077-31: Data and Text Mining the Electronic Medical Record to Improve Care and to Lower Costs

This analysis uses data-mining techniques on an electronic medical record in the Emergency Department of a hospital to improve care while lowering costs. All patients' records for a 6-month period were examined, and records of those patients who had an initial complaint of shortness-of-breath were extracted. The data-mining techniques of transactional time series in the HPF procedure and the association rules in SAS® Enterprise MinerTM were used to examine the data. Patients' orders, medications, and complaints were also examined using SAS® Text Miner to investigate relationships among the variable categories. The Association Node in SAS Enterprise Miner is applied to one target variable that uses a patient identifier to link orders, medications, and charges. Unfortunately, the Association Node is inadequate when there are too many choices for each target; it cannot relate different target values to each other. An alternative method is to change the observational unit to the patient by using the TRANSPOSE and CONCAT procedures. In this way, all patients' orders, medications, and changes are linked in text strings that can be examined and compared using SAS Text Miner. It was discovered that patients with similar complaints were treated very differently depending on the attending physician, and those differences can impact both costs and care in a hospital Emergency Department. INTRODUCTION The introduction of the electronic medical record allows for the examination of cost and quality in healthcare. Large numbers of patient records can now be examined in a hospital or private practice setting, and patients can be tracked longitudinally through the healthcare system. In the past, data mining techniques have rarely been used to examine the totality of patient records; matched cohorts have been preferred to examine specific hypotheses. This paper will examine physician decision-making to determine whether there is a general consensus as to the treatment of patients in the Emergency Department. All electronic records for a six month period were examined using SAS Enterprise Miner and SAS Text Miner. It is also the purpose of this paper to examine changes in diagnosis from initial complaint to final outcome in the hospital emergency department. In the Emergency Department (ED), those patients who are determined to have emergent conditions are seen before patients seen as having urgent or non-urgent conditions. Generally, the triage values of emergent, urgent, and non-urgent are assigned based upon the initial complaint. Patients identified as nonurgent can become emergent once the final diagnosis is made. EXAMINATION OF COMMON PROBLEM IN THE HOSPITAL EMERGENCY DEPARTMENT Chronic obstructive pulmonary disease (COPD) was chosen for examination. Over the six-month period, approximately 53 patients were diagnosed in the ED with COPD. Table 1 shows triage values, where patients were categorized as non-urgent, urgent, and emergent upon entry into the ED. Table 2 shows the relationship between the triage level and the patient disposition after treatment in the ED. Table 1. Patients Diagnosed with COPD as Triaged Upon Entry into the ED Level of Urgency Number of Patients Non-Urgent 12 Urgent 38 Emergent 3 Table 2. Relationship of Initial Triage to Final Disposition Patient Disposition NonUrgent Urgent Emergent Discharged Home 6 4 0 Admit for 23-hour Observation 4 22 3 Admit as Inpatient 1 7 0 Admit to ICU 0 5 0 Note that 72% of the patients were triaged as urgent while less than 10% were emergent. Since all 3 emergent patients were admitted for observation, it is clear that this must be a standard protocol. Note, however, that almost 50% of those initially classified as non-urgent were also admitted for observation. Data Mining and Predictive Modeling SUGI 31