Making strides toward seeing the future

Researchers develop a model to predict GVHD mortality rates in patients undergoing transplants
Dr. Wendy Leisenring seated at her computer
Dr. Wendy Leisenring, a biostatistician in the Clinical Research Division, led the development of a tool that can predict outcomes and help guide treatment decisions for patients with acute graft-vs.-host disease. Photo by Dean Forbes

Ask any physician and most will tell you, predicting a patient's outcome is one of the most challenging aspects of the job. Some cases can be very straightforward, others have more complicating factors, while others, still, are unpredictable wild cards. But thanks to a team from the Clinical Research Division, a crystal ball is in the making for one group of patients.

In the March 14 issue of the journal Blood, the research team, led by Dr. Wendy Leisenring, announced that they have developed a model to accurately predict mortality rates among patients who have acute graft-vs.-host disease (GVHD) during the first 100 days after transplant. Clinicians were able to assign evidence-based scores by evaluating patients every 10 days for specific signs and symptoms of GVHD. The score, or acute GVHD Activity Index (aGVHDAI), was then used to predict whether patients would survive.

"We can stand at the patient's bedside with this paper in hand, using numbers and tables to derive a score that predicts with reasonable accuracy the likelihood that the patient is going to be alive at 200 days from transplant," said Dr. Paul Martin, study co-author and Clinical Research Division investigator with a long-standing interest in GVHD.

Peripheral blood-stem cell and bone-marrow transplant patients run the risk of developing GVHD, a frequent and sometimes fatal disorder caused by donor immune cells that attack tissues in the recipient. GVHD encompasses a wide spectrum of symptoms, from mild rash to serious damage in the digestive tract. While the reaction can actually be helpful when the targets are malignant cells, GVHD can wreak havoc on normal tissue. Being able to predict future outcomes would help doctors determine the best course of action — serious cases could be treated more aggressively, while milder cases could forgo any risky treatments in favor of a less dangerous regimen — but until recently, doctors have had little to go by in making that decision.

Developing a crystal ball for this disease was a key goal for the study, said Dr. George McDonald, co-author and head of the division's gastroenterology/hepatology section. "The hardest thing that doctors have to do is predict the future. We all have descriptions of diseases that we treat, and we're really good at describing the diseases, but the most difficult thing that we do is to predict what is going to happen to the patients who have those disorders," he said.

"The traditional ways of trying to do this for patients with acute GVHD have not worked well for predicting mortality," Martin said, adding that the current system has been in place since 1973. That method is based on the peak severity of GVHD, whereas the aGVHDAI can be applied at any time during the first 100 days after transplantation.

Traditional limitations

The traditional system has significant limitations when it comes to predicting outcomes, and it overlooks several important factors. "In the same way that a novel can be a wonderful, but not very precise, description of events, the current GVHD grading systems are very descriptive and tell us what's happening, but their precision in predicting patient outcomes is lacking," McDonald said. "The traditional system was based on the best educated guesses of the leaders in the field at the time. It has served us well as a good descriptive tool, but not for gauging the risk of mortality for individual patients in real time."

To build a more sensitive model, clinicians collaborated with Leisenring, a Clinical Research Division biostatistician. "We listed a variety of different symptoms and wanted to determine which ones would be important for predicting future outcomes for that patient," Leisenring said. "I tested many different factors in the model and tried to identify the combination that allowed me to best predict mortality."

Leisenring was able to determine which of the variables made little or no difference in predicting mortality, and these variables were then eliminated from consideration. Surprisingly, there were only four factors that played a significant role: increased levels of bilirubin in the blood, indicating liver damage; decreased oral intake; requirement for treatment with prednisone; and overall activity level.

"What was nice about reducing the model to just four variables is that physicians could collect the relevant information every day for individual patients and then use it to create a risk score to predict how they might do in the future," Leisenring said.

Of the 386 patients studied, one version of the model developed by the team predicted mortality in 76 percent of those who died, with a false positive rate of just 12 percent. But there is still more work to be done.

"This is not the be all and end all of scoring systems," McDonald said. "This is our best effort in the year 2006. People will have to keep analyzing similar data and perhaps perfecting this kind of model. But the results show that when we use evidence-based methods, we can come up with a better mousetrap, as opposed to just guessing at what's important." He added that the system should be examined to see how it applies to other groups of patients, such as children and those at other institutions, to continue improving the accuracy of this new forecasting tool.

Help Us Eliminate Cancer

Every dollar counts. Please support lifesaving research today.