טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
Ph.D Thesis
Ph.D StudentBoldo Irith
SubjectKnowledge-Based Recognition of Clinical-Guideline
Application in Time-Oriented Medical Records
DepartmentDepartment of Industrial Engineering and Management
Supervisors Dr. Opher Etzion
Professor Yuval Shahar
Full Thesis textFull thesis text - English Version


Abstract

Clinical guidelines prescribe a set of policies that embodies the state of the medical art at a certain point of time regarding a particular clinical problem. However, physicians often deviate from the original prescribed specifications, believing their own therapy to be closer to the intentions of the guideline designer.

Determining by which guideline the patient is or was most likely to have been treated would support a suggestion how to continue in the application at the point of care. Furthermore, the determination of the degree of adherence to the guideline specification would support runtime quality assurance and retrospective quality assessment. Another benefit of automated guideline recognition is to support clinical research, for example, when analyzing a set of electronic medical records (EMRs).

In practice, there is no explicit indication within the EMR specifying which guideline the physician has chosen to follow. It exists only in the mind of the care provider.

No rigorous methodology for generating a partial matching between a clinical guideline’s description and an EMR currently exists. Thus, the focus of this research is proposing a solution to that problem. The output of the computational framework developed here is a degree of fit between each given record and each guideline and a justification for that degree. The problem can also be viewed as a highly specific type of a classification problem.

The challenges to this research included the development of a new methodology for matching vague information in the guideline, to exact data within the EMR (e.g., blood-glucose level), and specifying the similarity functions. In addition, we developed an abstract knowledge representation format for clinical guidelines as vectors of attributes.

A new computational tool was developed for the evaluation of the methodology. The tool was evaluated on a set of 773 records of hypertensive patients. A clinical expert assisted us in the formal representation of several guidelines for management of hypertension. A significant correlation was found between the expert’s judgment on a subset of the EMRs and the tool’s score.

We conclude that it is feasible to represent in abstract fashion a set of clinical guidelines to support the task of generation of partial-match scores between an EMR and each guideline, although full-fledged quality assessment might require additional information. 

A highly beneficial effect of the development of the tool was the generation of a data set that can support multiple avenues of future clinical research.