| Clinical data warehouses accumulate large amounts of terminology-coded data. In addition to increased accumulation of data, higher data granularity, and longer time-spans, there is also an increasing demand for analysis of this data. For a nonexpert, the ability to analyze this data unaided is very limited. To address this problem, I developed an analytical framework that works with flowchart models which can be extended with modular external applications and executed on retrospective data. This framework was inspired by emerging workflow technology. Workflow technology offers several tools which support modeling, execution, and extensive analysis of IT or organizational processes. The three specific aims of this dissertation were to review workflow technology and its current use, develop an analytical framework which utilizes graphical, process-based modeling, called RetroGuide (RG), and evaluate this framework using a series of case studies and a formal, comparison evaluation study.; RG's graphical representation format facilitates a stepwise, procedural approach to formulating analytical tasks. It uses a single patient execution model, and it resembles a manual chart review methodology. RG models can model complex temporal conditions and utilize external data manipulation, statistical, or reasoning technologies. The representation format is split into two layers, a flowchart and a code layer, which improves collaboration of analytical team members. Reports generated automatically by RG allow advanced drill-down capabilities, show in detail the model's execution trail for each analyzed patient, and support iterative model improvements.; Within this dissertation, three analytical domains of quality improvement, decision support development, and medical research were explored. Seven case studies which utilize the Enterprise Data Warehouse (EDW) at Intermountain Healthcare are described (e.g., quality improvement problems in osteoporosis and cardiovascular patients, analysis of a computerized glucose management protocol, a problem in adverse drug event monitoring, or a research analysis of cancer patients). These case studies demonstrate RG's ability to support a wide range of complex analytical tasks, facilitate iterative exploration and review of electronic health record data, and provide a testing environment for retrospective simulation of analytical or decision support processes (using data from a real, large EDW).; Finally, a formal comparison study involving modeling analytical tasks in RG and Structured Query Language (SQL), and a qualitative study of RG are presented. The results suggest that RG's modeling approach is intuitive and easy to use, enables better modeling of the evaluated set of analytical tasks, and is preferred over SQL by a group of nonexpert data analysts. |
