“Being a formally trained doctor, I always see things from the doctor’s point of view — having to follow guidelines, the need to follow evidence,” Krakow said. “But having a more personalized way to inform the choice of therapies is important.”
Cancer research is moving faster than ever, and this offers scientists and doctors a great opportunity to use evidence to guide their patients’ treatment. But the sheer scale of the data, the different levels of evidence available, and the complexity of each patient’s case makes it a daunting task even for expert minds trying to objectively choose the best of many treatment options for one patient at a particular time.
“No human can hold that many variables in their mind at one time and weight them appropriately,” Krakow explained.
“When doctors treat patients, we rarely make a once-and-for-all-time treatment decision. We integrate the evidence-based medical literature and our personal and collective anecdotal experiences with our knowledge of the individual patient’s clinical attributes, values and preferences,” she said. Then, doctors see how the patient does and modify therapy as needed.
As a result, she explained, “personalizing treatments is much more complicated than simply applying the results of a randomized clinical trial” — the so-called gold standard of medical research.
Enter “algorithm-informed treatment,” a term she coined to describe the approach she is developing. The idea is to harness computing power, including both traditional statistics and newer machine-learning methods, to do what the human mind cannot: Provide an unbiased, evidence-based look at the relative risks and benefits of a patient’s many different treatment options at a given moment, especially when there is no gold-standard data available, or when the patient differs in important ways from clinical trial participants.
“When doctors treat patients, we rarely make a once-and-for-all-time treatment decision. We integrate the evidence-based medical literature and our personal and collective anecdotal experiences with our knowledge of the individual patient’s clinical attributes, values and preferences."
Through it all, Krakow keeps her mom — and the helplessness she felt as her cancer spread — close. During her treatment, Krakow’s mom made a painting. It expresses her feelings during that time with a chaotic swirl of colors, from which a single word emerges: HELP.
Krakow hung the painting in her office, where she looks at it any time she needs to concentrate on her ultimate goal: improving, and saving, patients’ lives.
Most on her mind these days is group of patients whose leukemia has relapsed despite receiving an aggressive, potentially curative transplant of blood-forming stem cells. There is currently no standard of care for treating these patients. Different cancer centers and even different doctors take different approaches.
“Any time we deal with a post-transplant relapse, we’re skating on thin ice in terms of evidence,” Krakow said.
For example, her clinical team routinely has to consider between two very different treatments for these patients: a second transplant from a new donor, or an infusion of cancer-fighting immune cells from the first donor. These therapies pose very different risks, time investments and costs to patients.
“It’s nice to have evidence from randomized clinical trials to guide those decisions, but we just don’t,” she said.
But by using the wealth of detailed clinical data the Hutch collected over decades on transplant patients, Krakow is developing algorithms to help physicians and patients choose the treatments most likely to prolong their lives, and those most likely to achieve a cure — which, she noted, are not necessarily the same.
The algorithms could be updated as science advances and new treatments are developed, ensuring doctors have access to the most up-to-date, relevant information to help them make better-informed decisions for their patients at these kinds of high-stakes moments.
As Krakow works, a call for help echoes silently around her. She is determined to hear it.
— By Susan Keown, Oct. 1, 2018