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

Introduction

Erectile dysfunction (ED) affects men across age groups and is associated with significant physical, psychological, and relational consequences. Beyond impaired sexual function, ED often coexists with cardiometabolic disorders and may signal broader health risks (1,2). While pharmacologic therapies (e.g., phosphodiesterase-5 inhibitors) and non-pharmacologic approaches can improve symptoms for many patients, long-term outcomes depend on sustained follow-up after discharge or initial consultation (3). Follow-up allows clinicians to evaluate treatment response, identify adverse effects, address adherence problems, adjust therapy, and respond to psychosocial needs that frequently accompany sexual dysfunction (4).

In routine clinical practice, ED follow-up remains challenging. Many patients disengage from follow-up care, discontinue treatment without consultation, or provide limited feedback (5). Stigma, embarrassment, and concerns about masculinity can inhibit open communication, particularly when follow-up relies on in-person visits or telephone calls that may feel uncomfortable or time-consuming. These barriers can lead to incomplete clinical information, delayed optimization of therapy, and persistent symptom burden (6-8). Nurses often play a central role in follow-up pathways, including education reinforcement, symptom triage, adherence counseling, and coordination of referrals, yet their capacity is constrained by staffing pressures and limited contact opportunities (9).

Large language models (LLMs) can engage in naturalistic conversation, summarize information, and support structured data collection (10,11). In theory, LLM-based conversational agents may provide patients with a more private and non-judgmental channel to report symptoms and concerns, potentially lowering communication barriers in sexual health care (12,13). LLMs may also support frequent check-ins, reinforce discharge instructions, and identify early signs of treatment failure or adverse events. However, ED follow-up is safety-sensitive: incorrect advice about medications, psychological distress, relationship conflict, or comorbid conditions may cause harm (14,15). Therefore, LLMs should not be viewed as autonomous substitutes for clinical follow-up. Instead, safe implementation likely requires a human-in-the-loop model in which healthcare professionals remain responsible for monitoring, verification, and escalation (16,17).

This narrative review examines the main barriers to effective ED follow-up, the follow-up tasks that may be supported by LLMs, and the governance strategies needed for safe implementation. We synthesize representative evidence and conceptual work on digital follow-up, conversational AI, and safety governance to explore how LLM-assisted follow-up may be integrated into ED care while preserving clinical oversight and patient safety. We present this article in accordance with the Narrative Review reporting checklist (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-1-0097/rc).

Methods

Study design

This study was conducted as a narrative review to synthesise and critically discuss the potential role of LLMs in post-discharge follow-up care for patients with ED. Given the emerging nature of generative AI in clinical practice and the limited volume of ED-specific evidence, a narrative approach was chosen to integrate heterogeneous empirical and conceptual sources and to derive clinically transferable implementation principles rather than quantitative effect estimates.

Literature search strategy

We conducted a structured search of PubMed and Embase from database inception to 15 January 2026. Google Scholar was used for supplementary searching (citation tracking and identification of additional relevant records). Because ED-specific evidence on LLM-assisted follow-up remains limited, we used two complementary search streams to balance ED specificity with the identification of transferable follow-up workflows and safety governance:

• ED-specific stream (primary search): studies and commentaries explicitly addressing ED (or sexual dysfunction) follow-up in connection with LLMs, chatbots, or conversational AI.
• Transferable stream (supplementary search): LLM- or chatbot-enabled follow-up, remote monitoring, continuity-of-care workflows, and human-in-the-loop safety models in other chronic or sensitive conditions that share key follow-up task characteristics with ED (e.g., reliance on patient-reported outcomes, barriers to disclosure, psychosocial distress, behavioural support needs, escalation/triage requirements). Conditions were not restricted a priori and could include, but were not limited to, mental health disorders, chronic pain, sleep disorders, chronic urological symptoms, and other long-term conditions involving sensitive symptom reporting.

Search strategies combined controlled vocabulary (e.g., MeSH/Emtree) and free-text terms. The full database-specific search strategies are provided in Appendix 1, and a summary of the overall search approach is shown in Table 1. We also performed supplementary citation chasing to identify relevant governance, implementation, and safety literature that may not have been indexed with ED terms. Only English-language articles were considered.

Study selection and synthesis

We included publications that:

• Addressed the use of LLMs, chatbots, or conversational AI in post-discharge follow-up, remote monitoring, or continuity of care in ED/sexual dysfunction;
• Evaluated LLM- or chatbot-enabled follow-up interventions or workflows in non-ED conditions with follow-up characteristics comparable to ED;
• Provided conceptual, ethical, regulatory, or implementation-focused discussion relevant to risk stratification, escalation, and human-in-the-loop models for clinical follow-up.

We excluded studies focusing solely on technical model development without clinical workflow context, non-conversational AI tools not intended for patient follow-up interactions, and one-off diagnostic applications not linked to ongoing follow-up processes. Given the narrative design, we did not apply formal risk-of-bias tools; instead, we prioritised clinical relevance, transparency of methods, and alignment with follow-up workflow questions.

Data synthesis and analysis

No meta-analysis was planned because of heterogeneity in study designs, populations, interventions, and outcomes. Findings were synthesised narratively with emphasis on (I) communication barriers in ED follow-up, (II) candidate LLM-enabled tasks, and (III) safety governance and workflow design. Evidence from non-ED populations was used to derive transferable design principles rather than infer direct effectiveness in ED populations, and limitations of cross-condition transferability were explicitly considered.

Use of AI tools

ChatGPT (OpenAI) was used only for language editing and readability improvement. It was not used to generate scientific content, determine the review focus, select studies, perform analyses, or produce figures/tables. The authors reviewed and approved all edits and take full responsibility for the final manuscript.

Results

Overview of included literature

A total of 10,045 records were identified through database searching (PubMed: 2,793; Embase: 7,252), including 149 in the ED-specific stream (Path A) and 9,896 in the transferable-evidence stream (Path B). After removal of 1,384 duplicates, 8,661 records were screened and 468 full-text reports were assessed (Path A: 45; Path B: 423). Following full-text review, 280 sources were included in the narrative synthesis (Path A: 10; Path B: 270). Most exclusions reflected lack of relevance to follow-up workflows, absence of a clinical AI component, non-clinical content, or article types not eligible for narrative synthesis. Details of the study screening and selection process are presented in Figure 1.

Figure 1 PRISMA 2020 flow diagram of literature search and study selection. AI, artificial intelligence; ED, erectile dysfunction.

LLM evidence in the ED-specific stream

Direct empirical evidence for LLM-assisted ED follow-up was limited, and no prospective clinical studies of post-discharge ED chatbots were identified. Available ED-specific sources mainly involved three areas: evaluative studies of general-purpose LLMs in sexual health and urological queries, preliminary telemedicine or digital follow-up applications in urology and oncology, and LLM-assisted informatics tasks within urological workflows. Overall, the ED-specific literature remains early-stage and is focused more on performance evaluation and workflow support than on prospective implementation of LLM-assisted follow-up pathways.

LLM evidence in the transferable stream

The transferable stream yielded broader evidence relevant to implementation. Key themes included conversational symptom monitoring and patient-reported outcome collection, triage and risk stratification support, human-in-the-loop governance models, variable safety and guideline adherence in clinical query handling, and privacy, ethical, and accountability challenges in AI-mediated follow-up. This evidence informed the implementation framework and risk-stratified model proposed in this review, while cross-condition transferability limitations are considered in the Limitations section.

A comparison of LLM capabilities and current clinical practice in post-discharge ED follow-up is presented in Table 2.

Discussion

Communication challenges in post-discharge follow-up for ED

Effective post-discharge follow-up is central to long-term ED management because it enables assessment of treatment response, adverse effects, adherence, and psychosocial concerns (18). However, ED follow-up is often suboptimal in routine practice. Many patients disengage from care, provide limited symptom feedback, or discontinue therapy without informing healthcare professionals, reducing opportunities for timely intervention and individualized treatment adjustment (19-21).

A major barrier is impaired patient-provider communication. Sexual health remains a sensitive topic, and embarrassment, shame, or fear of judgment may prevent patients from openly discussing erectile difficulties, treatment failure, or emotional distress, particularly during follow-up encounters that require proactive disclosure (2,22). As a result, clinicians and nurses may receive incomplete information and underestimate ongoing symptoms or unmet needs. Conventional follow-up modalities, including outpatient visits and telephone calls, do not consistently overcome these barriers, especially when time constraints and lack of privacy further inhibit disclosure (23).

From a workflow perspective, ED follow-up is also resource-intensive and difficult to standardize. Follow-up contacts often focus on medication use, perceived efficacy, and side effects, yet the quality of information obtained depends heavily on patient engagement. Nurses may encounter vague or avoidant responses, making it difficult to distinguish benign concerns from clinically significant issues (24,25). Limited staffing and increasing outpatient workloads further restrict the frequency and depth of follow-up, contributing to missed opportunities for early detection of treatment failure, adverse events, or psychological distress (26,27).

These challenges reflect a mismatch between the sensitive, longitudinal nature of ED care and the episodic structure of conventional follow-up pathways. Patients may need repeated, low-threshold opportunities to communicate concerns privately, while healthcare teams need scalable tools to support ongoing monitoring without excessive burden (27,28). This gap provides a rationale for exploring alternative follow-up approaches, including generative AI and conversational systems, to improve communication, engagement, and safety in post-discharge ED care.

Potential roles of LLMs in ED follow-up

LLMs offer opportunities to address longstanding communication and continuity challenges in post-discharge follow-up for ED (29). Unlike rule-based digital tools or static questionnaires, LLMs can engage in flexible, conversational interactions that resemble natural dialogue. This feature is especially relevant in ED care, where patients may struggle to articulate concerns or may avoid structured, clinician-led questioning (30,31). By providing an anonymous and non-judgmental interface, LLM-based systems may lower psychological barriers and encourage more open disclosure of symptoms, treatment experiences, and emotional concerns (32).

One potential role of LLMs in ED follow-up is facilitating patient-reported outcome collection and symptom monitoring. Through regular conversational check-ins, an LLM-assisted system can prompt patients to describe treatment effectiveness, adherence, and side effects in their own words, rather than through rigid scales alone (33,34). Such interactions may capture clinically meaningful information that is often missed during brief outpatient visits or scripted telephone calls. From a nursing perspective, this approach may improve the completeness and timeliness of follow-up data, enabling earlier identification of patients who are struggling with treatment response, experiencing adverse effects, or disengaging from care (35).

LLMs may also support patient education and self-management following discharge. ED treatments frequently require ongoing guidance regarding medication timing, lifestyle modification, expectations of response, and management of common side effects. LLM-based follow-up tools can reinforce discharge instructions, respond to frequently asked questions, and tailor educational content to individual patient concerns. The ability to deliver information in plain language and an empathetic tone may enhance patient understanding and confidence, particularly for individuals who feel uncomfortable asking questions directly to healthcare professionals. In this way, generative AI can complement, rather than replace, the educational role traditionally performed by nurses during follow-up care (36-38). Recent adjacent evidence from andrological surgery also suggests that natural language processing (NLP)- and LLM-based tools may support patient education, improve workflow efficiency, and facilitate semi-automated follow-up and symptom triage, although direct evidence in ED-specific post-discharge follow-up remains limited (39).

In addition, LLM-assisted systems may function as an early triage mechanism within ED follow-up pathways. By continuously monitoring patient input, generative AI can help flag patterns or statements suggestive of treatment failure, significant adverse effects, or psychological distress. When integrated into a structured follow-up workflow, such systems could prompt escalation to human review when predefined thresholds or concerning cues are detected (40-42). Representative escalation triggers and corresponding follow-up actions are outlined in Table 3. This capability may improve safety and responsiveness in post-discharge care, while reducing the burden of routine follow-up tasks on clinical staff.

Taken together, these potential roles suggest that generative AI and LLMs may be well suited to support post-discharge ED follow-up by enhancing communication, continuity, and patient engagement. However, the same characteristics that make LLMs appealing also raise concerns regarding accuracy, appropriateness, and accountability. These risks require careful consideration before integrating such tools into clinical follow-up pathways in sensitive sexual health care.

Challenges and risks of generative AI-assisted ED follow-up

While LLMs may enhance communication and continuity in post-discharge ED follow-up, their clinical deployment raises distinct safety, ethical, and governance risks (30,37). These risks are amplified in sexual health care, where conversations are highly sensitive and the consequences of inappropriate advice—whether medical, psychological, or relational—can be substantial. A credible ED follow-up pathway therefore requires explicit boundaries regarding what an AI system is permitted to do, when human review is mandatory, and how accountability is maintained.

Clinical safety and content reliability

A primary concern is clinical reliability. LLMs generate probabilistic language outputs rather than clinically reasoned decisions, which can lead to omissions, overconfidence, or incorrect statements (43). In ED follow-up, seemingly “minor” inaccuracies may still carry risk. Examples include inappropriate reassurance in the presence of concerning symptoms, incomplete counseling about expected response and dosing, or failure to recognize contexts in which ED symptoms may indicate broader health deterioration. Importantly, ED follow-up often requires interpretation of symptom narratives (e.g., fluctuating response, mixed psychogenic and organic features, partner concerns) that can be difficult to standardize; without structured clinical oversight, an LLM may misinterpret these narratives and provide guidance that does not match the patient’s true situation (30,44).

Medication contraindications, interactions, and red flags

Medication-related harm is a specific and clinically important risk in ED care. First-line therapies, particularly phosphodiesterase type 5 inhibitors, involve contraindications and drug-drug interactions that must be carefully assessed (e.g., concurrent nitrate use, unstable cardiovascular status, severe hypotension, or complex polypharmacy). A conversational AI that does not have reliable access to an individual’s medication list and comorbidities may generate unsafe recommendations (45-47). Additionally, patients may present red-flag symptoms during follow-up (e.g., severe chest pain, syncope, sudden neurological symptoms, prolonged painful erection, or severe adverse reactions) that require urgent escalation rather than self-management advice. An ED follow-up system must therefore treat clinical triage as a core safety function, rather than an optional feature (48-50).

Privacy, confidentiality, and trust

Privacy and confidentiality represent another critical domain. Sexual health data are particularly sensitive, and patients’ willingness to engage with follow-up often depends on trust that information will remain confidential (51,52). The integration of generative AI introduces risks related to data storage, secondary use of conversation data, access control, and potential exposure through third-party platforms. From a clinical perspective, loss of confidentiality may lead not only to psychological harm but also to disengagement from care, undermining follow-up effectiveness (53,54). Transparent communication about data handling, clear consent processes, and secure deployment (rather than consumer-grade tools) are therefore essential prerequisites for implementation.

Ethical accountability and the limits of “simulated empathy”

Ethical considerations extend beyond privacy to accountability and the limits of AI-mediated care. Patients should be clearly informed when they are interacting with an AI system and what the system can and cannot do (55,56). Responsibility for clinical decisions must remain with licensed healthcare professionals, and the care pathway must avoid any ambiguity that could shift accountability to an automated tool (57,58). In addition, ED management often includes psychological support, expectation setting, and relationship-sensitive counseling. Although LLMs may produce language that appears empathic, such “simulated empathy” cannot replace professional judgment or appropriately respond to complex emotional distress. Over-reliance on AI may inadvertently reduce meaningful human engagement, particularly for patients with significant anxiety, depressive symptoms, or relationship conflict (59,60).

Taken together, these risks do not preclude the use of generative AI in ED follow-up; rather, they indicate that implementation must be designed around safety governance. In practice, this requires a risk-stratified model in which LLMs support low-risk and structured follow-up tasks, while predefined triggers mandate human review and escalation. Such a human-in-the-loop approach is critical to preserve patient safety, clinical responsibility, and trust in AI-assisted sexual health care.

Implementing a risk-stratified, human-in-the-loop model for ED follow-up

A practical pathway for integrating generative AI and LLMs into post-discharge ED follow-up should be built around two principles: (I) risk stratification of follow-up interactions and (II) human-in-the-loop oversight for clinically meaningful decisions. Rather than positioning an LLM as an autonomous clinician, a safer and more feasible approach is to use it as a communication and workflow tool that supports structured follow-up tasks, while reserving clinical judgment, prescribing decisions, and escalation for healthcare professionals. This model aligns with outpatient workload realities and the sensitive nature of sexual health care, in which trust, confidentiality, and individualized counseling remain central.

Patients may engage through scheduled check-ins or patient-initiated messages. The system can capture structured follow-up information and generate a concise summary with key points and risk flags for review. Interactions are then stratified into low-, moderate-, and high-risk tiers linked to predefined response pathways. When risk cannot be confidently classified as low, the system should default to human review rather than provide potentially unsafe guidance.

Risk stratification: matching oversight to clinical acuity

Risk stratification in AI-assisted ED follow-up should be operationalized through clinically defined low-, moderate-, and high-risk tiers, each linked to a prespecified response pathway. This approach allows the LLM to support communication and structured data capture while ensuring that clinically meaningful judgment remains under human oversight (42,61). A proposed risk-stratified, human-in-the-loop workflow for AI-assisted ED follow-up is presented in Figure 2.

Figure 2 Risk-stratified, human-in-the-loop pathway for AI-assisted ED follow-up. AI, artificial intelligence; ED, erectile dysfunction; FAQs, frequently asked questions; LLM, large language model.

Low-risk interactions typically involve routine check-ins, clinician-approved educational content, expected treatment-response questions, and non-urgent administrative support. Examples include stable treatment response, requests for general lifestyle advice, or mild expected side effects without alarm features. In this tier, the LLM may guide structured questioning, reinforce approved content, collect patient-reported outcome data, and generate a concise summary for later clinician review. Immediate human response is not required, although periodic audit of low-risk interactions should be performed for quality assurance.

Moderate-risk interactions involve concerns that require clinical judgment but do not indicate immediate danger. Examples include persistent non-response to treatment, repeated adverse effects, unclear non-adherence, emerging psychosocial distress, or uncertainty about whether a symptom is treatment-related. In such cases, the LLM may support structured data collection and provide neutral acknowledgment, but the pathway should require timely review by a nurse or clinical coordinator. The reviewing clinician can then determine whether medication adjustment, further counseling, referral, or physician assessment is needed.

High-risk interactions involve potential immediate harm or clinical urgency and should trigger escalation rather than extended automated dialogue. Examples include cardiovascular red flags, concurrent nitrate use, priapism, new neurological symptoms, acute suicidality, crisis-level psychological distress, or suspected abuse. In these situations, the LLM should immediately display standardized emergency guidance, classify the encounter as a critical escalation event, and notify the responsible clinical team without delay. These triggers should be hard-coded, non-bypassable, and subject to regular audit.

Escalation responsibility and oversight

Responsibility for defining escalation thresholds lies with the supervising clinical team rather than the AI system itself. The LLM should apply prespecified triage rules consistently, but clinicians remain accountable for determining which findings qualify as low, moderate, or high risk, for updating those rules as evidence and practice evolve, and for reviewing incidents in which escalation was delayed, missed, or inappropriate. Governance documentation should therefore assign explicit responsibility for threshold-setting, rule revision, and safety oversight to named clinical roles, such as the lead nurse and responsible physician, so that accountability for patient safety is never ambiguous.

Human-in-the-loop workflow: defining roles and escalation triggers

A human-in-the-loop workflow requires explicit definition of what the LLM can do independently and where human intervention is mandatory (62). In practice, the LLM can conduct standardized check-ins, collect and structure patient narratives, provide low-risk educational reinforcement, and generate concise summaries for review. Human clinicians—often nurses coordinating follow-up—should review moderate- and high-risk interactions, validate summaries, initiate escalation to physicians when medication review is indicated, and coordinate further clinical evaluation or intervention (63,64). The pathway should include predefined escalation triggers based on symptom content, medication risks, and psychosocial cues. When uncertainty exists, the system should default to human review rather than offering potentially unsafe guidance.

Patient-centered design and nursing practice integration

From a clinical and nursing perspective, the value of this model lies in improving both patient experience and workflow efficiency without compromising safety (65). Patients may benefit from a low-threshold channel to communicate sensitive information privately, receive timely clarification, and feel supported between visits. At the same time, nursing staff can shift from repetitive routine calls to a supervisory and intervention-focused role. For example, nurses could oversee a dashboard that displays (I) patient-reported outcomes over time, (II) flagged interactions requiring review, and (III) summaries of unresolved issues. This approach supports continuity by enabling proactive intervention based on trends (e.g., declining perceived efficacy, repeated non-adherence, escalating distress) rather than relying solely on sporadic clinic encounters. Patient consent and expectation-setting are essential: patients should be informed that the system is AI-assisted, does not replace clinical care, and that clinicians remain responsible for decisions.

Governance, documentation, and evaluation for safe implementation

Safe implementation requires governance structures that address privacy, documentation, and ongoing evaluation. ED follow-up systems should be deployed in secure clinical environments with access control, audit trails, and clear policies for data retention and secondary use (66,67). AI outputs should enter the clinical record when they inform care decisions, and conversation logs should be preserved to support accountability and quality monitoring. Continuous evaluation is needed to identify failure modes, including inappropriate reassurance, misleading education, and missed escalation, and to refine triage rules and prompts (68).

Beyond institutional governance, broader safeguards are needed for responsible deployment at scale. Healthcare institutions should clearly define intended use, scope limitations, and escalation protocols within their clinical governance framework, and regular audit should include retrospective review of escalated cases, missed triggers, and patient-reported concerns (69,70). Clinicians who use AI-generated summaries remain responsible for verifying their accuracy and completeness (70,71).

At the regulatory level, frameworks such as the Food and Drug Administration (FDA)’s action plan for AI/machine learning (ML)-based software as a medical device (SaMD), the European Commission’s AI Act, and NHS AI Lab governance guidance illustrate emerging infrastructure relevant to LLM-based follow-up (72). These approaches address pre-market evaluation, post-market surveillance, transparency, and continuous model updating, and institutions should monitor evolving requirements for explainability and bias auditing (73,74).

Patient-facing transparency is also essential. Patients should be informed that they are interacting with an AI system rather than a human clinician, should be able to reach a human at any point, and should receive clear information about how AI-generated data will be used, stored, and shared (75,76).

Outcome evaluation should extend beyond technical performance to include patient-centered and clinical measures such as follow-up engagement, adherence, timeliness of escalation, patient satisfaction, and clinician workload (77). Overall, a risk-stratified, human-in-the-loop model offers a feasible pathway for AI-assisted ED follow-up, but prospective real-world evaluation remains necessary to define consensus escalation criteria, minimum safety standards, and adaptable implementation guidance.

Regulatory and medico-legal considerations

The deployment of LLM-based follow-up systems in clinical sexual health care raises a distinct set of regulatory and medico-legal questions that are not fully addressed by existing AI governance frameworks and that warrant specific consideration in implementation planning.

Regulatory classification

In many jurisdictions, AI-enabled clinical tools that provide information intended to inform clinical decisions may be classified as SaMD. In the United States, the FDA regulates SaMD and has published guidance on AI/ML-based tools, including requirements for pre-market notification [510(k)] or de novo classification depending on the risk level and intended use. A post-discharge follow-up chatbot that provides triage recommendations or escalation flags could fall within the scope of SaMD regulation if it is intended to support diagnostic or therapeutic decision-making. Developers and healthcare institutions must conduct a regulatory assessment before deployment to determine applicable requirements. In the European Union, AI systems used in healthcare are subject to both the Medical Device Regulation (MDR) and the AI Act, which classifies high-risk AI systems—including those used in medical diagnostics and patient management—as subject to conformity assessment, technical documentation, and post-market monitoring obligations (72).

Medico-legal liability

Responsibility for harm arising from AI-assisted follow-up remains legally and ethically contested. When a patient experiences an adverse outcome following an LLM interaction—for instance, if a medication contraindication is not detected, or an escalation trigger fails to fire—questions of liability may involve the software developer, the deploying institution, the supervising clinician, and the reviewing nurse. Current medico-legal frameworks in most jurisdictions do not explicitly address AI-mediated harm in clinical follow-up, creating a responsibility gap that institutional policies must address proactively. Clear documentation of the human review chain, the scope of the AI system’s role, and the thresholds and rules embedded in the triage logic is essential for legal defensibility. Institutions should seek legal and clinical governance advice before deployment, and clinicians should be aware that professional indemnity may not automatically extend to harms arising from unreviewed AI outputs (70,71).

Data privacy and sexual health information

Follow-up data collected from ED patients constitutes sensitive personal health information and, in many frameworks, is subject to heightened protection. In the United States, Health Insurance Portability and Accountability Act (HIPAA) applies to protected health information (PHI) held by covered entities and their business associates, including AI system vendors. In the European Union, the General Data Protection Regulation (GDPR) imposes strict requirements on the processing of health data, including explicit consent, purpose limitation, data minimisation, and the right to erasure. Sexual health data may attract additional scrutiny given its sensitive nature and potential for discrimination or stigma if disclosed. AI system deployment agreements must include data processing agreements that specify data retention periods, secondary use restrictions, data storage location (particularly for cross-border data transfers), and breach notification obligations (78).

Informed consent

Patients enrolled in LLM-assisted follow-up pathways should provide informed consent that covers: (I) the nature of the system (AI, not a human clinician); (II) the types of data collected, processed, and stored; (III) how AI-generated outputs are used in clinical decision-making; (IV) the right to withdraw from AI-assisted follow-up and revert to standard care at any time; and (V) how data will be managed if the system is discontinued (76). Consent processes should be accessible, jargon-free, and available in patients’ preferred languages.

Integration into clinical workflows: electronic health record (EHR), nurse-led follow-up, and telemedicine pathways

Effective integration of LLM-assisted follow-up into clinical practice requires alignment with existing workflows, health information systems, and care delivery models. Three contexts are particularly relevant.

EHR integration

For LLM-generated summaries to support clinical decision-making, they must be accessible within the clinician’s primary working environment—the EHR (79,80). Integration may take the form of structured fields from chatbot interactions or summary notes entered after nurse review and approval. Because most commercial LLM platforms do not natively support Health Level Seven Fast Healthcare Interoperability Resources (HL7 FHIR) or Clinical Document Architecture (CDA) standards, integration often requires local infrastructure or middleware solutions. AI-generated content should not enter the clinical record without clinician review and authorization (71,79).

Nurse-led follow-up clinics

The proposed model is intended to augment rather than replace nurse-led follow-up. Nurses remain the primary human interface: they review flagged summaries, make triage decisions, communicate with patients regarding moderate-risk concerns, and coordinate physician escalation when indicated (81). LLM-assisted follow-up may reduce low-acuity contacts by handling routine check-ins and education delivery, but implementation should include nurse training and workload monitoring to avoid unintended burden (70,82).

Telemedicine pathways

LLM-based follow-up is conceptually well-aligned with telemedicine frameworks already used in urology and sexual health care, particularly post-COVID-19 expansions in virtual care (73,83,84). In settings with established telephone or video follow-up, chatbot check-ins may serve as interim contacts between scheduled teleconsultations. These interactions should supplement rather than replace scheduled reviews, with flagged moderate- or high-risk cases escalated promptly to clinician-led telehealth follow-up (82,83).

Technical considerations: prompt design, fine-tuning, and hallucination mitigation

The reliability of LLM outputs in follow-up settings depends heavily on system configuration. Prompt design can constrain responses to clinically appropriate, pre-approved content, while retrieval-augmented generation (RAG) can reduce hallucinated or off-guideline output by grounding responses in curated institutional knowledge (85,86). Fine-tuning may further improve relevance, although it adds validation and regulatory complexity.

A particular concern in medication-adjacent follow-up tasks is the generation of plausible but inaccurate dosing or drug interaction information—so-called “hallucinations” (87,88). Medication-related responses should therefore be restricted to pre-approved content or routed to human review, with automated filters used to flag uncertainty or complexity for nurse escalation. These measures do not eliminate hallucination risk entirely, but they reduce the likelihood of clinically consequential errors reaching patients without human oversight (88).

Future research agenda

Despite the conceptual promise of LLM-assisted ED follow-up, the evidence base remains nascent. Future research should prioritize the following agenda:

Feasibility and pilot studies. Prospective feasibility studies in real-world ED follow-up settings are needed to evaluate uptake, engagement, technical performance, and safety endpoints across different clinical contexts (89).

Patient acceptability and trust. Qualitative and mixed-methods research should examine willingness to disclose sensitive sexual health information to chatbots, trust in AI-generated advice, and preferences for human versus AI contact, particularly in populations with lower digital or health literacy (75).

Safety outcome evaluation. Prospective evaluation should include predefined safety outcomes such as missed escalation, harmful or inaccurate outputs, AI-attributable adverse events, and time-to-escalation for high-risk interactions, ideally with comparison to standard nurse-led follow-up (70,90,91).

Clinical effectiveness and governance. Comparative studies should evaluate adherence, symptom monitoring, patient satisfaction, and quality of life, while implementation science and consensus work should inform minimum governance and regulatory standards for safe deployment (69,82).

Limitations of this review

This narrative review has several limitations. First, the evidence base for LLM-specific applications in ED follow-up is limited, and no prospective clinical studies of LLM-based chatbots in post-discharge ED care were identified. Conclusions regarding implementation design therefore rely substantially on transferable evidence from non-ED populations, and cross-condition generalisability cannot be assumed. Second, the search was restricted to English-language publications and may have missed relevant reports from other settings. Third, as a narrative review, this study did not include a formal risk-of-bias assessment, and the synthesis remains subject to selection and interpretation biases. Fourth, given the rapid evolution of LLM technology and the clinical literature, some findings may become outdated quickly.

Conclusions

Generative AI, particularly LLMs, may support post-discharge ED follow-up by lowering communication barriers, improving continuity, and enabling more frequent symptom monitoring. The current evidence base for LLM-specific applications in ED follow-up is limited; no prospective clinical studies were identified, and this review draws substantially on transferable evidence from analogous conditions—a limitation that underscores the urgency of ED-specific feasibility research. In a sensitive domain where stigma limits disclosure, conversational systems may increase reporting completeness and facilitate earlier identification of treatment failure, adverse effects, or psychosocial distress. However, ED follow-up involves medication safety, contextual judgment, and trust-dependent counseling, which make autonomous deployment inappropriate in most real-world settings.

A risk-stratified, human-in-the-loop model offers a feasible pathway to balance scalability with safety. LLMs can support low-risk, structured follow-up tasks, while predefined triggers mandate human review and escalation for moderate- and high-risk scenarios. This model also aligns with nurse-led workflows in which nurses supervise flagged interactions, validate summaries, and coordinate interventions. Future work should prospectively evaluate feasibility, patient acceptability, safety outcomes, and workload impact, while developing governance standards for privacy, auditability, and accountability in AI-assisted sexual health follow-up.

Acknowledgments

The authors acknowledge the use of ChatGPT (OpenAI) solely for language editing (grammar, clarity, and readability). The authors confirm that the study topic, interpretation, and all substantive content were developed by the authors. All AI-assisted edits were reviewed and approved by the authors, who take full responsibility for the final manuscript.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-1-0097/rc

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-1-0097/prf

Funding: This work was supported by the Zhejiang Province Traditional Chinese Medicine Key Research Project (No. 2021ZZ016) and the National Natural Science Foundation of China (Nos. 82174376 and 82074433).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-1-0097/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Lue TF. Erectile Dysfunction. N Engl J Med 2000;342:1802-13. [Crossref] [PubMed]
2. McMahon CG. Current diagnosis and management of erectile dysfunction. Med J Aust 2019;210:469-76. [Crossref] [PubMed]
3. Pastuszak AW. Current Diagnosis and Management of Erectile Dysfunction. Curr Sex Health Rep 2014;6:164-76. [Crossref] [PubMed]
4. Brotto LA, Atallah S, Carvalho J, et al. Psychological and interpersonal dimensions of sexual function and dysfunction: recommendations from the fifth international consultation on sexual medicine (ICSM 2024). Sex Med Rev 2025;13:118-43. [Crossref] [PubMed]
5. Domes T, Najafabadi BT, Roberts M, et al. Canadian Urological Association guideline: Erectile dysfunction. Can Urol Assoc J 2021;15:310-22. [Crossref] [PubMed]
6. Rice SM, Oliffe JL, Kealy D, et al. Men’s Help-Seeking for Depression: Attitudinal and Structural Barriers in Symptomatic Men. J Prim Care Community Health 2020;11:2150132720921686. [Crossref] [PubMed]
7. Vergara TV, Magsanoc JM, Mendoza MJ, et al. Overcoming barriers to prostate cancer care in the Philippines. Front Urol 2023;3:1336179. [Crossref] [PubMed]
8. Yousaf O, Grunfeld EA, Hunter MS. A systematic review of the factors associated with delays in medical and psychological help-seeking among men. Health Psychol Rev 2015;9:264-76. [Crossref] [PubMed]
9. Murphy J, Mollica M. All Hands on Deck: Nurses and Cancer Care Delivery in Women’s Health. Front Oncol 2016;6:174. [Crossref] [PubMed]
10. Liu D, Bai J, Zhang Z, et al. TherAIssist: Assisting Art Therapy Homework and Client-Practitioner Collaboration through Human-AI Interaction. arXiv:2502.12443 [Preprint]. 2025. Available online: https://doi.org/10.48550/arXiv.2502.12443
11. Obradovich N, Khalsa SS, Khan W, et al. Opportunities and Risks of Large Language Models in Psychiatry. NPP Digit Psychiatry Neurosci 2024;2:8. [Crossref] [PubMed]
12. Mills R, Mangone ER, Lesh N, et al. Chatbots to Improve Sexual and Reproductive Health: Realist Synthesis. J Med Internet Res 2023;25:e46761. [Crossref] [PubMed]
13. Tu T, Schaekermann M, Palepu A, et al. Towards conversational diagnostic artificial intelligence. Nature 2025;642:442-50. [Crossref] [PubMed]
14. Hastings SN, Stechuchak KM, Coffman CJ, et al. Discharge Information and Support for Patients Discharged from the Emergency Department: Results from a Randomized Controlled Trial. J Gen Intern Med 2020;35:79-86. [Crossref] [PubMed]
15. Peng P, Simmons C, Villars M, et al. QR-code Cards for Residents: A Novel Method for Post-ED Visit Follow-up and Reducing Burnout. Research Square (Research Square). 2024. [Preprint]. Available online: https://doi.org/10.21203/rs.3.rs-4139691/v1
16. Artsi Y, Sorin V, Glicksberg BS, et al. Large language models in real-world clinical workflows: a systematic review of applications and implementation. Front Digit Health 2025;7:1659134. [Crossref] [PubMed]
17. Hager P, Jungmann F, Holland R, et al. Evaluation and mitigation of the limitations of large language models in clinical decision-making. Nat Med 2024;30:2613-22. [Crossref] [PubMed]
18. Capogrosso P, Albersen M, Burnett AL, et al. Erectile Dysfunction: Update on Clinical Management. Eur Urol 2025;88:388-99. [Crossref] [PubMed]
19. Abbas S, Kermode M, Khan MD, et al. What Makes People With Chronic Illnesses Discontinue Treatment? A Practice Theory Informed Analysis of Adherence to Treatment among Patients With Drug-Resistant Tuberculosis in Pakistan. Int J Health Policy Manag 2023;12:6576.
20. Corona G, Giorda CB, Cucinotta D, et al. Sexual Dysfunction in Type 2 Diabetes at Diagnosis: Progression over Time and Drug and Non-Drug Correlated Factors. PLoS One 2016;11:e0157915. [Crossref] [PubMed]
21. Kreyenbuhl J, Nossel IR, Dixon LB. Disengagement from mental health treatment among individuals with schizophrenia and strategies for facilitating connections to care: a review of the literature. Schizophr Bull 2009;35:696-703. [Crossref] [PubMed]
22. Allen MS, Wood AM, Sheffield D. The Psychology of Erectile Dysfunction. Current Directions in Psychological Science 2023;32:487-93.
23. Solberg LI, Trangle MA, Wineman AP. Follow-up and follow-through of depressed patients in primary care: the critical missing components of quality care. J Am Board Fam Pract 2005;18:520-7. [Crossref] [PubMed]
24. Moreno-Poyato AR, Abidi KE, Rodríguez-Nogueira Ó, et al. A qualitative study exploring the patients’ perspective from the ‘Reserved Therapeutic Space’ nursing intervention in acute mental health units. Int J Mental Health Nurs 2021;30:783-97.
25. Skorstad M, Vistad I, Fegran L, et al. Nurse-led consultations reinforced with eHealth technology: a qualitative study of the experiences of patients with gynecological cancer. BMC Nurs 2022;21:326. [Crossref] [PubMed]
26. McIntyre RS, Greenleaf W, Bulaj G, et al. Digital health technologies and major depressive disorder. CNS Spectr 2023;28:662-73. [Crossref] [PubMed]
27. Novara G, Checcucci E, Crestani A, et al. Telehealth in Urology: A Systematic Review of the Literature. How Much Can Telemedicine Be Useful During and After the COVID-19 Pandemic? Eur Urol 2020;78:786-811.
28. Arnon S, Shahar G, Brunstein Klomek A. Continuity of care in suicide prevention: current status and future directions. Front Public Health 2023;11:1266717. [Crossref] [PubMed]
29. Hartman V, Zhang X, Poddar R, et al. Developing and Evaluating Large Language Model-Generated Emergency Medicine Handoff Notes. JAMA Netw Open 2024;7:e2448723. [Crossref] [PubMed]
30. Williams CYK, Miao BY, Kornblith AE, et al. Evaluating the use of large language models to provide clinical recommendations in the Emergency Department. Nat Commun 2024;15:8236. [Crossref] [PubMed]
31. Williams CYK, Zack T, Miao BY, et al. Use of a Large Language Model to Assess Clinical Acuity of Adults in the Emergency Department. JAMA Netw Open 2024;7:e248895. [Crossref] [PubMed]
32. Malgaroli M, Schultebraucks K, Myrick KJ, et al. Large language models for the mental health community: framework for translating code to care. Lancet Digit Health 2025;7:e282-5. [Crossref] [PubMed]
33. Boyer L, Fernandes S, Auquier P, et al. Reimagining patient-reported outcomes in the age of generative AI. NPJ Digit Med 2025;8:624. [Crossref] [PubMed]
34. Serugunda HM, Jianquan O, Kasujja Namatovu H, et al. Using Large Language Models for Chronic Disease Management Tasks: Scoping Review. JMIR Med Inform 2025;13:e66905. [Crossref] [PubMed]
35. Bergholdt SH, Hansen DG, Johnsen AT, et al. A NEw MOdel of individualized and patient-centered follow-up for women with gynecological cancer (the NEMO study)—protocol and rationale of a randomized clinical trial. Trials 2023;24:74. [Crossref] [PubMed]
36. Reddy S. Generative AI in healthcare: an implementation science informed translational path on application, integration and governance. Implementation Sci 2024;19:27.
37. Stanceski K, Zhong S, Zhang X, et al. The quality and safety of using generative AI to produce patient-centred discharge instructions. NPJ Digit Med 2024;7:329. [Crossref] [PubMed]
38. Tischendorf T, Hinsche L, Hasseler M, et al. GenAI in nursing and clinical practice: a rapid review of applications and challenges. J Public Health (Berl.) 2025; [Crossref]
39. Grosso AA, Conte F, Mazzola L, et al. The Evolving Role of Artificial Intelligence in Andrological Surgery: Current Landscape and Future Direction. J Clin Med 2026;15:1473. [Crossref] [PubMed]
40. Sim KYH, Choo KTW. Envisioning an AI-Enhanced Mental Health Ecosystem. arXiv:2503.14883 [Preprint]. 2025. Available online: https://doi.org/10.48550/arXiv.2503.14883
41. Strachna O, Asan O, Stetson PD. Managing Critical Patient-Reported Outcome Measures in Oncology Settings: System Development and Retrospective Study. JMIR Med Inform 2022;10:e38483. [Crossref] [PubMed]
42. Swaminathan A, López I, Mar RAG, et al. Natural language processing system for rapid detection and intervention of mental health crisis chat messages. NPJ Digit Med 2023;6:213. [Crossref] [PubMed]
43. Haltaufderheide J, Ranisch R. The ethics of ChatGPT in medicine and healthcare: a systematic review on Large Language Models (LLMs). NPJ Digit Med 2024;7:183. [Crossref] [PubMed]
44. Silberlust J, Genes N. Clinical Integration of Local AI Assistants Into Emergency Department Discharge Documentation. JAMA Netw Open 2025;8:e2538437. [Crossref] [PubMed]
45. Chiang J, Yafi FA, Dorsey PJ Jr, et al. The dangers of sexual enhancement supplements and counterfeit drugs to “treat” erectile dysfunction. Transl Androl Urol 2017;6:12-9. [Crossref] [PubMed]
46. Gibson D, Jackson S, Shanmugasundaram R, et al. Evaluating the Efficacy of ChatGPT as a Patient Education Tool in Prostate Cancer: Multimetric Assessment. J Med Internet Res 2024;26:e55939. [Crossref] [PubMed]
47. Schwartz BG, Kloner RA. Drug interactions with phosphodiesterase-5 inhibitors used for the treatment of erectile dysfunction or pulmonary hypertension. Circulation 2010;122:88-95. [Crossref] [PubMed]
48. Friedman AB, Delgado MK, Weissman GE. Artificial Intelligence for Emergency Care Triage—Much Promise, but Still Much to Learn. JAMA Netw Open 2024;7:e248857. [Crossref] [PubMed]
49. Meer A, Rahm P, Schwendinger M, et al. A Symptom-Checker for Adult Patients Visiting an Interdisciplinary Emergency Care Center and the Safety of Patient Self-Triage: Real-Life Prospective Evaluation. J Med Internet Res 2024;26:e58157. [Crossref] [PubMed]
50. Raita Y, Goto T, Faridi MK, et al. Emergency department triage prediction of clinical outcomes using machine learning models. Crit Care 2019;23:64. [Crossref] [PubMed]
51. Barrow RY, Ahmed F, Bolan GA, et al. Recommendations for Providing Quality Sexually Transmitted Diseases Clinical Services, 2020. MMWR Recomm Rep 2020;68:1-20. [Crossref] [PubMed]
52. Gibbs J, Solomon D, Jackson L, et al. Measuring and evaluating sexual health in the era of digital health: challenges and opportunities. Sex Health 2022;19:336-45. [Crossref] [PubMed]
53. McGraw D, Mandl KD. Privacy protections to encourage use of health-relevant digital data in a learning health system. NPJ Digit Med 2021;4:2. [Crossref] [PubMed]
54. Wies B, Landers C, Ienca M. Digital Mental Health for Young People: A Scoping Review of Ethical Promises and Challenges. Front Digit Health 2021;3:697072. [Crossref] [PubMed]
55. Karaferis D, Balaska D, Pollalis Y. Enhancement of Patient Engagement and Healthcare Delivery Through the Utilization of Artificial Intelligence (AI) Technologies. Austin J Clin Med 2024;9:1053.
56. Powell J. Trust Me, I’m a Chatbot: How Artificial Intelligence in Health Care Fails the Turing Test. J Med Internet Res 2019;21:e16222. [Crossref] [PubMed]
57. Ueda D, Kakinuma T, Fujita S, et al. Fairness of artificial intelligence in healthcare: review and recommendations. Jpn J Radiol 2024;42:3-15. [Crossref] [PubMed]
58. Weiner EB, Dankwa-Mullan I, Nelson WA, et al. Ethical challenges and evolving strategies in the integration of artificial intelligence into clinical practice. PLOS Digit Health 2025;4:e0000810. [Crossref] [PubMed]
59. RoshanaeiMRezapourREl-NasrMS. Talk, Listen, Connect: Navigating Empathy in Human-AI Interactions. arXiv:2409.15550v1 [Preprint]. 2024.
60. Yirmiya K, Fonagy P. Mentalizing Without a Mind: Psychotherapeutic Potential of Generative AI. J Med Internet Res 2025;27:e79156. [Crossref] [PubMed]
61. Sharma A, Lin IW, Miner AS, et al. Human–AI collaboration enables more empathic conversations in text-based peer-to-peer mental health support. Nature Machine Intelligence 2023;5:46-57.
62. Liu Y, Li Y, Mayfield R, et al. Improving Emotional Support Delivery in Text-Based Community Safety Reporting Using Large Language Models. Proceedings of the ACM on Human-Computer Interaction 2025;9:1-31.
63. McCahon D, Denholm RE, Huntley AL, et al. Development of a model of medication review for use in clinical practice: Bristol medication review model. BMC Med 2021;19:262. [Crossref] [PubMed]
64. Tilson H, Hines LE, McEvoy G, et al. Recommendations for selecting drug-drug interactions for clinical decision support. Am J Health Syst Pharm 2016;73:576-85. [Crossref] [PubMed]
65. Romero-Brufau S, Rosenthal J, Kautz J, et al. Clinical implementation of a machine learning system to detect deteriorating patients reduces time to response and intervention. medRxiv (Cold Spring Harbor Laboratory). 2021. doi: https://doi.org/10.1101/2021.10.10.21264823.
66. El-Zammar D, Johal R, Eng J, et al. Implementation of an emergency department virtual follow-up care process in a community-based hospital: a quality improvement initiative. BMJ Open Qual 2022;11:e001782. [Crossref] [PubMed]
67. Faris N, Saliba M, Tamim H, et al. Electronic medical record implementation in the emergency department in a low-resource country: Lessons learned. PLoS One 2024;19:e0298027. [Crossref] [PubMed]
68. Xu Z, Lee YC, Stasiak K, et al. The Digital Therapeutic Alliance With Mental Health Chatbots: Diary Study and Thematic Analysis. JMIR Ment Health 2025;12:e76642. [Crossref] [PubMed]
69. Chamsi-Pasha MAR, Chamsi-Pasha H. Ethical Challenges and Current Opportunities of Artificial Intelligence in Cardiology. Avicenna J Med 2025;15:95-9. [Crossref] [PubMed]
70. Ebnali Harari R, Altaweel A, Ahram T, et al. A randomized controlled trial on evaluating clinician-supervised generative AI for decision support. Int J Med Inform 2025;195:105701. [Crossref] [PubMed]
71. Mess SA, Mackey AJ, Yarowsky DE. Artificial Intelligence Scribe and Large Language Model Technology in Healthcare Documentation: Advantages, Limitations, and Recommendations. Plast Reconstr Surg Glob Open 2025;13:e6450. [Crossref] [PubMed]
72. Jędrasiak K, Bijoch J. Generative AI-driven synthetic media risks in digital health: implications for telemedicine and teledentistry. Front Public Health 2026;14:1781216. [Crossref] [PubMed]
73. Pool J, Indulska M, Sadiq S. Large language models and generative AI in telehealth: a responsible use lens. J Am Med Inform Assoc 2024;31:2125-36. [Crossref] [PubMed]
74. Zendehbad SA, Ghasemi J, Khodadad FS. Trustworthy AI in Telehealth: Navigating Challenges, Ethical Considerations, and Future Opportunities for Equitable Healthcare Delivery. Healthc Technol Lett 2025;12:e70020. [Crossref] [PubMed]
75. Isaksen AA, Schaarup JR, Bjerg L, et al. Changes in public perception of artificial intelligence in healthcare after exposure to ChatGPT. NPJ Digit Med 2025;8:795. [Crossref] [PubMed]
76. Lawrence K, Kuram VS, Levine DL, et al. Informed Consent for Ambient Documentation Using Generative AI in Ambulatory Care. JAMA Netw Open 2025;8:e2522400. [Crossref] [PubMed]
77. Soe NN, Kusnandar II, Latt PM, et al. Use of AI in Identification of Sexually Transmitted Infections and Anogenital Dermatoses. JAMA Netw Open 2025;8:e2533512. [Crossref] [PubMed]
78. Montagna S, Ferretti S, Klopfenstein LC, et al. Privacy-preserving LLM-based chatbots for hypertensive patient self-management. Smart Health 2025;36:100552.
79. Deschepper M, Rogge H, Syx M, et al. Tailoring Discharge Summaries to Health Care Providers’ Needs (Part 1 of the Framework and Implementation of AI Tools Project): User-Centered Design Approach. JMIR Med Inform 2026;14:e80613. [Crossref] [PubMed]
80. Schmiedmayer P, Rao A, Zagar P, et al. LLMonFHIR: A Physician-Validated, Large Language Model-Based Mobile Application for Querying Patient Electronic Health Data. JACC Adv 2025;4:101780. [Crossref] [PubMed]
81. Geoghegan L, Scarborough A, Wormald JCR, et al. Automated conversational agents for post-intervention follow-up: a systematic review. BJS Open 2021;5:zrab070. [Crossref] [PubMed]
82. Olivella A, Méndez Fernández AB, García EG, et al. Conversational AI for remote monitoring in heart failure: a prospective controlled pilot study. Eur Heart J Digit Health 2026;7:ztag032. [Crossref] [PubMed]
83. Farmer H, Kreiner K, Schütz T, et al. The Evolution of Telehealth in Heart Failure Management: The Role of Large Language Models and HerzMobil as a Potential Use Case. Stud Health Technol Inform 2024;313:228-33. [Crossref] [PubMed]
84. Dal E, Srivastava A, Chigarira B, et al. Effectiveness of ChatGPT 4.0 in Telemedicine-Based Management of Metastatic Prostate Carcinoma. Diagnostics (Basel) 2024;14:1899. [Crossref] [PubMed]
85. Daulat SR, Dholaria N, Burnet G, et al. Prompt Engineering and Follow-Up Questioning Improves the Readability of Spine Surgery Questions in Large Language Models. World Neurosurg 2025;203:124423. [Crossref] [PubMed]
86. Xiong G, Jin Q, Wang X, et al. Improving Retrieval-Augmented Generation in Medicine with Iterative Follow-up Questions. Pac Symp Biocomput 2025;30:199-214. [Crossref] [PubMed]
87. Gamble JL, Ferguson D, Yuen J, et al. Limitations of GPT-3.5 and GPT-4 in Applying Fleischner Society Guidelines to Incidental Lung Nodules. Can Assoc Radiol J 2024;75:412-6. [Crossref] [PubMed]
88. Kaboudi N, Firouzbakht S, Shahir Eftekhar M, et al. Diagnostic Accuracy of ChatGPT for Patients’ Triage; a Systematic Review and Meta-Analysis. Arch Acad Emerg Med 2024;12:e60. [Crossref] [PubMed]
89. Tao X, Zhou S, Ding K, et al. An LLM chatbot to facilitate primary-to-specialist care transitions: a randomized controlled trial. Nat Med 2026;32:934-42. [Crossref] [PubMed]
90. de Pennington N, Mole G, Lim E, et al. Safety and Acceptability of a Natural Language Artificial Intelligence Assistant to Deliver Clinical Follow-up to Cataract Surgery Patients: Proposal. JMIR Res Protoc 2021;10:e27227. [Crossref] [PubMed]
91. Palmer WJ, Michlin D, Estephan L, et al. Utility and Safety of Artificial Intelligence for Patient-Initiated Contact After Functional Rhinoplasty. OTO Open 2025;9:e70170. [Crossref] [PubMed]