Patient acceptance of AI in breast disease care is strong, but trust still runs through the physician
A cross-sectional survey from Peking Union Medical College Hospital found broad patient openness to artificial intelligence in breast disease management, especially when AI was used for tasks that feel operational rather than deeply interpersonal. Among 457 patients who completed the questionnaire, 91.7% described AI tools as at least somewhat helpful. Even so, the numbers also showed a clear ceiling on enthusiasm: when AI and physician advice conflicted, 50.8% said they fully trusted doctors, while 43.3% wanted both inputs combined before making a decision.
One result stands out immediately. Patients welcomed AI most readily when it stayed in the lane of sorting, screening, and accelerating access, not when it moved into the core of diagnosis and treatment.
That pattern fits the way many patients experience clinical risk. In breast disease care, triage and pre-diagnosis can be framed as logistical support, while treatment advice touches fear, uncertainty, and the emotional burden of cancer suspicion or recovery. The survey captured that distinction in practical terms: willingness to use AI was 86.0% for pre-diagnosis and 92.1% for precise triage, but fell to 70.0% for clinical diagnosis and treatment assistance. The drop suggests patients are more comfortable letting algorithms help organize care than letting them shape decisive therapeutic judgment.
Emotional support remains the main weak point in human–machine care
The survey also exposed the most persistent reservation about AI in this setting. Two-thirds of respondents, 66.7%, identified lack of emotional support as a major drawback. In breast disease management, that concern is not abstract. Patients often need reassurance, interpretation of risk, and a clinician who can absorb anxiety as well as review imaging, pathology, and surgical plans. A machine may improve speed or consistency, but it cannot replicate the pacing of a difficult conversation, the response to hesitation, or the subtle reassurance that often changes whether a patient feels able to proceed.
Privacy concerns and accuracy worries also remained part of the picture, even as broad acceptance was high. These are not separate anxieties; they reinforce one another. If patients believe an AI system might mishandle sensitive data or produce an unreliable recommendation, confidence in the entire digital workflow weakens. In breast care, where decisions can affect surgery, surveillance, and adjuvant treatment pathways, trust is not a soft variable. It is the mechanism that determines whether the tool is used at all.
Age and education shaped sentiment in opposite directions
The study did not show a uniform patient response across demographic groups. Multivariable ordinal logistic regression found that patients older than 45 tended to hold more negative views of AI tools, while younger participants were more positive. Educational level moved in the same direction only partly: those with postgraduate education or above were also more likely to express negative sentiment toward AI.
Those findings are clinically interesting because they cut against the assumption that more education automatically produces more comfort with new technology. In this survey, greater educational attainment did not translate into unqualified enthusiasm. One plausible explanation is that patients with more schooling may also have a sharper sense of the risks around algorithmic overreach, data privacy, and false reassurance. Older patients may be less immersed in digital tools in daily life, but they may also be more likely to prefer established clinician-led decision-making in a high-stakes setting.
The age signal deserves particular attention. Younger patients often navigate digital systems with less friction, which may make AI feel familiar rather than disruptive. Older patients may be less willing to outsource any part of diagnostic interpretation to a system they cannot interrogate in the same way they question a physician. In breast disease care, where uncertainty already carries a heavy cognitive load, comfort with technology is only one piece of acceptance; perceived safety matters just as much.
What the survey suggests for clinical implementation
The practical message from this cohort is straightforward: patients are open to AI when it supports care rather than replacing the relational work of medicine. Precision triage, pre-diagnosis support, and workflow assistance appear to be the easiest entry points. Direct diagnostic authority and treatment guidance face more resistance, especially when the interaction lacks a human bridge.
For clinicians and health systems, that means implementation cannot be limited to model performance metrics. Even highly accurate tools may fail in practice if patients do not understand what the system does, who reviews its output, or how much weight it carries in the final decision. Disclosure protocols will matter. So will simple explanations about whether AI is screening for risk, prioritizing appointments, summarizing data, or contributing to a recommendation that a physician will still vet.
Breast disease care also places a premium on shared decision-making. Patients frequently need to weigh surgery, follow-up imaging, pathology findings, and treatment tradeoffs under emotional strain. In that setting, AI can be useful if it shortens delays, improves triage, or organizes information before the consultation begins. It becomes far more controversial if it appears to substitute for bedside judgment or remove the clinician from the center of the encounter.
Survey design limits how far the findings can travel
This was a cross-sectional, online questionnaire study conducted among patients who had undergone breast surgery at a single hospital in China between March and May 2025. Of 530 invited participants, 457 completed the survey, yielding an 86.2% response rate. The design is useful for capturing attitudes at one point in time, but it cannot show how those views change after repeated exposure to AI tools, direct education about model performance, or an actual clinical encounter in which AI is used.
The sample also reflects one institution and one clinical context. That limits how broadly the findings can be generalized to other hospitals, other countries, or patients who have not yet entered surgical care. Attitudes may shift in screening clinics, oncology wards, or primary care settings where the emotional and informational stakes differ. Even so, the core signal is difficult to ignore: patients are not rejecting AI. They are drawing boundaries around where it belongs.
Those boundaries define the next phase of implementation. If AI is going to be embedded in breast disease care, it will need transparency, privacy safeguards, and a workflow that leaves physicians visible at the point where reassurance, judgment, and accountability matter most.