The Nature and Determinants of Neuropsychological Functioning in Late-LifeDepression

Late-life depression (LLD) is a heterogeneous disorder that has becomea major public health concern as the population has aged. Late-life depressionis associated with significant morbidity and mortality and is both underrecognizedand undertreated.¹

Cognitive impairment in LLD is substantial and disabling. The cognitiveresponse to antidepressant treatment is variable, and impairments persisteven after effective treatment of depression.^{2 - 3} Thereis a substantially increased risk of developing a progressive dementia duringthe 2 to 4 years following a depressive episode.^{4 - 5}

Of the studies describing cognitive functioning in LLD, only a few^{6 - 9} haveincluded both a comprehensive assessment of cognitive domains and a healthycontrol group. Although the findings across these studies differ somewhat,they suggest that impairments exist in visuospatial ability, memory, speedof information processing, and executive functioning, with the executive deficitsbeing particularly related to late age at onset of first lifetime depressiveepisode (late-onset).^{9 - 10} Factorsthat are associated with neuropsychological impairment in LLD patients includemore severe depressive symptoms,^{6 ,11 - 12} moresevere anxiety,¹³ and/or vegetative symptoms.¹⁴ White matter hyperintensities are associated withpsychomotor slowing¹⁵ and executive functionimpairment, particularly in LLD patients with late-onset depression.^{9 ,16} In fact, 2 influential models of thecognitive effects of depression emphasize the role of subcortical-frontallobe circuit dysfunction,^{17 - 19} whichwould account for particular deficits in information processing speed andexecutive functions.

A variety of other risk factors may play a role in the likelihood thatLLD patients will exhibit cognitive impairment. For example, older age maybe associated with cognitive impairment in LLD,^{20 - 22} althoughthis is not always the case.^{23 - 24} Somestudies^{25 - 28} havefound no relationship between late-onset depression and greater risk for cognitiveimpairment during the index episode, although several studies^{16 ,29 - 31} havefound that they are related. Moreover, even if impairment were initially reversedwith successful treatment of depression, there may be an increased risk forsubsequently developing progressive dementia.⁴ Severalrecent studies^{5 ,32 - 35} havefound an association between late-onset depression and Alzheimer disease.Some studies,^{36 - 37} but not all,^{22 ,38 - 39} have found an increasein cognitive dysfunction in LLD with increasing medical burden. The apolipoproteinE (APOE) ϵ4 allele has been found to be relatedto poorer cognitive functioning in geriatric depression patients in one⁴⁰ but not all studies.^{41 - 43} Finally,there is evidence suggesting that serum anticholinergic burden negativelyaffects cognitive functioning, especially memory ability, in LLD.^{44 - 45} Many of these non–depression-relatedfactors also place so-called normal, elderly patients (ie, nonpsychiatricgroups) at risk for cognitive dysfunction^{46 - 48} andeven for dementia.⁴⁹

To our knowledge, no single study has provided a comprehensive analysisthat includes predictors and outcomes related to neuropsychological functioningin LLD. That is, although some studies have evaluated a single cognitive domainor a single risk factor, none have examined both risk factors and cognitivefunctioning in a comprehensive manner. Moreover, we were particularly interestedin examining the relationship among neuropsychological domains to determinewhether there might be a parsimonious mediating cognitive factor. Specifically,we hypothesized that information processing speed, which plays a criticalrole in the effects of normal cognitive aging⁵⁰ anddepression-related alterations in cognition,⁵¹ wouldmediate the association between biological and clinical risk factors and othercognitive domains.

We enrolled 140 patients 60 years and older with current unipolar majordepression (nonpsychotic). All patients had sought treatment at the WesternPsychiatric Institute and Clinic (Pittsburgh, Pa) and were enrolled in federallysponsored intervention studies conducted within the Intervention ResearchCenter for Late-Life Mood Disorders at the University of Pittsburgh Schoolof Medicine. Forty elderly control subjects with no psychiatric history wererecruited using local advertisements. The data for the present study werecollected during participants' baseline, pretreatment evaluation. Detailson 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 administeredby formally trained master's-level and doctoral-level clinicians and a consensusdiagnostic conference attended by the raters and at least 3 research geriatricpsychiatrists. Patients and control subjects were excluded if they had psychoticsymptoms or major unstable medical illnesses (eg, metastatic cancer). However,chronic diseases (eg, diabetes mellitus, hypertension) did not constituteexclusion criteria if the subject was medically stable. To study a broad,representative sample of those with LLD, ethnicity, insurance status, concurrentmedical and psychiatric conditions, and symptom profile did not constituteexclusion criteria. Additional exclusionary criteria for the present studyincluded the following: neurologic disorders or injuries known to have significantdirect effects on cognitive functioning (eg, traumatic brain injury, multiplesclerosis) (n = 10), uncorrectable sensory handicap (eg, blindness) (n = 2),a diagnosis of any type of dementia from the University of Pittsburgh's Alzheimer'sDisease Research Center either before or following resolution of the indexdepressive episode (n = 19), or a best (including pretreatment and posttreatment)Dementia Rating Scale (DRS)^{54 - 55} scoreof 129 or lower (n = 9). A DRS score of 129 is 3 SDs below the mean of ourage- and education-equated control group. These exclusionary criteria reducedthe original patient study group of 140 to 100 (composed of 4 inpatients and96 outpatients). After complete description of whichever parent treatmentstudy in which subjects were participating, written informed consent approvedby the University of Pittsburgh's Institutional Review Board was obtained.

Subjects had a broad-based pretreatment assessment that included clinical,psychosocial, biologic, and neuropsychological measures. (The neuropsychologicalmeasures are listed in Table 2.)Many of the nonneuropsychological measures assessed potential risk factorsfor cognitive impairment either in general or in LLD in particular. Thesemeasures, grouped by domain, included demographics (age, education, sex),depression-related factors (score on the 17-item Hamilton Depression RatingScale [HDRS],⁵⁶ age at onset of first lifetimedepressive episode), medical burden (score on Cumulative Illness Rating Scale–Geriatrics[CIRS-G]),⁵⁷ vascular burden (CIRS-G combinedheart and vascular scale scores, APOE allele type,^{40 - 43} serumanticholinergic burden⁵⁸ ), and structural brainabnormalities (cortical and ventricular atrophy and white matter hyperintensityburden⁵⁹ ).

Structural magnetic resonance (MR) imaging was conducted within 2 to3 weeks of the baseline evaluation. Each subject's imaging data were readindependently by 2 raters (including E.M.W. and C.C.M.) based on the CardiovascularHealth Study protocol.⁵⁹ The variables assessedwere ventricular atrophy, sulcal atrophy, and the total white matter hyperintensityburden on the T1-weighted, T2-weighted, and proton density images. As perthe Cardiovascular Health Study protocol, each variable was assigned a numericalrating by comparing each subject's imaging data to predefined visual standardsthat represent progressive severity on a 10-point scale (0 through 9). Ifthe 2 independent raters differed in their rating by 1 point, the final ratingwas the mean of the 2 values. A greater than 1-point difference between raterswas considered a major disagreement and was adjudicated by consensus. Fourraters were involved in assessing the MR imaging data; these raters achievedintraclass correlation coefficients of 0.71 for ventricular atrophy, 0.44for 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 wasperformed by 2 examiners who were highly trained and closely supervised bythe first author (M.A.B.). In most cases, patients were tested before antidepressanttreatment was initiated. All of the control subjects and 86 of the 100 depressedpatients were psychotropic drug free. Seven patients were taking low dosesof nortriptyline hydrochloride, which had been initiated 1 to 3 days beforethe 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 formany months and were in the process of tapering off the medication. One patienthad been taking mirtazapine for several weeks before testing. Importantly,all patients met clinical research criteria (ie, DSM-IV) for current major depression.

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

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

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

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

Age at onset of first depressive episode has frequently been found tobe 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 <60years of age, n = 43) and late-onset depression (onset at ≥60 years ofage, n = 57) and performed t tests or χ² analyses (where appropriate) comparing them on demographic, clinical,and biologic variables. MANCOVA (2 groups for 5 cognitive domains) comparingthe groups' performance on the cognitive domains while controlling for ageand serum anticholinergic burden was also performed.

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

There were no significant differences between the depressed patientsand control subjects on age, sex, education, CIRS-G combined heart and vasculardisease scores, frequency of APOEϵ4, level ofserum anticholinergic burden, or structural MR imaging measures of sulcalor ventricular atrophy (Table 1).Depressed patients reported more medical problems (CIRS-G) and also demonstratedhigher burden of white matter hyperintensities (MR imaging) and overall poorercognitive functioning (Mini-Mental State Examination⁶² andthe DRS).

MANCOVA revealed that CIRS-G was not a significant covariate (F_5,127 = 0.92, P = .47) and that depressed patientsperformed significantly worse than control subjects (F_5,127 = 2.80, P = .02). Two-sample t tests revealedthat the controls and LLD patients differed in all 5 domains (Table 2). The information processing speed domain was the only onein which subjects differed significantly on all 3 component measures. Subjectsdiffered on 2 of 3 tasks in the visuospatial domain, 2 of 4 tasks on the executivedomain, 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 controlgroup's mean z score is always 0. Depressed patients'and control subjects' raw scores on individual measures are presented in Table 3.

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

Comparison of early- and late-onset LLD patients revealed that the late-onsetsubgroup was significantly older at the time of neuropsychological evaluation(t₉₈ = −4.52, P<.001), had a higher percentage of males (χ²₁ = 5.42, P = .02), and had higher levels ofserum antiholinergic burden (χ²₁ = 4.05, P = .04). MANCOVA comparing the groups' performance on the cognitivedomains detected no significant differences between the groups.

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

To understand associations among the factors that affect cognitive functionin depressed patients without dementia, we used multiple regression analyses.Consistent with our initial hypothesis, the correlations between informationprocessing speed and the various risk factor variables (and other neuropsychologicaldomains) were uniformly high and significant. This observation supported ourhypothesis that information processing speed is a mediating factor in thecognitive functioning of the elderly, depressed patients and is consistentwith current models of cognitive aging.⁵⁰ Toexamine these relationships further, we regressed the language, visuospatial,memory, and executive domain scores on information processing speed and thoserisk factors that were significantly correlated (P<.05, r = .20) with the domain scores (ie, age, education, HDRSscore, age at onset of first depressive episode, and all 3 MR imaging measures).All of the predictor variables were entered simultaneously, and only thosethat significantly increased the explained variance (after controlling forall other variables) were retained. With one exception, information processingspeed was the sole significant predictor of each of the other neuropsychologicaldomain scores; this is represented in Figure2 by paths from speed to each of the other domains (with the associated βweights). We then regressed the speed domain score on age, education, andHDRS score, and each had a significant, independent association with speed.Education was the only predictor variable that had an independent associationwith one of the domain scores after accounting for the association with speed.Thus, the language domain score, but not the other domain variables, was significantlyassociated with both speed and level of education.

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

Compared with an age- and education-equated control group, largely unmedicated,depressed, elderly patients without dementia performed more poorly on allcognitive domains measured. The depth and breadth of impairment were substantiallygreater than previously suspected. More than half of LLD patients exhibitedclinically and statistically significant cognitive impairment. The most frequentlyimpaired cognitive domains were information processing speed, visuospatial,memory, and executive abilities. The depressed patients' impairments in theinformation processing speed and visuospatial domains were due to poor performanceacross nearly all of the individual measures that contributed to each of thesebroad domains. By contrast, the impairment in executive functioning was restrictedto 2 tasks that specifically measure set-shifting ability.

These findings on neuropsychological functioning in LLD are in generalagreement with the existing literature. Both Boone et al⁶ andLesser et al⁹ found that, compared with controlsubjects, mixed-age groups of mildly to moderately depressed patients wereimpaired in nonverbal intelligence, visual memory, and executive ability.The depressed patients in the study by Lesser et al were also impaired inconstructional ability and information processing speed, as were the mostdepressed patients in the sample studied by Boone et al. In a group of 10elderly, depressed patients, Hart et al⁷ founddeficits 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 comparedwith a control group, a group of 41 elderly, depressed patients exhibiteddeficits in memory, visuospatial ability, and information processing speed.

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

It is striking that the broad array of neuropsychological deficits wasexplained almost entirely by a fundamental deficit in information processingspeed (Figure 2). Other combinationsof predictor variables (eg, with executive function as a mediator) did notresult in such parsimonious relationships among predictors and outcomes. Thus,factors slowing information processing speed affect a range of cognitive functionsin LLD patients.

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

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

Some risk factors had no effect on cognitive performance in the presentanalyses. We found, as have others,^{22 ,38 - 39} noassociation between overall medical burden and cognitive ability. Therefore,the source of the cognitive dysfunction associated with LLD seems not to berelated to the excess medical burden that characterizes many elderly, depressedindividuals. Perhaps most notably, in our study group, we detected no effectof APOE allele type on the variance in cognitiveability. Several investigators,^{42 - 43} includingourselves,⁴¹ have found that the presence ofthe APOEϵ4 allele does not influence overallcognitive functioning in the context of LLD. The present study extends thiswork, particularly by specifying that the APOEϵ4allele does not influence any particular cognitive domain, especially memoryfunctioning, in LLD. This point is important because impairment in memoryfunctioning is the hallmark of early Alzheimer disease. The lack of influenceon memory ability suggests that LLD patients' increased risk of Alzheimerdisease is mediated by some other, non–APOE-related factor.

Our group has previously found that even minimal serum anticholinergicactivity has a negative effect on cognitive ability in both a community sample⁴⁴ and older, depressed patients without dementia.⁴⁵ Mulsant et al⁴⁴ studieda community sample that included individuals with dementia, whereas the presentstudy excluded patients with even mild dementia on an a priori basis. It ispossible that patients with preclinical or mild Alzheimer disease are themost vulnerable to deleterious anticholinergic effects, which would explainwhy Mulsant et al detected the relationship, whereas this study did not. Thepresent study group was larger, but otherwise its subjects had similar characteristicsto those studied by Nebes et al.⁴⁵ Nonetheless,since the study by Nebes et al⁴⁵ was conducted5 years earlier in the same research clinic as the present study, it is possiblethat clinicians became more vigilant in reducing anticholinergic burden inpatients identified clinically as being at risk for cognitive impairment.

In a secondary analysis, we examined how the MR image–derivedvariables affected cognitive outcomes. Although fewer subjects were enteredinto the multiple regression equations, β weights associated with thepatterns of associations between non–MR image variables did not changesubstantially; neither cortical atrophy nor white matter hyperintensity scoreswere associated with any outcomes, whereas ventricular atrophy was linkedto information processing speed and language domain scores. This finding isin accord with the general finding that structural neuroimaging abnormalitiesare associated with cognitive impairment in LLD.^{37 ,78 - 79} However,the lack of influence of white matter hyperintensities on any domain of cognitiveability contrasts with more recent studies.^{6 ,8 - 9 ,15 - 16 ,80} Thereason for this discrepancy is not clear. Our finding that white matter hyperintensitiesare not related to cognitive functioning is indirectly supported by the onlyneuropathologic study of LLD to date, in which O'Brien et al⁸¹ foundno relationship between cognitive functioning in LLD and either vascular orAlzheimer-type neuropathologic characteristics. This issue bears further study,especially in subjects with late-onset depression.

The general pattern of neuropsychological functioning found in thisstudy along with the model emanating from the risk factor analyses is relevantto the predominant cognitive theories of LLD.^{17 - 19} Massmanet al¹⁹ compared middle-aged patients withdepression and patients with cortical and subcortical dementia syndromes.They concluded that among cognitively impaired depressed patients, the patternof dysfunction resembles that associated with disorders that disrupt the frontostriatalpathways (eg, Huntington disease, Parkinson disease). Recent studies revealingevidence that disruption of prefrontal systems or their modulating subcorticalpathways may play an important role in LLD have led to suggestions that inmany older, depressed patients subcortical cerebrovascular disease disruptsfrontostriatal circuits, producing dysfunction in executive abilities.^{17 - 18} This executive dysfunction is seenas a major component of the various cognitive impairments associated withLLD.¹⁰

The present finding that LLD patients as a group are characterized byimpairments in information processing speed, visuospatial ability, and executivefunction supports the subcortical-frontal circuit dysfunction model of LLD.However, our findings diverge from the emphasis on executive dysfunction asthe primary deficit characterizing LLD in studies by Alexopoulos et al.^{17 - 18} Rather, the present findings, similarto those of Degl'Innocenti et al⁸² and Nebeset al,⁵¹ suggest that poor performance on measuresof executive dysfunction (even more so than other cognitive domains) is largelyaccounted for by slowed information processing speed. The conflicting viewson the nature of the fundamental deficit notwithstanding, there is growingevidence that both executive dysfunction and information processing speedmay be mediated by the striatum.^{83 - 84}

The present study represents an enhancement in methodologic rigor overprevious studies by (1) using minimal exclusionary criteria to maximize theheterogeneity among LLD patients, thus permitting evaluation of a range ofpotential risk factors for cognitive impairment; (2) using a large, exclusivelygeriatric, mostly outpatient group to maximize the generalizability of findings;(3) including elderly control subjects to account for normal age-related cognitivechanges; and (4) using a comprehensive neuropsychological battery to assessa range of cognitive functions. Nevertheless, there are a number of limitationsto 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 measurethe aspects most relevant to depression, reflected in the very low Cronbach αof the attention measures. Another limitation is that the control group ofnormal, elderly individuals was relatively small and had less medical diseaseburden than did the depressed group. Although several analyses demonstratedthat the difference in disease burden had no effect on the results, a controlgroup more closely equated to the LLD patients in terms of medical burdenwould be ideal. In addition, the exploratory nature of our multiple regressionanalyses limited our ability to determine the extent of multicollinearityamong our variables. Finally, although this study had a substantially largerpatient group than most in the literature, larger sample sizes, includinglate-onset depression, in both the depressed and control groups would permitmore extensive modeling of the relationship between risk factors and cognitivedysfunction in LLD.

Clearly, LLD is a multifactorial disorder with multiple clinically meaningfulphenotypes and trajectories.^{9 ,15 ,85 - 87} Thisstudy provides the most extensive characterization to date of the range, type,and depth of cognitive impairments in elderly patients who are in an episodeof major depression. In addition, the association between a broad range ofrisk factors and domain-specific cognitive functioning was demonstrated. Thedata show that cognitive impairment in LLD is prevalent and broad based, involvinghalf of a clinical study group and encompassing a broad range of deficits,notably speed of information processing, visuospatial ability, and executivefunction. Most of the variance in the cognitive domains measured in this studywas attributable to information processing speed and was not related to theburden of chronic medical illness, APOEϵ4 alleledistribution, or even serum anticholinergic burden. Further study is neededto clarify the persistence of deficits after successful treatment, their relationshipto Mild Cognitive Impairment and subsequent diagnosis of dementia, and theextent to which their progression can be slowed (or halted) with currentlyavailable therapies, including dopaminergic and cholinesterase inhibitor therapies.

Submitted for publication September 19, 2003; final revision receivedDecember 24, 2003; accepted January 20, 2004.

This research was supported by US Public Health Service grants MH01684,MH01077, MH65416, MH52247, MH43832, MH37869, and AG05133.

Some of the data in this article were presented at the Eighth InternationalConference on Alzheimer's Disease and Related Disorders; July 21, 2002; Stockholm,Sweden.

We thank Sati Mazumdar, PhD, and Patricia Houck, MS, for invaluablestatistical advice; Melanie Fukui, MD, and Joelle Scanlon, PhD, for ratingthe MR imaging scans; and the staff of the Mental Health Intervention ResearchCenter for Late-Life Mood Disorders for their help recruiting and evaluatingparticipants in this study.