Health Digital Twins (HDTs) rely on Machine Learning (ML) capabilities to provide insights and decision support for healthcare stakeholders. However, the complexity of ML models makes it challenging for non-technical stakeholders to understand the reasoning behind predictions, raising concerns about lack of transparency and trust in HDTs. To address these shortcomings, explainable AI (XAI) methods have been proposed to describe the predictions of ML models. Despite the efforts, the technical outputs of the XAI methods can be difficult for both expert and non-expert healthcare stakeholders to comprehend. In this paper, we propose a framework called Stress Management Digital Twin (SMDT) that integrates XAI methods with Large Language Models (LLMs) to generate natural language explanations for predictions of ML models. This enhances transparency and trustworthiness in HDTs. Specifically, we leveraged the Google Gemini and Mistral 7B to transform Shapley Additive exPlanations (SHAP) local explanation of stress management score prediction by Random Forest model into natural language narratives. From our experiments, the Google Gemini generated clear and concise narratives of the model’s decision while retaining the accuracy of the given SHAP values. The findings from this study demonstrate that the proposed digital twin can be used for what-if-analysis of stress management scores while providing user-friendly explanations to enhance transparency and trust in HDTs.