Data Mining Techniques in a

¶ … data mining techniques in a healthcare data warehouse" by Silver (et al.) deploys a case study approach to examine how data mining can be used in clinical support systems to reduce costs in an increasingly cash-strapped healthcare environment. The specific focus was upon unprofitable DRGs [diagnosis-related groups] and attempted to discern the negative financial impact of specific types of Medicaid and Medicare patients upon a facility (Silver et al. 2001, p. 159). The study utilized online analytical processing (OLAP) tools to facilitate effective data mining techniques.

Two techniques for data mining were involved: patient rule induction method (PRIM) and weighted item sets (WIS). The latter is a type of statistical associative technique used to find relationships between seemingly unrelated sets of data. " for example, ninety out of one hundred patients in the database with DRG '999' have a length of stay greater than or equal to ten days" (Silver et al. 2001, p. 157).

WIS is used to find general associations, and thus it can be difficult for computer neophytes to use because the conclusions are not self-evident, while PRIM directly "calculates a subset of data points with the highest average value, given a set of input attributes and a minimum size of the region specified by the user" (Silver et al. 2001, p. 157-158).

No specific focus on the types of relationships study was determined before the analysis; rather the utility of relationships for future business decisions were to be determined after the data was amassed. However, according to the authors, finding unexpected relationships can often be more fruitful than pre-determining what types of associations to make with the data sets.

The central problem, as determined by preliminary PRIM analysis using "loss" as the output variable found that the average loss associated with the inpatient visits in the hospital region was seven times larger than the inpatient visits outside the region and average length of stay for the inpatient visits in the region was twice as long as stays outside region (Silver et al. 2001, p.160).

Results also indicated that patients between the ages of forty-six and sixty-four incurred significantly larger financial losses for the facility than the rest of the patients and Medicare and Medicaid nonexempt patients admitted via routine admission had the highest average loss for inpatient visits. Also, on Tuesdays, the average loss per patient was significantly greater for patients in the region. Only with data-mining could such an apparently insignificant finding as the predominance of Tuesday admittances come to light.

The reason for this was not random: "Examination of Tuesday's admitting physicians" revealed that "several of the physicians were in the same medical specialty. This specialty cared for patients that typically required a high level of service intensity over a long period of time. The identification of a subset of patients with disproportionately high costs has prompted the institution to reevaluate its admission criteria to this unit" and also the diagnostic approach of the physicians in question (Silver 2001, et al. p. 160).

One potentially negative effect of this case study is that it merely targets which patients are.