Photo by Todd McNaught
Most homeowners sleep easier knowing that their household smoke detectors will wake them in the event of a fire. Yet few would feel safe if the alarm tripped each time the dinner-table candles were lit-or sat silent while a burned roast enveloped the kitchen in a smoky haze.
A similar balance between extremes of sensitivity is an equally important quality of a successful cancer-detection test, said Dr. Yingye Zheng, who joined the Public Health Sciences Division last fall.
As a biostatistician, Zheng works to make sure such tests meet stringent requirements before they ever wind up in a doctor's office. Although the task may mean sifting through thousands of data points, Zheng said the goal of producing reliable tests that spare patients the potential worry, risk and cost of an inaccurate diagnosis is well worth her effort.
"A test for early signs of cancer can't give too many false-positive results, because it might lead to unnecessary procedures that are costly and have risks of their own," she said. "On the other hand, a good test needs to identify those patients who really do have cancer so that they can receive appropriate care."
Zheng applies her statistical expertise to a number of Fred Hutchinson projects that seek to identify biomarkers-molecules in the body signifying the presence, type or stage of disease-which could be adapted to serve as the medical equivalent of smoke detectors.
The research could someday transform how doctors diagnose, treat and monitor the progress of patients who suffer from potentially deadly diseases. What's more, biomarker tests that identify precancerous conditions could even lead to strategies that prevent tumors from ever forming.
Zheng said her skills, and those of her fellow PHS biostatisticians, are valuable in each of the stages required to develop a successful biomarker.
"The first step is the discovery phase," she said. "This involves laboratory research to find molecules produced by individuals who have cancer or that distinguish different types of cancer."
New techniques-such as DNA arrays, which enable scientists to scan thousands of genes simultaneously, and highly sensitive mass spectrometry, which can sift through all the proteins in a blood sample-greatly speed the biomarker-discovery process.
Still, a molecule or gene that looks promising at first glance may not hold up to statistical scrutiny, Zheng said. To evaluate whether potential biomarkers warrant further study, she adapts conventional statistical formulas. These can reveal whether a protein is present at higher levels in blood samples from cancer patients than from healthy individuals.
"There's a very high-false discovery rate," she said. "At this stage of the biomarker development process, we usually have data from multiple measurements made on a very limited number of samples, which can complicate data analysis. It's my job to overcome this problem."
Her strategy is to apply her statistical formulas to an artificial situation that is similar to the actual experimental data. If the results from the simulation make sense, she'll apply it to the actual data.
"It's very analogous to discovery in the laboratory," she said. "You do a simulation based on a theory or intuition, and if it doesn't work, you revise it and test it again."
Zheng used this approach in her graduate research at the University of Washington to help identify biomarkers that distinguish between the liver disease cirrhosis and early-stage liver cancer more effectively than the conventional diagnostic test.
The liver study, as well as an esophageal-cancer project in which Zheng participates, are part of the Early Detection Research Network-a National Cancer Institute-sponsored consortium to develop biomarkers. Under the direction of PHS investigator Dr. Ziding Feng, Fred Hutchinson houses the Data Management and Coordinating Center for the entire network. Dr. Nancy Kiviat, professor of pathology at UW and a PHS investigator, directs one of the consortium's 18 Biomarker Development Laboratories.
Zheng also collaborates with Dr. Nicole Urban and colleagues in the Pacific Ovarian Cancer Research Consortium to validate potential biomarkers for early detection and prognosis of ovarian cancer.
Scientists use the terms sensitivity and specificity to rate a biomarker's utility, and a good biomarker test will have high rankings in both categories. Sensitivity is a measure of how good a test is at identifying those patients who do have a disease, such as cancer. Specificity reflects the test's ability to distinguish those who are healthy.
A test with high sensitivity and low specificity would yield too many false positives, a complication of the prostate-cancer screening procedure known as the PSA test. Over-diagnosis of prostate cancer due to widespread PSA testing has been blamed for unnecessary treatment of men who would likely never develop the disease.
Potential biomarkers that receive a thumbs-up in the discovery phase of analysis are further analyzed for their potential to be accurately detected in the bodily fluids most likely to be used in an actual diagnostic test in a doctor's office, such as blood or urine. During this stage of biomarker development, biostatisticians help to confirm whether such tests give reproducible results.
Once a reliable test procedure is confirmed, researchers must then evaluate a biomarker's potential for detecting disease in large population studies. Scientists begin with retrospective studies, in which they assess the biomarker's ability to be detected in samples from a group of individuals with cancer but not in a group of healthy individuals.
"The limitations of retrospective studies is that even if the biomarker can identify samples from cancer patients, we don't know how early the biomarker was present during the course of the disease," she said.
Ultimately, the biomarkers must be tested in prospective studies, in which samples are taken regularly from a large group of healthy individuals. Over time, some portion of the group will develop cancer, and with luck, the biomarker test will have detected the presence of the disease at its earliest stage.
Barrett's Esophagus Project
"You need very convincing data to initiate a prospective study, which takes many years and can be very expensive," Zheng said. "So the statistical analyses are very important."
One such prospective study at the center in which Zheng participates is the Seattle Barrett's Esophagus Project, led by Dr. Brian Reid of the Human Biology and PHS divisions. Participants in the study suffer from a precancerous condition known as Barrett's esophagus, which, in a small percentage of patients, will develop into esophageal cancer. Researchers in the study expect the identification of reliable biomarkers to predict who is likeliest to progress to cancer will allow doctors to intervene before the disease becomes advanced.
The prospect of benefiting those with serious illness is what Zheng finds most satisfying about her research.
"At a place like the center, a biostatistician's job is not just looking at numbers," she said. "The idea that my work could actually help people with cancer is very rewarding."