FDA Perspectives: An Initial Report of CDER's Recall Root Cause Research Project (Part II)

The authors discuss how their research will help FDA in its identification of areas of emphasis in pre- and postapproval evaluation of products and processes.
Jan 02, 2011
Volume 35, Issue 1

This is the second in a series of papers reporting on the Recall Root Cause Research (RRCR) project. The first article in this series provided an overview of the project and was published in Pharmaceutical Technology in August 2009 (1). These papers describe retrospective research on drug-product recalls performed by the US Food and Drug Administration's Center for Drug Evaluation and Research's (CDER) Division of Manufacturing and Product Quality (DMPQ) in collaboration with other CDER components including the Division of Compliance Risk Management and Surveillance, and Office of Pharmaceutical Science. DMPQ introduced the project and its objectives at the FDA/PDA Joint Regulatory Conference on Sept. 9, 2008, in Washington, DC (2).

Quality defects in pharmaceutical products may have no effect on consumers' health or can have effect ranging from a reversible adverse event to the gravest consequences. The consumer expects every unit of a drug product to be safe and effective, and it is therefore essential for a firm's daily operations to be robustly conducted so that each batch uniformly meets that standard.

The principle objective of the RRCR project is to retrospectively assess why significant defects happen. More specifically, our research endeavors to answer the following questions:

  • Which recalls have the greatest impact on patients relative to safety, efficacy, and availability?
  • What appears to be the manufacturing and quality root causes of drug recalls?
  • What can we learn by looking at recall case studies and patterns in summary statistics?

To answer these questions, DMPQ collected information on Class I and II1 drug recalls and selected defect types, and then compiled data for each recall by searching several databases.


The project had two especially complex aspects. First, the nature of the data posed a number of challenges. Most of the information in the databases studied was free-form text that ranged from several words to several thousand words. As a result, significant and time-intensive data mining was required. The text material included a combination of information supplied by the recalling firm and FDA investigation reports. This resulted in a mix of data, facts, symptoms, reasons, and hypotheses mingled with possible root causes. In addition, because the information used was primarily text, the same topic or concept could be expressed in many different terms. These terms were then used to search the data. For example, there are several ways to say a product lacks sterility assurance, positive sterility, microbial contamination, lack of sterility, contains bacteria, and sterility failure. To facilitate the data-mining process, the different terms were consolidated into a smaller list of free-form fields. These fields were then converted into fixed-form fields for tabulation and into case studies for illustration of the root causes.

It is important to note that some events leading to recalls are not inadvertent mistakes or errors but intentional fraud by unscrupulous entities. For example, FDA undertook sample collections of Chinese herbal over-the-counter (OTC) products. The FDA sampling program found some drugs had an undeclared active pharmaceutical ingredient (API) that enhances male performance. The undeclared API was the same as or similar (analog) to the active ingredient in approved drugs. A large percentage of Class I recalls of OTC and unapproved new drugs have occurred due to this intentional act of using illegal ingredients in dietary supplements. In contrast, most other recalls have been due to substandard product and poor process design or operational errors.

Study populations. The study population includes products regulated by CDER that have been recalled.

  • For Class I recalls, the population of interest comprises recalls for the period Dec. 7, 2000, through Sept. 12, 2008.
  • For Class II recalls, the authors studied a smaller population, ranging from Jan. 1, 2006, to Dec. 31, 2007.
  • For several defects selected for special focus, the authors again studied over a longer period, in this case spanning Jan. 1, 2000, to the data collation date (in 2008 or 2009).

Data acquisition. For purposes of this project, data acquisition means the collection of numerical and non-numerical data from the relevant databases and sources. Given the nature of the information collected, the primary activities were extracting the material, summarizing free-form text from the sources, and entering it into a data set for analysis. The recalls evaluated are considered the complete populations of interest, not samples from larger populations. Thus, sampling variability is not considered.

lorem ipsum