Laryngology: New Techniques in Vocal Cord Surgery Spare the Voice Dermatology: Sun May Cause Cancer by Targeting Rb Pathway Epidemiology: No Cancer Risk Found for Acrylamide Clinical Research: Statistical Approach Speeds Up Stent Trials Diversity: Teachers Institute Aims to Improve Diversity, Increase Science Literacy Integrated Medicine: First Osher Pilot Grants Announced for Studies in Complementary and Integrative Medicine   The Winter Bookshelf   Integrin Crystal Structure Reveals New Conformation Study Links Brain Attack to Oral Health Telomere Loss Acts Together with Atm Deficiency to Sap Stem Cell Supplies Blocking Protein Aggregates Reverses Disease in Mouse Model of Huntington's   New Appointments to Full and Named Professorships State Health Commissioner to Join HSPH Faculty Nathan to Receive Howland Medal for Pediatrics HSPH Launches Revamped Home Page Call for Papers on Medical Education in Poorer Countries Nominations Wanted for Excellence in Mentoring Awards Announcing the Second Year Show Honors and Advances News Briefs   Faculty, Students Build Bridges to Better Dental Health   Beyond Roe vs. Wade   Front Page

CLINICAL RESEARCH Statistical Approach Speeds Up Stent Trials New statistical methods devised by HMS researchers may allow smaller, faster clinical trials of improved stents for coronary disease--and with greater scientific validity. Richard Kuntz, James O'Malley, and Sharon-Lise Normand (l to r) are the cocreators of a new statistical model that can speed up approvals for new coronary stents and be adapted for other devices. (Kuntz photo by Pam Murray; Others by Jeff Cleary) The computation may be used in lieu of the control part of a study. Such trials then would be able to focus on the new device and reduce the time and cost of a randomized control design. Developed by statistician James O'Malley, assistant professor of health care policy at HMS, and his colleagues, the new model recycles some data from clinical trials of similar stents already approved by the Food and Drug Administration, much of which is stored at the data center of the Harvard Clinical Research Institute (see sidebar). The paper appears in the Jan. 30 Statistics in Medicine (online Dec. 18, 2002). "This is a major paper for the approval process dealing with equivalency and non-inferiority of new devices," according to an accompanying editorial by Boston University statistician Ralph D'Agostino Sr. "The issues and methods employed are very instructive for our thinking in the drug and biologics field." Little Digs Coronary stenting has become standard treatment in the vast majority of the 1.5 million patients worldwide per year who undergo catheter-based coronary treatment. In this country, threading a catheter up an artery from an incision in the upper thigh--usually to insert the small expandable metallic scaffolds to enlarge segments of plaque-narrowed arteries--has become the dominant way to widen clogged arteries, at a rate more than double the number of coronary artery bypass graft surgeries last year. Laura Mauri and James O'Malley tweaked the new method to discover that patients had better outcomes if cardiologists used stents that were slightly shorter than the lesion. Helping to drive the trend, new stent designs proliferate so fast they may be obsolete by the end of the two or more years required to enroll and evaluate 1,000 people at a nine-month endpoint for the standard evidence-based, randomized controlled trial. Medical device improvements tend to be incremental with small changes in a localized effect, said FDA statistician Gene Pennello, so there is usually an older version of the device, often with relevant clinical data. Bayesian statistical methods can take advantage of the historical information--often with more precise estimates of effectiveness and safety parameters, he said. The Clinical Trial Data Center Conveniently, much of the raw data from clinical trials of approved coronary stents in the last 10 years lives in the data center at the Harvard Clinical Research Institute (HCRI), a joint venture by Partners, CareGroup, and HMS. The data center began doing business 10 years ago as a nonprofit clinical research organization at Beth Israel Deaconess Medical Center to design and analyze clinical trials for stents and other interventional cardiovascular devices. The accumulating high-quality data from clinical trials of the devices could be used to investigate meaningful biological and medical questions. It was the brainchild of Richard Kuntz, interventional cardiologist at BWH and an HMS associate professor of medicine, who trained in biostatistics during a clinical fellowship at the Harvard School of Public Health, and interventional cardiologist Donald Baim, HMS professor of medicine at BWH. As the data center quickly grew--nearly cornering the market on clinical trials of stents and related devices--Kuntz developed new trial designs and statistical methods adopted by companies and the FDA. For example, Kuntz played a key role in introducing stent-vs.-stent equivalency trials about six years ago, which led to smaller trials and quicker approval times for a second generation of stents. He has advocated the idea of the objective performance criterion to drive a standard for new stents, as well as the criterion's new statistical underpinnings. Still developing, the HCRI aims to extend the support and facilitation of clinical trials across the Harvard community, said HCRI board chair Raphael Dolin, HMS dean for clinical programs and the Maxwell Finland professor of medicine at BWH. The FDA has already given the green light for some single-arm stent studies, testing only the experimental device in patients and using relevant data from trials of older versions as the control group. Under increased regulatory pressure to speed up the approval process, the FDA suggested that stent investigators pool their data for an updated version of the objective performance criterion, a concept first adopted to judge new heart valves in single-arm studies. The heart valve criterion was a single number representing the complication rate per year calculated from a meta-analysis of approved valve studies. In the context of modern statistics, the old standard for heart valves is all wrong. It does not allow for probability distributions that change with patient populations, O'Malley said. Even worse, the old model treats the historical data as a concurrent randomized arm rather than as an observational study with many related variables. Over time, each new trial compared to the criterion would tend to overstate the significance of the results. "Modeling is a way of describing the relationship of variables and outcomes mathematically," O'Malley said. "People tend to put full faith in existing statistical methods and treat numbers from those methods as exact, but they're only an approximation of what we observe in the world." Computing Standards To develop a new methodology for medical device clinical trials, Richard Kuntz, interventional cardiologist at Brigham and Women's Hospital and HMS associate professor of medicine, turned to Sharon-Lise Normand, associate professor of health care policy at HMS and a statistician with expertise in observational studies. Together, they recruited O'Malley, who developed the new method for computing industry standards during a two-year postdoctoral fellowship. O'Malley built the model using seven randomized trials of FDA-approved stents involving 5,806 patients stored at the HCRI. The new objective performance criterion is based on two measures of renarrowing nine months after stent placement. (Mortality is rarely used as an outcome in stent trials, because the progressive success in keeping vessels open at the stent site has had little impact on the underlying atherosclerosis process or long-term heart attack risk.) O'Malley found that in about 10 percent of patients, a treated vessel becomes reblocked and requires a repeat procedure; however, the mean ranges from 8.6 percent for non-diabetics with one diseased vessel to 16.3 percent for diabetics with three diseased vessels. Stented arteries experience about 38 percent renarrowing on average, ranging from 37.5 to 42.7 percent in the studied populations. More important than these means is the way the model's criterion shows the range of uncertainty inherent in statistical inferences from clinical trials. The range becomes increasingly wide as the subpopulation becomes more finely tuned. The model also can judge on the fly whether variables and outcomes are similar enough to be effective historical controls. The model's Bayesian statistical tools usually involve relatively complex computations requiring modern computer power but result in more intuitive data interpretation based on likelihood ratios rather than the often misinterpreted and convoluted logic of the P-value, said O'Malley, who also calls the results more accountable. Bayesian computations in the model can evaluate and compare variables measuring health of patients before stenting, characteristics of the disease, procedural variations, and how patients are recruited. The new model can be adapted to randomized trials needed to test more dramatic improvements in stents because of its potential to combine historical data with a smaller control group in a randomized, controlled trial. It can also be used to answer biological and medical questions. For example, lead author Laura Mauri, an HMS research fellow at BWH, and O'Malley tweaked the new method to discover that patients had better outcomes if cardiologists used stents that were slightly shorter than the lesion, according to a poster presented at the November meeting of the American Heart Association. --Carol Cruzan Morton