Large collections of matched genomic and clinical data could help oncologists choose the best therapies for patients with acute myeloid leukemia. New research shows that such databases can assist in determining who is most likely to benefit from risky but potentially curative stem-cell transplants.

After achieving initial remission, many patients with acute myeloid leukemia (AML) opt to receive an allogeneic stem-cell transplant to rid their bone marrow of any lurking cancer cells. This intensive treatment decreases the risk of disease recurrence compared with standard chemotherapy, but it also raises the risk of serious complications, such as graft-versus-host disease and death. Balancing the pros and cons can be challenging, but large collections of genomic data matched to clinical variables could aid in decision-making.

According to a new study, databases of this kind—coupled with statistical models that predict likelihood of remission, relapse, and mortality—could spare patients with AML a stem-cell transplant if they are unlikely to benefit, which could improve quality of life and lower health care costs while maintaining overall survival rates (Nat Genet 2017 Jan 16 [Epub ahead of print]). Such “knowledge banks” could also be used to personalize treatment for patients with different types of cancer.

“It's the only way forward in many senses,” says Peter Campbell, MD, PhD, of the Wellcome Trust Sanger Institute in Hinxton, UK, who co-led the study. “Drugs are not going to be a universal panacea across all patients with a particular type of cancer, and there might be ways to use molecular data to predict which patients are going to benefit.”

Campbell previously teamed up with Hartmut Döhner, MD, of Ulm University in Germany, to sequence the coding regions of 111 cancer-related genes from 1,540 patients with AML from a trio of treatment trials. By combining mutational data with cytogenetic measures and clinical outcomes, they stratified the disease into 11 subtypes, each with a distinctive constellation of clinical and genomic features (N Engl J Med 2016;374:2209–21).

Now, in one of the largest analyses of its kind, Campbell, Döhner, and their colleagues have used that dataset to develop a model to predict survival odds associated with different treatments. They based their algorithm on 231 variables in patients' genetic and clinical records, including copy-number alterations, point mutations, and demographic details. They then validated the model using data from an independent cohort of 186 patients from The Cancer Genome Atlas. The algorithm's predictions closely matched patient outcomes in each cohort.

The team concluded that up to one third of individuals could have their treatment altered if a knowledge-bank approach was implemented. For example, in the United States, about 44% of young adult patients with AML undergo a stem-cell transplant. By using the tool to tailor treatment decisions, that could drop to 35% without affecting overall survival.

“This is a powerful study which addresses how we can begin to apply data from population studies to individual patients,” says Jeffery Klco, MD, PhD, of St. Jude Children's Research Hospital in Memphis, TN, who was not involved in the research. Klco notes, however, that the findings rely on retrospective modeling. “Ultimately,” he says, “this algorithm is going to have to be evaluated appropriately in a prospective clinical study.”

According to co–first author Elli Papaemmanuil, PhD, of Memorial Sloan Kettering Cancer Center in New York, NY, the team is adding 3,000 more patient records to the database to refine the algorithm before prospectively testing it.

Papaemmanuil and her colleagues are also implementing the approach for other cancers, starting with myelodysplastic syndrome. The key, she says, will be building, maintaining, continuously updating, and properly analyzing large clinical–genomic databases for any given disease. “The data exist,” she says. “It's more of an organizational challenge.” –Elie Dolgin

For more news on cancer research, visit Cancer Discovery online at http://cancerdiscovery.aacrjournals.org/content/early/by/section.