Large language models for post-discharge follow-up in erectile dysfunction care: a narrative review.

Background and Objective: Erectile dysfunction (ED) is common and closely linked to cardiometabolic disease, psychological distress, and relationship problems. Although effective treatments are available, outcomes depend on sustained post-discharge follow-up to assess response, adherence, adverse effects, and psychosocial needs. In practice, ED follow-up is often hindered by stigma, embarrassment, limited access, and fragmented communication. This narrative review examines the potential role, risks, and implementation requirements of large language model (LLM)-assisted follow-up in ED care. Methods: We conducted a targeted literature search in PubMed and Embase up to 15 January 2026, supplemented by Google Scholar for citation tracking. We selected clinically relevant evidence and conceptual work on ED follow-up, conversational artificial intelligence (AI), and safety governance, and synthesized findings thematically. Key Content and Findings: Among 280 included sources, only 10 were ED-specific, while 270 provided transferable evidence from chronic disease, oncology, and mental health follow-up settings. LLMs may reduce communication barriers, enable low-threshold symptom check-ins, reinforce discharge education, and collect patient-reported outcomes more frequently than traditional follow-up. Potential clinical value is greatest for structured, low-risk tasks such as education reinforcement, adherence support, and routine symptom monitoring. However, ED follow-up includes medication safety, nuanced counseling, mental health and relationship contexts, and sensitive data governance. Key risks include inaccurate or unsafe recommendations, privacy breaches, inappropriate handling of mental health red flags, and unclear accountability. A risk-stratified, human-in-the-loop implementation model can balance feasibility and safety by reserving autonomous LLM interactions for low-risk content while triggering clinician review and escalation for predefined high-risk scenarios. Conclusions: LLM-assisted follow-up could become a useful adjunct in ED care if implemented with clear boundaries, auditability, privacy protection, and robust human oversight. A nurse-supervised, trigger-based governance model, which aligns with current follow-up workflows, may enhance continuity and responsiveness without compromising clinical accountability. Prospective evaluations should prioritize patient acceptability, safety outcomes, workload impact, and implementation feasibility.