Nicklas Linz, Levi Muyela, Elisa Mallick, Johannes Tröger, Anne Nyambura Njogu, Anne Gitere, Rachel Maina, Olivera Nesic-Taylor, Chinedu Udeh-Momoh, Vaibhav Narayan, Irene Meier, and Karen Blackmon
* Poster presented at the conference “The Future of Dementia in Africa: Advancing Global Partnerships”, Nairobi (Kenya)
Abstract
INTRODUCTION: Dementia remains a significant and growing health challenge in Africa, where multilingualism and low literacy often complicate diagnostic efforts. The animal fluency test is a widely used measure of rapid semantic retrieval, easily administered across cultures and suitable for low literacy populations. This study explores the feasibility of using an automatic bilingual analysis system to evaluate semantic verbal fluency (SVF) in both Swahili and English for early dementia screening.
METHODS: We present data from a Davos Alzheimer’s Collaborative funded study in Nairobi, Kenya. This feasibility analysis includes 24 participants (mean age = 52 years; 14 females; mean no. of languages spoken = 3; median educational level = 12 years). Participants completed a neuropsychological battery including animal fluency, naming as many animals as possible in 60 seconds. Code switching (using English, Swahili, or tribal languages) was allowed. Automatic analysis used ki’ SIGMA speech analysis pipeline to recognize correctly named animals in any language. Features extracted included traditional metrics (number of correctly named animals) and qualitative temporal (pauses, speech rate) or semantic features (semantic clusters, word frequency). Swahili answers were translated into English for qualitative analysis. We aim to develop a dementia screening score suitable for African settings, considering multilingualism and low education.
RESULTS: Participants produced between 3 and 16 correct answers in the animal fluency. Language switching between English and Swahili was present in 5/24 participants (21%) and occurred between 1 and 5 times within a single trial.
CONCLUSION: With our approach we were able to extract all qualitative features including semantic coherence metrics, which demonstrates the feasibility of extracting speech biomarkers in an African setting. As this is an ongoing study, we aim to present a bigger sample at the conference. We aim at incorporating a big Kenyan data set into screening algorithms for cognitive impairment, which will be an important step in reducing global disparities in the early detection of Alzheimer’s disease.
