Step-by-Step Guide to Develop a Language Model-Based Chatbot for Administering and Scoring the Morisky Scale

Step 1: Define Objectives and Scope
1. Purpose: Develop a chatbot to ask the Morisky scale questions one at a time, calculate scores, and create a tailored medication adherence action plan.
2. Key Features:
• Accurately present and interpret the Morisky Medication Adherence Scale (MMAS).
• Provide user-friendly interaction.
• Generate a tailored action plan based on adherence level.


Step 2: Conduct Research and Obtain Permissions
1. Research the Morisky Scale:
• Review peer-reviewed academic journals on MMAS for insights into its application and scoring.
• Ensure the selected scale version (e.g., MMAS-4 or MMAS-8) fits your intended use.
2. Obtain Permissions:
• Contact the copyright holders of the Morisky scale for licensing and use rights.


Step 3: Build a Development Framework
1. Programming Language and Framework:
• Choose a programming language (e.g., Python) and a chatbot framework like Rasa, Microsoft Bot Framework, or Dialogflow.
2. Infrastructure:
• Use cloud platforms like AWS, Azure, or Google Cloud for hosting the chatbot.


Step 4: Create a Training Dataset
1. Data Collection:
• Use academic journal data to gather examples of patient adherence discussions.
• Train the model on peer-reviewed, ethical, and high-quality healthcare datasets.
2. Annotation:
• Annotate datasets to teach the model how to handle context-sensitive queries, understand Morisky questions, and recognize user responses.


Step 5: Train a Language Model
1. Base Model Selection:
• Start with a pre-trained transformer model (e.g., GPT, BERT).
2. Fine-Tuning:
• Fine-tune the model using your annotated dataset.
• Ensure focus on understanding adherence-related language.


Step 6: Design Chatbot Flow
1. Question Sequencing:
• Implement a step-by-step flow to ask Morisky scale questions one at a time.
• Add validation to ensure users answer each question before moving forward.
2. Response Handling:
• Design natural language processing (NLP) logic to interpret various user responses.


Step 7: Implement Scoring Logic
1. Score Calculation:
• Add a backend algorithm to calculate the MMAS score based on user responses.
• Define adherence levels (e.g., low, medium, high) using MMAS thresholds.
2. Coding Responses:
• Program the chatbot to output a score and adherence level.


Step 8: Develop the Action Plan Generator
1. Tailored Plans:
• Create a database of tailored recommendations based on adherence levels.
• For example:
• Low adherence: Suggest reminders, education, or healthcare provider follow-ups.
• Medium adherence: Recommend simplifying medication regimens or addressing specific barriers.
• High adherence: Encourage ongoing adherence and positive reinforcement.
2. Dynamic Generation:
• Use user responses to customize action plans in real-time.


Step 9: Test the Chatbot
1. Pilot Testing:
• Test with healthcare professionals and a sample user group.
• Gather feedback to refine question flow, interpretation, and action plans.
2. Performance Evaluation:
• Use metrics like accuracy, user satisfaction, and healthcare outcomes to measure chatbot effectiveness.


Step 10: Deploy and Monitor
1. Deployment:
• Host the chatbot on a secure, HIPAA-compliant server if used in healthcare settings.
• Make it accessible through web, mobile, or integration with electronic health record (EHR) systems.
2. Monitoring:
• Continuously monitor chatbot performance.
• Update the model as new research on medication adherence or the Morisky scale becomes available.


Step 11: Ensure Compliance and Ethical Standards
1. Data Privacy:
• Encrypt all data and follow privacy regulations (e.g., HIPAA, GDPR).
2. Bias Mitigation:
• Regularly review the model to ensure fairness and avoid biases in responses or recommendations.


Step 12: Continuous Improvement
1. Feedback Integration:
• Use user feedback to refine the chatbot’s question flow, scoring accuracy, and action plans.
2. Update Based on Research:
• Stay updated on academic advancements in medication adherence and integrate them into the chatbot’s logic.




​