- Email: [email protected]
Sample sizes for constructing confidence intervals and testing hypotheses
Abstracts
345
rapidly ascertaining and documenting whether criteria for the diagnosis of Schizophrenia, Schizophreniform, or Schizoaffective disorders have been met. Employing a microcomputer based expert system development tool, modules for each of the relevant DSM-III-R categories were constructed. Employing a user-friendly interface, the computer displays each criterion and asks if it is met or whether no information is available. The n u m b e r of criteria that must be responded to before a diagnosis is either made or ruled out is markedly reduced by the expert system "inference engine" and an optimizing routine. A short written report is produced that documents the diagnosis, or a longer report can be generated which presents the diagnosis along with questions and answers utilized. The immediate utility of the system is that it: l) reduces the time necessary to arrive at a diagnosis; 2) is completely reliable in applying criteria to make the diagnosis; and 3) provides documentation of the specific criteria utilized.
Statistical Methods for Monitoring Recruitment in a Multicenter Clinical T r i a l B. M i c h e l e Melia, M a r i e D i e n e r - W e s t
The Johns Hopkins University, Baltimore, Maryland (P-30) Monitoring recruitment is an important task in a multicenter clinical trial. Two aspects of the monitoring process may include the assessment of interim recruitment goals and comparison of recruitment among the clinics. Interim recruitment goals are useful indicators of whether recruitment is proceeding according to schedule, or there are recruitment problems which could result in failure to meet the study's final goal. High recruiting clinics may be able to provide information regarding successful recruitment techniques; low recruiting clinics may use a disproportionate share of the study resources. Statistical methods which are helpful for objectively identifying clinics with significantly poorer or better recruitment than that of their fellow clinics, and for determining whether recruitment is adequate to meet a study's recruitment goals are discussed. The methods are illustrated using recruitment data from the Collaborative Ocular Melanoma Study (COMS) which has more than 30 participating clinics across the United States and Canada. A "Weighted" Rank Sum Test A l f r e d P. H a l l s t r o m
University of Washington, Seattle, Washington (P-31) A rank sum test is sometimes used w h e n ordinal and measured p h e n o m e n a need to be put on a common scale (e.g., when death interferes with a blood pressure measure). It can be argued that this approach does not give appropriate weight (e.g., should death have only the rank next to that of the highest blood pressure). A "weighted" rank sum test is proposed for which the relative weights are determined by expert panel and then scaled to provide a test sensitive to alternatives of interest. Development of an outcome measure for the Myocardial Infarction Triage Intervention (MITI) Trial is given as an example. In this study the outcome measure, left ventricular ejection fraction, is confounded by death and stroke. Sample Sizes for Constructing Confidence Intervals and Testing Hypotheses D a v i d R. Bristol
CIBA-GEIGY Corporation, Summit, New Jersey (P-32) Although estimation and confidence intervals have become popular alternatives to hypothesis testing and p-values, sample sizes for randomized clinical trials are usually determined by controlling the power of a statistical test at an appropriate alternative, even by those statisticians who recommend the use of confidence intervals for interference. As the technique used for analysis of the data should be the technique used for sample size determination, some comparisons of the sample size determined using the length of the confidence interval as the criterion and the sample size determined by controlling the power are presented.
Multiple Corresponding Analysis: An Illustrative Multivariate Method Applied to EORTC Clinical Trials M. V a n G l a b b e k e , M. S t a q u e t
EORTC Data Center, Brussels, Belgium (P-33) Multiple correspondance analysis (MCA) is a multivariate technique, widely used in psychosocial sciences. It can be seen as a multidimensional extension of correspondance analysis, recently
345
rapidly ascertaining and documenting whether criteria for the diagnosis of Schizophrenia, Schizophreniform, or Schizoaffective disorders have been met. Employing a microcomputer based expert system development tool, modules for each of the relevant DSM-III-R categories were constructed. Employing a user-friendly interface, the computer displays each criterion and asks if it is met or whether no information is available. The n u m b e r of criteria that must be responded to before a diagnosis is either made or ruled out is markedly reduced by the expert system "inference engine" and an optimizing routine. A short written report is produced that documents the diagnosis, or a longer report can be generated which presents the diagnosis along with questions and answers utilized. The immediate utility of the system is that it: l) reduces the time necessary to arrive at a diagnosis; 2) is completely reliable in applying criteria to make the diagnosis; and 3) provides documentation of the specific criteria utilized.
Statistical Methods for Monitoring Recruitment in a Multicenter Clinical T r i a l B. M i c h e l e Melia, M a r i e D i e n e r - W e s t
The Johns Hopkins University, Baltimore, Maryland (P-30) Monitoring recruitment is an important task in a multicenter clinical trial. Two aspects of the monitoring process may include the assessment of interim recruitment goals and comparison of recruitment among the clinics. Interim recruitment goals are useful indicators of whether recruitment is proceeding according to schedule, or there are recruitment problems which could result in failure to meet the study's final goal. High recruiting clinics may be able to provide information regarding successful recruitment techniques; low recruiting clinics may use a disproportionate share of the study resources. Statistical methods which are helpful for objectively identifying clinics with significantly poorer or better recruitment than that of their fellow clinics, and for determining whether recruitment is adequate to meet a study's recruitment goals are discussed. The methods are illustrated using recruitment data from the Collaborative Ocular Melanoma Study (COMS) which has more than 30 participating clinics across the United States and Canada. A "Weighted" Rank Sum Test A l f r e d P. H a l l s t r o m
University of Washington, Seattle, Washington (P-31) A rank sum test is sometimes used w h e n ordinal and measured p h e n o m e n a need to be put on a common scale (e.g., when death interferes with a blood pressure measure). It can be argued that this approach does not give appropriate weight (e.g., should death have only the rank next to that of the highest blood pressure). A "weighted" rank sum test is proposed for which the relative weights are determined by expert panel and then scaled to provide a test sensitive to alternatives of interest. Development of an outcome measure for the Myocardial Infarction Triage Intervention (MITI) Trial is given as an example. In this study the outcome measure, left ventricular ejection fraction, is confounded by death and stroke. Sample Sizes for Constructing Confidence Intervals and Testing Hypotheses D a v i d R. Bristol
CIBA-GEIGY Corporation, Summit, New Jersey (P-32) Although estimation and confidence intervals have become popular alternatives to hypothesis testing and p-values, sample sizes for randomized clinical trials are usually determined by controlling the power of a statistical test at an appropriate alternative, even by those statisticians who recommend the use of confidence intervals for interference. As the technique used for analysis of the data should be the technique used for sample size determination, some comparisons of the sample size determined using the length of the confidence interval as the criterion and the sample size determined by controlling the power are presented.
Multiple Corresponding Analysis: An Illustrative Multivariate Method Applied to EORTC Clinical Trials M. V a n G l a b b e k e , M. S t a q u e t
EORTC Data Center, Brussels, Belgium (P-33) Multiple correspondance analysis (MCA) is a multivariate technique, widely used in psychosocial sciences. It can be seen as a multidimensional extension of correspondance analysis, recently