September 3, 2009

The Group Health Biostatistics unit

Team science at its best

Group Health Epidemiologist and Senior Investigator Lisa Jackson, MD, MPH, had a hunch a few years ago that something was awry in studies suggesting that flu vaccine slashed seniors' risk for pneumonia and death. Most research had suggested the vaccine provided huge benefits. But she wondered whether these results might be skewed by not fully considering seniors' health status.

So she and colleagues collaborated with Associate Investigator Jennifer Nelson, PhD, Group Health Research Institute's lead biostatistician, to examine the issue more rigorously. Biostatisticians like Dr. Nelson specialize in developing and applying novel methodological approaches to address such questions. She and the team used the period before flu season as a "control," letting them focus on differences in health status between seniors who receive flu vaccine and those who don't. Then they measured how much those differences may contribute to the impression that flu vaccination saves lives in the elderly.

Surprisingly, the Institute's team found that this "healthy user bias" confounded prior studies and is likely large enough to account entirely for reports that the flu vaccine prevents deaths in seniors. Confounders are factors that falsely appear to cause the outcome being studied and can thus yield healthy user bias. In this case, Dr. Nelson's innovative statistical approach showed that death was less likely to occur in vaccinated seniors simply because they were healthier to begin with—not because of any protective benefit from the vaccine.

By filling in missing pieces of the puzzle—and discarding others that didn't belong—Dr. Nelson's work helped paint a more complete picture of an important epidemiologic question. It also dramatically illustrates how high-quality research depends on high-quality biostatistics. Without biostatistics, conducting scientifically sound health care research would be like trying to assemble a puzzle without all its pieces.

"Analyzing administrative or other observational data too quickly can lead to some strange results," Dr. Nelson explains. "Biostatistics is about recognizing when something looks off-base and then deriving and implementing methods to explain further what's really going on."

Growing a sophisticated team of biostatisticians has long been an important goal at Group Health Research Institute. Now 26 members strong, the Group Health Biostatistics Unit includes 7 PhD investigators, 10 masters-level biostatisticians, and 9 graduate students. As the unit has grown—nearly tripling in the last four years—so too has its reputation as a mainstay of research quality and innovation at Group Health and beyond.

From the perspective of 26-year veteran Lou Grothaus, MA—a Group Health biostatistician since the Institute's doors opened in 1983—this reputation is primarily due to the unit's "strong mathematical statisticians who communicate well and are interested in the science."

But the main reason for their success may be their commitment to team science. Whether leading nationwide research consortia, mentoring students in the University of Washington (UW) Department of Biostatistics, or supporting junior faculty, the unit's members are known for their collaborative spirit. Colleagues at the Institute and elsewhere view them as valuable scientific partners.

"The Group Health Biostatistics Unit is a high-quality, collegial group," comments Bruce Weir, PhD, biostatistics chair at the UW. "They are like an extension of our department—all are either affiliate faculty or regular faculty. They teach classes and supervise student projects. And individually, they are very good at developing new methods."

Partnership with Group Health's care-delivery system is another critical component of the group's success. In a region that includes heavy hitters such as the UW, Children's Hospital, and Fred Hutchinson Cancer Research Center, the Group Health Biostatistics Unit stands out for its direct link to Group Health's large, stable population of patients. This link provides the chance to apply statistical methods to real-world situations—an asset that is rare in most academic research settings and has proven attractive to young biostatisticians.

It's one of the main reasons Assistant Investigator Rebecca Hubbard, PhD, joined the Institute in 2008. A primary interest of Dr. Hubbard's is health services research—that is, studying how health care works.

"The Group Health population is like a health care laboratory," says Dr. Hubbard. "It provides a good opportunity for observing the real impact of things like burden of disease or risk adjustment. This work is theoretical in other places. It's real here."

The opportunity to pursue independent methodological research (see "Breakthrough biostatistical papers" below) is another of the unit's hallmark traits—leading to innovative research designed to address real-world problems encountered in applied studies. For instance, Dr. Nelson and Assistant Investigator Andrea Cook, PhD, are working with the CDC's Vaccine Safety Datalink and the Food and Drug Administration (FDA)'s Sentinel Initiative on real-time post-marketing safety analysis of new vaccines and other medical products (see "Biostatistical leadership in major nationwide consortia" below). The methodological goal is to improve statistical approaches that use observational data from electronic health care database networks for routine monitoring of the safety of medical products after they are licensed. The real-world goal is earlier and more accurate identification of potentially harmful products that are found to cause previously undetected adverse events after hitting the market.

In another expression of team science, all Group Health biostatisticians—from masters-level scientists to senior investigators—enjoy the freedom to explore a broad range of methods-based and applied research.

"Biostatisticians at GHRI can sample a wide variety of health care research areas, learn new things, then specialize in an area of particular interest if they so choose," states Dr. Nelson. "We also share an appreciation for contributions made along the entire biostatistics continuum—from developing new methods to applying them."

This freedom and diversity of research fosters an environment where biostatisticians can become experts in fields other than statistics—and where they can feel confident their work is making a difference.

For example, Senior Investigator and Biostatistician Diana Miglioretti, PhD, is widely known for her expertise in breast cancer screening. Dr. Miglioretti is the principal investigator of the Statistical Coordinating Center (SCC) for the National Cancer Institute's Breast Cancer Surveillance Consortium (BCSC) (see "Biostatistical leadership in major nationwide consortia" below). She oversees analyses of the nation's largest and most comprehensive collection of data on breast cancer screening. These data serve as the basis for national breast imaging quality standards and breast cancer screening guidelines such as the U.S. Preventive Services Task Force's new recommendations. Valuable contributions from other Group Health biostatisticians, including Drs. Hubbard and Cook and Associate Investigator Sebastien Haneuse, PhD, further bolster the SCC's leadership in breast cancer prevention.

Cutting-edge science has earned the unit noteworthy regard. But the scientists themselves have most helped the unit to flourish. Under past leadership from veteran investigators Bill Barlow, PhD, and Carolyn Rutter, PhD, the group evolved into a cohesive team that supports its young scientists. Now, with Dr. Nelson at the helm, the unit is poised for even more growth as funding from the American Recovery and Reinvestment Act (ARRA) pours in.

The unit's large size allows for a biostatistician on nearly every project at Group Health Research Institute. This helps boost the Institute's scientific rigor and ability to attain funding—attributes that make the unit extraordinarily valuable. Widespread appreciation for the group's contributions is obvious among their many internal and external scientific partners.

"Our biostatistics group has developed into a real center of energy and creativity," comments Group Health Psychiatrist and Senior Investigator Greg Simon, MD, MPH. "All of our research has advanced because of our biostatisticians' innovative approaches to practical problems."

"At the UW, we see Group Health biostatisticians as members of our family and important players in our work," adds Dr. Weir. "We are very proud of their accomplishments."

Approaching her second year at Group Health, Dr. Hubbard reflected on this unique collaborative environment and what it means to fellow biostatisticians and her. "Biostatisticians at Group Health are considered integral members of our research teams. We're involved in the entire research process—from conception to implementation. We're appreciated and respected. It's the culture of team science at its best."


by Jessica Ridpath

Breakthrough biostatistical papers

Barlow WE. Robust variance estimation for the case-cohort designBiometrics. 1994; 50:1064-72. PubMed

Rutter CM, Gatsonis CA. A hierarchical regression approach to meta-analysis of diagnostic test accuracy evaluations. Statistics in Medicine. 2001; 20:2865-84. PubMed

Wakefield J, Haneuse SJ. Overcoming ecologic bias using the two-phase study designAmerican Journal of Epidemiology. 2008; 167(8):908-16. Epub Feb 2008. PubMed

Rutter CM, Miglioretti DL, Savarino JE. Bayesian calibration of microsimulation modelsJournal of the American Statistical Association. Epub 29 Jul 2009.

Hubbard RA, Inoue LYT, and Diehr P. Joint modeling of self-rated health and functional statusJournal of the American Statistical Association. 2009; 104(487): 873-885.

Miglioretti DL, Haneuse SJPA, Anderson M. Statistical approaches for modeling radiologists' interpretive performance. Academic Radiology. 2009; 16(2):227-38. PubMed

Nelson JC, Jackson ML, Weiss NS, and Jackson LA. New strategies are needed to improve the accuracy of influenza vaccine effectiveness estimates among seniorsJournal of Clinical Epidemiology. 2009; 62:687-94. Epub Jan 2009. PubMed

Cook AJ, Li Y, Arterburn D, Tiwari RC. Spatial cluster detection for weighted outcomes using cumulative geographic residuals. Biometrics. 2009 Sep 14. [Epub ahead of print]. PubMed


Biostatistical leadership in major nationwide consortia

The Breast Cancer Surveillance Consortium's Statistical Coordinating Center

The Breast Cancer Surveillance Consortium (BCSC) is a National Cancer Institute-funded network of seven mammography registries linked to tumor and pathology registries. Group Health biostatisticians lead the consortium's Statistical Coordinating Center (SCC)—the repository and central analysis site for all BCSC data, which includes information on nearly 100,000 breast cancer cases and more than nine million breast imaging examinations performed in more than two million women.

The Cancer Intervention and Surveillance Modeling Network

The National Cancer Institute established the Cancer Intervention and Surveillance Modeling Network (CISNET) in 2000, creating a nationwide collaboration that uses comparative modeling to help understand cancer screening, prevention, and treatment. CISNET models are used to project trends and improve prevention strategies for breast, colorectal, lung, and prostate cancers. Group Health serves as the prime site for a current CISNET research to develop microsimulation models examining the effect of screening on colorectal cancer incidence and mortality.

The Food & Drug Administration's Sentinel Initiative

In 2008, the Food & Drug Administration partnered with more than 25 government departments and private health care organizations to launch the Sentinel Initiative—an effort aiming to establish an electronic monitoring system enabling real-time safety surveillance of new medical products. The system gives the FDA access to a broad array of information to facilitate quick identification of possible post-market adverse events. This unprecedented opportunity to detect problems as they first arise relies heavily on developing and applying sophisticated biostatistical methods, and Group Health biostatisticians are co-leading the Methods Core.

The Vaccine Safety Datalink

Founded in 1990, the Vaccine Safety Datalink (VSD) is a collaboration between the Centers for Disease Control and Prevention (CDC)'s Immunization Safety Office and eight health plans, including Group Health. The VSD combines data on immunizations and diseases for millions of people, allowing researchers to conduct scientifically rigorous studies of vaccine safety and effectiveness. Several Group Health biostatisticians have contributed to wide-ranging VSD research over the years, including a 2001 study showing that vaccine-related seizures are unlikely to cause lasting harm.


Jennifer C. Nelson, PhD

Director, Biostatistics; Senior Investigator
Kaiser Permanente Washington Health Research Institute


Andrea J. Cook, PhD

Senior Biostatistics Investigator
Kaiser Permanente Washington Health Research Institute

Gregory E. Simon, MD, MPH

Senior Investigator

KPWHRI; Psychiatrist, Washington Permanente Medical Group