Table 1 presents descriptive statistics on study variables at the index visit by cognition status at the index visit. Relative to subjects with normal cognition, those with dementia or MCI were more likely to be women, were somewhat older, had somewhat less educational attainment, and were less likely to live alone. As expected, there were strong differences between groups in their functional independence, with dementia patients much more likely to need some assistance, and virtually all normal cognition patients falling into the fully independent category. Patients with MCI were generally independent. On physical health background characteristics, differences among groups were not as dramatic, with relatively similar prevalence of cardiovascular disease and diabetes. Finally, at the index visit, patients with dementia or MCI were more than twice as likely to have a clinician-rated diagnosis of depression, as well as higher levels of current depressive symptomatology.

Table 2 contains results from logistic and negative binomial regressions for clinician-rated depression and depressive symptoms, respectively. Controlling for basic demographic characteristics, there is a strong association of both dementia and MCI with clinician-rated depression and depressive symptoms as reported by the GDS. Although the association is somewhat attenuated after including additional covariates, results still suggest a strong, contemporaneous relationship between cognition status and depression, with both dementia and MCI patients approximately 2.5 times more likely to have depression. We conducted an additional analysis of persons with a clinical diagnosis of depression at baseline only to examine whether cognition status is associated with increased depression severity as measured by the GDS. We found that, relative to normal controls, both MCI (relative risk [RR]: 1.34, 95% confidence interval [CI]: 1.27, 1.40) and dementia (RR: 1.14, 95% CI: 1.08, 1.21) report significantly more depressive symptoms.

The preceding analyses suggest a strong association between concurrent cognitive impairment and depression over time; however, the descriptive data in Table 1 suggested that dementia and MCI patients were more likely to have depression at the index visit. Thus, a final analysis examined a prospective association, by selecting the subset of individuals who did not have clinician-rated depression at the index visit (normal cognition: N = 8,352; dementia: N = 5,880; MCI: N = 4,610). Analysis of clinician-rated depression included all previous covariates and as well as interaction terms between time since index visit and cognition status. Because there was evidence of nonlinear trends, a quadratic term for time was included.

All time and cognition status × time interaction terms were significant (all p <0.001). Due to the interactions and nonlinear terms for time, predicted regression lines are presented in Figure 1 (with other covariates set to reference values). By definition, all patients in this subsample begin without depression, but there are fairly dramatic differences in proportion of patients with depression by cognition status that begin within the first year or two of follow-up and become more dramatic over time. At 2 years post index visit, 10% of patients with normal cognition are predicted to have developed depression, whereas 22% of MCI patients and 25% of dementia patients have developed depression. By 4 years post index visit, these have increased to 20% (normal cognition patients), 38% (MCI patients), and 43% (dementia patients). A similar set of analyses focused on depression symptoms as measured by the GDS-SF did not reveal a similar pattern; instead, there was a notable main effect of higher depression symptoms in MCI and dementia relative to normal cognition, but depression severity did not increase over time.

Results thus far suggest a strong association over time, as well as a prospective association between cognition status and future depression. Two additional confounds were considered in sensitivity analyses, however. First, there could be differential use of ADM, where underprescribing of ADM to MCI and dementia patients may explain their higher depression rates. This does not appear to be the case, however, as dementia patients are the most likely to report ADM use in subpopulations rated as being depressed (M_Dementia = .69, M_MCI = .59, M_{Normal Cognition} .59) and among those rated as not being depressed (M_Dementia = .23, M_MCI = .10, M_{Normal Cognition} .07). Regression analyses controlling for ADM yielded substantively identical conclusions to those presented earlier. Finally, the proportion of dementia patients who died during the study (3,082 of 10,486, 29.4%) was higher than the proportion for MCI (756 of 7,096, 10.7%) or normal cognition patients (740 of 10,194, 7.3%). Regression analyses stratified by death status revealed that the association of depression and cognition status was very similar within patients who died during study follow-up and those who did not. The only notable difference was that patients with normal cognition who died at some point during follow-up were somewhat more likely to have become depressed than those normal cognition patients who did not die during follow-up (see Fig. 2).

The results of this large and well-characterized sample, which includes patients with normal cognition, MCI, and dementia, present robust evidence that cognitive impairment is associated with significantly higher rates of depression. Prior research found no association of cognitive impairment with severe depression in a smaller sample with less–well-characterized dementia status,²⁵ and some work using only the Mini-Mental State Exam (MMSE) and GDS found increased odds of depression in those with MMSE scores below 24.²⁶ Although results are strongest using the clinician-rated depression variable, our finding is supported by significantly higher current depression on the GDS-SF. The finding of increased development of depression in subjects without depression at index suggests that the increased risk of depression in cognitively impaired subjects is not entirely due to continued presence of previous depression. The results of analyses of subjects with clinical diagnosis of depression at the index visit also suggest that the severity of depression over time is worse in subjects with MCI or dementia. Antidepressant use does not confound these results.

The results suggest that from a public health standpoint, it is important to consider depression not only as a predictor of dementia as established in prior research, but as a comorbid condition that will develop during the course of dementia—because the results demonstrate significantly higher rates of depression in cognitively impaired subjects who were not depressed at intake. Thus, community-based interventions for treatment of dementia will need to be prepared to address depression²⁷ and may need to include high levels of collaboration between nurse managers and physicians in order to significantly improve depressive symtpoms.²⁸ Zlatar and colleagues have made a similar recommendation around the importance of considering depression when evaluating subjects with subjective complaints of cognitive impairment who demonstrate normal cognitive testing.²⁹ Conversely, evidence-based approaches for treatment of depression in older adults will need to accommodate treatment of subjects with dementia-severity cognitive impairment. Currently, most community- and clinic-based geriatric depression effectiveness trials exclude subjects with significant cognitive impairment. Also, other factors such as race/ethnicity, education, income, language, acculturation, and family and community support need to be considered when examining the recognition and treatment of depression in diverse elders with cognitive impairment.³⁰

The continued association of depression and cognitive impairment after the development of MCI or dementia supports a possible common patho-physiological pathway of the conditions. Stress theory models suggest that exposure to significant stress increases corticotropin releasing factor and that impaired hypothalamic-pituitary-adrenal axis inhibitory feedback leads to elevated glucocorticoids, which are associated with smaller hippocampal volumes, depressive features, and dementia.^31,32 Similarly, research on inflammation models of the illnesses find increased inflammatory markers of interleukin and c-reactive proteins in depression and a role for inflammatory disease processes in Alzheimer and vascular dementia pathophysiology.³³ Heterogeneity of types of depression, combined with increased recognition of heterogeneity of types of Alzheimer dementia as well as other dementia, help explain the lack of consistent findings across studies of pathophysiology or treatment trials.^33,34

This complicates translation of effective treatments to these disease models. Several papers have looked at the prevalence and/or incidence of depression in persons with cognitive impairment. Lyketsos and colleagues³⁵ examined the prevalence of depression in persons with dementia and with mild cognitive impairment (N = 682). Depression was defined using the Neuropsychiatric Inventory³⁶ and 32% and 20% of persons with dementia or MCI, respectively, reported depressive symptoms. These rates are higher than our prevalence of depressive symptoms (as defined by the GDS-SF) of 17% and 14% for persons with dementia or MCI, and lower than our reported depression prevalence using a clinical diagnosis (43% for dementia and 35% for MCI). Our study differs from the Lyketsos paper as it includes a significantly larger sample (N = 27,776) of both persons with and without cognitive¼impairment. We also include information drawn from clinical depression diagnoses in addition to a depression scale instrument and are able to report rates of the development (incidence) of depression in those without depression at intake. Because the GDS scoring is responsive to change over time, we are able to analyze change in depression severity over time as well.

Peters³⁷ reported prevalence of depressive symptoms data for persons with dementia; with cognitive impairment, no dementia (CIND); and with normal cognition using data from the Cache County Study (N = 5,092). Their reported prevalence of depressive symptoms was comparable to ours for persons with normal cognitive function (4.9% in their study versus 5.0% for our study) and for persons with CIND/MCI (16.9% versus 14.0%), and were higher for persons with dementia (29.9% versus 17%). As with the Lyketsos paper, this study was cross-sectional and used the Neuropsychiatric Inventory and therefore was not able to comment on incident depression in persons with and without cognitive impairment, or on the change in depression severity over time. Lastly, Panza and colleagues³⁸ conducted a review of studies that compared the prevalence and incidence of depression for persons with MCI only. Most of the reported studies were cross-sectional and/or had a small sample size, and, as defined by their review criteria, none of the papers included persons with normal cognitive function or with dementia. They found a range of prevalence rates for depressive symptoms (4%–69%) largely due to the inconsistencies in how depression is assessed and MCI is defined. Other recent papers such as Reinlieb and colleagues³⁹ found elevated rates of depressive symptoms for persons with MCI as well compared with those without cognitive impairment. Two papers included in Panza et al.'s review³⁸ described incident depression, reporting rates of 29.6/100 person-years for the one population-based study and 11.7/100 person years for the one hospital-based study. We conducted a post hoc analysis to estimate our incident rates of depression per 100 person-years and found rates of 4.4/100 person-years for persons with normal cognition, 8.3/100 person-years for persons with MCI, and 11.1/100 person-years for persons with dementia, suggesting lower rates than the two studies described in the Panza review for persons with MCI. It should be noted that our definition of depression/depressive symptoms and our control group are different than those reported in the review by Panza and colleagues.

There are several limitations to this study. Although the GDS-SF is a valid, established, screening instrument, it is not a diagnostic instrument and so the GDSSF findings here should not be interpreted as specific to major depression. We do, however, include information about depression as defined by a clinical diagnosis as well and indicate where the data are consistent and where they differ. The clinical diagnosis was made using a standardized comprehensive history that included other items that may contribute to depressive symptoms, such as past history of bipolar disorder; medications such as beta-blockers for congestive heart failure or hypertension; and anemia, hypothyroidism, or other medical illnesses. In addition, we adjusted for many of these possible cofounders in our analysis. Dosages and durations of ADM were not available in the UDS and so it is uncertain whether differences in the adequacy of treatment contributed to increased rates of depression in patients with cognitive impairment. Further research is warranted to study the impact of appropriately prescribed and managed ADM in persons with and without cognitive impairment.

Despite these limitations, these results consistently show increased rates of depression in patients with dementia and mild cognitive impairment at cross-sectional intake evaluations and over longitudinal follow-ups, including analysis of incident depression. The results suggest that beyond being a predictor or prodromal condition, depression is an important comorbid chronic condition in patients with dementia and mild cognitive impairment.