Learning Deep Temporal Representations of Socioeconomic Deprivation.

This study focuses on AI-based quantification of socioeconomic disparities within society. Specifically, we investigate whether embeddings derived from a recurrent neural network (RNN), trained on the features of a deprivation index, capture more information on deprivation than traditional expert-based indices. Using the German Index of Socioeconomic Deprivation (GISD) as a benchmark, predictive performance is evaluated through survival analyses in three independent cancer cohorts. RNN embeddings consistently outperformed the GISD score. The largest improvement in overall survival prediction was observed in thyroid carcinoma (c-index +6.85%, from 0.555 to 0.593), followed by breast cancer (4.46%) and lung cancer (2.19%).