We developed an innovative technology that examines the patients’ Electronic Medical Records (EMR), extracts the patterns of care for these patients, and clusters them according to a combination of pattern of care, the subtype of MRSA, and the outcome.
The methodology involves several data mining steps:
1) Building Patient Care Models - sequences of patient care events for each patient.
2) Building Patient Care Descriptors summarizing each Patient Care Model.
3) Clustering of Patient Care Descriptors to detect clusters with similar patterns of care.
The resulting clustering with 6 clusters of patients will be described in detail. These results underwent review by a physician who received a small percentage of patient care models. The goal of this medical review, performed in a masked fashion, was to group EMRs by apparent type and importance of MRSA infection. The physicians grouped the EMRs into categories such as probable hospital acquired infection, probable community acquired infection, immunosupressed patients, invasive MRSA infections, cutaneous MRSA infections, MRSA infection subsequent to significant trauma, and deaths. The categories were not mutually exclusive. Once they had categorized all of the patient care models, we revealed the clustering results. Our findings demonstrate the utility of our new, automated technology to examine EMRs and stress why the EMR is an important investment for the future of public health practice.