DialysisBot: AI Chatbot for Hemodialysis Self-Management

Condition(s) studied

Investigational drug(s) / intervention(s)

DialysisBot: DialysisBot is a multiplatform, AI-based conversational agent (chatbot) providing personalized, self-management support to patients undergoing chronic hemodialysis. It combines a generative large language model for dialogue management, a fine-tuned BERT classifier for request-domain recognition (diet, fluid intake, vascular access care, dialysis organization), and a vector-similarity retrieval engine that restricts all responses to a clinically validated educational knowledge base preloaded by the research team. The system performs no autonomous web search. When query-to-source similarity falls below a predefined threshold, the system withholds a response and directs the patient to healthcare staff, serving as a safeguard against unvalidated or hallucinated content. Participants use the application as needed in their daily routine over a 3-month period, after receiving individual training on its use.

Study summary

End-stage kidney disease requiring hemodialysis is a chronic condition with substantial clinical, functional, and psychosocial burden. Maintaining an adequate health status between dialysis sessions depends largely on patients' ability to self-manage key aspects of treatment, including fluid restriction, dietary control, and vascular access care. Adherence in this population remains frequently suboptimal, and elevated interdialytic weight gain (IDWG), hyperphosphatemia, and hyperkalemia are among the most common complications associated with non-adherence.

DialysisBot is a multiplatform, artificial intelligence-based conversational agent designed to support hemodialysis patients in the day-to-day self-management of their disease. The system combines a generative large language model for natural dialogue management, a fine-tuned BERT classifier for recognizing the domain of each patient request (diet, fluid intake, vascular access care, or organizational aspects of dialysis), and a vector-similarity retrieval engine that limits every response to a clinically validated knowledge base preloaded by the research team and approved by the hospital institution. The system performs no autonomous web search. When the cosine similarity between a user query and the indexed reference documents falls below a predefined threshold, the system withholds a clinical answer and instead informs the patient that it cannot respond, directing them to contact healthcare staff; this mechanism is the main technical safeguard against hallucinated or unvalidated responses.

This is a prospective experimental pilot study evaluating the feasibility, acceptability, impact, and user satisfaction associated with DialysisBot as a self-management support tool for patients undergoing hemodialysis. Secondary objectives include assessing support for dietary management (phosphorus, potassium, and sodium restriction), support for interdialytic fluid intake control, whether system-provided guidance on vascular access management (arteriovenous fistula and central venous catheter) is put into practice and consistent with current standards of care, the impact on treatment adherence, and the barriers, facilitators, and overall user experience associated with the intervention.

The study is organized into two methodological phases: a quantitative longitudinal phase (T0 baseline, T1 at 1 month, T2 at 3 months) using standardized patient-reported outcome measures and routine clinical parameters, followed by a qualitative phase conducted after T2, based on semi-structured interviews analyzed through reflexive thematic analysis. A convenience sample of 20-40 adult patients undergoing chronic hemodialysis, recruited through a participating dialysis center and/or an online patient community, will be enrolled and trained on the application before use. As a pilot feasibility study, its results are intended to inform the design and sample size of future, larger-scale confirmatory trials.

Eligibility

Primary outcome measure(s)

• Feasibility of the DialysisBot Intervention — Baseline (T0) through 3 months (T2)
  Feasibility will be assessed through four indicators: recruitment rate (proportion of eligible patients who consent to participate), retention rate (proportion of participants completing the study at the final assessment), questionnaire completion rate (proportion of instruments adequately completed from the intermediate to the final assessment), and application engagement (mean number of weekly patient-system interactions, derived from anonymized system logs).
• Usability of DialysisBot — 1 month (T1)
  Usability will be assessed using the Italian version of the Chatbot Usability Scale (BUS-11), an 11-item, 5-point Likert scale (1 = strongly disagree to 5 = strongly agree) validated for AI-based conversational systems. Total score ranges from 11 to 55; higher scores indicate greater perceived usability. \[Time Frame: 1 month (T1)\]
• Change in Treatment Adherence Indicators — Baseline (T0), 1 month (T1), and 3 months (T2)
  Adherence will be measured through routinely collected clinical parameters: interdialytic weight gain (IDWG, from pre- and post-dialysis body weight), serum phosphate (mmol/L), serum potassium (mmol/L), and attendance rate at scheduled dialysis sessions (%).

Trial sites (1)

More University of Rome Tor Vergata trials in Italy

Other trials for the same condition