Objectives  To determine the characteristics and determinants of neuropsychological functioning LLD.

Design  Cross-sectional study of groups of LLD patients and control subjects.

Setting  Outpatient, university-based depression research clinic.

Participants  One hundred patients without dementia 60 years and older who met DSM-IV criteria for current episode of unipolar major depression (nonpsychotic) and 40 nondepressed, age- and education-equated control subjects.

Main Outcome Measures  A comprehensive neuropsychological battery.

Results  Relative to control subjects, LLD patients performed poorer in all cognitive domains. More than half exhibited significant impairment (performance below the 10th percentile of the control group). Information processing speed and visuospatial and executive abilities were the most broadly and frequently impaired. The neuropsychological impairments were mediated almost entirely by slowed information processing ( = .45-.80). Education ( = .32) and ventricular atrophy ( = .28) made additional modest contributions to variance in measures of language ability. Medical and vascular disease burden, apolipoprotein E genotype, and serum anticholinergicity did not contribute to variance in any cognitive domain.

Conclusions  Late-life depression is characterized by slowed information processing, which affects all realms of cognition. This supports the concept that frontostriatal dysfunction plays a key role in LLD. The putative role of some risk factors was validated (eg, advanced age, low education, depression severity), whereas others were not (eg, medical burden, age at onset of first depressive episode). Further studies of neuropsychological functioning in remitted LLD patients are needed to parse episode-related and persistent factors and to relate them to underlying neural dysfunction.

Cognitive impairment in LLD is substantial and disabling. The cognitive response to antidepressant treatment is variable, and impairments persist even after effective treatment of depression.^2-3 There is a substantially increased risk of developing a progressive dementia during the 2 to 4 years following a depressive episode.^4-5

Of the studies describing cognitive functioning in LLD, only a few^6-9 have included both a comprehensive assessment of cognitive domains and a healthy control group. Although the findings across these studies differ somewhat, they suggest that impairments exist in visuospatial ability, memory, speed of information processing, and executive functioning, with the executive deficits being particularly related to late age at onset of first lifetime depressive episode (late-onset).^9-10 Factors that are associated with neuropsychological impairment in LLD patients include more severe depressive symptoms,^{6, 11-12} more severe anxiety,¹³ and/or vegetative symptoms.¹⁴ White matter hyperintensities are associated with psychomotor slowing¹⁵ and executive function impairment, particularly in LLD patients with late-onset depression.^{9, 16} In fact, 2 influential models of the cognitive effects of depression emphasize the role of subcortical-frontal lobe circuit dysfunction,^17-19 which would account for particular deficits in information processing speed and executive functions.

A variety of other risk factors may play a role in the likelihood that LLD patients will exhibit cognitive impairment. For example, older age may be associated with cognitive impairment in LLD,^20-22 although this is not always the case.^23-24 Some studies^25-28 have found no relationship between late-onset depression and greater risk for cognitive impairment during the index episode, although several studies^{16, 29-31} have found that they are related. Moreover, even if impairment were initially reversed with successful treatment of depression, there may be an increased risk for subsequently developing progressive dementia.⁴ Several recent studies^{5, 32-35} have found an association between late-onset depression and Alzheimer disease. Some studies,^36-37 but not all,^{22, 38-39} have found an increase in cognitive dysfunction in LLD with increasing medical burden. The apolipoprotein E (APOE) 4 allele has been found to be related to poorer cognitive functioning in geriatric depression patients in one⁴⁰ but not all studies.^41-43 Finally, there is evidence suggesting that serum anticholinergic burden negatively affects cognitive functioning, especially memory ability, in LLD.^44-45 Many of these non–depression-related factors also place so-called normal, elderly patients (ie, nonpsychiatric groups) at risk for cognitive dysfunction^46-48 and even for dementia.⁴⁹

To our knowledge, no single study has provided a comprehensive analysis that includes predictors and outcomes related to neuropsychological functioning in LLD. That is, although some studies have evaluated a single cognitive domain or a single risk factor, none have examined both risk factors and cognitive functioning in a comprehensive manner. Moreover, we were particularly interested in examining the relationship among neuropsychological domains to determine whether there might be a parsimonious mediating cognitive factor. Specifically, we hypothesized that information processing speed, which plays a critical role in the effects of normal cognitive aging⁵⁰ and depression-related alterations in cognition,⁵¹ would mediate the association between biological and clinical risk factors and other cognitive domains.

METHODS

We enrolled 140 patients 60 years and older with current unipolar major depression (nonpsychotic). All patients had sought treatment at the Western Psychiatric Institute and Clinic (Pittsburgh, Pa) and were enrolled in federally sponsored intervention studies conducted within the Intervention Research Center for Late-Life Mood Disorders at the University of Pittsburgh School of Medicine. Forty elderly control subjects with no psychiatric history were recruited using local advertisements. The data for the present study were collected during participants' baseline, pretreatment evaluation. Details on subject recruitment, retention, and evaluation have been described elsewhere.^{2, 52} Diagnosis was established by a Structured Clinical Interview for Axis I DSM-IV Disorders (SCID-IV)⁵³ interview administered by formally trained master's-level and doctoral-level clinicians and a consensus diagnostic conference attended by the raters and at least 3 research geriatric psychiatrists. Patients and control subjects were excluded if they had psychotic symptoms or major unstable medical illnesses (eg, metastatic cancer). However, chronic diseases (eg, diabetes mellitus, hypertension) did not constitute exclusion criteria if the subject was medically stable. To study a broad, representative sample of those with LLD, ethnicity, insurance status, concurrent medical and psychiatric conditions, and symptom profile did not constitute exclusion criteria. Additional exclusionary criteria for the present study included the following: neurologic disorders or injuries known to have significant direct effects on cognitive functioning (eg, traumatic brain injury, multiple sclerosis) (n = 10), uncorrectable sensory handicap (eg, blindness) (n = 2), a diagnosis of any type of dementia from the University of Pittsburgh's Alzheimer's Disease Research Center either before or following resolution of the index depressive episode (n = 19), or a best (including pretreatment and posttreatment) Dementia Rating Scale (DRS)^54-55 score of 129 or lower (n = 9). A DRS score of 129 is 3 SDs below the mean of our age- and education-equated control group. These exclusionary criteria reduced the original patient study group of 140 to 100 (composed of 4 inpatients and 96 outpatients). After complete description of whichever parent treatment study in which subjects were participating, written informed consent approved by the University of Pittsburgh's Institutional Review Board was obtained.

MEASURES

Subjects had a broad-based pretreatment assessment that included clinical, psychosocial, biologic, and neuropsychological measures. (The neuropsychological measures are listed in Table 2.) Many of the nonneuropsychological measures assessed potential risk factors for cognitive impairment either in general or in LLD in particular. These measures, grouped by domain, included demographics (age, education, sex), depression-related factors (score on the 17-item Hamilton Depression Rating Scale [HDRS],⁵⁶ age at onset of first lifetime depressive episode), medical burden (score on Cumulative Illness Rating Scale–Geriatrics [CIRS-G]),⁵⁷ vascular burden (CIRS-G combined heart and vascular scale scores, APOE allele type,^40-43 serum anticholinergic burden⁵⁸), and structural brain abnormalities (cortical and ventricular atrophy and white matter hyperintensity burden⁵⁹).

View this table: [in this window] [in a new window] Table 2. Results of t Tests Comparing Elderly Control Subjects' and LLD Patients' z Score Performance on Cognitive Domains and Individual Neuropsychological Tests*

Structural magnetic resonance (MR) imaging was conducted within 2 to 3 weeks of the baseline evaluation. Each subject's imaging data were read independently by 2 raters (including E.M.W. and C.C.M.) based on the Cardiovascular Health Study protocol.⁵⁹ The variables assessed were ventricular atrophy, sulcal atrophy, and the total white matter hyperintensity burden on the T1-weighted, T2-weighted, and proton density images. As per the Cardiovascular Health Study protocol, each variable was assigned a numerical rating by comparing each subject's imaging data to predefined visual standards that represent progressive severity on a 10-point scale (0 through 9). If the 2 independent raters differed in their rating by 1 point, the final rating was the mean of the 2 values. A greater than 1-point difference between raters was considered a major disagreement and was adjudicated by consensus. Four raters were involved in assessing the MR imaging data; these raters achieved intraclass correlation coefficients of 0.71 for ventricular atrophy, 0.44 for sulcal atrophy, and 0.66 for white matter hyperintensities.

Age at onset was ascertained from multiple sources, including the SCID-IV, all available medical and psychiatric records, and caregiver interviews. Comprehensive neuropsychological assessment was performed by 2 examiners who were highly trained and closely supervised by the first author (M.A.B.). In most cases, patients were tested before antidepressant treatment was initiated. All of the control subjects and 86 of the 100 depressed patients were psychotropic drug free. Seven patients were taking low doses of nortriptyline hydrochloride, which had been initiated 1 to 3 days before the assessment. Six patients were taking paroxetine at the time of the assessment; 4 had their initial dose within 3 days and 2 had been taking paroxetine for many months and were in the process of tapering off the medication. One patient had been taking mirtazapine for several weeks before testing. Importantly, all patients met clinical research criteria (ie, DSM-IV) for current major depression.

PROCEDURES AND STATISTICAL ANALYSIS

Descriptive Analyses

We compared depressed patients and control subjects on each of the risk factors, using either t tests or nonparametric tests, when appropriate.

Neuropsychological Functioning in LLD

To characterize neuropsychological functioning in LLD, we first combined the measures within each of 6 domains to yield a single score that reflected distinct areas of cognition (attention; information processing speed; and visuospatial, executive, language, and memory ability). Initially, we transformed raw scores for individual variables into z scores, using the distribution of the elderly control sample,^{6, 60-61} and these z scores were then averaged within each neuropsychological area to produce domain scores. Internal consistency (Cronbach ) exceeded .70 for all but 2 domains. The was .67 for the memory domain and .30 for the attention domain, leading us to exclude the attention domain from further analyses.

We next performed a series of analyses comparing patient and control subjects' domain scores to characterize cognitive functioning in LLD. The descriptive analyses revealed that the patients' CIRS-G scores were significantly higher than controls. After verifying that there were no interactions between group and CIRS-G total score, we used multivariate analysis of covariance (MANCOVA) to model the main effect of group on the 5 cognitive domains controlling for CIRS-G scores. We performed t tests on each domain and on each of the individual measures (with Bonferroni-adjusted P values) to determine which measures accounted for differences between the groups.

To further delineate the cognitive patterns exhibited by both the depressed patients and control subjects, the number of domains in which each subject was impaired (defined as performance below the 10th percentile of the control group) was determined. The percentage of control subjects and LLD patients whose performance fell below the 10th percentile of the control group for each domain also was calculated.

Age at onset of first depressive episode has frequently been found to be an important factor in many aspects of LLD (eg, related to cognitive functioning, structural anatomy, response to pharmacotherapy). Therefore, based on convention, we divided our sample into those with early-onset depression (onset at <60 years of age, n = 43) and late-onset depression (onset at 60 years of age, n = 57) and performed t tests or ² analyses (where appropriate) comparing them on demographic, clinical, and biologic variables. MANCOVA (2 groups for 5 cognitive domains) comparing the groups' performance on the cognitive domains while controlling for age and serum anticholinergic burden was also performed.

Risk Factors for Cognitive Dysfunction

We examined the relationship among the domain scores and several potential risk factors for cognitive dysfunction in LLD patients. Univariate Pearson correlations among each of the potential risk factors and each of the 5 cognitive domains for the LLD group were calculated. We then performed a series of regression analyses using only those risk factors that had significant Pearson correlations (P<.05, r = 0.2).

RESULTS

There were no significant differences between the depressed patients and control subjects on age, sex, education, CIRS-G combined heart and vascular disease scores, frequency of APOE4, level of serum anticholinergic burden, or structural MR imaging measures of sulcal or ventricular atrophy (Table 1). Depressed patients reported more medical problems (CIRS-G) and also demonstrated higher burden of white matter hyperintensities (MR imaging) and overall poorer cognitive functioning (Mini-Mental State Examination⁶² and the DRS).

View this table: [in this window] [in a new window] Table 1. Characteristics of the Study Participants

NEUROPSYCHOLOGICAL FUNCTIONING IN LLD

MANCOVA revealed that CIRS-G was not a significant covariate (F_5,127 = 0.92, P = .47) and that depressed patients performed significantly worse than control subjects (F_5,127 = 2.80, P = .02). Two-sample t tests revealed that the controls and LLD patients differed in all 5 domains (Table 2). The information processing speed domain was the only one in which subjects differed significantly on all 3 component measures. Subjects differed on 2 of 3 tasks in the visuospatial domain, 2 of 4 tasks on the executive domain, only 1 task in the language domain, and no tasks in the memory domain. Table 2 presents the z score transformed data for the patients. By definition, the control group's mean z score is always 0. Depressed patients' and control subjects' raw scores on individual measures are presented in Table 3.

View this table: [in this window] [in a new window] Table 3. Comparison of Elderly Subjects' and LLD Patients' Raw Score Performance on Individual Neuropsychological Tests*

Table 4 shows the patterns of cognitive deficits exhibited by both depressed patients and control subjects. Only 39% of depressed patients performed within normal limits (greater than the 10th percentile of the control group in all domains). The overall distribution of cognitive performance differed between depressed patients and control subjects (Fisher exact P = .01); the depressed patients were impaired in more domains than were the control subjects (Wilcoxon exact P<.001). Figure 1 depicts the percentage of subjects whose performance was impaired on each cognitive domain. Information processing speed was the most frequently impaired domain, followed by the visuospatial, memory, executive, and language domains. More depressed patients than control subjects were impaired in all domains (Fisher exact P values: visuospatial, P = .01; memory, P = .01; executive, P = .02; information processing speed, P<.001) except for language (Fisher exact P = .10).

View this table: [in this window] [in a new window] Table 4. Percentage of Participants Exhibiting Cognitive Deficits (Defined as 10th Percentile Below Comparison Group)

View larger version (34K): [in this window] [in a new window] Figure 1. Performance on cognitive domains by patients with late-life depression. Impairment is defined as performance below the 10th percentile of the control subjects (see reference line).

Comparison of early- and late-onset LLD patients revealed that the late-onset subgroup was significantly older at the time of neuropsychological evaluation (t₉₈ = –4.52, P<.001), had a higher percentage of males (²₁ = 5.42, P = .02), and had higher levels of serum antiholinergic burden (²₁ = 4.05, P = .04). MANCOVA comparing the groups' performance on the cognitive domains detected no significant differences between the groups.

RISK FACTORS FOR COGNITIVE DYSFUNCTION

We examined univariate correlations and identified those with P<.05. Age and education were significantly associated with nearly all cognitive domains. The HDRS total score, number of medications, and age at onset of first depressive episode were each significantly correlated with 1 or 2 cognitive domains. Each of the 3 MR imaging variables was significantly correlated with 2 to 4 cognitive domains. Of note, each of the risk factors noted thus far was significantly correlated with information processing speed. Sex, CIRS-G total score, CIRS-G vascular scale, APOE allele type, and serum anticholinergic burden were not significantly correlated with any of the cognitive domains.

To understand associations among the factors that affect cognitive function in depressed patients without dementia, we used multiple regression analyses. Consistent with our initial hypothesis, the correlations between information processing speed and the various risk factor variables (and other neuropsychological domains) were uniformly high and significant. This observation supported our hypothesis that information processing speed is a mediating factor in the cognitive functioning of the elderly, depressed patients and is consistent with current models of cognitive aging.⁵⁰ To examine these relationships further, we regressed the language, visuospatial, memory, and executive domain scores on information processing speed and those risk factors that were significantly correlated (P<.05, r = .20) with the domain scores (ie, age, education, HDRS score, age at onset of first depressive episode, and all 3 MR imaging measures). All of the predictor variables were entered simultaneously, and only those that significantly increased the explained variance (after controlling for all other variables) were retained. With one exception, information processing speed was the sole significant predictor of each of the other neuropsychological domain scores; this is represented in Figure 2 by paths from speed to each of the other domains (with the associated weights). We then regressed the speed domain score on age, education, and HDRS score, and each had a significant, independent association with speed. Education was the only predictor variable that had an independent association with one of the domain scores after accounting for the association with speed. Thus, the language domain score, but not the other domain variables, was significantly associated with both speed and level of education.

View larger version (53K): [in this window] [in a new window] Figure 2. Relationships among cognitive domains and predictor variables. Thin arrows represent significant results from multiple regression analyses performed on the depressed subsample that underwent magnetic resonance imaging (n = 58). Bold arrows represent significant results from multiple regression analyses performed on both the subsample and the full sample of depressed patients. HDRS indicates Hamilton Depression Rating Scale.

When we repeated the analysis using the MR imaging variables from the patients who had undergone scanning, the path weights were virtually identical. We first regressed the domain scores on speed and ventricular atrophy (and the other predictor variables), and the extent of ventricular enlargement was significantly associated only with the language domain. When the MR imaging variables were regressed on the other predictors, only age was associated with the degree of ventricular enlargement. Age did not independently predict any domain scores, acting only through the mediation of ventricular atrophy and speed.

COMMENT

These findings on neuropsychological functioning in LLD are in general agreement with the existing literature. Both Boone et al⁶ and Lesser et al⁹ found that, compared with control subjects, mixed-age groups of mildly to moderately depressed patients were impaired in nonverbal intelligence, visual memory, and executive ability. The depressed patients in the study by Lesser et al were also impaired in constructional ability and information processing speed, as were the most depressed patients in the sample studied by Boone et al. In a group of 10 elderly, depressed patients, Hart et al⁷ found deficits compared with control subjects in measures of verbal and visual memory, construction, executive ability, and information processing speed. Finally, Kramer-Ginsberg et al⁸ found that compared with a control group, a group of 41 elderly, depressed patients exhibited deficits in memory, visuospatial ability, and information processing speed.

In the present study, relative to control subjects, LLD patients showed impairment in all 5 of the domains measured along with a strikingly wide range of overall cognitive ability. Some patients exhibited virtually no measurable impairment, whereas others were impaired across multiple cognitive domains. In comparison with earlier studies, the relatively large size of our depressed group and their relatively advanced age and severe depressive symptoms enabled us to better characterize the cognitive heterogeneity of the disorder.

It is striking that the broad array of neuropsychological deficits was explained almost entirely by a fundamental deficit in information processing speed (Figure 2). Other combinations of predictor variables (eg, with executive function as a mediator) did not result in such parsimonious relationships among predictors and outcomes. Thus, factors slowing information processing speed affect a range of cognitive functions in LLD patients.

The finding that cognitive dysfunction is largely accounted for by information processing speed is in accord with our previous work in geriatric depression.⁵¹ Thirty-eight of the 100 depressed patients in the present study also participated in the study by Nebes et al⁵¹; however, the overlap in measures was limited to 2 tests (Block Design and Digit Symbol subtests of the Wechsler Adult Intelligence Scale III). The present results also fit within the framework proposed by Salthouse⁵⁰ that the cognitive effects of normal aging are almost entirely accounted for by slowing of information processing. It is also noteworthy that prior work from our group³ has shown that this slowing of information processing persists even after depressed patients have responded to antidepressant treatment. Although there was some improvement in the performance of the depressed patients during treatment, the magnitude of this improvement was no greater than that produced in older control subjects by practice alone. Thus, the cognitive slowing that appears to be central to neuropsychological impairment of our depressed patients may well be a trait feature of geriatric depression.

Examination of a broad array of potential risk factors revealed that a few exerted their influence on cognitive function by altering speed of information processing. Nearly all of the effects that the various risk factors had on language, memory, visuospatial, and executive ability were accounted for by alterations in information processing speed. The strength and direction of the independent associations among age, education, and depression symptoms, and processing speed are all straightforward. Patients who were older or who had more severe depressive symptoms were cognitively slower. By contrast, education was a protective factor—greater educational achievement was associated with faster information processing speed. Education level was also independently associated with the language domain score, which is heavily dependent on semantic knowledge.

Some risk factors had no effect on cognitive performance in the present analyses. We found, as have others,^{22, 38-39} no association between overall medical burden and cognitive ability. Therefore, the source of the cognitive dysfunction associated with LLD seems not to be related to the excess medical burden that characterizes many elderly, depressed individuals. Perhaps most notably, in our study group, we detected no effect of APOE allele type on the variance in cognitive ability. Several investigators,^42-43 including ourselves,⁴¹ have found that the presence of the APOE4 allele does not influence overall cognitive functioning in the context of LLD. The present study extends this work, particularly by specifying that the APOE4 allele does not influence any particular cognitive domain, especially memory functioning, in LLD. This point is important because impairment in memory functioning is the hallmark of early Alzheimer disease. The lack of influence on memory ability suggests that LLD patients' increased risk of Alzheimer disease is mediated by some other, non–APOE-related factor.

Our group has previously found that even minimal serum anticholinergic activity has a negative effect on cognitive ability in both a community sample⁴⁴ and older, depressed patients without dementia.⁴⁵ Mulsant et al⁴⁴ studied a community sample that included individuals with dementia, whereas the present study excluded patients with even mild dementia on an a priori basis. It is possible that patients with preclinical or mild Alzheimer disease are the most vulnerable to deleterious anticholinergic effects, which would explain why Mulsant et al detected the relationship, whereas this study did not. The present study group was larger, but otherwise its subjects had similar characteristics to those studied by Nebes et al.⁴⁵ Nonetheless, since the study by Nebes et al⁴⁵ was conducted 5 years earlier in the same research clinic as the present study, it is possible that clinicians became more vigilant in reducing anticholinergic burden in patients identified clinically as being at risk for cognitive impairment.

In a secondary analysis, we examined how the MR image–derived variables affected cognitive outcomes. Although fewer subjects were entered into the multiple regression equations, weights associated with the patterns of associations between non–MR image variables did not change substantially; neither cortical atrophy nor white matter hyperintensity scores were associated with any outcomes, whereas ventricular atrophy was linked to information processing speed and language domain scores. This finding is in accord with the general finding that structural neuroimaging abnormalities are associated with cognitive impairment in LLD.^{37, 78-79} However, the lack of influence of white matter hyperintensities on any domain of cognitive ability contrasts with more recent studies.^{6, 8-9,15-16,80} The reason for this discrepancy is not clear. Our finding that white matter hyperintensities are not related to cognitive functioning is indirectly supported by the only neuropathologic study of LLD to date, in which O'Brien et al⁸¹ found no relationship between cognitive functioning in LLD and either vascular or Alzheimer-type neuropathologic characteristics. This issue bears further study, especially in subjects with late-onset depression.

The general pattern of neuropsychological functioning found in this study along with the model emanating from the risk factor analyses is relevant to the predominant cognitive theories of LLD.^17-19 Massman et al¹⁹ compared middle-aged patients with depression and patients with cortical and subcortical dementia syndromes. They concluded that among cognitively impaired depressed patients, the pattern of dysfunction resembles that associated with disorders that disrupt the frontostriatal pathways (eg, Huntington disease, Parkinson disease). Recent studies revealing evidence that disruption of prefrontal systems or their modulating subcortical pathways may play an important role in LLD have led to suggestions that in many older, depressed patients subcortical cerebrovascular disease disrupts frontostriatal circuits, producing dysfunction in executive abilities.^17-18 This executive dysfunction is seen as a major component of the various cognitive impairments associated with LLD.¹⁰

The present finding that LLD patients as a group are characterized by impairments in information processing speed, visuospatial ability, and executive function supports the subcortical-frontal circuit dysfunction model of LLD. However, our findings diverge from the emphasis on executive dysfunction as the primary deficit characterizing LLD in studies by Alexopoulos et al.^17-18 Rather, the present findings, similar to those of Degl'Innocenti et al⁸² and Nebes et al,⁵¹ suggest that poor performance on measures of executive dysfunction (even more so than other cognitive domains) is largely accounted for by slowed information processing speed. The conflicting views on the nature of the fundamental deficit notwithstanding, there is growing evidence that both executive dysfunction and information processing speed may be mediated by the striatum.^83-84

The present study represents an enhancement in methodologic rigor over previous studies by (1) using minimal exclusionary criteria to maximize the heterogeneity among LLD patients, thus permitting evaluation of a range of potential risk factors for cognitive impairment; (2) using a large, exclusively geriatric, mostly outpatient group to maximize the generalizability of findings; (3) including elderly control subjects to account for normal age-related cognitive changes; and (4) using a comprehensive neuropsychological battery to assess a range of cognitive functions. Nevertheless, there are a number of limitations to this study. One is the difficulty we had in measuring subjects' attention. Attention is a multifaceted concept, and it is possible that we did not measure the aspects most relevant to depression, reflected in the very low Cronbach of the attention measures. Another limitation is that the control group of normal, elderly individuals was relatively small and had less medical disease burden than did the depressed group. Although several analyses demonstrated that the difference in disease burden had no effect on the results, a control group more closely equated to the LLD patients in terms of medical burden would be ideal. In addition, the exploratory nature of our multiple regression analyses limited our ability to determine the extent of multicollinearity among our variables. Finally, although this study had a substantially larger patient group than most in the literature, larger sample sizes, including late-onset depression, in both the depressed and control groups would permit more extensive modeling of the relationship between risk factors and cognitive dysfunction in LLD.

CONCLUSIONS

From the Departments of Psychiatry (Drs Butters, Whyte, Nebes, Dew, Mulsant, Bhalla, Meltzer, Pollock, Reynolds, and Becker and Mss Begley and Zmuda), Radiology (Dr Meltzer), and Neurology (Drs Meltzer and Becker), University of Pittsburgh Medical Center, and the Geriatric Research Education Clinical Center of the Pittsburgh VA Healthcare System (Drs Whyte and Mulsant), Pittsburgh, Pa.

REFERENCES