Table 3.

Framework for health care chatbot evaluation.

Constructs (levels 1, 2, and 3)				Description
Safety, privacy, and fairness
	Safety		Prevent worse outcomes for the patient, provider, or health system from occurring as a result of the use of an MLa algorithm.
		Outcome proxies appropriateness	Use alternative measures or indicators that accurately reflect the desired health outcomes in the absence of direct measurements.
		Data provenance	Track and document the origin and history of data, including where it came from and how it has been handled. Data providers: assign roles and responsibilities to entities like hospital EHRsb and patient-generated health data for maintaining safe AIc. Data sources: include various origins of data, such as social media and clinical settings.
		Harm control	Reduce and manage potential risks and negative impacts associated with using a chatbot.
		Automation bias reduction	The tendency to accept automated suggestions without critical evaluation or questioning.
		Critical help	Provide necessary assistance and address negative and help-seeking information.
		Ethics	Principles and standards that govern the conduct of individuals and organizations, ensuring fairness, privacy, and respect in using ML algorithms in health care.
	Security		Maintain confidentiality, integrity, and availability through protection mechanisms that prevent unauthorized access and use
		Protection method	Implement techniques and tools to safeguard data from unauthorized access and threats.
		Security standard	Follow established guidelines and practices designed to protect data and systems from security breaches.
		Third-party reliability	Ensure the trustworthiness of external partners or services in maintaining data security and integrity.
	Resilience		Withstand unexpected adverse events or changes in their environment or use.
	Privacy		Protect privacy according to standards like HIPAAd and GDPRe, ensuring user autonomy and dignity.
		Data exchange	Maintain privacy standards for accessing and sharing data with third-party tools, cloud platforms, and other external systems.
		Data collection and storage	Maintain privacy standards for gathering and securely storing data for future use.
		Data usage	Maintain privacy standards for using collected data for analysis, decision-making, and improving chatbot algorithms.
		Privacy policy	Outline how an organization collects, uses, protects, and shares personal data.
		Data protection	Implement methods to ensure privacy and prevent unauthorized access and breaches.
	Fairness and bias management		Ensure the chatbots operate with minimized and acknowledged biases to ensure fair outcomes.
		Systemic bias	Address biases originating from societal norms and institutional practices.
		Computational and statistical bias	Manage biases arising from the way data is processed and algorithms are designed.
		Human-cognitive biases	Recognize biases stemming from individual or group perceptions and attitudes.
		Population bias	Address the issue where certain populations are underrepresented in data, leading to less accurate model performance for those groups.
Trustworthiness and usefulness
	Accountability		Ensure those involved in the chatbot’s lifecycle uphold standards of auditability and harm minimization.
	Transparency		Communicate clearly regarding the chatbot’s characteristics and performance throughout its lifecycle.
		Usage specification	Define how the chatbot should be used.
		Model characteristics	Describe the specific features and behaviors of the chatbot.
		Model availability	Ensure the chatbot is accessible as needed.
		Model limitations	Identify and communicate the boundaries and constraints of the chatbot.
		Data usage	Explain how data is used within the chatbot.
	Explainability and interpretability		Described below.
		Model explainability	Detail the internal mechanisms and decision-making processes of the chatbot.
		Model interpretability	Make the outputs of chatbots clear and meaningful to end-users.
	Beneficence		Ensure the chatbot positively impacts its intended outcomes, emphasizing measurable benefits over potential risks.
		Health outcomes	Focus on improving health results.
		Clinical evidence	Use rigorous methods like A/B tests or randomized controlled trials to validate effectiveness.
		Use behavior	Influence and improve user actions.
		Intervention	Apply targeted measures to achieve desired outcomes.
		Health care system	Integrate effectively within the broader health care environment
	Validity		Ensure the chatbot performs as expected in real-world conditions.
		Data relevance and credibility	Use high-quality, pertinent training data.
		Language understanding	Ensure the chatbot’s linguistic capabilities are robust.
		Information retrieval accuracy	Accurately retrieve relevant information.
		Outcome accuracy	Deliver precise and correct results.
		Task completion	Effectively complete required functions.
	Reliability		Ensure that the chatbot consistently performs as intended under various conditions and maintains dependable operation over time.
		Failure prevention	Prevent system failures to maintain functionality.
		Robustness	Handle unexpected inputs and diverse data without errors.
		Workflow integration	Fit seamlessly into existing processes.
		Reproducibility	Ensure consistent outcomes across different settings.
		Monitoring	Continually check chatbots to ensure proper operation.
		Up-to-dateness	Keep the system current with the latest information.
	Generalizability		Apply learned patterns to new, unseen data.
		Contextual adaptability	Function effectively in different environments or clinical contexts. Age group adaptability: cater to different age groups. Scenario adaptability: adapt to various situations.
		Novel data performance	Perform well with new, unseen data.
	Testability		Verify and meet standards for robustness, safety, bias mitigation, fairness, and equity.
		Verifiability	Ensure different attributes can be tested. Quantifiability: measure attributes precisely.
		Regular auditing	Measure attributes regularly.
Design and operational effectiveness
	Accessibility		Ensure the chatbot is usable by the intended users regardless of their abilities, devices, or technical skills, promoting inclusivity and ease of use.
		Versatile access	Provide multiple interaction methods to accommodate user preferences and needs. Multilanguage: enable interaction in multiple languages to cater to a diverse user base. Different input and output modes: accommodate various input and output methods, such as text, voice, and visual. Multiplatform: ensure functionality across different platforms, such as web, mobile, and desktop applications. Multidevice: provide compatibility with various devices, including smartphones, tablets, laptops, and desktop computers.
		User literacy	Ensure the system is usable by individuals with varying levels of technical knowledge and literacy.
		User experience	Create a pleasant and effective interaction for users. Likability: design the system to be appealing and enjoyable to use. Understood by the conversational agent: ensure clear communication between the user and the chatbot. User engagement: maintain user interest and active participation. Respectfulness: interact with users in a polite and respectful manner. Response appropriateness: provide suitable and contextually relevant responses. Credibility: ensure the chatbot’s reliability and trustworthiness.
		User interface design	Create an intuitive and easy-to-use interface for users.
		Simplicity and ease of use	Make the system straightforward and user-friendly, minimizing complexity and effort required from users.
	Personalized engagement		Tailor responses based on patient data and preferences.
		Personalization	Customized responses based on patient data and preferences.
		Anthropomorphism and relationship	Build a human-like relationship with users. Relationship building: develop a rapport with users. Empathy: show understanding and compassion. Humor: use appropriate humor to engage users. Identity: establish a clear and consistent chatbot persona.
		User adherence	Track and analyze how well users follow recommendations and adjust the chatbot’s strategies based on this data to improve compliance and outcomes.
		Feedback incorporation	Use user feedback to improve the system.
		Progress awareness	Monitor and respond to the conversation’s context and progress. Memory: support multiturn or multisession conversations. Strategy adjustment: adapt the conversation strategy as needed.
	Cost-effectiveness		Assess whether the chatbot delivers beneficial outcomes at a reasonable cost, providing a better or more economical solution compared to existing methods.
		Comparative effectiveness	Demonstrate that the chatbot is a better solution than previous methods.
		Economical viability	Ensure the system is cost-effective.
		Environmental viability	Minimize environmental impact.
		Task efficiency	Perform tasks quickly and effectively. Appropriate response time: provide timely responses. Response conciseness: give clear and succinct information. Response relevance: ensure responses are pertinent to the query. Response practicality: offer practical and actionable information.
		Workflow considerations	Integrate smoothly into existing systems.

^aML: machine learning.

^bEHR: electronic health record.

^cAI: artificial intelligence.

^dHIPAA: Health Insurance Portability and Accountability Act.

^eGDPR: General Data Protection Regulation.