Am I Going to Die? What That Question Actually Asks
A woman with a new diagnosis of ovarian cancer sat across from her gynecologic oncologist. The staging was not yet complete. The prognosis was uncertain. She asked one question: am I going to die .
That question is not a request for information. It is a request for presence. There is a clinician who can meet it, and one who cannot. The difference between them is not communication skill. It is compassion, and compassion is not what our AI empathy studies have been measuring.
The Distinction That Changes Everything
Empathy is the perception of another person’s emotional state and some resonance with it. It is cognitive and affective. It can be trained, scripted, and, as recent research confirms, convincingly simulated by a large language model. Patients shown AI responses to medical questions rated them as more empathic than physician responses. That finding has been replicated in oncology.
Compassion is something else. It is empathy plus the moral commitment to act on it. The word means to suffer with. A compassionate oncologist does not just recognize her patient’s terror at a new cancer diagnosis. She is moved by it. She stays in the room with it. She follows this patient through four rounds of carboplatin and paclitaxel, through the scan that shows progression, and sometimes through the death. That staying is not a communication technique. It is a moral relationship.
You can sound empathic and be compassionless. You can produce the right words and be entirely absent. Patients in cancer care know the difference. They remember it. Gynecologic cancer diagnoses are recalled with the kind of detail that marks events that have reordered a person’s life. What patients remember is not tone. It is whether someone was actually with them.
— PAYWALL BREAK — Free readers see above. Paid subscribers continue below.
ObGyn Intelligence: The Evidence of Women’s Health is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
What the Empathy Scores Are Actually Measuring
Bioethicist John Lantos argues that most empathy research captures communicative empathy: warm tone, verbal acknowledgment, scripted validation. [2] An LLM trained on millions of human conversations produces exactly that output reliably. The scales were not designed to measure what Jodi Halpern calls emotional reasoning: a disciplined, medically-informed attunement to what illness means in a specific patient’s life.
Ovarian cancer has a five-year survival rate of approximately 49 percent for all stages combined, with stage III and IV disease making up the majority of diagnoses. [1] That number sits behind the question am I going to die? A patient who asks it is not asking for a warm acknowledgment that her feelings are valid. She is asking whether the person across from her can tolerate the uncertainty with her and whether she has found someone who will not flinch. An empathy scale cannot detect that.
The AI finding tells us that physicians are not giving patients enough communicative empathy. That is real and worth fixing. It tells us nothing about whether an algorithm can substitute for the gynecologic oncologist who has committed to being with this patient through whatever comes.
What AI Can Actually Do in Oncology
The contribution of AI to gynecologic oncology care is real and underappreciated in these debates. Clinical tools that synthesize trial data, flag deviations from treatment algorithms, identify patients who have not been offered germline testing for BRCA1 and BRCA2 variants, and generate plain-language summaries of complex pathology findings are genuinely useful. They do not replace the oncologist. They make the oncologist more accurate and better prepared for the conversation that matters.
The clinician who walks into a room knowing that her patient’s BRCA status is pending, that the staging CT is equivocal at the para-aortic nodes, that this patient has a 14-year-old daughter, and that her last three portal messages were about fertility preservation, is a different clinician from one who enters without that synthesis. AI can provide that synthesis. Reducing documentation burden means more time in the room. More time in the room creates the conditions for compassion to function.
What AI cannot do is replace the moment when the oncologist says: I do not know exactly what this will look like for you, but I am going to be here for all of it. That sentence, when it is true, is not a communication strategy. It is a compassionate commitment. The machine cannot make it, and the patient knows the difference.
The Training Problem in Oncology
Lantos observes that medical training systematically erodes empathy through structural mechanisms: the hidden curriculum, evaluation systems that reward diagnostic competence over relational quality, and role models who demonstrate efficiency rather than presence. [2] In gynecologic oncology, the consequences are specific.
Fellows learn to manage technically complex surgical and chemotherapy decisions under intense time pressure, with almost no protected time for the relational skills the work requires. A fellow who can deliver a technically perfect radical hysterectomy but cannot sit in silence with a frightened patient has learned half the job. The half she has not learned is where compassion lives. It cannot be outsourced to a chatbot.
Conclusion
AI empathy scores in oncology measure communicative form. They do not measure compassion, which is the moral commitment to be present to and act on a patient’s suffering across the full arc of a cancer diagnosis and treatment. AI tools can and should improve accuracy, reduce burden, and ensure that no patient with ovarian cancer goes without genetic counseling. The relationship between a cancer patient and her clinician, built across treatment cycles and recurrences, requires a human being who has made a genuine commitment to stay. That is not a sentiment. It is a clinical standard, and it is the one worth measuring.