Design and evaluation of a multilingual chatbot for clinical communication in urban clinics

Thesis Abstract

The rapid growth of urban populations in multilingual settings poses significant communication barriers in clinical encounters, leading to misdiagnosis, reduced patient satisfaction, and disparities in care. This study addresses the problem by designing, implementing, and evaluating a multilingual chatbot to support clinical communication in urban clinics, with the aim of improving information exchange between patients and healthcare providers across language groups. The specific objectives are to (1) identify language- and culture-specific communication barriers in urban clinic workflows, (2) develop a multilingual chatbot prototype capable of facilitating symptom elicitation, appointment scheduling, and patient education in English, Spanish, Mandarin, and Arabic, (3) integrate culturally sensitive negotiation and clarification strategies informed by sociolinguistic theory and patient-provider communication frameworks, (4) evaluate usability, acceptance, and impact on communication outcomes among patients and clinicians, and (5) assess potential effects on patient comprehension, satisfaction, and perceived quality of care. A mixed-methods research design will be employed, combining iterative design-based research for the chatbot development with a quasi-experimental evaluation in two urban clinics serving diverse populations. The population comprises patients and clinicians in these clinics, with purposive sampling of 200 patients (50 per language group English, Spanish, Mandarin, Arabic) and 40 clinicians. Data collection instruments include (a) a needs assessment survey and semi-structured interviews with 40 patients and 12 clinicians to identify communicative barriers and requirements; (b) task-based usability testing with 40 patients to refine chatbot interfaces; (c) pre- and post-implementation patient comprehension tests, satisfaction scales, and perceived quality of care measures; (d) clinician workload and perceived efficiency metrics; and (e) system logs capturing interaction quality, completion rates, and error types. Validity and reliability will be established through pilot testing, expert review, and triangulation across qualitative and quantitative data. The analytical framework will integrate thematic analysis of interview data (following Braun and Clarke) to identify barriers and facilitators, with regression analyses to examine relationships between chatbot use and patient outcomes, and ANOVA to compare satisfaction and comprehension across language groups. Additionally, an interrupted time series analysis will assess changes in clinic workflow metrics pre- and post-implementation. A conceptual model drawing on communicative action theory and the Technology Acceptance Model (TAM) will guide feature design and interpretation of adoption patterns. Expected findings include (i) statistically significant improvements in patient comprehension scores and satisfaction following chatbot-assisted interactions, particularly for non-English speakers; (ii) higher rates of accurate symptom elicitation and reduced information loss in multilingual encounters; (iii) positive perceptions of chatbot usability and perceived usefulness among patients and clinicians, moderated by language proficiency and prior technology exposure; (iv) evidence that culturally adapted clarifying strategies embedded in the chatbot reduce miscommunication episodes; and (v) manageable impact on clinician workload with potential reallocation of time to complex clinical decisions. The study will contribute to knowledge by bridging natural language processing, intercultural communication, and health informatics to create scalable solutions for multilingual clinical settings. It will offer a validated design framework, including linguistic and cultural adaptation guidelines, evaluation metrics, and an implementation blueprint for urban clinics, with implications for policy on language access and digital health equity. The main conclusion is that a well-designed multilingual chatbot can enhance communicative accuracy and patient satisfaction in urban clinics without imposing substantial additional workload on clinicians, provided that the system is co-created with stakeholders, aligned to clinical workflows, and continuously refined through user feedback. Recommendations include expanding language offerings, integrating with electronic health records with robust privacy safeguards, conducting longitudinal studies to assess long-term health outcomes, and exploring deployment in other high-traffic clinical environments such as emergency departments and diagnostic centers.