Cognitive Domains and Functional Assessment Questionnaire (CDFAQ): Accuracy Across Educational Levels

Background The Cognitive Domains and Functional Assessment Questionnaire (CDFAQ) assess cognitive and functional decline for Neurocognitive Disorders based on the DSM‐5 criteria (1). It’s accuracy to the Informant Questionnaire on Cognitive Decline in the Elderly ‐ Long Version (IQCODE‐LV) has been assessed (2), and was translated and validated into English. The informant version (CDFAQ‐IV) assess: Complex Attention (CA), Executive Functions (EF), Learning and Memory (LM), Language (L), Perceptual‐Motor (PM) and Social Cognition. Methods To evaluate the CDFAQ‐IV’s accuracy to IQCODE‐LV across participant’s educational levels. Both questionnaires were applied to 196 informants of older adults. Older adults were stratified by educational levels into three groups: 73 participants had low educational level (0‐3 years), 71 had middle educational level (4‐7 years), and 52 with high educational level (8 or more years). Accuracy assessment was made by Receiver Operating Characteristic (ROC) and cut‐off values defined by Youden’s J statistic. This study was approved by the ethics committee of UFMG. Results The low educational group had good accuracy (Area Under the Curve = 0.879) for the CDFAQ‐IV total score, excellent accuracy for the EF domain (AUC = 0.907), good accuracy for LM domain (AUC = 0.822) and poor accuracies for other domains. For the middle educational group, CDFAQ‐IV (AUC = 0.908) and LM domain (AUC = 0.914) showed excellent accuracies, while CA (AUC = 0.876), L (AUC = 0.832) and SC (AUC = 0.814) domains showed good accuracies. The others domains showed poor accuracies. For high educational group, LM domain (0.924) had excellent accuracy, while CDFAQ’s total score (AUC = 0.894) and L domain (AUC = 0.898) showed good accuracy. Poor accuracies were found for the other domains. Conclusions CDFAQ’s total score shows good accuracy in samples of low and high educational levels and excellent accuracy in samples with 4 to 7 years of schooling. Interestingly, the LM domain shows excellent accuracy (AUC > 0.9) in samples with higher levels of education. Brazilian older adults have lower educational levels and more heterogeneity in schooling quality compared to those from high‐income nations. Assessment of psychometric properties according to education is needed, as well as more scenarios such as studies in community and clinical settings.