Dementia is a public health priority and its importance is projected to increase in coming decades, particularly in low-to middle-income countries. A description of the methodological challenges of observational studies and the limitations of previous attempts to combine the published literature leads me to discuss ascertainment of dementia cases and the suitability of dementia mortality as an outcome. I report the findings of a memory clinic study where 71.5% of 502 deceased individuals with probable Alzheimer dementia had dementia correctly recorded on their death certificate, which is an improvement on similar results from two decades earlier. I review the evidence for geographical variation in dementia and discuss the implication that such variation might point towards potentially modifiable risk or protective factors for dementia. I have attempted to overcome the methodological challenges alluded to above by only examining within-study comparisons. A metaanalysis of rural-urban comparisons reveals some evidence of increased prevalence (odds ratio; 90% confidence interval (CI): 1.11; 0.79, 1.57) and incidence (1.20; 0.84, 1.71) of dementia in rural areas. These associations were stronger for Alzheimer dementia and particularly so in studies which identified early life rural residence (prevalence 2.22; 1.19, 4.16; incidence 1.64; 1.08, 2.50). Since there are no effective treatments, there is an obvious need to focus on prevention and an urgent need to improve our understanding of the aetiology of dementia in order to attempt to prevent or delay its onset. However, it is clear that prevention must begin sufficiently early in life to have an effect – intervening in later life might be too late. I describe a body of work using the Health Survey for England cohort studies examining the association between a series of risk factors and later dementiarelated death, including cardiovascular disease risk factors, psychological distress, and socioeconomic status. For example, there is a dose-response relationship between increasing psychological distress and dementia death (12-item General Health Questionnaire score 1-3 vs 0 age- and sex-adjusted hazard ratio; 95% CI: 1.44; 1.17, 1.78; score 4-12 vs 0: 1.74; 1.36, 2.22). I conclude by summarising the contribution these publications have made to the field of dementia epidemiology and by outlining ongoing and future projects building on the work presented in this thesis.

Acute hospitals have seen unprecedented demographic changes, where older age, frailty and cognitive impairment now characterise the majority of health service users. Delirium is very common in this setting, and adverse outcomes are well described. However, studies investigating cognitive outcomes after delirium in unselected samples have been lacking. This thesis had four objectives: (1) To estimate the prevalence of delirium in the general population (2) To assess the association of delirium with cognitive outcomes (3) To investigate how these associations relate to underlying dementia pathology (4) To develop novel methods for retrospectively ascertaining delirium. Methods: Data from three population-based neuropathology cohort studies were used: Vantaa 85+; Cambridge City over-75s Cohort (CC75C); MRC Cognitive Function and Ageing Study (CFAS). (1) To ascertain the prevalence of delirium in the general population, a measure of delirium was developed using data recorded in standardised interview schedules, with criterion validity evaluated through the association with mortality and dementia risk. (2) The association with cognitive outcomes was tested in a series of logistic regression models, where delirium was the exposure and dementia (or worsening dementia severity) was the outcome. In addition, the association with change in Mini-Mental Status Examination (MMSE) score was assessed using random-effects linear regression. (3) In brain donors from all three cohorts, the independent effects of delirium, dementia pathology, and their interaction, were investigated using the same approach. (4) A chart-based method for deriving a retrospective diagnosis for delirium was developed, validated against bedside psychiatrist diagnosis. Vignettes from the medical record were abstracted and delirium status decided by expert consensus panel. Results: (1) Age-specific prevalence in CFAS increased with age from 1.8% in the 65-69 year age group to 13.5% in the ≥90 age group (p<0.01 for trend). (2) Delirium was consistently associated with adverse cognitive outcomes: new dementia (OR 8.7, 95% CI 2.1 to 35); worsening dementia severity (OR 3.1, 95% CI 1.5 to 6.3); faster change in Mini-Mental Status Examination (MMSE) score (1.0 additional points/year, p<0.01) (3) In the neuropathology analyses, decline attributable to delirium was -0.37 MMSE points/year (p<0.01). Decline attributable to dementia pathology was -0.39 MMSE points/year (p<0.01). However, the combination of delirium and dementia pathology resulted in the greatest decline, where the interaction contributed a further -0.16 MMSE points/year (p=0.01), suggesting that delirium worsened cognitive trajectories in dementia, but through distinct pathophysiological pathways not accounted for by Alzheimer’s, vascular or Lewy body pathology. (4) The chart abstraction method yielded a sensitivity of 0.88 and specificity 0.75 for ‘possible delirium’, with lower sensitivity (0.58) and higher specificity (0.93) for ‘probable delirium’ (AUC 0.86, 95% CI 0.82 to 0.89). This thesis adds to the small body of work on delirium in prospective studies, with the first ever analyses conducted in whole populations. The findings suggest new possibilities regarding the pathology of cognitive impairment, positioning delirium and/or its precipitants as a critically inter-related mechanism.

Introdução: O AVC (acidente vascular cerebral) é uma das mais importantes causas de alterações neuropsicológicas. Uma avaliação cognitiva inicial realizada por telefone implicaria em um diagnóstico mais precoce de prejuízo cognitivo e demência, reduzindo custos e tempo. Objetivo: Examinar as propriedades psicométricas da versão brasileira da Entrevista Telefônica para Avaliação do Estado Cognitivo - Modificada (TICS-M) em pacientes pós-AVC. Métodos: Previamente à validação da TICS-M em indivíduos acometidos por AVC, foi realizada tradução para o Português do Brasil e adaptação transcultural da versão original da TICS-M em uma amostra de 30 sujeitos não clínicos. Após esta fase, um subgrupo de 61 pacientes com AVC, participantes do Estudo da Mortalidade e Morbidade do AVC (EMMA) que ocorre no Hospital Universitário da Universidade de São Paulo, foram convidados a participar da validação da TICS-M, seis meses após o evento agudo. A TICS-M foi aplicada em três momentos: avaliação inicial (entrevista presencial), uma e duas semanas após a primeira avaliação. Na avaliação inicial, além da TICS-M, questionários adicionais foram aplicados para avaliar a cognição: MoCA (Montreal Cognitive Assessment), MEEM (Mini Exame do Estado Mental); e para a depressão, HDRS (Hamilton Depression Rating Scale). Todos os questionários foram aplicados por duas entrevistadoras treinadas para o estudo. A confiabilidade intra-observador da TICS-M foi testada através dos coeficientes de Pearson, Intraclasse e alfa de Cronbach. As características internas do TICS-M também foram avaliadas através de uma análise exploratória utilizando o método Análise de Componentes Principais. A validade discriminatória do instrumento para rastreamento de demência pós-AVC foi avaliada em comparação a MEEM pela análise da área sob a curva (AUC) determinada pela curva ROC. Foram calculadas sensibilidade e especificidade para o ponto de corte ideal para rastrear demência. Resultados: De maneira geral, a TICS-M traduzida para o português apresentou um bom entendimento dos itens na mostra de indivíduos não clínicos. Foi observada uma frequência de 23% sugestiva de demência pós-AVC. O nível de escolaridade esteve positivamente associado ao estado demencial rastreado pelo MEEM. O estado depressivo assim como outras características de base não se associou à demência sugerida pelo MEEM. A confiabilidade teste-reteste intra-observador revelou taxas quase totais nos três momentos avaliados (Pearson Coeficiente > 0,85, Coeficientes de Correlação Intraclasse > 0,85 e Coeficiente alfa de Cronbach: 0,96). A análise fatorial determinou três domínios: memória de trabalho e atenção; memória recente e de evocação e orientação. A área sob a curva (AUC) determinada para a TICS-M em comparação com MEEM foi de 0,89 (intervalo de confiança 95%: 0,80-0,98). O ponto de corte sugerido para TICS-M foi de 14 pontos (escala de 0-39 pontos) para rastrear demência com sensibilidade de 91,5% e especificidade de 71,4%. Resultados semelhantes foram observadas com o MoCA. Conclusão: A versão brasileira da TICS-M sugere ser um instrumento de pesquisa útil e confiável para rastrear demência em pacientes pós-AVC / Introduction: Stroke is one most important cause of neuropsychological disorders. An initial cognitive assessment performed by telephone resulting in an early diagnosis of cognitive impairment and dementia, reducing costs and time. Objective: To examine the psychometric properties of the Brazilian version of the Modified Telephone Interview for Cognitive Status Assessment (TICS-M) for assessment of dementia in post-stroke patients. Methods: Prior to validation of TICS-M in post-stroke patients, translation was performed for the Brazilian-Portuguese and cross-cultural adaptation of the original version of TICS-M in a non-clinical sample of 30 subjects. After this phase, 61 stroke patients enrolled in the Stroke Mortality and Morbidity Study (The EMMA study) that occurs at the University Hospital of the University of São Paulo, were invited to participate in this sub-study to validate the TICS-M six months after the acute event. The TICS-M was applied in three moments: first evaluation (personal interview), one and two weeks after of the first evaluation. At the first evaluation, beyond the TICS-M, additional questionnaires were applied to assess cognition: MoCA (Montreal Cognitive Assessment), MMSE (Mini-Mental Status Examination), and for depression, HDRS (Hamilton Depression Rating Scale). All questionnaires were administered by two trained interviewers for the study. Reliability of the TICS-M was tested by intra-observer rates using Pearson, Intraclass and Cronbach´s alpha coefficients. The internal characteristics of TICS-M were also evaluated by an exploratory analysis using Principal Component Analysis. The discrimination validity of the instrument to assess dementia was evaluated by comparison to the MMSE analysis of the area under the curve (AUC) determined by the ROC curve. Sensitivity and specificity for the ideal cutoff to assess dementia were calculated. Results: In general, the TICS-M translated into Portuguese version showed a good understanding of the items in non-clinical individuals. A frequency of 23% suggestive of post-stroke dementia was observed. The level of education was positively associated with dementia status assessed by MMSE. The depressive status, as well as, other baseline characteristics was not associated with dementia suggested by MMSE. Test-retest reliability intra-observer revealed almost total rates in the three evaluation moments (Pearson coefficient > 0.85, Intraclass Correlation Coefficient > 0.85 and Cronbach\'s alpha coefficient: 0.96). The factorial analysis determined three domains: working memory and attention, recent and recall memory and orientation. The area under the curve (AUC) determined by TICS-M compared to MMSE was 0.89 (95% confidence interval: 0.80-0.98). The cutoff suggested for TICS-M was equal or greater than 14 points (range 0-39 points) to assess dementia (91.5% sensitivity, 71.4 % specificity). Similar results were observed with the MoCA. Conclusion: The Brazilian version of TICSM suggests being a useful and reliable research instrument to evaluate dementia in poststroke patients in epidemiological studies

Introdução: O AVC (acidente vascular cerebral) é uma das mais importantes causas de alterações neuropsicológicas. Uma avaliação cognitiva inicial realizada por telefone implicaria em um diagnóstico mais precoce de prejuízo cognitivo e demência, reduzindo custos e tempo. Objetivo: Examinar as propriedades psicométricas da versão brasileira da Entrevista Telefônica para Avaliação do Estado Cognitivo - Modificada (TICS-M) em pacientes pós-AVC. Métodos: Previamente à validação da TICS-M em indivíduos acometidos por AVC, foi realizada tradução para o Português do Brasil e adaptação transcultural da versão original da TICS-M em uma amostra de 30 sujeitos não clínicos. Após esta fase, um subgrupo de 61 pacientes com AVC, participantes do Estudo da Mortalidade e Morbidade do AVC (EMMA) que ocorre no Hospital Universitário da Universidade de São Paulo, foram convidados a participar da validação da TICS-M, seis meses após o evento agudo. A TICS-M foi aplicada em três momentos: avaliação inicial (entrevista presencial), uma e duas semanas após a primeira avaliação. Na avaliação inicial, além da TICS-M, questionários adicionais foram aplicados para avaliar a cognição: MoCA (Montreal Cognitive Assessment), MEEM (Mini Exame do Estado Mental); e para a depressão, HDRS (Hamilton Depression Rating Scale). Todos os questionários foram aplicados por duas entrevistadoras treinadas para o estudo. A confiabilidade intra-observador da TICS-M foi testada através dos coeficientes de Pearson, Intraclasse e alfa de Cronbach. As características internas do TICS-M também foram avaliadas através de uma análise exploratória utilizando o método Análise de Componentes Principais. A validade discriminatória do instrumento para rastreamento de demência pós-AVC foi avaliada em comparação a MEEM pela análise da área sob a curva (AUC) determinada pela curva ROC. Foram calculadas sensibilidade e especificidade para o ponto de corte ideal para rastrear demência. Resultados: De maneira geral, a TICS-M traduzida para o português apresentou um bom entendimento dos itens na mostra de indivíduos não clínicos. Foi observada uma frequência de 23% sugestiva de demência pós-AVC. O nível de escolaridade esteve positivamente associado ao estado demencial rastreado pelo MEEM. O estado depressivo assim como outras características de base não se associou à demência sugerida pelo MEEM. A confiabilidade teste-reteste intra-observador revelou taxas quase totais nos três momentos avaliados (Pearson Coeficiente > 0,85, Coeficientes de Correlação Intraclasse > 0,85 e Coeficiente alfa de Cronbach: 0,96). A análise fatorial determinou três domínios: memória de trabalho e atenção; memória recente e de evocação e orientação. A área sob a curva (AUC) determinada para a TICS-M em comparação com MEEM foi de 0,89 (intervalo de confiança 95%: 0,80-0,98). O ponto de corte sugerido para TICS-M foi de 14 pontos (escala de 0-39 pontos) para rastrear demência com sensibilidade de 91,5% e especificidade de 71,4%. Resultados semelhantes foram observadas com o MoCA. Conclusão: A versão brasileira da TICS-M sugere ser um instrumento de pesquisa útil e confiável para rastrear demência em pacientes pós-AVC / Introduction: Stroke is one most important cause of neuropsychological disorders. An initial cognitive assessment performed by telephone resulting in an early diagnosis of cognitive impairment and dementia, reducing costs and time. Objective: To examine the psychometric properties of the Brazilian version of the Modified Telephone Interview for Cognitive Status Assessment (TICS-M) for assessment of dementia in post-stroke patients. Methods: Prior to validation of TICS-M in post-stroke patients, translation was performed for the Brazilian-Portuguese and cross-cultural adaptation of the original version of TICS-M in a non-clinical sample of 30 subjects. After this phase, 61 stroke patients enrolled in the Stroke Mortality and Morbidity Study (The EMMA study) that occurs at the University Hospital of the University of São Paulo, were invited to participate in this sub-study to validate the TICS-M six months after the acute event. The TICS-M was applied in three moments: first evaluation (personal interview), one and two weeks after of the first evaluation. At the first evaluation, beyond the TICS-M, additional questionnaires were applied to assess cognition: MoCA (Montreal Cognitive Assessment), MMSE (Mini-Mental Status Examination), and for depression, HDRS (Hamilton Depression Rating Scale). All questionnaires were administered by two trained interviewers for the study. Reliability of the TICS-M was tested by intra-observer rates using Pearson, Intraclass and Cronbach´s alpha coefficients. The internal characteristics of TICS-M were also evaluated by an exploratory analysis using Principal Component Analysis. The discrimination validity of the instrument to assess dementia was evaluated by comparison to the MMSE analysis of the area under the curve (AUC) determined by the ROC curve. Sensitivity and specificity for the ideal cutoff to assess dementia were calculated. Results: In general, the TICS-M translated into Portuguese version showed a good understanding of the items in non-clinical individuals. A frequency of 23% suggestive of post-stroke dementia was observed. The level of education was positively associated with dementia status assessed by MMSE. The depressive status, as well as, other baseline characteristics was not associated with dementia suggested by MMSE. Test-retest reliability intra-observer revealed almost total rates in the three evaluation moments (Pearson coefficient > 0.85, Intraclass Correlation Coefficient > 0.85 and Cronbach\'s alpha coefficient: 0.96). The factorial analysis determined three domains: working memory and attention, recent and recall memory and orientation. The area under the curve (AUC) determined by TICS-M compared to MMSE was 0.89 (95% confidence interval: 0.80-0.98). The cutoff suggested for TICS-M was equal or greater than 14 points (range 0-39 points) to assess dementia (91.5% sensitivity, 71.4 % specificity). Similar results were observed with the MoCA. Conclusion: The Brazilian version of TICSM suggests being a useful and reliable research instrument to evaluate dementia in poststroke patients in epidemiological studies