Penn Medicine uses AI chatbot “Penny” to improve cancer care

Over the past 25 years, the field of oncology has witnessed a dramatic shift in cancer care delivery with the development of new therapies, a desire to move care from hospital and clinic settings to home settings, and the widespread implementation of electronic health records.


One area of ​​innovation has been the development of oral anticancer drugs, including cytotoxic chemotherapies that patients can self-administer at home. An estimated 25% of new anticancer agents in development are for oral administration NCCN Task Force Report: Oral Chemotherapy.

Using these drugs allows patients to become more autonomous in their care and allows them to spend more time at home, said Dr. Parul Agarwal, an assistant professor of clinical medicine at the University of Pennsylvania Health System who is part of the “Penny” team, notably taking the lead in its role in gastrointestinal cancers.

“However, there are several issues to consider when administering oral chemotherapies at home,” he noted. “Medication schedules can be complex, and there is concern about compromised efficacy if patients take less than the prescribed dose and an increased risk of potentially life-threatening toxicity if more than the prescribed dose is given.

“Recognizing these challenges, several mitigation strategies have been evaluated, including intensive, provider- and pharmacist-led, ambulatory monitoring programs,” he continued. “Furthermore, measuring medication adherence and appropriately identifying and managing toxicities for patients undergoing oral chemotherapy is challenging.”

Capecitabine is a particularly complex oral chemotherapy that serves as the treatment backbone for many gastrointestinal malignancies. It is manufactured in two pill sizes and prescribed twice daily, dosed based on body surface area, and treatment regimens are cyclical, leading to on and off days.

“Our Penny pilot using an augmented intelligence chatbot is a new strategy to improve medication adherence and manage oral chemotherapy agent-related toxicities with the primary goal of improving patient outcomes and the secondary goal of improving health care utilization metrics.

Dr. Parul Agarwal, University of Pennsylvania Health System

“Furthermore, capecitabine is often administered with other intravenous chemotherapies, other oral chemotherapy regimens, or concurrently with radiation which can lead to further complications with the administration,” Agarwal explained. “Typically, monitoring these agents to ensure safe handling requires intensive use of resources and vendors, and the literature documents frequent occurrences of errors and subsequent avoidable complications.

“Our Penny pilot using an augmented intelligence chatbot is a novel strategy to improve medication adherence and manage oral chemotherapy agent-related toxicities with the primary goal of improving patient outcomes and the secondary goal of improving health care utilization metrics,” he added.


In order to improve medication adherence and monitor and manage toxicities with the ultimate goal of providing high-quality patient care, Penny, the algorithm-based augmented intelligence chatbot was proposed.

Penny uses two-way text-based conversational interactions to guide patients through potentially complex regimens and reduce the potential mistakes they may make during chemotherapy cycles.


Patients with gastrointestinal cancer receiving oral chemotherapy with capecitabine alone or capecitabine in combination with other anticancer drugs or radiation were eligible to participate. The Penny chatbot was bi-directional and messages could be initiated by the chatbot or by the patient.

To address medication adherence and dosing issues, the chatbot sent patients medication reminders twice a day about the dose of capecitabine and when to take it. In the first phase of the pilot, participants received a message asking them to respond to TAKEN after each dose of capecitabine. This was later changed to a weekly survey to assess medication adherence.

To address the toxicities associated with oral anticancer agents, the chatbot also launched a weekly survey to elicit patient-reported symptoms. These symptoms were assessed using a combination of algorithmic surveys and natural language processing, a form of artificial intelligence.

Participants can also initiate a message with the chatbot at any point during the process to report a symptom. All interactions between the participants and the chatbot were monitored in real time by the study team.


“The main outcome of this study was to evaluate the feasibility of using the AI ​​chatbot from a patient safety perspective,” Agarwal explained. “Nearly 4,000 drug-related text messages were exchanged and approximately 93% were interpreted accurately by Penny.

“A medication adherence rate of approximately 70 percent was measured in participants who responded to TAKEN, although the actual medication adherence rate was likely much higher,” he continued. “Over 500 symptom-related text messages were exchanged and about 98% were interpreted correctly by Penny.”

Through an iterative process, patient engagement in symptom assessment increased from 25% to approximately 70%, and most of the messages that were misinterpreted were accounted for by the research team in real time, he added.

Over 50% of participants completed qualitative exit interviews. These interviews demonstrated an overwhelming level of patient engagement with clear feedback that Penny provided an extra level of support, increasing their confidence in taking their medication and interacting with their care team, she said.


“The iterative way in which this intervention was implemented led to several improvements in Penny’s natural language processing function and resulted in higher levels of patient engagement as the pilot progressed, which is critical to the innovations health care in this area,” he said Agarwal.

“This process has required a high level of commitment from all parties involved, the healthcare team, the research team and the Memora Team, with whom we have collaborated on a technology platform,” he continued. “Penny identified care delivery inconsistencies with capecitabine schedule, dosing, and monitoring among gastrointestinal cancer providers, leading to initiatives to reduce unnecessary variations in care.”

A patient journey dashboard was created to clearly outline the schedule and dosing with the goal of facilitating the delivery of high-quality care.

“Moving forward, this dashboard will be integrated into the EHR to become a standard tool for clinical teams to use to monitor their patients and adjust dosing as needed,” he said Agarwal.

Follow Bill’s HIT coverage on LinkedIn: Bill Siwicki
Email him:
Healthcare IT News is a HIMSS Media publication.

#Penn #Medicine #chatbot #Penny #improve #cancer #care

Leave a Comment