Brief
Newly diagnosed oncology patients need access to the right provider quickly to establish a plan for treatment, to mitigate any disease progression and reduce stress. Duke Cancer Institute supports easy access to the right provider for these patients, via the new patient access center and a hotline number for prospective patients to call. Scheduling these patients can be time consuming due to manual chart review and complex patient cases. To support our front-line schedulers and sustain the growth of the program, we use artificial intelligence to automatically review a patient’s chart and quickly determine the appropriate provider and time window for scheduling.
Challenge
Ease of access to care for newly-diagnosed oncology patients is vital for both clinical outcomes and patient experience. A majority of newly diagnosed patients experience high levels of psychological distress combined with a desire for information about their care and treatment options at the time of diagnosis.1 Initial experience is a key component in a patient’s choice for where they seek treatment. This is evidenced by studies that show a ‘fast track’ oncology clinic offering results in significant increases in new patient volume.2 As part of new patient intake, a New Patient Coordinator (NPC) reviews the patient’s chart to obtain relevant clinical data, and then applies a set of scheduling rules to determine the correct provider and the correct timeframe for scheduling an initial appointment. This chart review is a manual, time-consuming step that limits the capacity of the patient access center to service a growing volume of prospective patients in need. Additionally, there is a substantial maintenance cost to train new NPCs to perform this scheduling task effectively. There is an opportunity to use artificial intelligence to expedite the chart review and decision-making required to determine the appropriate provider and time window for scheduling.
Solution
To address this challenge, Duke Cancer Institute (DCI) and the Duke Institute for Health Innovation (DIHI) formed a transdisciplinary team to develop and implement the Oncology Access Assistant (OAA). The OAA uses artificial intelligence to automatically review a patient’s chart and determine the appropriate provider and time window for scheduling. When an NPC is triaging a prospective patient, they access the OAA via a secure Web App, type in the patient’s MRN, and click “Calculate scheduling requirements.” The OAA automatically retrieves all the patient’s clinical notes, imaging result narratives, pathology results, lab results, and diagnosis codes from Maestro Care (Duke’s instance of Epic, an electronic health record) over the previous 18 months. It also searches for and retrieves notes from other health systems for the patient via the Care Everywhere functionality within Epic. The Care Everywhere data incorporation is critical for a large portion of these prospective patients, who have received some type of relevant care prior to seeking treatment at Duke. The OAA then feeds the data, along with a refined prompt that includes the scheduling rules, into a large language model (LLM), to produce instructions on which provider and within what timeframe to schedule an appointment. It also provides justification for those results, citing relevant clinical data used to make its determination. To evaluate the solution, 30 retrospective cases were adjudicated by clinical leads on the project team to establish ground truth “schedule with” and “schedule within” answers along with supporting data and reasoning per the scheduling logic. The OAA solution separately assessed the ground truth cases and produced its determination. The initial phase of the OAA solution implementation focuses on support of NPCs who triage new patients who have received a diagnosis related to breast cancer. The future state workflow maintains a “human in the loop” step wherein the NPC reviews the OAA solution output before finalizing the scheduling step.
Outcomes
The evaluation of the Oncology Access Assistant showed a 97% accuracy of the solution. 29 of 30 cases were successfully identified by the OAA solution for “schedule with” and “schedule within” answers and the supporting rational and documentation. Assessment of the case that failed revealed ambiguous language in the documentation about the recency of imaging documentation, which was a contributing point in the scheduling logic for determining the correct answers. The OAA solution went live on March 3, 2026. At the three-week mark post-go live, the OAA solution had been used to support scheduling tasks for 77 new patients. On average, it took the OAA solution 14 seconds to provide the “schedule with” and “schedule within” answers, along with rational and supporting documentation references from the patient’s chart. This approach is more than fifty times faster than a manual review by a NPC, which takes 10-15 minutes per new patient triage. The total time saved by using the OAA solution to review the 77 new patients’ documentation is estimated to be 15.74 hours.
Next Steps
The Oncology Access Assistant’s impact will continue to be assessed for time saved and for accuracy of scheduling recommendations. The project team is expanding the OAA to support new GI Oncology patient triage using the same methods, with a go-live goal of summer 2026.
References
- Whelan, Timothy J., E. Ann Mohide, Andrew R. Willan, Andrew Arnold, Michelle Tew, Scott Sellick, Amiram Gafni, and Mark N. Levine. “The Supportive Care Needs of Newly Diagnosed Cancer Patients Attending a Regional Cancer Center.” Cancer 80, no. 8 (1997): 1518–24. https://doi.org/10.1002/(sici)1097-0142(19971015)80:83.0.co;2-7.
- Basta, Y L, K M Tytgat, H H Greuter, J H Klinkenbijl, P Fockens, and J Strikwerda. “Organizing and Implementing a Multidisciplinary Fast Track Oncology Clinic.” International Journal for Quality in Health Care 29, no. 7 (2017): 966–71. https://doi.org/10.1093/intqhc/mzx143.
