The promotion of healthy aging is one of the main challenges of our society in the face of the aging population phenomenon. Early detection of frailty, which is a geriatric syndrome that alters the lifestyle of the elderly and threatens their desire to live independently, is a challenge, and to date there is no consensual method for this detection. In this paper, we analyze the transcribed discourse of older adults, in response to a series of open-ended questions. To this end, the answers of 34 subjects aged over 80 years were transcribed into texts. The objective of this study is to evaluate the ability of discourse analysis to separate robust from pre-frail individuals. Explicitly, we ran through two text classification approaches: (i) the application of a linear model on the TF-IDF technique and (ii) the application of a recurrent neural network after encoding the words with the Word2Vec algorithm using pre-learned information. Our models identify potentially informative words in the corpus of older individuals recruited in this study. This study suggests that text analysis provides promising diagnostic utility in the analysis of frailty, and opens the door for further research.