Methods Research in the Women's Health Initiative: Addressing Bias, Messy Data, and Multiplicity Issues in Cohort Studies

Michael Pennell Chair
The Ohio State University
 
Michael Pennell Organizer
The Ohio State University
 
Thursday, Aug 8: 8:30 AM - 10:20 AM
1672 
Topic-Contributed Paper Session 
Oregon Convention Center 
Room: CC-D135 

Applied

Yes

Main Sponsor

Section on Statistics in Epidemiology

Co Sponsors

Biometrics Section
Caucus for Women in Statistics
ENAR

Presentations

The Women's Health Initiative as a platform for methodologic development

The Women's Health Initiative (WHI), a large, population-based study sponsored by the National Heart Lung and Blood Institute, has stimulated a plethora of statistical innovations over the past 3 decades. The initial design included a partial factorial randomized trial testing three chronic disease prevention hypotheses in 68,132 postmenopausal women and a parallel observational study of over 93,000 similar women. To support its broad scientific mandate and diverse investigator interests, the WHI has collect multiple types of data, ranging from biomarkers and genomics, behavioral and livestyle factors, to time-to event and longitudinal outcomes using diverse sources (self-report, medical records, data linkages, laboratory and clinical assessments). Analyses of these data are often made "interesting" by design factors (e.g., data collection limited to subsamples), inherent biases (e.g., self-reports of behavior or informative missingness), or by the challenges of multiplicity. These data analytic challenges have spawned a range of innovative methodologic responses, some of which will be discussed in this session.  

Speaker

Garnet Anderson, Fred Hutchinson Cancer Research Center

Selection Bias in Cancer Survivorship Studies and its Remediation

With improvements in screening and treatment, the number of cancer survivors is growing, increasing the need for research into long-term health and well-being after diagnosis. Longitudinally collected data from the Women's Health Initiative (WHI), and associated Life and Longevity After Cancer (LILAC) Study, enable critical evaluation of time-varying health disparities between survivors and cancer-free individuals. However, traditional assessment of observed data may distort the effect of cancer and its treatment on trajectories of aging. Bias can result when there is differential loss to follow-up and death between survivors and cancer-free individuals, and when other characteristics are predictors of our outcome and censoring. Using simulations informed by real data, we will illustrate underlying causal structures that can introduce selection bias into studies of cancer survivorship and demonstrate methods to remediate this bias. These considerations will guide discussion of LILAC data analysis, and general recommendations for minimizing bias in future studies of aging trajectories among cancer survivors.  

Co-Author(s)

Joshua Nugent, Kaiser Permanente Division of Research
Alexandra Binder, University of Hawaii Cancer Center

Speaker

Joshua Nugent, Kaiser Permanente Division of Research

Addressing measurement error in diet assessment in the Women's Health Initiative: the use of biomarker-based calibration to mitigate measurement error

Self-report dietary data are prone to both systematic and random sources of measurement error, impacting both the monitoring of nutritional status in the population and quantifying diet-disease relationships, making diet a difficult exposure to study. Researchers from the Women's Health Initiative (WHI) have conducted biomarker studies and calibration studies nested within WHI to be able to use statistical methods to adjust for measurement error. In addition to measurement error, dietary data are prone to skewness and excess zeroes in some cases, which must also be addressed. Using examples from WHI, the impact of measurement error on power and paramter estimates will be described, and techniques for mitigating this error will be illustrated, including both study design and statistical methods. 

Co-Author(s)

Janet Tooze, Wake Forest School of Medicine
Marian Neuhouser, Fred Hutchinson Cancer Center

Speaker

Janet Tooze, Wake Forest School of Medicine

Variable selection with FDR control for noisy data – an application to screening metabolites that are associated with breast and colorectal cancer

The rapidly expanding field of metabolomics presents an invaluable resource for understanding the associations between metabolites and various diseases. However, the high dimensionality, presence of missing values, and measurement errors associated with metabolomics data can present challenges in developing reliable and reproducible approaches for disease association studies. Therefore, there is a compelling need for robust statistical analyses that can navigate these complexities to achieve reliable and reproducible disease association studies. In this paper, we construct algorithms to perform variable selection for noisy data and control the False Discovery Rate when selecting mutual metabolomic predictors for multiple disease outcomes. We illustrate the versatility and performance of this procedure in various scenarios, dealing with missing data and measurement errors. By applying our method to the Women's Health Initiative data, we successfully identify metabolites that are associated with either or both of breast cancer and colorectal cancer, demonstrating the practical utility of our method in identifying consistent risk factors and understanding shared disease mechanisms. 

Speaker

Runqiu Wang

The 2016 ASA statement on p-values in practice: Perspective of a collaborative biostatistician in the setting of a large, secondary analysis

The ASA statement on p-values was released nearly a decade ago and built on decades of existing literature, yet there remains work to be done. Practical approaches to address the issues raised in the statement will be discussed. This presentation focuses on the setting of secondary analyses conducted using data collected from a large, prospective cohort study - the Women's Health Initiative - but the approaches discussed are broadly relevant. Topics covered will include 1) approaches to reduce the number of p-values presented and 2) approaches to address multiple comparisons.  

Speaker

Haley Hedlin, Stanford University