416 Abstracts

A42 THE PUBLICATIONS AND PRESENTATIONS

COMMITTEE: WHAT MAKES IT WORK?

Susan E. Margiti~ Bowman Gray School of Medicine

Winston-Salem, North Carolina

The Publications and Presentations (P & P) Committee has assumed an important role in most multicenter clinical trials. Serving as a central clearinghouse for trial-related publications and presentations, the P & P Committee was designed to assure: 1) fair authorship representation among trial sites, 2) uniform terminology and study descriptions, and 3) opportunities for new investigators to become involved in the publication process. The P & P Committee additionally tries to 4) eliminate duplication or overlap of analyses and publication efforts, 5) facilitate realistic production schedules for the Data Coordinating Center to perform prioritized analyses and 6) maintain a high standard of quality for reporting trial results. To achieve these goals, a P & P Committee ideally should be chaired by an investigator with previous experience in collaborative trials who is wilting to take the time to ensure the smooth operations of the P & P process. The P & P policy involves developing guidelines for submission of paper proposals, prompt circulation of these proposals for clearance through the P & P Committee, generation of periodic status reports on trial papers in progress, solicitation of co-authors for recently approved papers as well as solicitation of proposals of new papers. This presentation will further elaborate on the role of the P & P Committee in a multicenter clinical trial and will describe the evolution of a paper through the Committee, from proposal status to acceptance for publication by a journal. Additionally, guidelines for order of authorship on a publication and special situations and actual dilemmas faced by the P & P process will be described.

A43 LONG-TERM SURVIVAL FOLLOW-UP IN

INTERNATIONAL CLINICAL TRIALS

Margaret C. Haugh, Catherine Cornu and Jean-Pierre Boissel Unit~ de Pharmacologic Clinique

Lyon, France

Trial organizers wishing to include long-term survival data as an endpoint in international clinical trials may encounter administrative difficulties since the methods to obtain this information varies from country to country. We have decided to review the procedures in as many countries as possible so that information concerning what patient information is mandatory, who should be contacted, if prior agreement is necessary and what level of information will be available, i.e., dead/alive on a given date, or date of death, or date and cause of death, can be summarized for each country. We wrote to scientists in various countries known to be involved in clinical trials and Health Departments or Ministries to ask for details of the procedures, if known, or the name of someone likely to be able to provide these details.

About 20 countries have been contacted, and a brief summary of the differences and similarities of the procedures will be presented. For example, in some countries (UK and USA) government departments run a fee-paying service to provide a copy of the death certificates for named patients. In other countries, enquiries to the town hall of birth or residence provides only information on the vital status of the named patients. Details of the patient data necessary to obtain information from different systems will be given. The results from this survey will help investigators planning international clinical trials.

A44 REDUCING PATIENT WITHDRAWALS IN A LARGE

MULTI-CENTER CLINICAL TRIAL

Jennifer Lee, Chris Granger, Ingrid Moffie, Lynn Woodlief, Kerry Lee

Duke University Medical Center Durham, North Carolina

To understand the magnitude of the problem of patient withdrawals in a large clinical trial and to minimize their occurrence, we collected information prospectively during the GUSTO Trial. GUSTO is a mortafity trial, with anticipated enrollment of 41,000 patients in 15 countries, comparing four thrombolytic regimens for the treatment of acute myocardial infarction. We collected information on patient withdrawals in North American only.

Abstracts 4 1 7

A patient not receiving any of the assigned therapy, in a situation where administering the assigned therapy was physically possible (i.e., drug kit available and patient alive) was considered a patient withdrawal. The reason for each patient withdrawal and the subsequent treatment were collected. Of 16,661 patients with CRFs entered, 3.1% did not receive any of the assigned therapy, and 1.8% were patient withdrawals. The most common reasons for withdrawal were physician preference for alterative therapy (21%), resolution of ECG or chest pain (indicating probable spontaneous thrombolysis) (18%), and patient of family refusal of participate (11%).

A concerted effort was made to reduce the number of patient withdrawals in early 1992. The following steps were instituted. A knowledgeablephysician was on call 24 hr/day through a telephone "hotline" enabling investigators to call and discuss their treatment concerns. During these calls, the "intention-to-treat" principle was explained and the importance of protocol adherence was emphasized; the investigators were often dissuaded from withdrawing the patient. When a patient was withdrawn without a "hotline" call, we followed up with the site, emphasizing the same principles. Reducing withdrawals was further emphasized in the GUSTO Gazette, a monthly newsletter sent to participating site personnel.

We have seen a decrease of patient withdrawals from the first to the second half of 1992, from 2.2% (175/8105)to 1.7% (143/8521), p=0.024. We conclude that on-line intervention and education of study site personnel can reduce the number of patient withdrawals.

A45 INTERIM ANALYSIS METHODS IN THE COLON

CANCER CONTROL STUDY

Timothy Church, Fred Edercr and Jack Mandcl Universily of Minnesota Minneapolis, Minnesota

The Colon Cancer Control Study, a trial of fecal-occult-blood testing as a screen for colorectal cancer, began in 1976, randomizing over 45,000 Minnesotans ages 50 to 80 to three treatment arms: annual screening, biennial screening and control. After the planned 5-year screening phase, the investigators extended the screening and follow-up with the goal of observing sufficient events to obtain the originally planned power against the original alternative. Data monitoring was begun using two frequentisit approaches for interim looks: group-sequential and stochastic curtailment. To accommodate uncertainty in the number of events from analysis to analysis, a constant spending function was used to formulate boundaries to reject the null hypothesis. Boundaries to "accept" the null hypothesis used stochastic curtailment based on the B-value given by Lan and Wittes. Stopping due to crossing one of these boundaries required adjustment of the estimates of effect based on the methods of Kiln and DeMets. In addition to these methods, "prior" beliefs about the effects of screening were elicited from the investigators before the first interim analysis, so that posterior distributions based on the d~_t~_ at each time of analysis show how the Bayesian approach compares to the frequentist approaches.

A46 ON CLOSED TEST PROCEDURES FOR

DOSE RESPONSE ANALYSIS

Dror M. Rom, Robert J. Costello and Laura Connell Rhtne-Poulenc Rorer Central Research

CollegeviUe, Pennsylvania

Clinical studies commonly involve the administration of several doses of a test drug and a control treatment. Experience with the drug and/or theoretical consideration may lead one to postulate a priori that the response should be monotonic with increasing doses. Although this is often true, there are situations when high doses of a drug are counter-productive and cause a down-turn in the dose response curve.

This presentation discusses a general procedure for analyzing dose response studies when either a monotonic or a down-turn in the dose response curve are anticipated. Our procedure, based on the closure method due to Perits (1970) provides information on all portions of the dose response curve while controlling the overall type-one error. It allows one to reach conclusions regarding those doses that are significantly better than the placebo, as well as those doses that are significantly better than other doses, thus enabling one to determine an 'optimal dose'. The procedure is illustrated on real data.