An NYSPI Biostatistics Seminar Series Presentation:
Sample Size Determination under Finite Patient Horizon:
Potential Implications for Implementation Studies
April 24, 2012 3:30-4:30 pm
New PI 6th Floor Multipurpose Room (6602)
Coffee: 3:00-3:30pm, Reception: 4:30-5:00 pm
Click here to see the videotaped presentation.
Naihua Duan, Ph.D.
Departments of Psychiatry and Biostatistics, Columbia University
Research Scientist, New York State Psychiatric Institute
Ying Kuen Ken Cheung, Ph.D.
Department of Biostatistics, Columbia University
Implementation studies have emerged in recent years as an important component of public mental health programs, to improve quality of care and patient outcomes through the implementation of therapies known to be efficacious or effective but not yet implemented efficiently in community practice settings. As implementation programs often require the adaptation of existing protocols or the improvisation of customized protocols to accommodate the idiosyncrasies of individual practice settings, there is a need to conduct local quality improvement investigations to produce local knowledge to inform such adaptations and improvisations.
The design of local investigations such as those for implementation studies poses important challenges and opportunities for statistical methods such as sample size determination. Those investigations often differ from traditional clinical research in an important aspect regarding patient horizon, namely, the total number of patients with the clinical condition being studied.
For traditional clinical research, there is usually an implicit assumption that patient horizon is very large, or effectively infinite. Under this assumption, it is often reasonable to enroll a large number of patients into a clinical trial, exposing many of them to a treatment option that might turn out to be inferior, so as to minimize the uncertainty in the ultimate conclusions from the trial. The ethical cost of the suboptimal treatment for those patients is justified on the ground of the enormous benefit that will accrue to future patients who will receive the treatment option identified in the trial as superior with a high level of statistical confidence.
The assumption of an effectively infinite patient horizon might not hold for implementation studies focused on producing local knowledge to inform implementation decisions for an individual practice setting, with a finite number of patients with a specific clinical condition under study. If the local investigation is designed with the usual sample size determination procedures with a 5% type I error rate and a target power of 80%, there might be few if any “future patients” left to benefit from the local knowledge produced, invalidating the argument that the ethical cost born by trial participants is justified by the benefits accrued on future patients.
In this presentation, we will discuss sample size determination procedures under the assumption of a modest patient horizon, and take into consideration the welfare for both trial participants and future patients in local investigations designed to inform implementation decisions.
Dr. Naihua Duan is Professor of Biostatistics (in Psychiatry) in the Departments of Psychiatry and Biostatistics at Columbia University, and the Director of the Division of Biostatistics in the New York State Psychiatric Institute (NYSPI). He received a B.S. in mathematics from National Taiwan University, an M.A. in mathematical statistics from Columbia University, and a Ph.D. in statistics from Stanford University. He is an accomplished practicing biostatistician with research interests in health services research, prevention research, sample design and experimental design, model robustness, transformation models, multilevel modeling, nonparametric and semi-parametric regression methods, and environmental exposure assessment. He has led the statistical work on a number of prominent field studies, including the Health Insurance Experiment, the HIV Costs and Service Utilization Study, the National Latino and Asian American Study, the Recovery After Initial Schizophrenic Episode study, and the Optimizing Treatment for Complicated Grief study. He has published more than 180 papers in leading journals in statistics, psychiatry, public health, epidemiology, etc. He is an elected fellow of the American Statistical Association and the Institute of Mathematical Statistics, and a former associate editor for the Journal of the American Statistical Association.
Dr. Ying Kuen K Cheung's primary research interests include experimental design, with an expertise in early phase clinical trials in cancer and stroke, both significant public health concerns especially in the aging population. He is a biostatistical consultant to the Office of Clinical Trial at the NINDS. He currently holds NIH R01 awards to develop statistical designs for clinical trials and community-based studies. He is the statistician for NeuSTART, a translational program studying the use of high-dose statins in acute stroke patients. Dr. Cheung has served on the steering committee or as an investigator in numerous Columbia and commercially-sponsored studies.