Jump to content

Wake Forest Baptist Health builds chatbot to match cancer patients with clinical trials

An online virtual assistant built with Dialogflow facilitates enrollment in clinical cancer trials during the pandemic.

Umit Topaloglu believes in the power of technology to help solve complex healthcare problems. By combining his two areas of expertise in cancer and informatics, Topaloglu was able to envision an innovative solution to matching cancer patients with clinical trials that could potentially provide life-saving treatments. Working closely with a cancer survivor group and technology partners at Google and Quantiphi, he and his team at the Wake Forest Baptist Comprehensive Cancer Center built a chatbot tool into their website to inform users about available trials at Wake Forest Baptist Health and screen them for eligibility criteria. Launched in May 2020, the initial implementation allows patients to search for active clinical trials for lung, breast, liver, pancreatic, colorectal, and prostate cancers.

There are always some trials that can’t accrue enough patients and therefore have to extend the timeline or modify their design due to lack of enrollment. We realized we could apply machine learning to identify patients and help solve that problem, while complementing existing methods.

Umit Topaloglu, Ph.D., Associate Professor of Cancer Biology and Associate Director of Informatics, Wake Forest Baptist Comprehensive Cancer Center and Center for Biomedical Informatics

Solving problems for both patients and researchers

Randomized clinical research studies are the gold standard for researchers to find new ways to prevent, diagnose, and treat diseases. As an academic health system, Wake Forest Baptist Health conducts hundreds of such research studies every year, with the goal of improving patient outcomes. Data shows that participants in clinical trials receive on average equal or better health results than those who don’t. But there is no single comprehensive source that consumers and patients can use to find clinical trials in the United States. And the challenge of managing trials has become even more urgent during the pandemic, when the healthcare system is strained and more healthcare is remote. “It’s a problem for many academic research institutions,” Topaloglu says. “There are always some trials that can’t accrue enough patients and therefore have to extend the timeline or modify their design due to lack of enrollment. We realized we could apply machine learning to identify patients and help solve that problem, while complementing existing methods.”

Wake Forest Baptist already had a relationship with Google Cloud, and Topaloglu knew about Dialogflow, Google’s natural language processing tool, from his son, who used it for a high school science project. According to Topaloglu, the team considered other vendors but returned to Dialogflow, which offered strong support for multiple languages. Through Google, the team found Quantiphi, a Google Premier Partner for Data Analytics, Machine Learning, and Marketing Analytics, which has already helped other academic institutions solve problems with Dialogflow.

Using machine learning to innovate in healthcare

From the start, the team faced a few challenges. First, in order to design a conversational flow of questions and answers, they needed to anticipate what patients knew and didn’t know about their own condition and treatment, so they could report it. For example, the chatbot asks patients to narrow down their cancer type, subtype, and stage of disease and always gives them the option “I don’t know.” Patients also need to understand definitions of clinical terms, so the team built in definitions too. “Our biggest challenge was designing the conversation flow,” Topaloglu explains. “We worked closely with a focus group of cancer patients and learned a lot from them. That’s a lot of gain.”

Quantiphi trained Dialogflow to analyze the patient’s input as natural language, categorize the questions, and gather relevant information quickly from an evolving knowledge base. They designed a custom interface to generate clear, concise, and standardized responses for thirty to forty questions in real time. The chatbot can also refer patients to a live adviser or help them request an appointment with a clinician. Any personal health information remains confidential to Wake Forest Baptist staff, though patients can choose to download the transcripts of their sessions for future reference.

With clinical studies opening and closing every month, the team has already updated the knowledge base and conversation flow since the chatbot went live. Next, they would like to reach out to more potential patients by increasing the number of cancers covered and adding multi-language support—from Spanish to Arabic. Eventually Topaloglu sees the potential to scale the project beyond Wake Forest Baptist to other institutions. “It’s not easy for patients to find reliable information,” he says. “And technology can help.”

To explore Wake Forest Baptist Health’s chatbot, click here. To read more about similar projects, see how Google Cloud and Quantiphi helped Penn State create a chatbot to support student advising.

Sign up here for updates, insights, resources, and more.