Checks of Case Record Forms Versus the Database for Efficacy Variables When Validation Programs Exist
- PDF / 429,201 Bytes
- 7 Pages / 504 x 719.759 pts Page_size
- 95 Downloads / 209 Views
0092-8615199 Copyright 0 1999 Drug Information Association Inc.
Printed in the USA. All rights reserved.
CHECKS OF CASE RECORD FORMS VERSUS THE DATABASE FOR EFFICACY VARIABLES WHEN VALIDATION PROGRAMS EXIST DAMIANJ. MCENTEGART, BSc, MSc, SIMONP.JADHAV, BSc, MSc, PHD, J. CHANNON, MA, DIPMATHSTAT TANYABROWN,BSc, AND EDWARD Statistics and Data Management, Knoll Pharmaceuticals, Nottingham, United Kingdom
It is common practice in the pharmaceutical industry to perform checks of the computer database for a clinical study against the data on the case record forms. This check generally encompasses primary efficacy variables which are also the subject of programmed validation (edit) checks. The value of such checks for efficacy variables when validation programs exist is investigated under certain assumptions and conditions including a range of different data error rates and sample sizes. Simulations are used for smaller samples and an exact formula is derived from characteristic function theory for larger samples. Even in the face of high data error rates, the value of database checks is notjustified for moderate or larger sample sizes. For data error rates typically observed in the industry, the loss of study power is minimal. Key Words: Database check; Validation check; Error rate; Mixture distributions
INTRODUCTION MANY COMPANIES AND contract research organizations (CROs) in the phannaceutical industry perform a check of the computer database for a clinical study against the data on the case report forms (CRFs) (1,2). This check may be part of quality control by the data management team or part of an independent audit for quality assurance. An unpublished mail questionnaire of European companies including CROs was organized by the Association of Clinical Data Management in 1994. Checks of CRFs against the database were performed in 22 out of the 23 companies which responded. Sixteen compaReprint address: Mr. Damian J. McEntegart. Statistics and Data Management, Knoll Pharmaceuticals, Newland House, 49 Mount Street. Nottingham, NGl 6BP, United Kingdom.
nies performed checks for all studies, one for pivotal studies only, and five for randomly selected studies. Eight companies performed checks on an ongoing basis, six as a final data management activity, five did both of the aforementioned, and three did the checks at other times (presumably after breaking the randomization code). At least 13 of the companies checked the fields relating to the primary efficacy variables (note: the questionnaire was not designed to elicit this information and thus, this figure may be an underestimate). Data were checked on a sample basis in the majority of instances. Fifteen of 21 responding companies worked to defined error rates and only one of these companies took no action if these error rates were exceeded. Prior to the investigation reported here, Knoll’s practice at its United Kingdom site was to perform a 100% check of the data
I01
Downloaded from dij.sagepub.com at MCMASTER UNIV LIBRARY on April 8, 2015
102
D. J. McEnteg
Data Loading...
