Preface

Introduction

The use and importance of observational data—such as naturalistic real-world trials, patient registries, and health care claims database analyses—have grown in recent years. Observational research produces real-world data—information on how treatments or policies work in practice—that is critical to consumers interested in actual practice information. Such data are now commonly used by researchers and health care decision-makers to assess treatment strategies and make policy decisions. However, techniques and standards for statistical analysis are less well-developed for observational data compared with randomized clinical trials. Quality analyses of observational data are more challenging due to such issues as selection bias. Literature reviews of recent manuscripts from various types of observational data have criticized the lack of quality, consistency, and transparency of observational data reporting. Low-quality analyses and limited experience with such data by many decision-makers has led to a lack of optimal use and even mistrust of such work. Several research groups, recognizing these analytical and reporting issues, are starting to provide general guidance on improving the quality of such analyses. However, there is still a lack of practical detailed guidance on implementing such methodology.

Our goal for creating this book was to provide a resource that would make high-quality, thorough, transparent analysis of observational data easy to perform. Each chapter includes background information, data examples, SAS code, output, and references to allow for implementation of these methods in accordance with the highest quality manuscripts and guidelines that exist for these topics. As a result, this book should be beneficial to a wide variety of researchers who use observational data (from sources such as prospective and retrospective observational studies, patient registries, survey research, and claims [billing] databases) for analyses and decision-making. Given the breadth of observational research, potential users include statisticians, health outcomes researchers, epidemiologists, medical researchers, health care administrators, statistical programmers, analysts, economists, professors, and graduate students, among others.

Outline of the Book

The main objective of this book is to provide information allowing researchers to perform high-quality analyses of observational data, to present the data in a transparent manner, and to ensure accurate interpretation and appropriate decision-making based on the data. To achieve this, the book includes detailed sections on the general methodological issues of both cross-sectional and longitudinal bias adjustment, followed by shorter sections focused on claims database analyses, economic analyses, and design of observational research.

Chapter 1 provides an overview of the issues involved in analyzing observational data. From a statistical perspective, the most common challenge in observational data analysis is addressing the selection bias (confounding) resulting from a lack of randomization. When the groups of interest are not randomized, they are likely to differ on many key characteristics and might not be comparable. Thus, standard statistical methods will most likely produce associations, not causal inferences. This book includes a detailed section with six chapters covering the core issue of bias adjustment. This section includes the commonly used propensity scoring approaches (applied via regression, stratification, and matching), as well as alternatives of doubly robust estimation, instrumental variables, and newly developed approaches such as local control. In addition, a separate chapter discusses the practical issue of addressing missing covariate data.

Cross-Sectional Selection Bias Adjustment:

• Chapter 2, “Propensity Score Stratification and Regression”

• Chapter 3, “Propensity Score Matching for Estimating Treatment Effects”

• Chapter 4, “Doubly Robust Estimation of Treatment Effects”

• Chapter 5, “Propensity Scoring with Missing Values”

• Chapter 6, “Instrumental Variable Method for Addressing Selection Bias”

• Chapter 7, “Local Control Approach Using JMP”

The analysis of longitudinal naturalistic data includes these challenges along with the potential for patients to switch treatments, time-dependent confounding, and censored records. The next section of the book provides details on performing four different analysis methods designed for longitudinal naturalistic data:

• Chapter 8, “A Two-Stage Longitudinal Propensity Adjustment for Analysis of Observational Data”

• Chapter 9, “Analysis of Longitudinal Observational Data Using Marginal Structural Models”

• Chapter 10, “Structural Nested Models”

• Chapter 11, “Regression Models on Longitudinal Propensity Scores”

Claims database analyses have become a growing area of research because such billing databases provide immediate access to data from thousands of patients. Such databases are being used more frequently to assess questions such as resource utilization, patient outcomes, treatment costs, and safety information that can not be practically addressed in clinical trials. For instance, detecting rare events requires many more patients than are typically included in clinical trials. Also, issues such as drug combinations and patient subsets might be assessed in such data. This book includes a chapter providing general guidance on such analyses as well as a chapter demonstrating a safety analysis from a health care database:

• Chapter 12, “Good Research Practices for the Conduct of Observational Database
Studies”

• Chapter 13, “Dose-Response Analyses Using Large Health Care Databases”

With rising medical costs becoming a major issue, another growing area of research with observational data is cost and cost-effectiveness analyses. Clinical trials are typically not the best source of data for assessing real-world costs that health care payers must cover—trials have strict entry criteria, restrictions on co-morbidities, polypharmacy, concomitant medications, required compliance, and mandatory visits and procedures. Thus, observational data are the preferred source for such information. However, proper analysis and presentation of cost data can be difficult, and a recent literature review has raised issues with the lack of consistency of methodology and the quality of reporting of cost analyses.¹

• Chapter 14, “Cost and Cost-Effectiveness Analysis Using Propensity Score
Bin Bootstrapping”

• Chapter 15, “Incremental Net Benefit”

• Chapter 16, “Cost and Cost-Effectiveness Analysis with Censored Data”

While the majority of the material focuses on analytical methods for existing data, researchers often are faced with the challenge of designing observational studies. The last section includes guidance on sample size determination and dealing with broader issues such as measurement and sponsor bias:

• Chapter 17, “Addressing Measurement and Sponsor Biases in Observational Research”

• Chapter 18, “Sample Size Calculation for Observational Studies”

List of Contributors

Aristide Achy-Brou, Johns Hopkins University
Daniel Almirall, University of Michigan
Peter Austin, University of Toronto, Institute for Clinical Evaluative Sciences
Heejung Bang, Weill Medical College of Cornell University
John Brooks, University of Iowa
Maria Chiu, Institute for Clinical Evaluative Sciences
Cynthia Coffman, VA-Durham and Duke University
William Crown, i3 Innovus
Peter Davey, University of Dundee
Marie Davidian, North Carolina State University
Elizabeth DeLong, Duke University
Douglas E. Faries, Lilly USA
Constantine Frangakis, Johns Hopkins University
Michelle Funk, University of North Carolina
Hassan Ghomrawi, Weill Medical College of Cornell University
Ron Goeree, McMaster University
Michael Griswold, Johns Hopkins University
Josep Maria Haro, St. John of God, Barcelona
Don Hedeker, University of Chicago
Dave Hutchins, Advanced PCS Health Systems
Sin-Ho Jung, Duke University
Zbigniew Kadziola, Eli Lilly
Dennis T. Ko, Institute for Clinical Evaluative Sciences
Taiyeong Lee, SAS Institute
Andrew C. Leon, Weill Medical College of Cornell University
R. Scott Leslie, MedImpact Healthcare Systems
Chunshan Li, Weill Medical College of Cornell University
Xiang Ling, Amgen
Ilya Lipkovich, Eli Lilly
Bradley Martin, University of Arkansas Medical School
Brenda Motheral, Care Scientific
Susan A. Murphy, University of Michigan
Robert L. Obenchain, Risk Benefit Statistics
Xiaomei Peng, Lilly USA
Yongming Qu, Eli Lilly
Paul Stang, West Chester University
David Suarez, St. John of God, Barcelona
Jack V. Tu, Institute for Clinical Evaluative Sciences, University of Toronto
Honkun Wang, University of Virginia
Ouhong Wang, Amgen
Chris Weisen, University of North Carolina
Daniel Westreich, University of North Carolina
Andrew Willan, University of Toronto
Will S. Yancy Jr., Duke University
Hongwei Zhao, Texas A&M University

Author Pages

Each SAS Press author has an author page, which contains several features that relate to the author, including a biography; book descriptions for upcoming titles and other titles by the author; contact information; links to sample chapters, and example code and data; events and extras; and more.

You can access the author pages from http://support.sas.com/authors.

For an alphabetical listing of all books for which example code is available, see http://support.sas.com/bookcode. Select a title to display the book’s example code.

If you are unable to access the code through the Web site, send e-mail to [email protected].

Acknowledgments

The authors would like to thank the reviewers who provided multiple quality suggestions to improve this text. In addition, the authors would like to thank Alex Dmitrienko, research advisor at Eli Lilly and Company, for his inspiration to start this work. Special thanks go to George McDaniel, acquisitions editor at SAS Press, for his guidance and hard work in helping us complete this project. Also, we would like to thank the entire SAS Press team who made this possible. These individuals include Kathy Restivo, copy editor; Mary Beth Steinbach, managing editor; Candy Farrell, technical publishing specialist; Patrice Cherry, cover designer; Jennifer Dilley, designer; and Stacey Hamilton and Shelly Goodin, marketing specialists.

Additional Resources

SAS offers a rich variety of resources to help build your SAS skills and explore and apply the full power of SAS software. Whether you are in a professional or academic setting, we have learning products that can help you maximize your investment in SAS.

Comments or Questions?

If you have comments or questions about this book, contact the authors through SAS as follows:

Mail:

SAS Institute Inc.
SAS Press
Attn: <Author’s name>
SAS Campus Drive
Cary, NC 27513

E-mail: [email protected]

Fax: (919) 677-4444

Please include the title of the book in your correspondence.

For a complete list of books available through SAS Press, visit http://support.sas.com/publishing.

SAS Publishing News: Receive up-to-date information about all new SAS publications via e-mail by subscribing to the SAS Publishing News monthly eNewsletter. Visit http://support.sas.com/subscribe.

¹See Doshi, J.A., H.A. Glick, and D. Polsky 2006. “Analyses of cost data in economic evaluations conducted alongside randomized controlled trials.” Value in Health 9: 334-340.