From Evidence to Empathy: AI-Powered Support for Breast Cancer Patients
Authors: Prof Yanni Wu1,2,3, Meiqin Luo
1. JBI Nanfang Nursing Centre for Evidence-Based Practice
2. Nanfang Hospital
3. Southern Medical University
‘I’m so scared about my surgery tomorrow. Any tips?’
‘My third chemo session is the day after tomorrow. I overthought it again last night and hardly slept.’
‘I finished my treatment months ago, but I still feel anxious every night.’
These inner struggles are commonly shared across online breast cancer support groups. Beyond physical issues and treatment side effects, people face ongoing emotional hardship after a diagnosis. Anxiety, low mood, fear of recurrence, worries about the future, and social isolation are battles that many breast cancer survivors fight for the rest of their lives.
The gap between evidence and experience
Although evidence-based guidelines for breast cancer survivorship care and psychological support already exist, emotional distress in breast cancer survivors often remains under-recognised in routine care. Many patients continue to struggle silently between intermittent clinical visits, yet feel reluctant or embarrassed to take the initiative to seek professional psychological help. Even when their emotional distress becomes overwhelming, timely support is often inaccessible. The fundamental challenge, therefore, is no longer whether credible evidence exists, but whether it reaches patients when they need it most.
Can AI help healthcare listen better?
Many instances of emotional distress among breast cancer patients arise outside clinical settings. These intermittent and subtle issues are hard for medical staff to detect and identify, which poses a major challenge for conventional health care systems. In that context, the integration of AI and medical services offers an effective solution. Large language models excel at contextual comprehension. They can listen to users, recognise emotions continuously, and deliver responses and personalised emotional support, extending care beyond the hospital walls. With this technology, we expect to enhance the overall response efficiency, service accessibility, and outreach of mental health care.
From evidence delivery to empathetic support
We must remember that emotional support goes beyond simply delivering information. While traditional patient education often provides standardised recommendations, patients’ emotional experiences are dynamic, personal, and constantly changing. What patients want is not only evidence-based information but also empathetic, responsive support tailored to their daily lives and emotional needs.
My team and I have been engaged in research related to breast cancer for many years. We have found that emotional distress of patients with breast cancer is often expressed through fragmented online conversations rather than clinical scenarios. To capture these real-life experiences, we have fully involved patients and the public in the development of our AI intervention tools, enabling them to act as active participants rather than passive recipients of care.
For accurate identification of patients’ emotional categories, we developed a breast cancer emotional lexicon based on patient-generated social media texts. Three credible methods were adopted to gather authentic patient narratives. First, expressive emotional writing was completed by 150 female breast cancer patients across different disease stages. Second, in-person semi-structured interviews were carried out with 17 participants, with all recorded conversations fully transcribed into textual data. Third, original posts and comments from breast cancer-related Weibo super topics were retrieved to capture patients’ spontaneous and genuine daily emotional expressions.
Moreover, we developed the ‘Non-pharmacological Management of Breast Cancer Patient and Public Version Guideline’ to provide materials. During the initial stage of selecting health concerns, we adopted semi-structured interviews and emotional lexicon analysis to directly identify patients’ real and core needs, thereby ensuring that the guideline targets their most concerning problems. In the subsequent plain-language transformation stage, we invited patients to review the guideline and provide practical feedback. Their suggestions helped us simplify expressions, improve readability and acceptability, and make the guideline more practical and patient-friendly. These patient participant approaches mean our AI-based intervention can fit patients’ real emotional states and meet their actual needs.
While our AI intervention was co-designed with patient input, its real-world use has allowed us to chart authentic patient experiences and key practical challenges encountered so far. In routine app use, most participants have reported positive feedback regarding its accessible emotional support and easy-to-understand guideline-based AI replies, which have helped to relieve cancer-related stress and confusion in their daily lives. However, noticeable experiential gaps also exist in practice. Middle-aged and elderly patients often struggle with unfamiliar digital operations, which has meant they have adapted slowly to the intelligent intervention system, even with auxiliary guidance materials and online assistance.
Furthermore, traditional fixed-mode AI interactions provide only one-way passive content delivery, and static response rules cannot fully accommodate patients’ fluctuating emotional states and diverse psychological needs across diagnosis, postoperative recovery, and chemotherapy stages. Additionally, these firsthand patient experiences demonstrate that one-time design optimisation is insufficient; continuous patient and public participation is critical to dynamically refining the AI system, shifting patients from passive recipients to active participants in personalised mental health intervention.
Importantly, this ongoing work (supported by the National Natural Science Foundation of China, Grant No. 72304131) is not aimed at replacing health care professionals or human empathy. Rather, it is an attempt to help evidence reach patients in a more responsive, accessible, and compassionate way.
What empathy means in AI-supported care
Empathy in health care is far more than simple emotional comfort. What breast cancer patients need most is not an optimal treatment plan, but simply to be heard, seen, and acknowledged. Online, patients tend to feel more comfortable expressing their vulnerable emotions and reaching out for support. In this context, AI makes it possible for patients’ inner feelings to be recognised and their voices truly heard. Rather than substituting for a clinical diagnosis or human nursing care, AI functions alongside medical staff to offer sustained long-term emotional companionship and empathy.
Challenges beyond technology
Trust and consent
Many patients hesitate to open up to AI. Unprompted emotional recognition and digital intervention also raise concerns regarding informed consent.
In the early version of our app, we displayed real-time emotional recognition results directly for patients to see. However, after usability evaluation, both clinical experts and participants advised us to remove this function. They disliked it, in particular, because occasional inaccurate emotion predictions undermined patients’ trust in the tool and reduced their long-term adherence to follow-up. This flawed initial design stood as a valuable failure, guiding our later revisions to strike a balance between AI transparency and user trust.
Credibility concerns
Patients hold varying levels of trust in AI-generated advice.
We integrated a large language model-powered AI Q&A assistant into our app to help patients quickly access information and receive psychological support. During sensitivity testing, we noticed frequent AI hallucinations, meaning the model would generate untrue or unsubstantiated content. To mitigate this risk, we optimised system parameters and confined all AI responses strictly to evidence from our pre-established guideline: Non-pharmacological Management of Breast Cancer Patient and Public Version Guideline. This adjustment guaranteed the accuracy and safety of outputs to prevent misleading patients.
Crisis safety
AI cannot properly manage severe distress, such as intense depression or suicidal thoughts, leading to potential safety risks.
Our early AI model could only send generic comforting words if users mentioned thoughts of self-harm. It had no function to alert clinical staff right away. This was a risky flaw in our first design. We later added an automatic emergency module, meaning the tool instantly notifies our research nurse once it picks up high-risk words typed by patients. The nurse will call the patient the same day to provide psychological support or arrange clinical referrals as needed.
Digital equity
Older adults and those in under-resourced areas, with limited device access and digital literacy, may be left behind.
Breast cancer patients are mostly middle-aged and elderly. Even when we prioritised participants with higher education and smartphone skills during recruitment, many still struggled to operate the app. To address this challenge, we gave every participant in the intervention group a printed app user guide when they enrolled. We also created short tutorial videos to walk users through key operations. We set up a dedicated patient chat group as well. A designated staff member stays online in the group all day to answer questions and offer real-time guidance, helping patients use the app smoothly throughout the trial.
Moving from evidence to humanity
Evidence-based nursing is evolving from evidence-led practice to humanised care via extensive patient, public and community engagement in AI self-care development. Accordingly, beyond clinical data, authentic patient insights from expressive writing, interviews, and online community input inform professional emotional lexicons and patient-oriented guideline optimisation. Continuous user feedback further refines AI’s interactive and safety functions. This multi-stakeholder participation empowers patients as active co-designers, enabling AI nursing tools to deliver personalised, community-adapted and human-centred self-care support.
References
Boyes, A. W., Girgis, A., D'Este, C. A., Zucca, A. C., Lecathelinais, C., & Carey, M. L. (2013). Prevalence and predictors of the short-term trajectory of anxiety and depression in the first year after a cancer diagnosis: a population-based longitudinal study. Journal of Clinical Oncology, 31(21), 2724–2729. https://doi.org/10.1200/JCO.2012.44.7540
Caruso, R., Nanni, M. G., Riba, M. B., Sabato, S., & Grassi, L. (2017). The burden of psychosocial morbidity related to cancer: patient and family issues. International Review of Psychiatry (Abingdon, England), 29(5), 389–402. https://doi.org/10.1080/09540261.2017.1288090
Downes, S., Krys, T., O'Hara, K., Western, M., Thompson, L., & Brigden, A. (2026). Conversational, Longitudinal, Ecological Assessment (CLEA): exploring a new AI-driven method for qualitative data collection in a behavioural health context. PLOS Digital Health, 5(5), e0001216. https://doi.org/10.1371/journal.pdig.0001216
Fann, J. R., Vanderlan, J., Brewer, B. W., Corbett, C., Keller, J., Lahijani, S., Niazi, S. K., Riba, M. B., Andersen, B., Atreya, C. E., Braun, I., Breitbart, W. S., Channa, Y., Farabelli, J., Fleishman, S., Garcia, S., Greenberg, D. B., Handzo, R. G. F., Horyna, A., Huang, C. H., … Darlow, S. (2026). NCCN Guidelines® insights: distress management, version 1.2026. Journal of the National Comprehensive Cancer Network: JNCCN, 24(4), e260018. https://doi.org/10.6004/jnccn.2026.0018
King, R., Stafford, L., Butow, P., Giunta, S., & Laidsaar-Powell, R. (2024). Psychosocial experiences of breast cancer survivors: a metareview. Journal of Cancer Survivorship: Research and Practice, 18(1), 84–123. https://doi.org/10.1007/s11764-023-01336-x
Li, C., Fu, J., Lai, J., Sun, L., Zhou, C., Li, W., Jian, B., Deng, S., Zhang, Y., Guo, Z., Liu, Y., Zhou, Y., Xie, S., Hou, M., Wang, R., Chen, Q., & Wu, Y. (2023). Construction of an emotional lexicon of patients with breast cancer: development and sentiment analysis. Journal of Medical Internet Research, 25, e44897. https://doi.org/10.2196/44897
Panteli, D., Adib, K., Buttigieg, S., Goiana-da-Silva, F., Ladewig, K., Azzopardi-Muscat, N., Figueras, J., Novillo-Ortiz, D., & McKee, M. (2025). Artificial intelligence in public health: promises, challenges, and an agenda for policy makers and public health institutions. The Lancet Public Health, 10(5), e428–e432. https://doi.org/10.1016/S2468-2667(25)00036-2
Schmidt, M. E., Goldschmidt, S., Hermann, S., & Steindorf, K. (2022). Late effects, long-term problems and unmet needs of cancer survivors. International Journal of Cancer, 151(8), 1280–1290. https://doi.org/10.1002/ijc.34152
To link to this article - DOI: https://doi.org/10.70253/XOWL4781
Disclaimer
The views expressed in this World EBHC Day Blog, as well as any errors or omissions, are the sole responsibility of the author and do not represent the views of the World EBHC Day Steering Committee, Official Partners or Sponsors; nor does it imply endorsement by the aforementioned parties.
