1565 Abstracts

P74 DATA OPERATIONS MANAGEMENT WITH CLINICAL DATAFAX:

THE PROSTATE CANCER PREVEWI’ION TRIAL (PO EXPERIENCE

Anne M. Ryan, Lynda Emel and Phyllis Goodman Southwest Oncology Group

Seattle, Washington

The PCPT is a chemoprevention trial of healthy men whose primary objective is to test the difference in the biopsy-proven prevalence of prostate cancer between a group of participants treated with finasteride and a group treated with placebo. All 18,000 randomized participants will be followed quarterly for 7 years at 221 Study Sites. Data documenting follow-up are received, entered and quality-controlled (QC) at the PCPT Statistical Center via Clinical DataFax, an image-based data management system using FAX technology and Optical Character Recognition (OCR) to create data records from faxed pages of data. For 7 years of follow-up an estimated 243 pages will be received per participant, or 4.5 million pages total. An average of 12,500 faxed pages are received weekly.

PCPT Data Operations (DO) is organized to support the timely movement of data records or pages through required review levels of DataFax. This organization was prompted by the rapid accrual of a large number of participants over a short time period, and is required to support the weekly receipt of large volumes of faxed pages. Staff assignments are determined by volume of pages received at the Statistical Center and/or accrual at the Study Sites.

Data Entty Qnerators (DE01 complete validation of key fields on each page, i.e., page #, ID#, contact#, date. This limited data entry supports the identification of a large volume of faxed pages in a short time period, and minimizes unnecessary requests for missing pages. A DE0 can validate > 25,000 pages monthly.

Data Control Techs (DCT) verify data in key fields, correct any OCR-generated errors or omissions, enter pertinent narrative data, and perform an abbreviated QC review. A status code indicating the data quality if applied to each page. Each DCT can validate > 7000 pages per month.

Data Coordinators (DC) tinalize the review of data for eligibility, as well as for adequate and accurate documentation of adherence, toxicity, endpoint occurrence, and general compliance to protocol and procedures. A DC can validate > 6000 pages per month.

This organization of PCPT DO is basic to the success of the data management of this large clinical trial. It permits timely access to data needed for QC review, reporting, and required analyses.

P75

CHALLENGES IN IMPLEMENTING A TELEPHONE TOUCH-TONE RANDOMIZATION SYSTEM IN THE PRIMARY CARE SETTING

Susan Anthony, Stephanie Reece, Scott Rushing, Susan Margitik, and Cheryl Albright, for the ACT Investigators

Bowman-Gray School of Medicine Wake Forest University

Winston-Salem, North Carolina

Touch-tone telephone randomization systems are an easily accessible and cost-effective method of randomizing participants in multicenter clinical trials. In the Activity Counseling Trial (ACT), health educators utilize one such system in the primary care setting to randomize eligible participants into the study.

Abstracts 157s

ACT is among the first studies to evaluate the efficacy of primary care-based physical activity behavioral interventions. Three U.S. clinical centers each have multiphysician-based practices where ACT health educators deliver various interventions messages to promote increased physical activity in sedentary participants. Immediate accessibility of the system by phone allows the health educators to remain independent of clinical measurement sites and requires no action by the Coordinating Center when randomizii participants. A special access code is utilized by the health educators which verifies eligibility based on the data collected at the measurement site. and provides stratification information to the Coordinating Center for randomization. Assignment of treatment arm is provided to the health educators usually within thirty seconds. Methods have been developed to provide confirmation in both the Coordinating Center and the clinical measurement sites when participants have been randomized.

A manual back-up system for randomization has been developed by the Coordinating Center for use by the health educators when problems with the system are encountered. Among some of the system problems encountered include system disconnection and invalid access code generation. Nonsystem problems include rescheduling of patient/physician appointments and delays in data entry. Thus far, the manual system has been utilized in less than two percent of total randomizations.

This presentation will illustrate in more detail the problems and solutions, and advantages and disadvantages in implementing a touch-tone randomization system in the primary care setting.

I76 STATISTICAL ISSUES IN SYSTEMATIC REVIEWS OF CHRONIC DISEASES

Deborah Ashhy, Carrol Preston, Katherine Cheng and Rosalind Smyth University of Liverpool

Liverpool, England

The Cochrane Collaboration aims to review systematically evidence from randomized controlled trials categorized by speciality and sub-speciality. Statistical methodology is well- established for meta-analysis of trials with binary outcomes or for continuous outcomes measured once. However, for chronic diseases, these measures have severe limitations. The appropriate analyses for single trials can be addressed using methods for longitudinal data, although in practice results form most trials have been presented using more naive methods.

Cystic fibrosis is an example of a chronic disease, where interest often focuses on measures such as FEVl, which is a measure of lung function which predicts longer term prognosis. The Cochrane Cystic Fibrosis Group has identified 350 trials, and is now undertaking reviews of specific interventions such as the use of oral steroids. Using these reviews for illustration, we will firstly discuss the choice of outcomes measures for evaluating therapies for chronic diseases, such as absolute level or change over a period, and then outline problems that face those attempting reviews. Directions for solutions will be suggested. These include summary statistics which can be extracted or imputed from trial reports, and implications for the reporting of individual trials to facilitate systematic reviews in chronic diseases. Appropriate models for a me&analysis with individual patient data will be briefly reviewed, to give a framework for evaluating how far it is possible to replicate results using summary statistics only.