Paper
Saturday, July 16, 2005
Using Clinical Data to Generate Expert Knowledge and Build Decision Support
Kathy H. Bowles, RN, PhD1, John H. Holmes, PhD2, Matthew Liberatore, PhD3, Mary D. Naylor, RN, PhD, FAAN1, Robert Nydick, PhD3, Eric Heil4, and Kimberlee Clark, BA1. (1) School of Nursing, University of Pennsylvania, Philadelphia, PA, USA, (2) Department of Biostatistics and Epidemiology, University of Pennsylvania, Philadelphia, PA, USA, (3) Decision and Information Technologies, Villanova University, Villanova, PA, USA, (4) Systems Engineering, University of Pennsylvania, Philadelphia, PA, USA
| Learning Objective #1: State the steps in the expert knowledge discovery process to develop decision support |
| Learning Objective #2: Describe how to use a nursing conceptual framework to guide the knowledge discovery process |
Expert knowledge is needed to develop a decision support system for discharge referral decision making for older adults. This knowledge was elicited from a panel of clinical experts in nursing, medicine, social work, and physical therapy using a unique combination of knowledge acquisition and processing methods. No previous studies reported in nursing have used this combination of methods. The Orem Self Care Framework was used to organize the chart abstraction process from which over 200 case studies were generated. The case study sample was drawn from existing records from three clinical trials conducted with hospitalized older adults and from recently hospitalized patients. The case studies were sent for judgment to the panel of multidisciplinary experts to determine the need and the reasons for a post discharge referral. Consensus was reached on 72% of the cases in the first round and the reasons for referral or not was provided. For those cases without consensus on the decision or site of referral, three Web-assisted online Delphi rounds were conducted to achieve expert consensus. Each round resulted in agreement on approximately 25% more of the cases. The information was coded using an ontology of over 200 codes that represent the reasons developed through focus groups with the clinical experts and validated by the scholars. Using See5, a well-known decision rule inducer, we identified factors that were associated with a decision to refer to home care or not. Patterns emerged in the data that correspond with the domains of Orem's framework and identify the critical factors to determine the need for discharge referrals. Next steps are to continue to refine and validate the decision rules. These methods are useful to researchers who develop decision support systems or seek knowledge from experts for evidence based practice.