Enhancing cognitive impairment screening by means of conversational agents and machine learning

• Integration of ML techniques into DigiMoCA, a T-MoCA-based conversational agent. • Models yielded an R 2 of 0.81 and RMSE of 1.84 in predictive validity. • MCI detection showed an AUC of 0.98, with 100% sensitivity and 93% specificity. • DigiMoCA offers a self-administered, accessible, and cost-effective alternative. • The integration of ML techniques significantly outperforms previous related work. Background: This study investigates the feasibility of integrating machine learning techniques into DigiMoCA, a T-MoCA-based conversational agent for the early screening of cognitive impairment in senior adults. Method: A sample of 46 participants with varying cognitive decline levels, age, and gender was recruited. Each participant undertook several gold standard tests as well as two administrations of DigiMoCA. Then, a data processing architecture was designed and implemented to test different Machine Learning (ML) algorithms. Results: Predictive validity testing using the Telephone Montreal Cognitive Assessment (T-MoCA) yielded promising results, with the best-performing algorithm achieving an R 2 = 0.81 and RMSE of 1.84. Criterion validity testing for discerning between individuals suffering from mild cognitive impairment (MCI) and non-MCI individuals showed an area under the ROC curve of 0.98 in the best case, with 100% sensitivity and 93% specificity. Thus, the inclusion of machine learning allows to exploit more effectively the information captured with DigiMoCA, improving its MCI screening potential. Conclusions: Integrating machine learning techniques into DigiMoCA showed promising predictive and criterion validity, indicating the efficacy of the approach in discerning cognitive decline levels and facilitating timely interventions for individuals at risk of cognitive decline. Despite limitations in sample size and heterogeneity, the findings notably outperform previous related work, laying a foundation for future research and showcasing the potential of conversational agents for this purpose.