Digital language measures capture episodic memory disruptions in people with human immunodeficiency virus: A machine learning study

Abstract

Objective: Human immunodeficiency virus (HIV) often affects episodic memory. Yet, standard measures of this domain are derived from clinicians’ simple counts of recalled and omitted pieces of information, undermining robustness, informativeness, and scalability. Here, we present an automated natural language processing (NLP) approach that tackles such limitations. Methods: We recruited 92 participants (50 people living with HIV and 42 controls), who performed a story retelling task. Using NLP tools, we compared the retellings and the original story in terms of verbosity, semantic acuity, and organizational structure. Results: We found that people living with HIV produced fewer nouns and had poorer semantic acuity and organizational similarity. Moreover, machine learning classifiers robustly differentiated between the two groups. Conclusion: These results suggest that our digital approach can reveal fine-grained episodic memory alterations in people living with HIV, offering an objective, scalable, and cost-effective complement to standard cognitive testing.