospital admission was CAD $83,251 ($107,926) for patients in the intervention group and CAD $87,044 ($104,406) for the pre-intervention group, though the difference ($3,793 [$7,897]) was not statistically significant. Costs were higher in the pre-intervention group for antibiotics, readmissions, and C. difficile infections. The intervention group had a lower mean expected cost; $110,580 ($108,917) compared with pre-intervention ($125,745 [$113,210]), with a difference of $15,165 ($8278). There were no statistically significant differences in quality adjusted life years (QALYs) between groups. The iNMB of the intervention group compared with pre-intervention was greater than $15,000 for willingness-to-pay (WTP) per QALY values of between $0 and $100,000. In our sensitivity analysis, the intervention was most likely to be cost-effective in roughly 56% of simulations at all WTP thresholds. Conclusions Our bundled intervention of ASP, PCT, and BCID among adult critically ill patients with sepsis was potentially cost-effective, but with substantial decision uncertainty.
Hans Chinese consume less O2 for muscular work than european-american
(2024-11-21) Guo, Mei-Han; Montero, David
Annual report 2023-24, Alberta Gambling Research Institute
(Alberta Gambling Research Institute, 2024-11-22) Alberta Gambling Research Institute
Ontology-Enhanced Automated Machine Learning
(2024-11-20) Davies, Cooper T.S.; Denzinger, Jorg; Maurer, Frank; Jacob, Christian; Walker, Robert; Dick, Scott; Boyd, Jeffrey
This thesis addresses the challenge of bridging the gap between traditional Problem-Specific Machine Learning (PSML) and Automated Machine Learning (AutoML) systems. While PSML offers high accuracy but demands substantial expertise, AutoML aims to auto-mate the process of building a machine learning (ML) model but often lacks domain-specific knowledge. To address this, we propose Ontology-Enhanced AutoML, a novel approach that integrates domain knowledge from ontologies into the AutoML pipeline. We first examine the current landscape of AutoML, highlighting the complexities faced by a system in selecting appropriate algorithms and hyperparameters. We identify the limitations of existing AutoML systems, particularly their blind reliance on datasets, which often leads to poor performance and lengthy training times. Our thesis presents experiments demonstrating the effectiveness of Ontology-Enhanced AutoML in mitigating these challenges. By incorporating mechanisms for ontology-based feature extraction and example filtering, we demonstrate significant improvements in accu-racy and optimization time compared to traditional AutoML. These results highlight the potential of Ontology-Enhanced AutoML to provide a wide range of systems lying between the extremes of PSML and AutoML. This thesis contributes not only a technical solution but also a conceptual framework for understanding ML as a spectrum. We discuss implications for future research and the potential for further advancements in bridging the gap between domain expertise and ML proficiency.
Single-player to Two-player Knowledge Transfer in Atari 2600 Games
(2024-11-18) Saadat, Kimiya; Zhao, Richard; Abou-Zeid, Hatem; Aycock, John
Playing two-player games using reinforcement learning and self-play can be challenging due to the complexity of two-player environments and the potential instability in the training process. It is proposed that a reinforcement learning algorithm can train more efficiently and achieve improved performance in a two-player game by leveraging the knowledge from the single-player version of the same game. This study examines the proposed idea in ten different Atari 2600 environments using the Atari 2600 RAM as the input state. The advantages of using transfer learning from a single-player training process over training in a two-player setting from scratch are discussed, and the results are demonstrated in several metrics, such as the training time and average total reward. Finally, a method for calculating RAM complexity and its relationship to performance after transfer is discussed. Results show that in most cases transferred agent is performing better than the agent trained from scratch while taking less time to train. Moreover, it is shown that RAM complexity can be used as a weak predictor to predict the transfer's effectiveness.