Joint Research Conference
June 24-26, 2014
Medical device manufacturers are highly regulated. As part of the required due diligence, devices are tested multiple times throughout development. These tests include formal verification and validations tests. During pre-production and production runs, quality control (QC) data is produced. Companies end up storing “big data” test result data sets (billions or more rows of data). When the test data is combined in storage, it no longer meets the definition of experimental data, and therefore becomes observation data. Observational data is less desirable for traditional statistical analysis as the conditions of the data creation are not controlled. This paper asks if medical device development and QC test data can be conglomerated and used as observational data in data mining activities (such as recursive partitioning and neural networking) that can lead to meaningful predictive (regression) models. In order to answer this question, stored data from a medical device manufacturer will be analyzed with these data mining techniques. The results will be compared, and the paper will conclude with a recommended method for using recycled data to create predictive models, thereby adding value to data stored by medical device manufacturers.
Please note: Due to the proprietary and confidential data used in this study, data and specific values of outcomes will not be disclosed. Comparisons of results will be presented instead. The initial version of this paper was written as a project for Westminster College’s MBA 623 Simulation Modeling course in April, 2014. It has since been updated.
The author’s credentials may be reviewed on LinkedIn (https://www.linkedin.com/in/anitadalrymple) or by resume upon request.