The U.S. Food and Drug Administration has long used randomized statistical tests as the gold standard in deciding whether drugs or medical procedures are safe and effective for the population at large.
“Personalized medicine” is based on the hope that such tests and other statistical methods can help doctors to determine how best to treat any given patient at any given time.
Marie Davidian offered a glimpse of what that might look like last week at the final event of this year’s Wittenberg Series.
The title for her IBM Endowed Lecture in the Sciences was “The Right Treatment for the Right Patient (at the Right Time): Personalized Medicine and Statistics.”
The William Neal Reynolds Professor of Statistics at North Carolina State University told her audience in Hollenbeck Bayley Auditorium the personalized medicine movement currently remains a “quest,” although one with tremendous promise.
In the giddy first days after scientists successfully mapped the human genome, she said, “there was the hope” that researchers might be able to sort out which patients needed which treatments based on their genetic profiles alone.
Although this works in a limited number of situations and with a limited number of drugs, in most instances, finding a “bio-marker” — the piece of evidence that identifies key differences — “may be much more complicated.”
Just as doctors use combinations of age, race, gender, medical history and genetics to make diagnoses and devise treatments, there may be a cluster of bio-markers involved in diseases and their treatments.
With so many variables at work, Davidian said, “routine personalized medicine may be decades away.”
But on the positive side, she said, a vast amount of medical data is becoming available through electronic medical records — data researchers at universities, drug companies and other medical concerns hope they can harness with the power of computers and analyze with powerful tools developed by statisticians.
If they can isolate the “bio markers” that play key roles in diseases and disease processes, they hope to be able to use them to tailor tests and treatments to individual needs.
Davidian said one goal of the overall effort is to provide meaningful enough data to help physicians make the best choice for their patients at each decision point along the way.
She offered as an example a patient with cancer.
After the initial diagnosis, a doctor has to choose which drug to use.
If drug A eradicates the cancer on the first try, Davidian said, the doctor then must decide what treatment, if any, is needed to maintain that success.
If drug A does not work, the doctor may have to choose among choices B, C and D, and if the first again doesn’t work, make additional decisions from there.
Meanwhile, even if drug A was effective, there may be a range of choices along that route of care.
Statisticians hope a method they call the Sequential Multiple Assignment Randomized Trial, or SMART, can track down the variables involved with all these choices at each decision point and provide a statistically significant means of determining which choices work out best.
For instance, if drug A is not effective, the information may allow the doctor to decide whether drug B, C or D is most likely to be effective.
If successful in doing that, she said, statisticians may be able to give doctors better information for designing more formalized regimes of treatment.
“I would never see purely … data driven (measures) being widely used” as the sole means of making medical decisions, Davidian said.
“I hope to see these kinds of things inform” doctors and so they can better decide about the right treatment for the right patient at the right time.