The UCSF Health problem: Cancer survivors at UCSF Health have unmet needs significantly impacting their health-related quality of life (HRQOL). These include managing ongoing side effects of treatment, surveillance for cancer recurrence, informational needs, health promotion and coordination of care. A Survivorship Care Plan (SCPs) is a document that is completed at the completion of curative cancer treatment when the patient is transitioning to survivorship care. SCPs are a patient activation tool that provides patients with important information about the details of their cancer care, follow-up plan, delayed and long-term side effects, and health maintenance. SCPs are considered an important component of comprehensive survivorship care by NCI(1) and Commission on Cancer (COC) that accredit cancer programs. Furthermore, SCPs can also serve as communication tool between primary care and oncology(2). However, the completion of SCP by healthcare providers has been low despite the fact that cancer survivors and primary care physicians (PCPs) consider SCP as an important tool in addressing informational needs of cancer survivors. At UCSF Health most patients do not get SCPs. The major barriers to SCP implementation are: 1. Creation of SCPs is time-consuming for the healthcare provider, 2. These complex documents require manual entry of many details leading to possible inaccuracies, and 3. SCPs are static documents and cannot be easily tailored to cancer types and include updated guidelines with advances in cancer care(3). Current literature suggests that it takes approximately 40 minutes to create a survivorship care plan, and that leaves very little time for discussion of the actual plan in a clinic encounter(4). The discussion between patient and provider of the SCP is the most important part of SCP process which is often unfortunately shortchanged in SCP delivery(5).

How might AI help?  AI/LLM are increasingly being used in many aspects of health care including medical documentation, radiology and pathology diagnostic reporting, patient communications, and decision supports. We expect that the time-consuming process of SCP/ treatment history generation could be taken over by AI, which will allow clinicians to spend more time with patients on discussion of SCP. We can optimize EPIC tools to provide automation of patient information such as demographics, health care team members, diagnosis and staging, and treatment received. We would utilize AI to provide multiple patient-friendly personalized components of the SCP. These components would include: 1. Long-term side effects of specific treatments received and of the cancer diagnosis itself (eg: cardiotoxicity, neuropathy, lymphedema, fear of recurrence), 2. UCSF resources relevant to the patients' diagnosis and demographics, including website and phone numbers, 3. Current NCCN Surveillance guidelines, based on patients' diagnosis, 4. Risk reduction strategies and resources based on the patients' diagnoses, 5. USPTF Age-appropriate healthcare maintenance screening recommendations based on the patients' demographics such as age and gender. We expect that this will include information from reputable organizations such as American Cancer Society, NIH, NCCN, USPTF and local UCSF resources. We expect that AI may be able to create these plans in five to ten minutes instead of 40 minutes allowing encounter to focus on discussion between patient and provider.
How would an end-user find and use it?
The SCPs  will be created by Survivorship/Oncology Healthcare providers who will be trained in the AI enabled process. The SCP will be delivered in the clinic and will be available to patients via MyChart. We anticipate SCP to be accessible to all providers and in the Survivorship section on the oncology snapshot page and attached to the problem in the Problem List. AI will allow these plans to be dynamic and easily updated with the most current resources and guidelines. AI support will be most impactful at the time of the SCP creation, which will be delivered at transition to survivorship care after completion of treatment for patients treated with curative intent. We expect two sets of end users. The first set will be patients who will be sent the SCP via My Chart and we aim to empower patients to manage their surveillance schedule, health care maintenance, and take steps to optimize their wellbeing. The other end user will be the PCPs who will be sent the SCP via MyChart or fax and will utilize the information to provide tailored appropriate primary care to cancer survivors. Finally, we hope that AI enable process will allows us to easily translate these into multiple languages allowing more patients to benefit from this process.

Embed a picture of what the AI tool might look like.   

We anticipate two distinct steps in the creation and implementation of the tool. Step 1 will include knowledge base creation step in which we will create a knowledge base to add to existing data in EPIC/APeX which will include key guidelines for cancer survivorship, relevant medical literature, and chemotherapy database including expected side effects and management of side effects(6). Based on the current literature we will divide this knowledge into the domain of survivorship care including 1. Surveillance, 2. Management of side effects of chemotherapy, 3. Health maintenance, 4. Management of comorbidities, and 5. Coordination of care. Step 2. will be creation and implementation of tool, in this step once clinician identifies a patient for SCP creation, the APeX tool will prepopulate the SCP template with details of treatments, and side effects from problem list, and the AI-enabled SCP will be created using the APeX data and knowledge base. Once the SCP is generated, the clinician will verify its accuracy and share it with the patient. We also expect a task-creation step including surveillance imaging, screening for cancer, and nudges for referrals for services. We will continue to iterate the SCP creation and delivery process to make it patient friendly and helpful to PCPs. We will actively seek feedback from survivorship expert group at UCSF and a patient advisory board to improve the SCP. We expect the SCP to be available in the Oncology Snapshot or problem list in APeX (see below). For patients, it will be available through MyChart.

 Once the SCP generation process is standardized, we will generate and deliver the first 20 SCPs and will evaluate the process with our survivorship expert group and patient advisory board. We will iteratively improve the process to optimize he SCP generation and delivery process.

What are the risks of AI errors?  The biggest risk would be hallucinations, as AI may generate recommendations that are not evidence-based. We will train clinicians who will participate in the pilot to double-check recommendations and links to ensure that recommendations are within guidelines. Site-specific survivorship providers will vet all initial AI generated SCPs to ensure that all SCPs adhere to the guidelines for surveillance and follow up. Finally, we will do a set of 20 SCPs which will be assessed for accuracy and consistency by investigators.   

How will we measure success?  We will use specific metrics to evaluate success. 1. Evaluation of AI-SCP for accuracy and consistency by the investigators and end users, 2. The number of SCP created and delivered to patients, 3. The time needed to create SCP’s. Delivery of SCPs will help us meet the NCI and  COC requirements for Cancer Survivorship Care. Finally, we will evaluate the SCP implementation success using the four item measures of Acceptability of Intervention Measure (AIM), Intervention Appropriateness Measure (IAM), and Feasibility of Intervention Measure (FIM)(7). This will be done as a brief survey for providers creating the SCPs.  

6.A. Measurements below are already being collected in APeX  

1. Number of patients eligible for SCPs: those treated with curative intent 

2. SCP delivery: the numerator is patient receiving SCPs, the denominator is eligible patients. We will evaluate current overall numbers, and site-specific data. What disease sites are not using the SCP currently.  

3. Capturing treatment details including prior chemotherapy, surgery, radiation oncology.  

4. Capturing prevalence of persistent side effects.   

6.B. A list of other measurements you might ideally have to evaluate success of the AI  To evaluate the success, we will need to include measures such as cancer surveillance, HRQOL measures, needs assessment, comorbidity care measures such as Hemoglobin A1C, lipid profile, screening for other cancers in the last one year for patient who have received SCPs.   

4. Describe your qualifications and commitment: Our applicant team includes Dr Niharika Dixit and NP Angela Laffan and cancer survivorship provider group. Dr Dixit is a medical oncologist and Professor of Medicine focused on breast cancer and survivorship care. She provides clinical care at ZSFG and is the Physician-lead of the UCSF Survivorship and Wellness Institute. Dr. Dixit focuses on improving care of cancer survivors across UCSF and Affiliate sites. She has conducted several research projects related to cancer survivorship and cancer survivorship care planning and is committed to optimal delivery of cancer survivorship care using innovative approaches in diverse clinical settings. Dr Dixit additionally serves as co-chair of a Cancer Survivorship Task Force for ASCO and is the co-lead for the UCSF Survivorship and Symptoms Science Research Hub promoting research in cancer survivorship and symptom science. NP Laffan is an Oncology Nurse Practitioner in the GI Medical Oncology and GI Survivorship program at UCSF. NP Laffan is the co-creator of the GI Oncology Survivorship Program and has an active clinical practice providing survivorship care to patients who have completed treatment for GI related cancers.  NP Laffan is also the Clinical Lead in the UCSF Survivorship and Wellness Institute focusing on program development, survivorship research initiatives and clinician education. We are also supported by other survivorship care clinicians who have site specific expertise such as Lung, breast and colon cancer etc. This group meets who meet once a month to discuss survivorship care across UCSF. We will present our planned intervention in these meeting to solicit ongoing feedback, As part of this endeavor we will work with a Patient advisory board to iteratively refine SCPs creation and delivery. Together, we believe that we are a strong team with extensive knowledge of survivorship care including barriers and exciting areas of opportunity. We are energized by this exciting opportunity to work with the health AI team and APeX enabled research team to address a health care issue that has been limited predominantly by the time intensity and complexity of the task. We are confident that AI can assist in rapidly creating SCPs that are patient friendly, easy to access, tailored and actionable which will lead to improved survivorship patient care.

References:  

1.    Mollica MA, McWhirter G, Tonorezos E, Fenderson J, Freyer DR, Jefford M, et al. Developing national cancer survivorship standards to inform quality of care in the United States using a consensus approach. J Cancer Surviv. 2024 Aug;18(4):1190–9.

2.    Hayes BD, Young HG, Atrchian S, Bennett EV, Haynes EMK, Loader A, et al. Optimizing the integration of family physicians into cancer survivorship care in the BC Interior: a mixed methods study of physicians’ opinions and experiences. J Cancer Surviv. 2025 Feb 4;

3.    Birken SA, Deal AM, Mayer DK, Weiner BJ. Determinants of survivorship care plan use in US cancer programs. J Cancer Educ. 2014 Dec;29(4):720–7.

4.    Birken SA, Mayer DK, Weiner BJ. Survivorship care plans: prevalence and barriers to use. J Cancer Educ. 2013 Jun;28(2):290–6.

5.    Dixit N, Sarkar U, Trejo E, Couey P, Rivadeneira NA, Ciccarelli B, et al. Catalyzing Navigation for Breast Cancer Survivorship (CaNBCS) in Safety-Net Settings: A Mixed Methods Study. Cancer Control. 2021;28:10732748211038734.

6.    Pradeepkumar J, Pankaj Kumar S, Reamer CB, Dreyer M, Patel J, Liebovitz D, et al. Survivorship Navigator: Personalized Survivorship Care Plan Generation using Large Language Models. medRxiv. 2025 Mar 28;

7.    Weiner BJ, Lewis CC, Stanick C, Powell BJ, Dorsey CN, Clary AS, et al. Psychometric assessment of three newly developed implementation outcome measures. Implement Sci. 2017 Aug 29;12(1):108.