Screening for Proteinopathic-related Dementias in Low-resource Clinical Contexts: A Machine Learning Approach
| dc.contributor.advisor | Thomas, Kevin | |
| dc.contributor.author | Lewis, Raphaella | |
| dc.date.accessioned | 2024-05-14T12:56:53Z | |
| dc.date.available | 2024-05-14T12:56:53Z | |
| dc.date.issued | 2023 | |
| dc.date.updated | 2024-05-14T12:23:08Z | |
| dc.description.abstract | Background: The prevalence of Alzheimer's disease (AD) and other subtypes of proteinopathic-related dementias (PRDs) is increasing rapidly in low- or middle-income countries (LMICs). The wide-ranging social and economic consequences of PRDs means there is an urgent need for clinical services dedicated to their early and accurate detection, particularly in low-resource contexts. The aim of the present study was to use machine learning techniques to identify a minimum number of clinical variables (neuropsychological test data and vascular risk factor information) required for accurate classification of PRDs in an older adult sample from an LMIC population and to derive a decision tree algorithm that diagnoses these types of dementias. Methods: The present study used data from a memory clinic sample of 253 South African older adults (130 with PRDs, 123 without). Information from 20 clinical variables were used as features for the analysis. We used C5.0 algorithms to identify the most important features for PRD diagnosis and to derive an algorithm that could accurately diagnose these types of dementia. Results: The C5.0 algorithm reduced the number of clinical variables for screening PRDs from 20 to 9 (Repeatable Battery for the Assessment of Neuropsychological Status [RBANS] Figure Recall, vascular risk factor, phonemic verbal fluency, RBANS List Recall, RBANS List Recognition, Digit Span Backward, CLOX1, CLOX2, and ∆CLOX2-1), and classified the validation sample with an accuracy exceeding chance performance. Accuracy, sensitivity, and specificity values were all greater than 70%. Performance on tests assessing memory and executive functioning were the features that predominantly distinguished the PRD and comparison groups from one another. Conclusions: The derived decision tree is a suitable and easy-to-interpret approach for PRD screening in LMICs. Its utility as part of standard clinical practice has the potential to free up strained resources and to allow clinical expertise to be employed more selectively. | |
| dc.identifier.apacitation | Lewis, R. (2023). <i>Screening for Proteinopathic-related Dementias in Low-resource Clinical Contexts: A Machine Learning Approach</i>. (). ,Faculty of Humanities ,Department of Psychology. Retrieved from http://hdl.handle.net/11427/39616 | en_ZA |
| dc.identifier.chicagocitation | Lewis, Raphaella. <i>"Screening for Proteinopathic-related Dementias in Low-resource Clinical Contexts: A Machine Learning Approach."</i> ., ,Faculty of Humanities ,Department of Psychology, 2023. http://hdl.handle.net/11427/39616 | en_ZA |
| dc.identifier.citation | Lewis, R. 2023. Screening for Proteinopathic-related Dementias in Low-resource Clinical Contexts: A Machine Learning Approach. . ,Faculty of Humanities ,Department of Psychology. http://hdl.handle.net/11427/39616 | en_ZA |
| dc.identifier.ris | TY - Thesis / Dissertation AU - Lewis, Raphaella AB - Background: The prevalence of Alzheimer's disease (AD) and other subtypes of proteinopathic-related dementias (PRDs) is increasing rapidly in low- or middle-income countries (LMICs). The wide-ranging social and economic consequences of PRDs means there is an urgent need for clinical services dedicated to their early and accurate detection, particularly in low-resource contexts. The aim of the present study was to use machine learning techniques to identify a minimum number of clinical variables (neuropsychological test data and vascular risk factor information) required for accurate classification of PRDs in an older adult sample from an LMIC population and to derive a decision tree algorithm that diagnoses these types of dementias. Methods: The present study used data from a memory clinic sample of 253 South African older adults (130 with PRDs, 123 without). Information from 20 clinical variables were used as features for the analysis. We used C5.0 algorithms to identify the most important features for PRD diagnosis and to derive an algorithm that could accurately diagnose these types of dementia. Results: The C5.0 algorithm reduced the number of clinical variables for screening PRDs from 20 to 9 (Repeatable Battery for the Assessment of Neuropsychological Status [RBANS] Figure Recall, vascular risk factor, phonemic verbal fluency, RBANS List Recall, RBANS List Recognition, Digit Span Backward, CLOX1, CLOX2, and ∆CLOX2-1), and classified the validation sample with an accuracy exceeding chance performance. Accuracy, sensitivity, and specificity values were all greater than 70%. Performance on tests assessing memory and executive functioning were the features that predominantly distinguished the PRD and comparison groups from one another. Conclusions: The derived decision tree is a suitable and easy-to-interpret approach for PRD screening in LMICs. Its utility as part of standard clinical practice has the potential to free up strained resources and to allow clinical expertise to be employed more selectively. DA - 2023 DB - OpenUCT DP - University of Cape Town KW - Psychology LK - https://open.uct.ac.za PY - 2023 T1 - Screening for Proteinopathic-related Dementias in Low-resource Clinical Contexts: A Machine Learning Approach TI - Screening for Proteinopathic-related Dementias in Low-resource Clinical Contexts: A Machine Learning Approach UR - http://hdl.handle.net/11427/39616 ER - | en_ZA |
| dc.identifier.uri | http://hdl.handle.net/11427/39616 | |
| dc.identifier.vancouvercitation | Lewis R. Screening for Proteinopathic-related Dementias in Low-resource Clinical Contexts: A Machine Learning Approach. []. ,Faculty of Humanities ,Department of Psychology, 2023 [cited yyyy month dd]. Available from: http://hdl.handle.net/11427/39616 | en_ZA |
| dc.language.rfc3066 | Eng | |
| dc.publisher.department | Department of Psychology | |
| dc.publisher.faculty | Faculty of Humanities | |
| dc.subject | Psychology | |
| dc.title | Screening for Proteinopathic-related Dementias in Low-resource Clinical Contexts: A Machine Learning Approach | |
| dc.type | Thesis / Dissertation | |
| dc.type.qualificationlevel | Masters |