Fiona Matthews

Statistical epidemiology in ageing research:report on progress 2001-2006

• Abstract
• Introduction
• Medical Research Council Cognitive Function and Ageing Study (CFAS)
  • Data and analysis
  • Dementia and cognition
  • Depression
  • Healthy Active Life Expectancy, functional disability and survival
  • Health and health policy
  • Ethics and research governance
  • Benefits of Unit's role within the CFAS co-operative
• Healthy active life expectancy and population burden of disease
• Longitudinal estimation in complex study designs
• Ophthalmology
• Summary of major achievements
• Publications from this programme

Abstract

My research programme has developed from my position as statistician for the MRC Cognitive Function and Ageing study (CFAS), and the application of my methodological focus remains within the field of ageing. Progress has concentrated in understanding the methodological complications of CFAS-type designs and how these can impact on estimates of disease rates. This internationally renowned study has in recent years provided the first population estimates relating neuropathology and in-life data, UK-specific prevalence and incidence of dementia and the first quantification of dementia at death. Attrition causes problems for most longitudinal studies, though it is of particular importance within the field of ageing. Therefore we have undertaken a detailed investigation of attrition within CFAS and other studies of the elderly and have developed methodology to adjust for problems of attrition.

CFAS has had a fundamental impact on the understanding of the changes in the ageing brain and their relationship to dementia. The CFAS brain donation programme is unique internationally and has allowed interrogation of assumptions about normal and abnormal ageing and the value of specification of dementia subtype in the common dementias of old age. Information from my research programme has provided population levels of dementia in terms of prevalence, incidence and the analysis of risk. Risk factors for dementia proposed by other studies that have been less representative of the complete elderly population (including those living in residential care) have limited population impact. This has been found for both environmental and genetic risks; such findings have implications for the preventative potential of dementia. Morbidity and the health of your life are just as important as preventing death. I have undertaken enumeration of differentials in healthy life by region, gender and disease state.

Introduction

My role within the Unit is principally as head statistician for the MRC Cognitive Function and Ageing study (CFAS). This involves running the data archive and statistical analysis core group, planning study designs and overseeing all analysis of CFAS data. The use of the study data has expanded dramatically and I now undertake study design (nested case control studies), power calculations (for grant applications and ethical approvals), analysis advice and monetary implications both before and after studies start, plus statistical critique of all analyses arising from the study. All proposals are evaluated to decide whether the core team undertakes, or alternatively supports, the analysis. There are many methodological complications that arise from analysis within the study data and these have developed into various methodological interest areas which are detailed below. Since 2001 I have employed or had statistical/data management oversight for ten full time members of staff, nine PhD students and eight masters students.

Medical Research Council Cognitive Function and Ageing Study (CFAS)

MRC CFAS was initiated in the late 1980s to investigate the prevalence and incidence of dementia and cognitive decline in a representative sample of over 18,000 individuals from the elderly population in England and Wales [02.407]. The study was initially designed as a two-wave two-phase study where a short initial screen is undertaken on the whole study, a stratified random sample of individuals are selected for a more detailed interview, in which the dementia (and other diseases) are diagnosed. This study design raises interesting methodological complexities which have been detailed below. Since this initial phase the study has, every two years, undertaken interviews on either all or selected subgroups of respondents and is currently interviewing a subset of respondents at fourteen years after they were first seen. The study is run from the department of Public Health and Primary Care, Cambridge and data are held in the Unit. This team, who together with the main collaborators (Newcastle, Liverpool, Nottingham, Oxford, Sheffield, Leicester, Edinburgh) plus lay members control the scientific direction of the study.

CFAS has had core research funding since the late 1980s and as a MRC Co-operative group since 2001, renewed in 2005. In 2001 I developed a scientific strategy to enable CFAS to concentrate on its strengths. The Unit has been the data centre since inception, but now has a more formal data archiving role with increasing usage of the CFAS data resource. To stimulate new proposals and to facilitate management of the increasing diversity of collaborative projects we have divided CFAS into five major and two supporting theme groups (each with a lead member): dementia and cognition (Fiona Matthews ), depression (Michael Dewey, Institute of Psychiatry), disability and healthy active life expectancy (Carol Jagger, Leicester), health and health policy (John Bond, Newcastle), cohort and new developments, ethics and governance (Bronwyn Parry, UCL), and data and analysis methods (Tony Johnson). We have linked existing component grants and collaborations to these themes, and we continue to seek new components and collaborations within each. Each theme group developed a set of aims and summary of intentions and research activities.

The ongoing fieldwork in CFAS is supported within the Unit, and the Unit holds the CFAS data archive and undertakes responsibility for the full audit and documentation of the data from over the fourteen years of interviews. In addition, data from fieldwork have been made available within weeks of completion ensuring timely analysis. The data management team codes and checks all scales embedded in the interviews using original publications. We release these into a dataset alongside the versioned interview data, together with notes and references on the website (http://www-cfas.medschl.cam.ac.uk); we send data to collaborators tailored to their requested analysis format preserving labels and missing value indicators. The CFAS data archive currently consists of documented and released data for all interviews undertaken on respondents (47,000 interviews), informants (7,500 interviews), together with death data (17,300 deaths, with ICD codes), medication data (43,900 records), genetic information (2000 genes measured) and neuropathological data (458 brains collected). The core team provides support for all analysis and I chair the CFAS Analysis Group. Analyses that link phenotypic and genetic information are not released to external collaborators and instead are analysed within the Unit using visitor computer accounts. The main outputs of the study are summarised below by theme. The study has received detailed scrutiny from the Multi Regional Ethics Committee (MREC) and has its own biological resource advisory committee.

Data and analysis:

Support for data and analytical aspects of CFAS is key to the Co-operative and this group draws on the long-term involvement of core members. I have developed methods of incidence estimation in studies with longitudinal sub-sampling [05.803; 05.068], and extended this to weighting probabilities for all analysis stages that take into account both the population sampling and the stratified nature of the interview structure [05.803; 05.107]. In addition methods of using multiple imputation rather than weighting have been evaluated for use to estimate prevalence of both dementia and depression [05.803] (Chatfield, 2002). More methodological detail is included in the specific sections below.

Dementia and cognition:

We have found that 60% of individuals in institutions have dementia, which was rather more than expected for in care settings [02.060], and that incidence of dementia does not vary by centre across England and Wales and continues to increase even at the oldest ages in both sexes. We found no evidence of the plateau suggested by earlier studies [05.068]. This analysis uses methodology specifically designed for the study and analyses rates both with and without adjustment for additional mortality [05.803; 05.068]. I have developed methods to estimate incidence of dementia that adjust for study design and potential bias through missing response data and further development will follow [05.803]. Analysis of risk factors for incident dementia at two and six-year follow-up shows stroke, self-perceived health and age to be the strongest [06.078].

We showed that neuropathology of ageing is more complex than previously thought; aspects such as plaques and tangles are related to dementia. However, more individuals have pathology than their cognitive level suggests and likewise some individuals with a large amount of pathology remained cognitively intact [02.063; 01.058]. We demonstrated that measurement of white matter lesions is feasible using magnetic resonance imaging (MRI) from fixed brain slices thereby opening up a new area for research into archived brain collections [06.035; 04.042; 04.041; 03.043]. We found that white matter lesions are common in brains of the elderly and are still related to dementia after adjusting for other pathological burdens; they are also weakly associated with Alzheimer's type pathology and vascular pathology. We have also found that whilst white matter lesions in different brain areas share some common pathology, different biological responses are demonstrated [06.035] and that proteins in glial cells are associated with Alzheimer type changes [05.122]. Many non-population based studies have found very strong relationships between Alzheimer's disease (AD) and genetic factors (in particular ApoE and ACE). CFAS demonstrated that these and other genes were not major risk factors for dementia and cognitive decline in a population setting [02.018; 02.062; 01.064; 01.090; 01.091; 02.109]. Further investigations showed that ApoE appears, rather, to be related to specific aspects of cognition [04.100].

CFAS undertook detailed cognitive analyses and these have provided useful population norms for clinical use [05.039; 02.088; 03.102]. Population norms, whilst usually derived from cross-sectional data, are generally used in the clinical setting, with individuals being plotted along a trajectory. The use of longitudinal data should therefore provide a better indication of the pathways of real individuals, though attrition plays a detrimental role. Longitudinal studies all suffer from attrition problems, where individuals who dropout from the study through death [04.076; 01.068] or refusal/non-contact [05.014; 04.076] differ with respect to their characteristics from those who return. In studies of older populations, these individuals are more likely to be older and cognitively frail [05.014; 04.076]. Many estimates of disease processes would therefore be biased if these informative missing data are ignored.

Depression:

Depression within CFAS has been measured using the assessment subsample and hence has been used for methodological investigation. We found that methods for estimating prevalence of depression using multiple imputation had larger estimates of error than other simpler methods and could not be expanded into an incidence framework for modelling depression in CFAS (Chatfield, 2002). Hence simple analysis using inverse probability weighting has been used.

CFAS neuropathological data have also been used to investigate differences in serotonin and noradrenalin which have been shown in hospital-based studies to be associated with late life depression. Population based neuropathology did not detect this difference [05.111].

Healthy Active Life Expectancy, functional disability and survival:

As populations age, the focus of health research shifts more from the question "what causes death" to "what causes healthy life". This subtle change has important implications since factors which have high prevalence but low associated mortality (e.g. arthritis) can have much more impact on overall health than factors which are more generally associated with poor survival (e.g. heart disease) [05.107]. Many government proposals (DH, 1999; DoETR, 1999; DSS, 1999) are now focused on healthy living rather than life extension, particularly within old age. Methods are therefore essential to understand differences in healthy life now and the potential for improvement for the future. We examined the relationship between levels of disability and death to find which factors have most impact on both morbidity (measured using functional impairment) and death. We showed that there is a trade off between rare but severe conditions, and common but less disabling conditions that cause a great deal of morbidity. We found important differences in healthy life expectancy by sex, functional and cognitive impairment and by region across England and Wales [02.043; 01.045; 06.069]. We found a very strong effect of cognition on survival both over the short term [04.076] and longer term [01.068]. This result has led to further research to quantify the difference in life expectancy with cognitive impairment [01.045; 06.069; 01.068].

Health and health policy:

MRC CFAS has provided much information about the costs, both formal and informal, involved in care [01.057]. CFAS data have been used to extend the Personal Social Services Research Unit (PSSRU) long term care financing model to examine cognitive impairment and its implications for future demand for services and costs [03.019; 01.020].

The relationship between immune measures and health have been investigated finding that there are strong relationships between some markers of immune function, chronic disease and survival [03.040; 03.041; 04.088]. Diabetes has been implicated as a risk factor for both dementia onset and cognitive decline (Biessels, 2005). We found no increase in risk of dementia onset with self-reported diabetes history [05.009]. At six-year follow-up interview a subset of the study population gave a blood sample and levels of HbA1c were measured. The analysis of these data (adjusting for study design) indicates that over all the range of values (not just clinical diabetes) individuals with higher HbA1c have higher death rates [05.009; 05.802].

The Organisation for Economic Co-operation and Development (OECD) commissioned a report on dementia care across nine member countries; the UK part was compiled by CFAS researchers. Information from CFAS, the Resource Implication Study (RIS) CFAS and Department of Health were combined to investigate the different types of dementia care seen across the UK [04.702].

Ethics and research governance:

CFAS has been at the forefront of ethics, particularly with the tissue resource. A recent investigation has been undertaken to investigate peoples perceptions of their own self, when is their body no longer part of themselves [03.064]? When it is tissue, when it is blood, when it is DNA? Individuals attitudes to blanket consent have also been investigated and it was found that individuals do not require constant re-enforcement of the details of the consent and would find this intrusive [05.006].

Benefits of Unit's role within the CFAS co-operative:

Population ageing, dementia and related disorders of ageing, remain amongst the most important issues to be faced by individuals, societies and governments over the next decades. All CFAS activity is aimed at contributing sound evidence in order to enable informed decision making for the population, now and in the future. We have made substantial progress on long-term archiving of medical data and creating web access, which will be developed further with new grants. This has been a consequence of the continuing role of the Unit within the Co-operative. This ongoing role has allowed statistical development of methods for complex studies [05.803; 05.068] and analyses which would not normally occur [06.035; 01.058]. Studies of CFAS-type design have not previously addressed these design and analytical issues that extend into the analysis of risk. CFAS has acted as a leader in the field of data archiving and collaboration with complex confidential information. The CFAS team act as expert advisors to other studies across the UK to pass on best practice.

Healthy active life expectancy and population burden of disease

Methods for investigation of healthy active life expectancy (HALE) have been based on cross-sectional prevalence rates of disease and application of these to population life expectancy tables; called Sullivan's method, this has been used extensively to estimate various life expectancy differences. We have employed it in CFAS for investigating differences in HALE by region [06.069], health [01.045] and gender [02.043]. Sullivan's method relies on population data for the groups of interest; however, many groups do not have specific life expectancy data available. These include both social class and education groups. We have used existing methods that combine Markov Chain methodology to calculate transition rates and multi-state life table approaches to calculate longitudinal healthy active life expectancies [05.107]. These methods are beginning to be used more readily and methodological developments within the Markov modelling field [03.044] can be incorporated into the longitudinal HALE field.

Longitudinal estimation in complex study designs

Methods proposed to enable incidence estimation of dementia in studies with two-phase two-wave designs (Clayton et al, 1998) were restricted to one specific study design. We have investigated methodology used by others within incidence of dementia estimation [05.803]. We found that many studies inadequately investigate the incidence of dementia within their study design settings and we have developed a more flexible simple method of estimating incidence of dementia that can be applied in most two-phase two-wave designs. In addition to relating screening information to the diagnostic information appropriately, our methods can model missing data using either missing at random (MAR) or informative missing mechanisms (NMAR) [05.803]. The effects of not adjusting for study design and informative missing data cause estimates of dementia incidence to be too low and too precise [05.803].

Ophthalmology

Over the years I have found the statistical challenges of measuring disease and health within the eye stimulating. The correlation between the two eyes in individuals, but the potential to get isolated diseases, creates challenges. Investigating ophthalmic epidemiology in the developing world has additional epidemiological challenges with problems of study design in the absence of population enumeration, work migration and isolation. I am the statistical supervisor of a PhD student who has been investigating trachoma and blindness in Southern Sudan [06.082; 06.083; 06.081].

Summary of major achievements

• CFAS neuropathology paper in the Lancet that is one of the most highly cited in the field;
• Widespread press interest and public engagement following publication of the first robust estimates of dementia incidence estimation in different regions in England and Wales;
• Production of estimates of disease, including cognitive change, that allow for wide usage including in the Department of Health projections of cost of care;
• Introduction of methodological improvements to the longitudinal healthy active life expectancy modelling mechanisms;
• Fiona Matthews is a contributing editor of the Lancet.

Publications from this programme

• Peer-reviewed publications
• Books, book chapters and reports
• Other non-peer-reviewed publications
• Theses

Peer-reviewed publications

• Publications - 2001
• Publications - 2002
• Publications - 2003
• Publications - 2004
• Publications - 2005
• Publications - 2006

Books, book chapters and reports

Other non-peer-reviewed publications

Theses