Exploratory Data Analysis in Observational Data Utilizing Machine Learning--Based Approaches

• Activity Number: 15
• Type: Topic Contributed
• Date/Time: Sunday, August 9, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Andrew Bate*
• Companies: Pfizer Inc.
• Keywords: machine learning ; exploratory data analysis ; safety surveillance ; real world data ; data mining ; big data

Abstract

Confirmatory data analysis through Epidemiological analysis of Real World Data is a well-established and core component of the drug lifecycle. Exploratory data analysis (EDA), particularly the search for complex patterns, is a much more undeveloped research area. Examples of machine learning for EDA includes the use of a recurrent Bayesian neural network and Bayesian clustering algorithm based on mixture models and the Expectation Maximization algorithm for unsupervised pattern recognition within haloperidol reported outcomes. Bayesian shrinkage regression and association rule analyses approaches for outlier detection have also been used. Analytics performance has been limited by sparse data capture in structured data. A Natural Language Processing approach for extracting concepts from free text on Electronic Medical Records can improve the timeliness and accuracy of Acute Liver Disease detection compared to structured data alone. There is great potential for further advances in the use of EDA in observational data, and machine learning algorithms, with strategies focussing on the process of knowledge discovery and interpretability and usefulness of outputs can contribute to this.