Advanced Detection of Nausea/Vomiting and Anxiety Symptoms in Patients with Cancer

This study aims to improve the detection of nausea/vomiting and anxiety in patients with cancer by further pretraining and fine-tuning the Bio-Clinical BERT (Bidirectional Encoder Representations from Transformers) model. The performance will be benchmarked and compared against two other advanced models: DeBERTa (Decoding-enhanced BERT with Disentangled Attention) and RoBERTa (A Robustly Optimized BERT Pretraining Approach). The study uses a gold-standard corpus and synthetic clinical notes to evaluate the performance of the models. The results show that Symptom-BERT excels at detecting cancer symptoms in actual EHR and synthetic notes, outperforming DeBERTa and RoBERTa across all evaluation metrics.