Within the confines of this longitudinal, community-based sample of elderly adults, we did not find strong evidence to support the hypothesis that mood disturbance was linked with the development of dementia. In the first analysis we identified CHS-CS participants in whom the scores on the 3MSE and the CESD were strongly negatively correlated. That is, among these participants when their CESD scores increased (indicating greater mood symptoms), their scores on the 3MSE at the same study visit decreased (indicating poorer cognitive functions). Our assumption was that in these participants the link between mood state and cognitive state was stronger than it was in other participants, and that this might indicate a greater vulnerability to dementia due to a decrease in cognitive reserve. However, there was no association between this correlation and subsequent cognitive decline. In the second analysis we defined mood disturbance based on the CESD score, and the persistence of any elevation during the five years prior to our observation period. Again, we did not find a significant link between persistently low mood state and the development of dementia. We thus conclude, as have others [9
], that while depression may be transiently association with poor cognition, it does not increase the rate at which people develop AD.
Our findings generally argue against the vulnerability hypothesis of depression in the elderly. Because there was a 3 year period between the last clinic contact in 1999 and the evaluation in 2002-03, we were expecting that the persistence of depression (from 1992 to1999) or a strong covariation between depression and cognitive functioning would have resulted in an increased rate of dementia; however, we did not find such an association. We did find, as we have in the past, that there is a group of cognitively normal individuals whose global cognitive functions (i.e., 3MSE scores) declined when depressive symptoms increased (or improved with decreased symptoms of depression), but this relationship did not convey any additional risk to develop dementia over time. To the extent that the such changes in global cognition may be severe enough to warrant a classification of Mild Cognitive Impairment (especially in the context of depression), this may explain why we find a link between depressive symptoms and Mild Cognitive Impairment in cross-sectional studies [41
], or in studies with short-term follow-up[43
Establishing that a personal characteristic such as depressed mood serves as a risk factor, rather than a correlate or outcome of cognitive impairment, requires that participants be free of cognitive impairment at the point(s) of risk factor assessment. When the normal/abnormal classification of dementia is made based on a cut-off of a global cognitive measure (e.g., [2
]), there is a risk of including in the “normal group”, individuals with early dementia. Even if the “miss rate” of a given two-stage screening protocol is known (e.g., [13
]), the fact remains that the group classified as cognitively normal likely includes individuals in the earliest stages of dementia. Thus, mood symptoms that accompany the development of memory loss and other symptoms of the impending dementia may be found in participants who, although classified as cognitively normal based on the screening tests, are actually in the earliest stages of clinical dementia. In that case, we would predict that the risk of dementia in such cases would actually decline
over time - once the early AD cases (who were considered normal) had developed dementia, there would be no additional increased rate of conversion. This is exactly what Andersen and colleagues found; the association between history of depression and incident AD after two years of follow-up (OR: 1.9, 95% CI: 1.0-3.3) was attenuated after 5 years follow up (OR: 1.6, 95%CI: 0.9-2.7) [44
In the CHS the mean
3MSE score of the AD cases was 88.1 [45
], which is well above the standard cut-off of normal cognition (i.e., greater than 80/100 [46
]). Had we used the standard criterion for determining normal (or “not impaired”) cognitive state, we would have under-estimated the true prevalence of impairment, but more important our “normal” group would have contained mild AD cases thus biasing studies of incident AD such as this one. Because we have historical information on the cognitive functions of the study participants, we can determine whether or not there is evidence of decline in cognitive functions - even if the summary measures remain within normal limits. shows that there was no decline in 3MSE scores across the years prior to the observation period as a function of whether the participants reported persistent symptoms of mood disturbance. Thus, we feel confident that our group of cognitively normal participants was as free from early clinical dementia as we could reasonably expect. We not only have a less biased group of cognitively normal participants at the beginning of the observation period, but we have good ascertainment of the dementia outcome at its earliest onset.
3) 3MSE Score over Time by Depression Group. Scores on the 3MSE (vertical axis) in the five years prior to the observation period (horizontal axis) as a function of the persistence of depressed mood.
There are several strengths of to this study. First, the original CHS participants were selected based on their inclusion on the Medicare lists, and not because of any mood or cognitive complaints or symptoms [24
]. Second, there was a wealth of data for the 10 years prior to the beginning of the observation period that were used to control for possible co-morbid factors that could affect the risk of dementia. Third, the ascertainment of AD was done using a neuropsychological assessment and neurological exam, coupled with interviews with informants. Fourth, the participants in the study had documented normal cognition at the beginning of the observation period, and for a period of time prior. (See ).
However, there are also limitations to these data that need to be kept in mind. First, we did not ascertain syndromal Mood Disorder using standardized interviews (e.g., SCID [47
], or MINI [48
]). Thus, while we have excellent visit-to-visit measurement of mood state, we do not have research-level diagnoses of any mood disorders. Second, the sample size is small by the standards of epidemiological studies (n ~ 300); however, with our sample size we were able to detect a HR of 3.2 with 80% power and a 2-sided alpha of 0.05. Said another way, assuming that HRs that we measured are valid indicators of the true HR, then we would have needed a sample of in excess of 4100 participants to detect them as statistically significant. Thus, we feel that we have adequate power to detect relevant effects, and that what we report is, in fact, a “null” finding. Third, the average age of the participants was 77 years, and our findings may be affected by a survivor bias. In order to be included in this study, the participants had to be cognitively normal in 1998/99. Thus, individuals who had developed dementia prior to 1998/99, or who had Mild Cognitive Impairment were excluded from this study. If depression were to exert effects to alter the risk of dementia, it would do so at an earlier age; those individuals who survived until their late 70's or early 80's without a cognitive impairment would be viewed as relatively immune to the additive effects of depression. This hypothesis is consistent with recent findings from the Rotterdam study [49
], although the nature of that two-stage screening protocol makes it difficult to draw firm conclusions.
One finding that bears discussion was our failure to find an association between incident dementia and level of education. Within the CHS study [50
] differences in incidence rates as a function of education level were significant only among white participants. Whites with less than high school education had an age-adjusted dementia incidence of 41.9 per 1,000 person-years, compared with 36.6 for high school graduates and 30.6 for individuals with at least some college education (p<.0001). Age-adjusted incidence of dementia for African Americans was 67.4 for those with less than high school, 55.8 for high school graduates, and 42.1 for college attendees (p=.79). Further, the age of this sample again raises the possibility that survivor bias may be affecting the results.
Geriatric depression is associated with an increased rate of medical comorbidities, and an increase in premature death [51
]. This explains why, in the earlier study from the CHS [12
] depression-dementia links were fully mediated by measurers of cardio-and cerebrovascular health (cf., [52
]). In the present study, there were no differences in the rates of medical problems between the study groups, which suggests the possibility that the sicker, depressed participants had either already developed dementia (as above) or had died. Thus, our failure to find a link between depression and incident dementia could be viewed as evidence that depression exerts its “effect” on cognition by having the unwanted fellow travelers of medical disease. The clinical ramification of this conclusion is that physicians caring for elderly patients with depression need to pay particular attention to caring for the conditions/diseases that affect the cardio- and cerebrovasculature. By doing so, they may reduce the medical burden of their patients and reduce the likelihood of their developing cognitive dysfunction.
One other important characteristic of the sample that bears emphasizing is that the rate of use of anti-depressant medications was low. Among the individuals classified as persistently depressed, only 2/27 (7.4%) were being treated. While this may mean that these individuals did not, in fact, meet criteria for Major Depressive Disorder, it is more likely the case that their mood state was not appreciated by the treating physician [51
These data emphasize the importance of attending to critical differences between studies conducted in referral research clinics and those conducted using community-based samples. In addition to questions related to the extent to which the samples accurately reflect the characteristics of the population of interest, the “portal of entry” into the study may be critical for the results of any studies of associations. For example, while participants in referral memory disorder clinics (e.g., Alzheimer's Disease Research Centers) by definition have complaints of cognitive impairment, fewer than 10% of the CHS-CS participants classified as demented had been previously diagnosed [13
]. Thus, the CHS-CS is identifying AD in its mildest (or earliest) manifestations. The same could hold true for studies of mood disturbance. Participants in typical research clinics are recruited based on the presence of diagnosed mood disorder, which means that either the patient or the family felt that the dysphoria was severe enough to warrant care at a specialty clinic (as opposed, for example, to treatment by the Primary Care Physician). By the same token, individuals with severe depressed mood may be the ones who are less likely to agree to enroll in a study such as the CHS-CS, or who will have the lowest rate of follow-up.
Taken in context, these data suggest that the link between mood disturbance and the development of dementia is not as strong as might be expected. In our view, the data that are needed to fully address this issue simply do not exist. In studies such as the CHS-CS that focused on cognitive outcomes, the assessment of diagnosable mood disorders is sub-optimal. Among those studies using research-level psychiatric diagnoses, the ascertainment of cognitive impairment is not as strong as it should be. Thus, there is a critical need for a single study that does a good job of assessing both types of outcomes. When coupled with a good evaluation of cardiovascular and cerebrovascular factors (including MRI scans of the brain), and when begun prior to the development of even the earliest symptoms of age-associated cognitive dysfunction (e.g., 45-55 years old at time of enrollment), that type of large scale study could provide a wealth of critical information regarding the earlier life events that affect the risk of developing clinical dementia.