Ordinal data is prevalent in medical outcomes research; for example, gender, living condition, and use of assistive devices are all both critical explanatory factors in determining the efficacy of medical procedures and are nominal or ordinal data. One basic objective of this study was to compare four assignment (ordinal analysis techniques') prediction of expected patient resource requirements in a specialty ambulatory (outpatient) health care setting. Data from 2427 patient discharges from 7 specialty low vision clinics were collected. Biographical and discharge characteristics of patients were used to develop homogeneous patient groups on the basis of resource use. Resource-use features were then stripped from the data. The four ordinal analysis techniques were subsequently applied to the reduced data set to predict iso-resource group membership for each patient in the data. Chance criterion was used as a benchmark in gauging the predictive ability of each ordinal analysis technique. Prediction results obtained were clinic-specific. This may largely be explained by the fact that the initial iso-resource groups were unique to each clinic and no meaningful iso-resource groups could be obtained from combined data across clinics. No technique was found to be universally superior at all clinics, however, each technique's performance across clinics was consistently better than the benchmark. Contrary to initial expectations, neural networks, in some cases, significantly underperformed the more traditional techniques.
- Health Care Resource Utilization
- Ordinal Analysis Techniques
ASJC Scopus subject areas
- Discrete Mathematics and Combinatorics
- Applied Mathematics