TWIN-GPT is a large language model-based approach for creating personalized digital twins to enhance clinical trial outcome prediction. The method leverages the comprehensive medical knowledge embedded in large language models (LLMs) to generate accurate and personalized digital twins for individual patients, even with limited data. TWIN-GPT can establish cross-dataset associations of medical information, generating unique digital twins that preserve individual patient characteristics. Comprehensive experiments show that TWIN-GPT outperforms existing prediction approaches in clinical trial outcome prediction. Additionally, TWIN-GPT can generate high-fidelity trial data that closely approximates specific patients, aiding in more accurate result predictions in data-scarce situations. The method also provides practical evidence for the application of digital twins in healthcare, highlighting its potential significance. TWIN-GPT addresses challenges such as data gaps and inconsistencies in electronic health records (EHRs) while ensuring patient privacy. The model is fine-tuned on a pre-trained LLM (ChatGPT) on clinical trial datasets to generate personalized digital twins for different patients. TWIN-GPT demonstrates enhanced personalization and accuracy by leveraging the vast medical knowledge embedded within ChatGPT. It also protects patient privacy by generating virtual patient data and simulating personalized physiological measures over time. The model is versatile and can be applied across various clinical trial scenarios, accelerating clinical trials and enhancing medical research and patient care. TWIN-GPT's performance is evaluated on real-world datasets, showing that it can account for individual patient variations and disease complexities, producing data that closely aligns with diverse real-world scenarios. The model's evaluation demonstrates high fidelity in adverse event prediction and severe outcome prediction, with AUROC scores close to those of real data. TWIN-GPT also shows strong privacy protection, with low sensitivity scores in presence disclosure, attribute disclosure, and nearest neighbor adversarial accuracy risk. The model's explainability and ablation study further confirm its effectiveness and robustness. Overall, TWIN-GPT provides a novel and effective solution for generating personalized digital twins in clinical trials, enhancing the accuracy and efficiency of clinical trial outcome prediction while ensuring patient privacy.TWIN-GPT is a large language model-based approach for creating personalized digital twins to enhance clinical trial outcome prediction. The method leverages the comprehensive medical knowledge embedded in large language models (LLMs) to generate accurate and personalized digital twins for individual patients, even with limited data. TWIN-GPT can establish cross-dataset associations of medical information, generating unique digital twins that preserve individual patient characteristics. Comprehensive experiments show that TWIN-GPT outperforms existing prediction approaches in clinical trial outcome prediction. Additionally, TWIN-GPT can generate high-fidelity trial data that closely approximates specific patients, aiding in more accurate result predictions in data-scarce situations. The method also provides practical evidence for the application of digital twins in healthcare, highlighting its potential significance. TWIN-GPT addresses challenges such as data gaps and inconsistencies in electronic health records (EHRs) while ensuring patient privacy. The model is fine-tuned on a pre-trained LLM (ChatGPT) on clinical trial datasets to generate personalized digital twins for different patients. TWIN-GPT demonstrates enhanced personalization and accuracy by leveraging the vast medical knowledge embedded within ChatGPT. It also protects patient privacy by generating virtual patient data and simulating personalized physiological measures over time. The model is versatile and can be applied across various clinical trial scenarios, accelerating clinical trials and enhancing medical research and patient care. TWIN-GPT's performance is evaluated on real-world datasets, showing that it can account for individual patient variations and disease complexities, producing data that closely aligns with diverse real-world scenarios. The model's evaluation demonstrates high fidelity in adverse event prediction and severe outcome prediction, with AUROC scores close to those of real data. TWIN-GPT also shows strong privacy protection, with low sensitivity scores in presence disclosure, attribute disclosure, and nearest neighbor adversarial accuracy risk. The model's explainability and ablation study further confirm its effectiveness and robustness. Overall, TWIN-GPT provides a novel and effective solution for generating personalized digital twins in clinical trials, enhancing the accuracy and efficiency of clinical trial outcome prediction while ensuring patient privacy.