Make an impact in your community by donating or registering for Relay For Life.
Stay informed and inspired!
Subscribe to our monthly e-newsletter×
When cancer spreads, it becomes more difficult to treat. At this stage, doctors often no longer treat the disease with the intent to eliminate it, but to manage the symptoms and reduce pain.
As with so many aspects of cancer, not every advanced cancer acts the same. Some people may live for years after their cancer has spread, while others face a far worse outcome. Yet, predicting how long someone may live is important, not just for the individual to know and make plans, but also to help doctors recommend appropriate treatments such as referring to palliative care.
Currently, computer predictions of how long a patient may live after diagnosis are based on a limited set of data, most of which need to be manually added to the program one by one, limiting the program’s usefulness and accuracy. As a result, many patients receive treatments near end of life that don’t benefit them and are not referred to services that could improve their quality of life.
A team of researchers from Stanford University recently published a study describing a new computer model that more accurately predicts a patient’s survival using thousands of variables included in their electronic health records.
Computer program predicts survival using electronic health records
The researchers built a computer program that scanned the electronic medical records of more than 12,000 people diagnosed with advanced cancer over a ten-year period. They split the group into two, one large set and one smaller set, and used the larger set to “train” the computer to predict survival.
The medical records included thousands of variables including age and sex, the treatments they received and how they responded, and how long they survived after diagnosis. Importantly, this program was able to include doctors’ notes written in the record, which often included details about the patient’s symptoms and their observations about how the patient was responding to treatment. In fact, the doctors’ notes were the most valuable to train the computer to make an accurate prediction of survival.
By analyzing all these variables and linking them to overall survival, the computer program “learned” which were most important in predicting survival.
The researchers then used the program to analyze the records of the smaller group of people to test if the computer could accurately predict how long an individual would survive after diagnosis of advanced cancer.
The researchers found that their program was more accurate than others currently in use, largely because, unlike current programs, it included doctors’ notes in the analysis along with so many more factors. In addition, their program could also be adjusted as patients lived longer and as new information was added to their medical record. It was programmed to weigh recent information more heavily than past information in the analysis, allowing it to be a dynamic, up-to-date prediction.
Accurate prediction may improve quality of life
An accurate prediction of survival when someone is diagnosed with advanced cancer may seem like a lower priority than eliminating the cancer itself. But at this stage of the disease, when eradicating cancer may no longer be feasible, preserving and enhancing quality of life become more important.
Many people with advanced cancer go through aggressive treatments near the end of their lives. While the doctors have good intentions in prescribing these treatments, these interventions may be difficult and painful for the patient and actually decrease their quality of life. Other times, doctors might be overly optimistic about a patient’s prognosis and wait too long before referring them to palliative care.
A computer model that takes into account all of the variables for an individual – their medical history, treatments, response to treatments, and many more – may be a far more accurate way to predict survival and could help doctors make more appropriate recommendations for treatment and care.
Eileen Hoftyzer, BSc