Values; and New Techniques in Clinical Pathology Research. Each talk has been documented as a reference to further advance our educational endeavors. In February 1992, we distributed the first edition of The Data Manager's Manual which was developed by experienced DMs. The manual is designed as a training guide for DMS and includes a general overview of specific topics of relevance to DMs throughout MH. Finally, we are in the process of developing a User's Manual for MH's Clinical Research DataBase (CRDB) and an Orientation-Training Program for new DMs. This program encompasses a broad array of services throughout MH. Services include: recruiting and training of new DMs upon request, separate computing and DM hotlines for assisting new DMs, logistical assistance in flow of data and locating data within MH, and evaluations of equipment and connections for setting up the new CRDB. The development of an EP has been an important step in the standardization and professionalization of DM practices at MH. In this presentation we will discuss the logical steps involved in developing an EP in addition to describing topics of the Grand Rounds, the Data Manager's Manual, the User's Manual and the Orientation Training Program.
P55 DESIGNING A REGISTRATION SYSTEM FOR THE
CLINICAL RESEARCH DATABASE SYSTEM
Q. Pan, D. Wu, M. Sun, C. Houston and C. Begg Memorial Sloan-Kettering Cancer Center
New York, New York
We have developed a computerized registration system as a component of our clinical research database at Memorial Hospital. The system is designed for registration of patients on research protocols, and for registering new protocols as they are activated. This presentation describes the functionalities of the system and the method of implementation using relational database technology. The registration system includes two major programs: the protocol registration program and the patient registration program. The protocol registration program is capable of 1) defining protocol- specific eligibility criteria, 2) defining a list/range of valid input for each eligibility criterion, 3) assigning an efigibility flag for each valid input of a criterion, 4) defining what the next criterion should be for each criterion based on the input value, 5) defining specific/sophisticated eligibility criteria involving calculations in a modular al>proach, 6) defining randomization stratification variables, 7) defining the randomization characteristics such as the block size and a flag indicating if the block size should be randomly generated, and 8) SUl~porting protocols which require multiple registration procedures. The patient registration program is capable of 1) allowing only those users who have appropriate access privileges to register the patient, 2) prompting the efi~.b.ility criteria defined through the protocol registration program, 3) determining the patient's eligibility after all criteria have been entered, 4) prompting the user to register the patient, 5) randomizing the treatment assignment if required, 6) allowing registration of previously randomized patients while maintaining treatment balance, 7) recording all attempts at patient registration, 8) ensuring a patient can only be registered to a protocol once. The entire registration system is implemented using a relational d~tA_base. In our presentation, we will provide details of this system, and describe the technical and practical challenges presented by the project.
P56 USING DATA MANAGEMENT STATISTICS AND SYSTEM
RESOURCES STATISTICS FOR QUALITY ASSURANCE
Nancy H. Remaley, William P. Amoroso, Kimberly C. Beringer, Gerald L. Swanson, Jeffrey P. Martin, Kevin L. Mowry,
David W. Burry, Emil A. Maurer and the Systems and Programming Group
University of Pittsburgh Pittsburgh, Pennsylvania
Statistics regarding d~_t~_ management and the use system resources are maintained by most, if not all, large data centers. These d~ta generally remain unpublished and unused. At the Epidemiology Data Center (EDC), a Vital Statistics Report (VSR) is published monthly by the Systems and Programming group to disseminate information captured by PARENT, the internally developed data management system, and the VAX/VMS accounting program. This report was formatted to showcase project related data for the fiscal year by month, by study, site, form, and in many cases, field level. The EDC is directly involved with the data collection and data management for over 10 clinical trials and registries and is the coordinating center for 5 multicenter trials. Rates are provided that show the usage over the total study record base. By providing this data regularly, study personnel can evaluate trends, identify problem areas, and provide a comparative view of an individual study's usage with
other studys' usage. The VSR is also a vehicle to communicate system updates, bug fixes, new features, new
developments in the computing industry, and new software to the user community. To relieve the tedium of so many numbers, comic relief is provided on the back page showing that systems people do have a sense of humor. The report also serves as a quick reference when informarion is requested for grants renewals or internal reports. The VSR provides a mechanism to distribute information regarding a study's use of system and data management resources which contributes to the efficient use of these resources.
!'57 BMDPLOT: PUBLICATION-QUALITY SURVIVAL
PLOTS FROM BMDP OUTPUT
S. Corley and K. Davis University of Washington
The BMDPLOT program produces publicarion-quality survival plots from line-printer out-put generated by BMDP survival analysis programs 1L and 2L on VAX VMS computers. The plots may be printed on any PostScript-type laser printer or on an HP 7550 pen plotter. The program is written in VAX FORTRAN with calls to the GKS (Graphical Kernel System) run-rime library of graphics functions. The computer on which the program is to be run must be equipped with GKS.
BMDP program 1L estimates a survivor function (i.e., survival curse) using either of two methods-actuarial life table or product-limit. BMDP program 2L performs survival analysis with covariates. BMDPLOT can produce five types of plots depending on the BMDP program 1L using the life table method, one type (cumulative survival) for BMDP program 1L using the product-limit method, and two types (Cox-Snell residual and cumulative hazard) for BMDP program 2L. BMDPLOT can process up to 99 problems (or subproblems) per BMDP job and up to six gr.oups and six strata per problem or sub-problem. BMDPLOT allows the user to specify various plot characteristics (such as plot rifle, axis labels, and axis lengths) or to accept the default characteristics determined by the program.
P58 CLUSTER SAMPLING TO ESTIMATE KEY ENTRY
Lesly A. Pearce Statistics and Epidemiology Research Corporation
Quality control in a clinical trial includes estimating the key entry error rate. The number of fields on a form, the number of expected records in the database, and also the error rate are dependent on the form type. We sampled 249 forms (5.2%) containing 9815 fields (3.8%) entered in a distributed data ...... entry__ system_ as _part of _quality control in the initial _Phase of the Stroke Prevention in Atrial Fibrillation (SPAF) study. Individual forms, or clusters of fields, were sampled randomly from within each stratum, or form type. Some or all forms of a given form type were sampled. Unbiased estimates of the error rate and variance were then computed using cluster sampling techniques within a stratum, weighting and summing stratum estimates to then compute the overall rate and variance. Estimation by strict random sampling would have required a fewer number of fields to be sampled (2.3 %), however, a much larger number of forms (approaching 100%) would have been solicited from the clinical sites. While simple random sampling does provide a basis for determining the number of forms necessary to sample within a stratum, cluster sampling is clearly a more efficient method for estimating the error rate and its variance.
i'59 A QUALITY ASSURANCE PROGRAM IN
CANCER CLINICAL TRIALS
Joyce Niland, Tamara Odom-Maryon and Christine Blevins City of Hope National Medical Center
The City of Hope Department of Biostaristics provides data management support for 200 cancer clinical trials. A Quality Assurance (QA) program has been implemented to monitor investigator compliance to protocols, and the quality of data collection. The Protocol Compliance Committee