The Computer That Goes to Medical School

IBM Watson can help physicians make decisions in the treatment of breast and lung cancer

(RxWiki News) New cancer medications and tests are in the news every day. What if a computer could keep track of all this information and help doctors offer you the best possible treatment options?

You may have heard about IBM’s Watson, the supercomputer that became a Jeopardy champ. Watson is now making big strides with regards to cancer treatment, according to a recent study.

Results of the study suggest that Watson can help doctors offer personalized diagnostic and treatment advice to cancer patients.

"Discuss all your treatment options with your oncologist."

The study was conducted by Peter Bach, MD, MAPP, attending physician at Memorial Sloan-Kettering Cancer Center, along with collaborators.

The objective of the study was to examine the accuracy at which Watson, IBM’s famed supercomputer, made treatment recommendations for breast and lung cancer.

Watson uses natural language processing, which means that doctors can ask the computer questions as they would to a human. It then gives the doctors recommendations that they can choose from, thus aiding them in the treatment process.

Watson does not make any clinical decisions on the doctors’ behalf. It is only a support tool.

Watson has had a lot of medical records, guidelines and scientific information fed into it over the years. It makes treatment recommendations based on analysis of all the data it has. It also ranks the treatment options in order of what it believes will work for a particular patient.

The researchers also included anonymous patient treatment plan data for breast and lung cancer from Memorial Sloan-Kettering Cancer Center as well as best practices at the center. They used data from around 1,000 cases to “train” Watson.

Then, the researchers started having Watson make treatment recommendations on new cases. After each round of treatment recommendations made, they added additional elements to train the computer further.

On analysis of the treatment recommendations, they found that Watson’s accuracy in making these recommendations improved from 40 percent accuracy in earlier stages to 77 percent after several rounds of recommendations.

According to the authors, the results suggest that Watson is capable of being trained to make accurate treatment plan recommendations for cancer patients.

It must be noted that more refinement and training is needed before Watson can be widely used in decision support for doctors.

Benjamin Smith, MD, Assistant Professor at the University of Texas MD Anderson Cancer Center, Houston, TX and DailyRx contributing expert finds this study very interesting.

"Practicing oncology is extremely complex and it is difficult, if not impossible, to master the nuances of all different disease sites. Any effort to leverage the massive computing power currently available to design smart applications to help physicians and their patients devise or double check treatment plans should be welcomed. It will be interesting to see whether this strategy can provide sufficient nuance to meaningfully assist good clinicians in the provision of cancer care," Dr. Smith said.

The abstract for this study was presented at the annual meeting of the American Society of Clinical Oncology in June 2013. All findings should be considered preliminary until published in a peer-reviewed journal.

The study was sponsored by IBM in collaboration with Memorial Sloan-Kettering and WellPoint. Dr. Bach did not disclose any conflicts of interest or financial relationships. Some of the collaborators had stock ownership and employment relationships at WellPoint and IBM.

Review Date: 
June 6, 2013