Detecting emotional distress in cancer patients through sentiment analysis

Abstract

Cancer patients often experience profound emotional distress throughout diagnosis, treatment, recurrence, and survivorship. Feelings of fear, sadness, uncertainty, and psychological burden frequently appear in cancer-related narratives, and these emotional challenges also extend to caregivers and support networks. Understanding these expressions is important for developing a broader view of the emotional landscape surrounding cancer. Existing sentiment analysis techniques face important limitations in this domain. Many studies rely on a single computational approach, offer limited sensitivity to context, and do not adequately address the complexity of emotionally rich cancer narratives drawn from heterogeneous online sources. As a result, different methods may produce inconsistent interpretations, which limits the reliability of emotion-related insights. This study investigates sentiment and emotion patterns in publicly accessible online cancer-related text instances using a multi-method natural language processing framework. Text collected from more than 200 online sources was analyzed with NRCLex, VADER, LSTM, BiLSTM, DistilBERT, and RoBERTa to examine how different methods capture emotional variation across cancer-related contexts. The findings reveal a predominance of negative and uncertainty related expressions, particularly in diagnosis and advanced disease contexts, while fear, trust, caring, and realization emerged frequently in the emotion analysis. Overall, the study shows that a multi-model strategy offers a broader and more informative exploratory view of emotional distress in cancer discourse. Findings are consistent with, rather than novel additions to, the existing psycho-oncology literature on emotional trajectories across the illness course and are presented here as corpus-level discourse patterns rather than clinical discoveries. Future research should strengthen validation, improve source stratification, and further align computational findings with psycho-oncology knowledge and ethically grounded research practice.