ORACLE BRIEF

Early Automation of Data Mining at the

FDA using Oracle Health Sciences
Empirica Signal and Study

The U.S. Food and Drug Administration (FDA) has published a white paper titled
Data Mining at FDA. The paper summarizes past and current data mining
activities at the FDA and addresses data miners in all sectors anyone interested
in the safety of FDA-regulated products, and those with a general interest in FDA
activities. The topics covered in the paper include:

routine and developmental data mining activities

short descriptions of the mined FDA data

advantages and challenges of data mining at FDA

future directions of data mining at FDA

This brief references the content of the white paper and draws upon the findings
specifically the FDAs use of Oracle Health Sciences Empirica Signal and Study
solutions in their data mining activities. The full white paper can be downloaded from
http://www.fda.gov/ScienceResearch/HealthInformatics/ucm446239.htm.

Data Mining What If?

Data Mining in real-time has been the ideal for pharmacovigilance professionals since the
advent of safety databases. However, this has been constrained by technology and labor
requirements. What if technology could be upgraded along with the near elimination of labor
resources? Then we would have an automated data mining system. Such a system is in its
early stages at the FDA using Oracle Health Sciences Empirica Signal and Study.

Data Mining at the FDA

The FDA Data Mining Council

promotes the improvement of
data mining to support FDAs
mission of protecting and promoting
public health.
It will be important for the FDA to
structure its IT systems so that data
can be submitted, retrieved,

In response to the need to develop an FDA-wide data mining collaboration and strategy, the

processed, and evaluated in a

FDA Data Mining Council (DMC) was formed in 2007. The DMC is collaborative and explores

standardized manner.

methods and best practices recommended by experts from other federal agencies, industry,
and academiaall of whom have analogous experience in knowledge discovery through
various data mining approaches. The Council serves as a forum for FDA scientists to share
their experiences and challenges in analyzing data contained in the vast databases the FDA
maintains to discuss new methods for such analyses. The FDA currently receives
approximately two million adverse event, use error, and product complaint reports each year
from consumers, healthcare professionals, manufacturers, and others. Since the early 1990s,
the FDA has advocated data mining to the industry in an effort to better understand the signals
within the safety data. Now, the FDA data mining experts have expanded their attention to
adding more sophisticated data mining methods and applying data mining to other types of
product safety-related FDA and non-FDA databases.
The Proportional Reporting Ratio (PRR) is the foundational concept for many disproportionality
methods. However, because this method does not adjust for small observed or expected

Regardless of the analytical tools used,

visualization of the data is paramount.
Because of the volume and complexity
of the data, extremely helpful graphical
tools used at FDA include heat maps
and sector maps.
SOURCE: DATA MINING AT FDA
WHITE PAPER

ORACLE BRIEF

numbers of reports of the product-event pair of interest, other more advanced statistical
methods are employed, such as the Multi-Item Gamma Poisson Shrinker (MGPS), which
produces Empirical Bayesian Geometric Mean (EGBM) scores. Several FDA Centers, including
CDER, CBER, and CFSAN, use the MGPS algorithm for their routine surveillance activities.
Various commercially available software programs generate PRR and/or EBGM scores, e.g.,
Oracle Health Sciences Empirica Signal. Oracle Health Sciences Empirica Signal is utilized by
the FDA for routine mining of drugs, foods, cosmetics, and dietary supplements, as shown in the
table below.
Product type

Drugs

Foods,
Cosmetics, and
Dietary
Supplements

Database features as of Spring 2014

Data mining method

Current #
reports
received
770,000 in
2013

Database
start date

Cumulative
# of reports

Stage of
use

Method or
tool

1968

>7,000,000

Routine

6,000 in
2013

2002

40,500

Routine

MGPS
with
Empirica
Signal
MGPS
with
Empirica
Signal

KEY POINTS

The FDA has recommended the use

of data mining to the drug industry

FDA data mining experts have

expanded their attention to adding
more sophisticated data mining
methods and applying data mining to
other types of product safety-related
FDA and non-FDA databases

CDER has applied software

packages, including Oracle Health
Sciences Empirica Study, to analyze
drug clinical trial data in either New
Drug Applications or supplemental
applications

Oracle Health Sciences Empirica

Signal is utilized by the FDA for
routine mining of drugs, foods,
cosmetics, and dietary supplements

Encouraged by the success of using data mining methods for safety report analysis, FDA
experts have started to apply the techniques to other types of data.
Type of
data
Clinical
study data
in drug
applications

Stage of use
of data mining
Routine

Data mining method or tool

Empirica Study creation of a wide
set of automatically generated
analytical outputs and tailormade, reusable tables and
graphs

Data mining
purpose
Save reviewers
from having to
create the tables
and graphs

CDER has applied Oracle Health Sciences Empirica Study to analyze drug clinical trial data in
either New Drug Applications or supplemental applications. Oracle Health Sciences Empirica
Study interfaces with data that conforms to the standardized Study Data Tabulation Model
(SDTM) of the Clinical Data Interchange Standards Consortium (CDISC) data standards to
create a wide set of automatically generated analytical outputs and tailor-made, reusable
tables and graphs. These outputs have helped reviewers to more efficiently analyze potential
safety issues in the clinical trial data of drugs approved by the FDA. The FDA notes the
benefits of data mining with these tools in the areas of standard processes (because data
mining is automated, the outputs are statistically objective and devoid of manual analyses),
simultaneous analysis (across an entire database at once), efficiency (analyses computed in
minutes), and the benefits of automated signal investigations (transparency with audit trails,

CONNECT W ITH US

oracle.com/healthsciences

drill-down capability, observation of signals over time, and study of a product in populations).
healthsciences_ww_grp@oracle.com

These tools are in the early stages of the automation efforts at the FDA. In the meantime, until
technology advancements and standardization practices proliferate, hands-on case reviews,
analysis of other data sources (e.g., FDA regulatory databases, the World Health Organization
drug safety report database, public scientific literature, and public knowledge databases) and
further epidemiologic assessments are necessary to characterize the clinical and public health
significance of signals generated by data mining analyses.

youtube.com/user/oraclehealthsciences
facebook.com/oraclehealthsciences
twitter.com/oraclehealthsci
blogs.oracle.com/health-sciences

Source: http://www.fda.gov/ScienceResearch/HealthInformatics/ucm446239.htm
To find out more and to access the full report, please click on the above link.

FOR MORE INFORMATION

Contact: 1.800.633.0643

Copyright 2016, Oracle and/or its affiliates. All rights reserved. Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names
may be trademarks of their respective owners. Intel and Intel Xeon are trademarks or registered trademarks of Intel Corporation. All SPARC trademarks
are used under license and are trademarks or registered trademarks of SPARC International, Inc. AMD, Opteron, the AMD logo, and the AMD Opteron
logo are trademarks or registered trademarks of Advanced Micro Devices. UNIX is a registered trademark of The Open Group. 0116