The goal of the Probability Sampling Study (P.S.S.) is to evaluate risk factors that contribute to the state of regulatory
compliance for medical device clinical investigations. The results of the P.S.S. can be used to direct limited FDA resources
into areas where there may be enhanced concern for human subject protection and/or clinical data quality systems. The data
from the probability sample can be extrapolated to the entire universe of medical device clinical investigators.
FDA inspected 200 randomly selected clinical investigators as part of the P.S.S. between August 4, 2003, and March 20, 2007.
Ultimately, FDA issued eight Warning Letters (WLs)1 to clinical investigators. At the eight investigator sites receiving WLs, a statistical analysis of variables demonstrates
a strong correlation between: the site being inspected for the first time by FDA, investigators that did not receive sufficient
training on the conduct of the study, and the site being monitored by the sponsor of the investigation.
This correlation and relationship to WLs supports the need to increase education and training for both clinical investigators
and sponsors.
BackgroundIn January 2001 the FDA's Center for Devices and Radiological Health (CDRH) Bioresearch Monitoring Division (BIMO) and the
Office of Surveillance and Biometrics (OSB) applied for and received additional needs funding to develop the Probability Sampling
Study (P.S.S.) for the Center's clinical investigator inspection program.
BIMO implements its device investigator inspection program in several ways. One of the approaches involves the inspection
of clinical investigators once the study is completed and the Premarket Approval Application (PMA)2 is submitted to the FDA. This type of inspection generally focuses on data verification.
A second approach, the early intervention program, is designed to inspect clinical investigators of on-going Investigational
Device Exemptions (IDEs). Early intervention inspections focus on human subject protection, protocol adherence, and data quality.
CDRH BIMO believes that early intervention inspections will result in higher quality submissions through the identification,
correction, and prevention of regulatory deviations.
The P.S.S. focused exclusively on inspecting clinical trial investigators that were conducting on-going IDEs through the early
intervention program.
Meaningful data
Probability sampling is defined as any valid method of sampling that uses some form of random selection. A sample analysis
is done after random selection to draw well-founded conclusions, which are representative of the whole population. The P.S.S.
data can be used to assess the device research community's compliance strengths and weaknesses.
Using the identified trends, BIMO believes that the device research community will be better equipped to identify and implement
improvements to their Good Clinical Practice (GCP) systems.3 BIMO has also used the information gathered from the P.S.S. to develop educational outreach for clinical investigators to
focus on the areas of greatest need.
Further, FDA can potentially assess the effectiveness of "early inspections" by FDA field investigators on the quality of
device marketing applications to determine if the inspections reduced the need for additional BIMO inspections at the marketing
phase (i.e., PMA or 510(k) Submission).
Method
CDRH estimated that a total of approximately 20,000 investigators are associated with the 1100 active IDE applications, based
upon the assumption that there are approximately 20 sites for every IDE. Westat, a statistical services firm that CDRH contracted
to draw the random samples, selected 158 out of the 1100 IDE files, and extracted the names and locator information of 974
physicians conducting the clinical investigations. The information gathered was entered into a database, which generated random
numbers for each investigator.
Procedure
A two-phase stratified random sampling, sometimes called proportional or quota random sampling, was used.
This sampling method divided the IDE submissions into homogeneous subgroups and then took a simple random selection from each
subgroup. The first phase involved stratifying the IDEs by two subgroups: device area and participating clinical investigators.
A sampling of IDEs within each device area was selected to ensure that the probability sample was allocated across all types
of medical devices in a proportionally appropriate manner. A sampling of investigators was selected to ensure variability
control in the estimates if investigators are homogeneous within an IDE.
Figure 1
|
The second phase sample was conducted after all investigators within the sampled IDEs were selected. The number of investigators
was stratified by region and different sampling rates were applied in each region to equalize the workloads for the FDA inspectors.
The sample was stratified geographically between 19 district office locations (see Figure 1) and 11 device types. Ultimately,
the sample encompassed 74 IDE applications and 200 investigators.
The inspectional assignments issued as part of the random sample were identical to all other normal BIMO inspectional procedures
in order to appropriately compare the results from the random sample to the results of the nonrandomly sampled inspections
conducted during the same time period.
