Improving MedDRA/J Coding Accuracy with a Fine-Tuned Text Embedding Model.

MedDRA coding in Japan can be tedious when source expressions differ from dictionary terms. We fine-tune a text embedding model on in-house adverse event (AE) data to enhance search accuracy. Compared to baselines, our model significantly improves MedDRA term ranking and recall (nDCG@20 = 76.2%, Recall@20 = 90.8% at 50K in-house entries). This approach suggests that using real-world AE data with advanced embeddings can greatly benefit MedDRA/J coding.