Harvard School of Public Health Names Robert L. Strawderman its 2008 Distinguished Biostatistics Alum
Harvard award winner Professor Robert L. Strawderman, who joined Cornell in 2000, was previously a faculty member in the Department of Biostatistics at the University of Michigan. His major research area is survival analysis, a branch of statistics that deals with characterizing the time until an event, such as the death of an organism or the failure of a machine, occurs. Professor Strawderman's particular research interests lie in the study of events that can recur, such heart attacks or epidemics. He collaborates extensively with subject matter specialists in applying these and other statistical methods to problems in health services, cardiology, epidemiology, demography, and veterinary medicine.
In connection with his alumni award, Strawderman was asked to deliver a public lecture at Harvard University, primarily for the benefit of the students, on his career since receiving his Ph.D. from the Harvard School of Public Health. He observed that "I found this surprisingly hard to do," but eventually settled on a topic that "reflects the major cornerstones of my responsibilities as a faculty member:" consulting, education and research. "The problem initially arose as a consultation, led to a masters thesis for a Cornell student and has generated two nice publications," he said.
His lecture, on May 28, 2008, was called "Profiling Pharmacy Expenditures in Managed Care: Some Statistical Considerations and Limitations." He noted that "profiling" has become an increasingly important topic in health care and other industries that are looking for ways to control costs, increase profitability, and increase service quality. In this context, profiling is the process of measing and ranking service providers - for example physicians - by various performance measures on the premise that health care costs such as those related to prescriptions can by better controlled by rewarding or penalizing physicians. The underlying assumption is that physicians have the discretion to modify prescribing practices so as to reduce costs, and will respond to incentives towards doing so.
In the particular case described in Strawderman's lecture, a large midwestern health management organization (HMO) used an annual analysis of data on prescription costs to determine which physicians should receive rewards or penalities, based on a targetted percentage of their deviation from the average for the HMO (adjusted for physician age and gender). The deviation was computed using an ordinary least squares regression approach. The physicians, who belonged to a physician-hospital organization (PHO) that had a contract to provide patient care services to the HMO, questioned the fairness of this arrangement, especially because it didn't account for differences in the mix of patients each sees.
The vice president for performance improvement of the PHO asked Strawderman to help evaluate the limitations of the arrangement by determining the extent to which variation in performance could be attributed to discretionary prescribing by its member physicians rather than other factors. The resulting analysis, which used a richer class of statistical models, showed that the variation in performance was due primarily to other factors, such as the difference in the patient mix, demographic differences among physicians, and random sources of variability. Two statistical approaches reflecting this analysis projected significantly smaller rewards and penalties that were more focussed on discretionary prescribing in comparison with an analysis conducted using ordinary least squares.
Consequently the PHO recommended that the HMO focus its efforts elsewhere, including designing a pay for performance program based on quality, not mere cost-containment. (The HMO did not change the system and the PHO eventually dissolved their relationship with the HMO for this and other reasons). One of the two approaches to profiling utilized by Strawderman and Cowen, hierarchical regession modelling, is now considered the 'standard of practice' in profiling.Strawderman noted that this work, which led to a 2002 publication with Dr. Mark E. Cowen in the journal Medical Care, was an improvement compared to the ordinary least squares approach but had several limitations. Some of these shortcomings were subsequently addressed through a so-called Bayesian model in a 2006 paper -- with Cowen, Strawderman's student Min Zhang (now an Assistant Professor at Purdue), and ORIE field member Martin T. Wells -- in the Journal of the American Statistical Association.
Harvard's Distinguished Alum Award recognizes an individual in government, industry or academia who has impacted the theory and practice of statistical science. "I was honored to receive this award," said Strawderman.
Strawderman is on the faculty of two departments at Cornell, Biological Statistics and Computational Biology (BSCB) and Statistical Science. With David Ruppert, he advised the Ph.D. thesis of recent ORIE Ph.D. graduate Emmanuel Sharef. Strawderman has a BA in Mathematics from Rutgers.
For additional information, click here.