L’impact du niveau d’études sur les performances cognitives est largement connu. Aussi, l’évaluation des sujets de bas ou très bas niveau d’études est un réel challenge pour les cliniciens. D’une part, les outils habituellement utilisés dans le bilan du sujet âgé ne sont pas adaptés aux individus analphabètes ou de bas niveau d’éducation, et, d’autre part, les normes nécessaires à l’interprétation des scores cognitifs ne prennent pas suffisamment en considération cette population de bas niveau scolaire. Ainsi, la question que pose ce travail de thèse est de savoir comment optimiser l’évaluation neuropsychologique de ces sujets de bas niveau scolaires. L’objectif poursuivi par la première étude de cette thèse est de déterminer s’il existe un effet propre du fait de savoir lire et écrire indépendamment de celui de la scolarisation. Pour répondre à cette question, nous avons pu accéder aux données recueillies dans la cohorte mexicaine de Coyoacán qui a la particularité d’avoir un échantillon important de sujets n’ayant jamais été scolarisés. Cette première étude a montré que des sujets n’ayant jamais accédé au système éducatif mais ayant des notions rudimentaires de lecture et d’écriture acquises de manière informelle, ont des performances plus élevées à tous les tests considérés, hormis pour le Set test d’Isaacs (IST), que des sujets ne sachant pas lire et écrire. Ainsi, ces résultats illustrent l’effet des capacités de lecture et d’écriture, un effet distinct de l’effet du niveau de scolarisation. Une seconde difficulté à laquelle doivent faire face les cliniciens est l’absence de normes adaptées aux sujets de bas niveau d’études. Ainsi, un travail autour de l’élaboration de normes adaptées aux sujets de bas niveau d’études a été réalisé dans la cohorte de Coyoacán et nous a permis d’élaborer des normes pour le Mini Mental State Examination (MMSE), le Rappel libre/Rappel indicé 16 items (RL/RI-16) et l’IST, jusque-là inexistantes pour la population âgée mexicaine et qui a la particularité de présenter une forte proportion de sujets de bas niveau d’études. Un second travail de normalisation a été réalisé dans la cohorte AMI, une cohorte menée en milieu rural dans le département de la Gironde dans laquelle la proportion de sujets de bas niveaux est plus élevée qu’en population générale, pour un nouveau test de mémoire visuo-spatiale, le test des gobelets pour lequel nous avons également étudié sa validité dans la détection de la démence. Ces normes classiques, corrigées pour des variables démographiques sont essentielles à l’interprétation des scores cognitifs. Dans le même temps, dans le cas de la démence où l’âge et le niveau d’études sont deux facteurs de risque majeurs, cette pratique habituelle de corriger pour ces variables afin d’établir ou de prédire un diagnostic de démence peut être remise en question. Si des travaux antérieurs ont montré que l’utilisation de scores corrigés diminue la qualité de détection de la démence, la dernière étude de cette thèse a montré qu’il en est de même lorsqu’il s’agit de prédire la démence : les sujets qui deviennent déments à court terme sont mieux classés lorsque les scores ne sont pas corrigés que lorsqu’ils sont corrigés pour l’âge et le niveau d’études. Ainsi, loin de régler le vaste problème de l’évaluation du sujet de bas niveau d’études, cette thèse tente d’apporter des éléments pragmatiques au clinicien sous la forme de tests ou de normes adaptées, mais aussi des éléments de réflexion sur l’utilisation de ces scores. / The effect of education on cognitive performances and neuropsychological assessment outcomes has been well documented so far. Indeed, the assessment of cognitive performance of individuals with low and very low educational level remains a major clinical challenge for several reasons. Firstly, conventional cognitive assessment tools used with elderly patients are not suitable for illiterate or poorly instructed individuals. Secondly, traditional norms used to identify strengths and weaknesses of cognitive performance are not sensible enough to detect cognitive impairments among illiterate or scarcely instructed individuals. Therefore, how to improve the neuropsychological assessment of individuals with low formal educational level is the main research question of this doctoral thesis. To adequately answer to this prior question, four studies have been conducted. The aim of the first study presented in this thesis was to investigate the specific effect of literacy acquisition on cognitive performances independently of education. For this work, we used data collected from the Coyoacán study, a Mexican population-based cohort which presents an important sample of uneducated participants. The main finding of this study was that participants who never attended school but counting with informal literacy abilities, presented better performances for all cognitive tests, except for the Isaacs Set test (IST), compared to their uneducated illiterate counterparts. These findings illustrate the specific effect of literacy skills acquisition on cognitive performances independently of the influence of schooling. The lack of adapted comparative norms is the second major challenge of neuropsychological assessment of individuals with low-educational level. Consequently, our second study aimed at establishing comparative norms for the MMSE, the Free and Cued Selective Reminding Test and the IST, three tests widely used for cognitive evaluation in elderly people. Of note, these norms were specifically established for the Mexican elderly population, which presents a high rate of illiteracy and low-educational level. Finally, a third study aiming at establishing comparative norms for a new visuo-spatial memory-test: the goblets test. For this study, we used data collected from the AMI cohort study which is conducted in rural areas in the South-West of France. Within this third study we also studied the validity of this test in detecting dementia illness. These comparative norms are necessary to interpret cognitive scores. However, as age and education are major risk factors for dementia, correcting for these demographic variables to improve the accuracy of detection or prediction of dementia may be questionable. Consistently with other studies showing that dementia detection accuracy is compromised when corrected scores are used; findings of our last study showed that regarding dementia prediction accuracy, participants developing dementia, in the short term, are better classified when using uncorrected scores for age and education than the corrected ones. To conclude, cognitive assessment of elderly individuals with low-educational level remains still a major clinical barrier to correctly diagnose dementia. However, besides providing clinicians with several pragmatic inputs such as cognitive tests and appropriated comparative norms, we believe that our findings will encourage clinical reflection regarding the use of these scores.

Diagnostic biases against individuals of advanced age or few years of formal education exist because age and education are commonly related to performance on cognitive tests, thus, demographic corrections for these tests are used. Corrections are complicated, however, by an association between demographic variables and dementia diagnoses. This dissertation examined the dementia diagnostic accuracy of demographic corrections for cognitive tests. Experiment I tested whether, in the context of skewed tests that violate the statistical assumptions of linearity and homoscedasticity, the accuracy of demographically-corrected test scores would be reduced. Experiment II tested whether demographic corrections would only be appropriate for biased factors instead of the total score for multifactorial tests. Experiment III explored whether demographic corrections would be inappropriate under conditions where the dementia pathology overrides the association between cognitive test scores and demographic variables. Experiment IV explored whether demographic corrections would be inappropriate in conditions where the demographic variables were, in themselves, risk factors for dementia, as this would remove predictive variance. Experiment V explored aspects particular to regression-based demographic corrections that might adversely affect diagnostic accuracy. Experiments I to V were simulation-based; consequently Experiment VI explored replication of these findings using regression adjusted scores in a previously collected clinical database. Finally, Experiment VII used clinical data in conjunction with published clinical normative data with demographic-stratification to test the generalizability of these findings to clinical practice. Using area under the receiver operating characteristic curve comparisons, the use of demographically-corrected scores repeatedly failed to improve upon the dementia diagnostic accuracy of uncorrected cognitive test scores, regardless of whether these corrections were regression-based or based on demographically stratified normative data. Demographic corrections reduced dementia diagnostic accuracy when cognitive test scores were skewed or when adjustments were regression-based and demographic variables were risk factors for dementia. The use of demographic corrections when dementia pathology supersedes any association between cognitive test scores and demographic variables does not impact the relative diagnostic accuracy of demographically-corrected versus uncorrected test scores. Overall, these results suggest that the use of demographic corrections for cognitive test scores is highly cautioned when the goal is to maximize dementia diagnostic accuracy.

Diagnostic biases against individuals of advanced age or few years of formal education exist because age and education are commonly related to performance on cognitive tests, thus, demographic corrections for these tests are used. Corrections are complicated, however, by an association between demographic variables and dementia diagnoses. This dissertation examined the dementia diagnostic accuracy of demographic corrections for cognitive tests. Experiment I tested whether, in the context of skewed tests that violate the statistical assumptions of linearity and homoscedasticity, the accuracy of demographically-corrected test scores would be reduced. Experiment II tested whether demographic corrections would only be appropriate for biased factors instead of the total score for multifactorial tests. Experiment III explored whether demographic corrections would be inappropriate under conditions where the dementia pathology overrides the association between cognitive test scores and demographic variables. Experiment IV explored whether demographic corrections would be inappropriate in conditions where the demographic variables were, in themselves, risk factors for dementia, as this would remove predictive variance. Experiment V explored aspects particular to regression-based demographic corrections that might adversely affect diagnostic accuracy. Experiments I to V were simulation-based; consequently Experiment VI explored replication of these findings using regression adjusted scores in a previously collected clinical database. Finally, Experiment VII used clinical data in conjunction with published clinical normative data with demographic-stratification to test the generalizability of these findings to clinical practice. Using area under the receiver operating characteristic curve comparisons, the use of demographically-corrected scores repeatedly failed to improve upon the dementia diagnostic accuracy of uncorrected cognitive test scores, regardless of whether these corrections were regression-based or based on demographically stratified normative data. Demographic corrections reduced dementia diagnostic accuracy when cognitive test scores were skewed or when adjustments were regression-based and demographic variables were risk factors for dementia. The use of demographic corrections when dementia pathology supersedes any association between cognitive test scores and demographic variables does not impact the relative diagnostic accuracy of demographically-corrected versus uncorrected test scores. Overall, these results suggest that the use of demographic corrections for cognitive test scores is highly cautioned when the goal is to maximize dementia diagnostic accuracy.