Understanding Breast Cancer Treatment Through AI

Share article:

Following evidence-based treatment guidelines is the highest standard of care for the treatment of cancer. However, it’s important to know how patients in our region respond to treatment, outside of a research study. A study led by Dr. Mark Levine, a medical oncologist at Hamilton Health Sciences (HHS) Juravinski Cancer Centre (JCC), is underway to conveniently understand the clinical course of breast cancer patients.

“When I began practice in 1981, if I wanted to understand how a group of my patients were doing on a specific treatment, I would do a chart review. This was a task I usually assigned to my residents or research fellows,” said Dr. Levine.

“With all of the technological advances we have today, we still have to resort to the same, labour-intensive, manual process of chart reviews.”

Many different computer systems used to collect information

Currently, in the clinic or hospital, many computer systems collect and store different types of patient information including patient visits at the JCC, chemotherapy orders, hospital admissions, laboratory tests and x-ray results. Unfortunately, these systems, which collectively are referred to as the electronic health record (EHR), often do not communicate with each other.

However, the information stored in the different systems is in the same digital format.

Dr. Levine had an idea that EHR was a treasure trove of patient information and that digital technology, such as artificial intelligence (AI) could be used to tap into the EHR to yield information on the clinical course of a patient’s illness, including diagnosis, treatment, and outcomes. The data provided would be in real-time and would better inform patient care.

“We base treatment decisions on randomized-control trials that are published in oncology journals. To provide the best treatment, we need to know how our patients are doing in the real world, compared to the experience of those on the research studies,” said Dr. Levine.

In 2017, a strategic alliance was formed between IBM and HHS to use AI applied to the EHR to solve the problem of how to collect patient data without doing a manual chart review. Roche Pharmaceuticals provided funding to support the research project.

Using natural language processing to recreate the patient journey

Fifty patients with stage III breast cancer who had completed treatment and had two years of follow-up care constituted the study sample. Data from six different computer systems were selected for each patient. A form of AI called natural language processing (NLP) was used to select specific keywords such as breast cancer, lumpectomy, from doctor’s notes within a patient’s chart. The computer was then trained to bring these keywords together to recreate the patient’s journey.

In a fraction of the time, the team was able to recreate the patient story from the time of the first abnormal mammogram, throughout treatment and the status of the cancer and whether it recurred or not, two years following diagnosis.

The researchers then identified a second sample of 19 patients with stage III breast cancer. The NLP process was applied to their data to describe 10 key breast cancer features for each patient, e.g. histology, stage, estrogen receptor, and type of surgery.

A manual review of the same charts was done to compare the results from the AI system with the gold-standard data in the chart. The AI process was 97% accurate as compared to the actual chart review.

As a result of the success of the pilot study, Dr. Levine and his team at the Escarpment Cancer Research Institute (ERCI), along with Dr. Jeremy Petch at the HHS Centre for Data Science and Digital Health (CREATE) are currently working with Pentavere, a company that specializes in AI on the next phase of the research program. In this next study, 2500 to 3000 breast cancer patients seen at the JCC will be included to show that the technology can be applied to large numbers of patients. The aim is also to expand the program to other hospitals in our region.

“The ability to use data to tell the patient story and understand outcomes in real-time helps doctors manage their patients more effectively. We have an eye to make this approach accessible for hospitals in our region and beyond,” said Dr. Levine.