A recent partnership between University of Cincinnati Cancer Institute Director and UC Radiation Oncology Chair William Barrett, MD, and dunnhumbyUSA, a marketing research firm, provided insight that will help inform prostate cancer patients of potential health conditions after undergoing brachytherapy treatment—radioactive seeds placed directly in a tumor to kill cancer cells.
The initiative involved 12 dunnhumbyUSA teams made up of 40 staff members from offices in Cincinnati, Chicago, New York and Sacramento who were given de-identified datasets taken from a UC prostate cancer database to predict how likely a prostate cancer patient was to experience health conditions after going through brachytherapy treatment.
These health conditions included complete elimination of disease, bowel function, pain with urination and impotence.
"Prostate cancer is the most common cancer among American men, and about 1 in 7 men will be diagnosed with prostate cancer during his lifetime,” Barrett says. "Thankfully, multiple treatment options exist with similar success rates for men diagnosed with early-stage prostate cancer. However, the potential for side effects is a major consideration when selecting the treatment process, which is why a project like this is so important.”
Teams were given very detailed pretreatment disease and patient medical characteristics and treatment outcomes. They worked to develop statistical models that could correctly predict outcomes based on the pretreatment characteristics.
The three teams with the best accuracy in predicting health outcomes for the de-identified patients were selected to present in front of a panel of judges and Barrett, and a winner was selected based on their solution approach and model results.
The winning solution used a combination of two model types, known as a regularized regression model and a random forest model. For the regularized regression model, the main factor in predicting the health outcomes was the patient’s age during treatment. The team found that the older the patient was during treatment, the more likely he would experience certain predicted health conditions. The random forest model consisted of 13 variables and the factors differed depending on which health condition was predicted. An example of variables that influenced the model is the prostate cancer stage at the time of the patient’s diagnosis and the percentage of biopsies taken that contained a tumor.
"This model and its insights were provided to our institute to help inform prostate cancer patients of potential health conditions after undergoing brachytherapy treatment,” Barrett continues. "If we use it clinically, it will be used in discussions with newly diagnosed prostate cancer patients. We routinely have a long discussion with each patient about their treatment options including no treatment, surgical prostatectomy, external radiation therapy brachytherapy, hormone therapy or a combination of those.
"We would use the chosen model to tell them that we would predict a higher susceptibility of those four particular health outcomes three years post-treatment if they were to be treated with brachytherapy with the caveat that this tool is preliminary and has not yet been validated in studies but could provide useful information.”
"Our company is embedded in the Cincinnati community. Through our Helping Hands program, we choose to serve many nonprofit agencies with our time, talent and dollars. Our work with UC is an extension of our desire to give back in meaningful ways,” says Stuart Aitken, CEO, dunnhumbyUSA. "Through our customer science approach, we are able to leverage our expertise to effect change in health care.”
"Regardless of whether we choose to use this model or not, this was a successful partnership and certainly something we hope upon which to build in the future,” Barrett says. "We will be exploring ways to collaborate in additional areas.”