Participants

Chinese participants were those from our Hefei cohort study in Anhui Province, China. The methods used in the study have been fully described elsewhere. ^{Reference Chen, Hu, Qin, Xu and Copeland14} Briefly, in 2001 we randomly selected 1810 people aged ⩾65 years from the residency committee lists who had lived for at least 5 years in Yiming district of Hefei city, Anhui. Permission for interview and informed consent were obtained from each perosn but if that was not possible, from the closest responsible adult. Refusals were respected. Ethical approval was obtained from Anhui Medical University and the district government.

There were 1736 people who agreed to participate in the study (a response rate of 95.9%). Participants were interviewed at home by a trained survey team from the School of Health Administration, Anhui Medical University. The main interview materials were the GMS – a comprehensive semi-structured mental state interview ^{Reference Copeland, Prince, Wilson, Dewey, Payne and Gurland17} and a general health record that included risk factors. ^{Reference Chen, Hu, Qin, Xu and Copeland14} Blood pressure and physical measurements were taken. The validation study of depression cases was carried out by two consultant psychiatrists. A year after the baseline interview, 1293 participants (74.5%) were successfully re-investigated, using the same protocol.

British participants were those from the MRC–ALPHA study, ^{Reference Wilson, Chen, Taylor, McCracken and Copeland10} part of the MRC Cognitive Function and Ageing Study (www.cfas.ac.uk). The study methods have been fully described elsewhere. ^{Reference Copeland, McCracken, Dewey, Wilson, Doran, Gilmore, Scott and Larkin18} In brief, a sample of 6035 older people aged ⩾65 were randomly selected using the Liverpool Family Practitioner Committee central computerised list of general practice patients in 1989. Of these, 5222 participants were interviewed with a response rate of 86.5%. The main interview materials included the GMS and the Minimum Data Set. ^{Reference Saunders, Copeland, Dewey, Davidson, McWilliam, Sharma and Sullivan19} Two years later, 3519 (67.4%) participants were successfully re-interviewed, and 4 years later 2238 (42.9%) participants were re-investigated. Ethical approval was obtained from the multicentre and local ethical committee.

Assessment of syndromes and cases of depression and dementia

A computer program assisted diagnosis – the Automated Geriatric Examination for Computer Assisted Taxonomy (AGECAT) – and was used to analyse the information from the GMS to identify the principal mental disorders in participants. ^{Reference Copeland, Prince, Wilson, Dewey, Payne and Gurland17,Reference Copeland, Dewey and Griffiths-Jones20} It was developed using a theoretical model and tested against its success at replicating diagnoses on samples diagnosed by psychiatrists. It attempts to replicate the process by which a psychiatrist achieves a syndromal diagnosis followed by a differential diagnosis. Geriatric Mental State symptoms are coalesced into a 150 ‘symptoms components’. In stage I the symptom components are brought together into groups which typify the major symptom areas of each diagnostic syndrome. The scores on these individual groups determine the final syndromal level of ‘confidence of diagnosis’. Thus, the system uses both quantitative and qualitative measures when allocating participants to the levels of confidence; required for its construction are many hundreds of clinical decisions on the placement of groups of symptom components on the syndrome levels. Individual participants are allocated to levels of confidence of diagnosis (0–5) on each of the eight diagnostic syndromes: organic disorder, depression, mania, schizophrenia and paranoid, obsessional, phobic, hypochondriacal, and general anxiety. In stage II the various syndrome levels are compared to derive a final differential diagnosis, a level of confidence of diagnosis from 0–5. A level of ⩾3 in most circumstances designates a ‘case level’ that has been shown to correspond with what psychiatrists usually recognise as a case for intervention. Levels 1 and 2 are designated as ‘sub-cases’, and level 0 (no confidence level on any syndrome) is classified as ‘well’. ^{Reference Copeland, Chen, Dewey, McCracken, Gilmore, Larkin and Wilson21} Although the levels are intended to be levels of confidence of diagnosis, it has been our observation that, in those conditions where diagnosis tends to be on a continuum with normality, for example in cognitive decline and depression, the different levels tend also to reflect levels of severity in terms of type and numbers of symptoms. Here we use ‘severity’ in terms of the symptoms themselves not in terms of their impact on daily living. In community-based samples the proportion of organic cases not resulting from dementia is small; acute confusional states have been estimated in about 0.5% of the community sample and the proportion of participants with depression sufficiently severe to be diagnosed as an organic state at case level also appear to be small. It has been the practice in community studies to ignore them.

A number of studies comparing AGECAT diagnoses for depression and dementia with psychiatrists' diagnoses using DSM–III and DSM–IV or ICD–10 criteria achieved a good level of agreement, ^{Reference Copeland, Dewey and Griffiths-Jones20} including in older Chinese people. ^{Reference Liu, Li, Zhang and Chen22,Reference Kua23} The AGECAT also divides depression into psychotic and neurotic types that appear to equate well with DSM–III ²⁴ /DSM–IV ²⁵ major affective disorder and dysthymia without the 2-year rule. ‘Type of depression’ refers to diagnostic cases which emerge after stage II where syndrome diagnoses have been converted into a single differential diagnosis. The symptoms defining depression and dementia are not similar. The GMS–AGECAT diagnosis has been the most widely used international community-based study method for investigations of depression and dementia in older people. ^{Reference Copeland, Prince, Wilson, Dewey, Payne and Gurland17}

Statistical analysis

Data were analysed for those who did not have dementia at baseline and were followed up (1254 Chinese and 3341 British). The associations of baseline depressive syndromes with subsequent dementia at follow-up were explored using a Cox regression model with adjustment for age, gender, educational level and CVD. In examining the risk of dementia in relation to depression, baseline sub-cases of dementia were further allowed in the multiple regression model for analysis. Differences in the relationship between subgroups of age, gender, ethnicity, CVD comorbidity and type of depression were tested. ^{Reference Altman and Bland26} All analyses were performed in SPSS version 13.0 for Windows.

Strengths and limitations

The strengths of the study were that we employed two markedly different populations in terms of levels of socio-economic status, social support, CVD and depression; the two cohorts had high response rates and the number of participants was relatively large, which gave sufficient power to test the relationship, including that in the subgroups; and the GMS–AGECAT, potentially measuring syndrome and diagnosis at 5 levels, offered a unique opportunity to examine dementia in relation to depressive severity (rather than simply as a dichotomous variable of depression).

The limitations of the study were that in the Chinese cohort, the incident dementia may be over-diagnosed because it included some illiterate participants ^{Reference Chen, Hu, Qin, Xu and Copeland14} (although the GMS–AGECAT was satisfactory for high-income countries with literate populations). ^{Reference Prince, Acosta, Chiu, Scazufca and Varghese29,Reference Prince, Acosta, Chiu, Copeland, Dewey, Scazufca and Varghese30} The follow-up in the Chinese cohort was relatively short (1 year), and the numbers with depression and dementia were small (and thus it was not possible to examine differences in the baseline covariables between people with depression levels 3 and 4, and differences in the effect of level 4 depression on dementia between the subgroups of participants). However, the findings were consistent and supported by the large British data with the long-term 4-year follow-up. Another limitation was that we could not separate dementia into Alzheimer's disease, vascular dementia, or both in order to examine their associations with depression. Thus, we could not directly link level 4 depression cases to Alzheimer's disease or vascular dementia. However, our data showed that depression at level 4 increased the risk of dementia in both participants with and without CVD, suggesting that the most severe cases of depression may be related to both vascular dementia and Alzheimer's disease. The finding that depression increased the risk of Alzheimer's disease has been shown in a recent meta-analysis paper, which did not examine the effects of the severity of depression on Alzheimer's disease. ^{Reference Ownby, Crocco, Acevedo, John and Loewenstein8}

Effects of depression on incident dementia

To our knowledge, our study is the first to examine the relationship between depressive syndromes and the risk of dementia, and report that within clinical depressions, the less severe cases of depression did not increase any risk for developing dementia, but the most severe cases did so significantly. The MRC–ALPHA data showed that individuals with level 4 depression at baseline were more likely to have level 4 depression within 2 years than those with baseline level 3 (data on request), suggesting depression lasted longer at level 4 (although we cannot exclude some participants who may have recovered and then relapsed in the meantime). This longer duration may be a reflection of the level of depression severity. Several epidemiological studies have examined depression as a risk factor for dementia with conflicting results. ^{Reference Andersen, Lolk, Kragh-Sorensen, Petersen and Green4} In a meta-analysis, Jorm concluded that a history of depression increased the risk for developing dementia by 87%. ^{Reference Jorm3} Our data suggest that the most severe level of depression increased the incidence of dementia to a greater degree than those in previous studies. ^{Reference Jorm3,Reference Ownby, Crocco, Acevedo, John and Loewenstein8} We believe this is because we separated severe from less severe depression. This may explain the variable conclusions in those studies which grouped severity levels of depression together. Failure to confirm this finding may have been partly due to a lower proportion of severe depression in their samples.

Why does the GMS–AGECAT level 4, not case level 3 depression appear to increase the risk of incident dementia? Such a pattern even existed in our sensitivity analyses. It seems not to arise from differences in those covariables at baseline. It could not be entirely explained by the lower proportion of depressive neurosis (or higher depressive psychosis) in participants with level 3 depression at baseline compared with those with level 4, since level 4 depressive psychosis itself increased the risk of dementia. We considered that some symptoms or other unmeasured variables might have been included in level 4, but not in level 3, depression and/or simply the severity of depression could play a role. Perturbations of the hypothalamic–pituitary–adrenal axis in depression can lead to hypercortisolemia-associated hippocampal atrophy which may also explain why the most severe cases of depression are those more likely to increase the risk of dementia. Nevertheless, the GMS–AGECAT diagnosis, adapted for clinical use, could be used to help identify those at higher risk in older populations in an attempt to prevent dementia. The strong effect found in younger elderly people and the long-term effect may imply that depression at a younger age is more important for developing dementia, suggesting that early treatment of depression and prevention of depression in the younger population may effectively reduce dementia in later life.

The increased risk of incident dementia in relation to only the most severe syndromes or cases of depression is unlikely to result from chance or bias. Our study urges that the aetiological roles of the severity of depressive symptoms, syndromes and clinical cases on dementia need to be addressed. ^{Reference Chen, Hu, Wei, Qin and Copeland31} Early treatment and prevention of depression and relieving its severity could reduce the burden of dementia in both high-income and low- and middle-income countries.