SUNDAY, July 20, 2008 (HealthDay News) -- A sweeping genetic analysis suggests that the activity of certain genes might someday allow doctors to predict which lung cancer patients need more aggressive therapies and which do not.
But the findings also underscore the difficulty of making such predictions, especially in the case of people with the earliest forms of the disease, when aggressive therapies could be of greatest value.
The goal is to build effective predictors based on gene expression (activity) and use them prospectively to guide treatment decisions, experts said.
However, to do that, "you have to know what are the potential issues that might influence how well gene expression might predict," said researcher David Beer, a professor in the department of thoracic surgery at the University of Michigan. "I guess the bottom line from this study is that because of the heterogeneity of lung adenocarcinoma, it is not an easy problem. There are still significant issues."
Still, this study -- the most comprehensive yet to date -- could pave the way to more tailored lung cancer treatment based on gene expression profiles, said one expert.
"The goal is five years from now, if I had this data on a stage 1 or stage 2 lung cancer patient, that I could say, 'Hey, you have a very low-risk profile, you don't need chemotherapy' and vice-versa, of course," said Dr. Edward Kim, an assistant professor of medicine in the department of thoracic/head and neck oncology at the University of Texas M.D. Anderson Cancer Center, in Houston.
"It is no different from what we do for breast cancer, where we use certain markers to help doctors make decisions about what treatment they need," Kim said. "This is a step in that direction for lung cancer."
The results were published online July 20 in Nature Medicine.
Beer, along with James Jacobson of the U.S. National Cancer Institute, led the study under the auspices of the NCI Director's Challenge Consortium for the Molecular Classification of Lung Adenocarcinoma, which also includes researchers at the H. Lee Moffitt Cancer Center in Tampa, Fla., the Memorial Sloan-Kettering Cancer Center in New York, the Dana-Farber Cancer Institute in Boston, and the Ontario Cancer Institute in Canada.
The consortium first compiled 442 lung adenocarcinoma samples from six institutions and then divided them into four test sets. For each sample, they collected gene expression data on some 22,000 genes found in these cancer samples. They also looked over clinical information, such as the stage of the cancer and the patients' outcomes.
Consortium members then used two of the test sets, including outcome data, to develop prognostic "classifiers" -- collections of genes whose changes in activity (expressing or producing proteins, for example), whether up or down, predict patient outcome.
Then, the researchers applied these classifiers -- eight were developed overall -- to the remaining two test sets in a so-called validation step. Unlike during the initial "training phase" of the study, patient outcome data at this stage was "blinded." That meant that the researchers had to let their gene signatures (with and without the aid of clinical data) predict patient outcome. Those predictions were then checked against the actual clinical data to measure their accuracy.
The results, said Beer, were mixed.
"We found some [classifiers] work well on one test set but not both, and very few worked well on both, and some of the published signatures did not work very well at all," he said.
Performance was better for tumors of all disease stages than when focusing exclusively on stage 1 disease, he noted. But, in most cases, the addition of clinical data substantially improved the predictions.
For Beer, the data highlight the difficulties of working with such a variable disease as lung cancer, which stems from both genetic and environmental (i.e., smoking) factors.
"It would be wonderful if this was very easy, and you could do it very accurately, but in reality it doesn't work as well as hoped, and we are trying to understand why that is the case," he said. "Why does it work well in some patients but not in others? How do you improve it? How do we identify genes that are prognostic for everybody, or at least for specific subgroups of patients?"
But Dr. Arul Chinnaiyan, a cancer microarray expert at the University of Michigan who was not affiliated with this research, praised the study's design -- particularly its size, use of blinded samples, and multi-institutional format. He also applauded the team's ability to develop and identify gene signatures that work across the various testing sites.
"Many biomarkers as developed often don't hold up across institutions," he said. "Early on, studies are done in an unblended way, at one institution. Often, when another researcher does this, it doesn't validate. That is what is so impressive, that it held up at all these institutions. That points to the robustness of the signature they identified, that it probably will hold up in a clinical setting."
Kim agreed that the study's strength lay in its numbers.
"This is extremely important [work] because they brought everyone together, they have 442 samples for which they have very good gene expression data and clinical data," he said. "And the goal is to grow this so it can be used in a prospective study and hopefully, then, be integrated into our daily clinical practice."
According to Chinnaiyan, the new data suggest that a lung cancer prognosis, like that of breast cancer, could be predicted from gene expression data via a diagnostic test. Two clinical tests, Agendia's MammaPrint and Genomic Health's Oncotype DX, already use the expression of 70 or 21 genes, respectively, to predict which breast cancer patients are likely to suffer a recurrence of disease, and thus might benefit from more aggressive therapies.
The hope is that similar strategies might work for an even bigger killer, lung cancer.
"This is very analogous," Chinnaiyan said.
For more on lung cancer, visit the American Lung Association.