Artificial intelligence and patient narratives: A novel approach to assessing hope in patients with cancer

The present study aimed to evaluate the feasibility of patient-generated text and its interpretation by artificial intelligence (AI) as a valid correlate of hope levels in patients with cancer. For this purpose, four medical centers recruited consecutive patients with cancer and the patients were administered a questionnaire to collect data on patient characteristics and a shortened version of the Adult Trait Hope Scale (s-ATHS). Additionally, all participants provided written text on their hope levels, which was then analyzed by a deep neural network model. AI predicted hope labels, as well as numerous patient, disease and center features which were then associated with the scores from s-ATHS using univariate and multivariate gamma regression analyses. The present study comprised 461 patients with cancer, 194 (42.1%) of whom had metastatic disease. Multivariate gamma regression analysis identified three variables independently associated with hope index scores (s-ATHS): Treatment center (B=-0.09, Wald=4.77, P=0.029), Eastern Cooperative Oncology Group (ECOG) performance status (B=-0.09, Wald=47.41, P<0.001) and AI-predicted hope level (B=0.06, Wald=44.24, P<0.001). The results revealed that cases from one of the centers in the present study, a university hospital located in a different city than the other centers, exhibited higher hope levels. Additionally, a poorer ECOG performance status and lower AI-predicted hope levels were associated with reduced hope index scores (s-ATHS). On the whole, the present study demonstrates that AI-predicted hope levels are associated with hope index scores (s-ATHS), suggesting that monitoring AI-predicted hope levels may provide valuable insight in the practice of oncology.