Purpose
The purpose of the project is to visually depict the utilization pattern for various procedure codes contained in the CMS data for year 2012 released in 2014. Utilization analysis is a signficant topic in health care due to the rising healthcare cost and the need for resource allocation. In this environment, physicians are being profiled with increasingly sophisticated altgorithms. Advances made in the field of data anlayics can be used to proactively perform such analysis by physicians.
Methods
K-means clustering algorithm was applied to the CMS dataset which was was stored in a MYSQL database. A query interface was created utilizing PHP, MYSQL and various JAVASCRIPT libraries including Jquery, Bootstrap and Angular. The user was presented with a list of Ophthalmology codes from which a dataset could be downloaded in .csv format. In order to determine the number of clusters present, a scree plot was performed uilizing R statsitical software package. The "elbow" of the plot was determined and utilized as the starting point for the k-means clustering algorithm. This dataset was then presented graphically using scatter plot along with plotting of the mean and median. The parameters examined evaluated raw number of procedures, raw number of patients and the ratio of treatments per patient. This technique was applied to intravitreal injections given its financial and health impact on the society. However, it can be applied to any code present in the dataset,
Results
The results demonstrated a two clustered group based on the scree plot. The distribution of the procedure quantity was distributed with a power law distribution with the distribution plot showing significant outliers. THe treatment ratio demonstrated a normal distribution.<br /> <br /> Correlation test demonstrated a correlation between number of treatments and number of patients. However, no other correlations were seen.
Conclusions
Significant variation exists in the utilization pattern. No signiciant correlation was seen except between the number of treatments vs. number of patients. This study was a demonstration of a technique for analyzing utilization data. . As such, the portal created to perform this task as well as the initial description of a code was successful.<br />