Factors Associated With 1-Year Outcome of
Major Depression in the Community
J. Kent Sargeant, MD; Martha L. Bruce, PhD; Louis P.Florio, MS; Myrna M. Weissman, PhD
\s=b\ Evidence from outcome studies of major depression indi-
cates a high rate of relapse and chronicity, and that prior chronic-
ity, recurrent episodes, and the presence of psychosocial stress-
ors are associated with a poor outcome. However, the
generalizability of these findings is limited because most studies
have focused on treated samples; thus, these studies may have
been biased toward more chronic or severe illnesses. In prospec-
tively surveying a large probability sample of the general popula-
tion, the Epidemiologic Catchment Area program offers the op-
portunity to investigate prognosis without selection bias. In this
study, the Epidemiologic Catchment Area subjects with a diag-
nosis of Major Depressive Disorder at first interview (n 423) =
were categorized according to their diagnostic status 1 year later.
The results confirmed a high rate of nonrecovery, with clinical
features associated with a poor outcome that resembled those
identified in previous clinical studies. Overall, clinical factors
were more important prognostically than were sociodemogra-
phic characteristics. However, there was some evidence that a
poorer outcome in older women may partially explain the greater
female prevalence of depression in the community.
(Arch Gen Psychiatry. 1990;47:519-526)
Tradi t i o nal
been an l y , episodic
the
important
nature of affective disorders has
feature that distinguishes them from
the
schizophrenia. Although origin of this view has often been
ascribed to Kraepelin,1 he, in fact, noted a great variation in
the course of manic-depressive psychosis. Nevertheless, de¬
pression, in particular, has been perceived as a disorder in
which basically a good function is punctuated by periodic
lapses into illness with a subsequent recovery of function.2,3
Since the development of explicit diagnostic criteria, there
has been a renewal of interest in the course and outcome of
mental disorders. Clinical optimism regarding the course of
depression has been challenged by a number of investigations
that have documented a poor outcome in a significant portion
of depressive subjects. Follow-up studies of patients who have
Accepted for publication June 30,1989.
From the Department of Psychiatry, University of British Columbia, Van-
couver, Canada (Dr Sargeant), the Department of Epidemiology and Public
Health, School of Medicine, Yale University, New Haven, Conn (Dr Bruce and
Mr Florio), and the Division of Clinical-Genetic Epidemiology, New York State
Psychiatric Institute and College of Physicians, Columbia University, New
York, NY (Dr Weissman).
Reprint requests to Department of Psychiatry, University of British Colum-
bia, 2255 Westbrook Mall, Vancouver, British Columbia, Canada V6T 2A1 (Dr
Sargeant).
Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Pittsburgh User on 12/06/2015
met DSM-III criteria or Research Diagnostic Criteria for
major depression have indicated that chronic depression may
develop in more than 20% of patients after an index episode.
Partial recovery will occur in 15% to 20% of patients, and 15%
to 25% of patients who recover will relapse within 1 year. Only
a minority of patients achieve and maintain a complete recov¬
ery."
Similar results have been obtained with community samples
of depressive subjects. High recurrence and chronicity rates
have been reported from both Great Britain10 and the United
States,11 and together with patient data, point to enduring
morbidity among depressive subjects after an index episode.
Despite this agreement, there are problems in interpreting
these findings. First, subject selection bias limits the generali-
zability of results. This has been a particular difficulty in
patient-based studies in the United States since a minority of
those with major depression receive treatment.12,13 Although
service utilization is influenced by many factors, tertiary cen¬
ter patients may represent a more severely depressed group
that is prone to more recurrence and chronicity than the wider
population of depressed persons.
In studies of community cases, treatment bias effects are
removed, but other selection biases may be substituted. Sur¬
tees et al,10,14,15 for example, studied only women. As reviewed
by Hirschfeld and Cross,16 other studies have frequently com¬
promised the clinical applicability of findings by using more
economical nondiagnostic measures of depressed psycho¬
pathology. Finally, most studies have focused on the influence
of either clinical or sociodemographic factors on outcome.
These factors have rarely been assessed simultaneously.
The Epidemiologie Catchment Area (ECA) Program" repre¬
sents the first opportunity to investigate these issues in a large
probability sample of the general population. Subjects included
both treated and untreated individuals who exhibited a range of
depression severity and comorbidity patterns. This, together
with the use of specific diagnostic criteria and standardized
methods of assessment in a prospective design, combined the
strengths of previous clinical and community follow-up studies.
In this study, ECA subjects with major depression at a first
interview were identified, and their depression status was
assessed 1 year later at a second interview. Research ques¬
tions of interest were as follows: (1) Do community cases of
major depression show recovery rates similar to patient
groups? (2) How do sociodemographic factors such as sex and
age, known to influence depression prevalence,16,18,19 affect out¬
come? (3) Do clinical characteristics identified in patient stud-
 íes as being important determinants of outcome hold in a
Table 1 .—Demographic Characteristics of ECA Subjects
community sample? and (4) Does psychiatric comorbidity in¬ Meeting Criteria for Major Depression at Initial Interview (T1)
fluence depression outcome?
(Five Sites)*
SUBJECTS AND METHODS % of Subjects
(n 423)
=

The ECA project is a five-site collaborative study with the primary Sex, F 77.1
aim of obtaining prevalence and incidence data on specific psychiatric Age, y
disorders in the general population. Its design has been described in 18-29 27.4
detail elsewhere and will only be reviewed in brief here. Stratified 30-44 35.0
sampling procedures were used to select 18 572 persons for inter¬ 45-59
60+
16.6
views from the community. (This report excludes institutionalized 21.0
Marital
subjects.) Subjects were assessed at two time points, 1 year apart, by status
trained lay interviewers who used the Diagnostic Interview Schedule
(DIS). The DIS is a semistructured interview designed for use by Single 23.6
Married 33.1
nonclinicians; it elicits information that may be used to generate Divorced 18.9
DSM-III diagnoses by a computer algorithm.20 Investigations of the Separated 9.9
DIS in a number of designs21"28 have found acceptable reliability Widowed 14.4
ratings for depression.24 Education, y
0-9
In this study, subjects from all five sites who met DSM-III criteria 24.3
for 10-12 40.7
major depression at a first interview (time point 1 [Tl]) were 13+ 34.5
identified; only those subjects with current depression were re¬ Race
tained. In keeping with previous ECA analyses,18, current disorder W 64.1
was defined as an episode that occurred in the 6 months before the 23.6
initial interview. Depression that resulted solely from grief was Other 10.6
included if the duration of depression exceeded 1 year. Socioeconomic
Since we were interested in investigating the effect of comorbidity statust
on outcome, all diagnostic hierarchies were suspended and multiple 1 23.2
2 36.9
diagnoses for a current disorder were allowed. The one exception to 3 29.3
this was the exclusion of those subjects who ever met criteria for 4 10.4
bipolar disorder. There is considerable evidence for a distinction
between bipolar and unipolar illness from treatment, course of ill¬ *ECA indicates Epidemiologie Catchment Area; T1, time point 1.
ness, and genetic studies.3 Also, while it is conceptually possible for tQuartile categories based on income, education, and occupation scores
individuals to have more than one disorder (eg, depression and schizo¬ (see text).
phrenia), the clinical meaning of comorbidity between bipolar illness
and depression is unclear. The exclusion of the group with bipolar
illness was intended to avoid this nosologie conundrum and give a
more homogeneous study sample. Table 2.—ECA Subjects With Major Depression at First
The specific outcome variable for analysis was the presence or ab¬ Interview: Depression Status After 1 Year (Five Sites)*
sence of major depression after 1 year (time point 2 [T2]). A 6-month
current disorder was used to define subjects who again met criteria for
Diagnostic
major depression, as well as those subjects with other disorders. Status No. (%)
The outcome of subjects was analyzed in four phases. First, the Not depressed
characteristics of those subjects who were lost to attrition were
263 (62.2)
Major depression 81 (19.1)
evaluated for possible bias effects. Second, the outcome of subjects Attrition 79 (18.7)
was analyzed for the effects of socioeconomic variables as assessed at Total 423 (100.0)
Tl. The variables selected—sex, age, race, site, marital status,
education, income, and employment status—represented those that *ECA indicates Epidemiologie Catchment Area.
previously9 have been tested for an association with depression preva¬
lence.14,18, Third, the Tl clinical characteristics of prior depression
history, Tl episode severity, and comorbidity were evaluated for The use of unweighted data here resulted in sample characteristics
outcome effects. Tests with the 2 statistic were used to detect
that did not strictly reflect those of major depressive subjects in the
significant associations in these initial analyses, thus providing candi¬ community. Thus, for example, there was an excess of females (fe¬
date variables for subsequent analysis.
In the final phase of analysis, models of depression outcome were male-male ratio of 3.5) that was somewhat greater than previous
generated using logistic regression to examine the simultaneous community prevalence estimates,19,25 since the original ECA survey
effects of sociodemographic and clinical characteristics identified contained more females than males. In addition, oversampling of
above. Logistic regression is an appropriate technique for analyzing elderly persons gave rise to elevated frequencies of less educated
the effects of categorical or continuous independent variables on a persons and persons of lower socioeconomic status.
dichotomous outcome.26 It has the attractive property of yielding Most persons in this depressed sample were married or single.
odds ratios as measures of association between variables, thus facili¬ Separated persons were least frequent, and in subsequent ana¬
tating clinical interpretation. lyses, these were combined with divorced and widowed subjects to
contrast those persons with unstable vs stable relationship
Previous ECA reports of prevalence and incidence have used
histories.
weighted data for estimates of population statistics. This study, Socioeconomic status is shown for quartile categories, ordered from
however, involved the aggregation across five sites of a small subset lowest to highest. Subjects were assigned by averaging education,
of the original sample. The appropriate method to adjust available
household income, and occupation percentiles. Education and income
weight factors for attrition effects under these circumstances is a percentiles were based on 1980 US census data. Occupation percen¬
complex and unresolved issue. Accordingly, it was decided to control tiles were derived following the procedure of Nam et al27 that catego¬
for factors that were relevant to the sampling design in the analysis,
rather than to weight data before the analysis. rizes job status according to the average of income and educational
percentiles for each occupational category. Occupation scores and,
therefore, overall socioeconomic status are thus dependent on nation¬
RESULTS al patterns of education and income and do not incorporate traditional
notions of "prestige."
Sample characteristics for 6-month current Tl depressive subjects Subjects who were selected for 6-month current depression at the
is given in Table 1. From all five sites, 423 persons met DSM-III first interview were assessed for their depression status 1 year later
criteria for major depression in this interval. (T2). Most subjects (62.2%) were noncases, while approximately one

Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Pittsburgh User on 12/06/2015

 Table 3.—Persistence of Major Depression After 1 Year by ECA Site (Five Sites)*
ECA Site

New Haven, Baltimore, St Louis, Durham, Los Angeles,

Conn Md Mo NC Calif
Initial No.
depressed at T1 91 58 80 44 71
% with persistent
depression at T2 29.7 8.6 23.8 25.0 26.8
*ECA indicates Epidemiologie Catchment Area; T1, time 1
point T2,
; time point 2. 2 =
9.32 and =
.05.

Table 4.—Sociodemographic Characteristics of ECA Subjects With Persistent Depression After 1 Year (Five Sites)*
Male Female Total

% With % With % With

Persistent Persistent Persistent
Depression Depression Depression
ge. y
18-29 28 17.9 65 15.4t 93 16.1t
30-44 22 18.1 100 29.0 122 27.0
45 + 26 15.3 103 28.2 129 25.6
Marital status
Married
single 52 19.2 164 19.9f 198 19.7t
Divorced-widowed
separated 24 12.5 122 32.0 146 28.8
Education, y
0-9 18 11.1 62 37.1t 80 31.2t
10 + 58 19.0 204 22.1 262 21.4
*ECA indicates Epidemiologie Catchment Area. Note: age contrasts are younger than 30 years vs older than 30 years. The and percent of patients with
persistent depression were as follows: male, 76 and 17.1 ; female, 268 and 25.4; and total, 344 and 23.6.
tP<05byx2test.

fifth (19.1%) again met DSM-III criteria for major depression (Table
2). A substantial number of subjects were lost to attrition via refusal,
failure to contact, or death (18.7%).
It is important to note that T2 current cases represented two major
groups: (1) those subjects with chronic, unremitting depression since
Tl and (2) those subjects who did recover but relapsed in the interven¬
ing time. The ECA/DIS data do not provide detailed information
regarding the episode course and therefore cannot differentiate these
groups. Nevertheless, for the purposes of this article, T2 current
cases will be referred to as persistent depressions.
Effect of Attrition
Since a significant minority of subjects was unavailable for follow-
up, it was important to investigate whether attrition was biased with
respect to the Tl sociodemographic and clinical characteristics of
interest in this study. Accordingly, the distribution of missing sub¬
jects was analyzed by site, age, sex, race, socioeconomic status,
depression history, and current episode severity. No comparisons
reached statistical significance (the P= .05 level), indicating that the
loss of subjects was likely to have been independent of these variables.
Subsequent analyses focused on comparisons between persistent and
recovered depressive subjects.
Comparison Between Persistent and
Recovered Depressive Subjects
Overall, the persistence of major depression in the community
sample was high, with 23.6% of those subjects who were followed up
meeting DSM-III criteria for depression 1 year after an index
episode. This is consistent with figures reported from patient studies
(see "Comment" section). Tables 3 through 5 present the results of
comparisons between persistent and recovered depressive subjects.
Sociodemographic Characteristics
Site differences in outcome were significant at the 5% level (Table
3), due primarily to the low frequency of persistence at the Baltimore,
Md site. Whether this was due to real intersite differences in outcome
or from méthodologie variation could not be ascertained with these
data. Additional analysis (not shown) of subject distribution at ECA
sites by sex, age, marital status, and education gave no evidence of
underrepresentation at Baltimore for those subjects at a high risk of
depression persistence. The issue of site differences is discussed further
below. For ease of subsequent presentation, results are not stratified by
site although this variable was controlled in logistic models.
Table 4 presents data on case persistence by sex and age. Overall,
females demonstrated a higher rate of persistence than males (25.4%
vs 17.1%). However, this difference failed to reach statistical signifi¬
cance (P=.13). For males, there was no evidence of variation in
persistence with age. Females, in contrast, showed a significant
increase in depression persistence after the age of 30 years (28.6 vs
15.4%; P<.05), with younger women and males of all ages showing
comparable rates. The absence of a statistically significant sex differ¬
ence in the rates of persistence throughout all ages may thus be due to
two factors: (1) Higher rates of persistence for females aged older than
30 years are obscured when combined with the lower rate in younger
women. (2) Unequal sample sizes with low numbers of male persistent
depressive subjects lead to unstable rate estimates and reduced sta¬
tistical power.
Other sociodemographic characteristics, analyzed for the persis¬
tence of depression (Table 4), indicated that both an unstable relation¬
ship history and less than a high school education were associated with
poor outcome (P=.05 for both in the overall sample). As with age
effects, males failed to show a significant variation in depression
persistence with these factors.
Socioeconomic characteristics unrelated to depression outcome
were race, global socioeconomic status score, and financial compo¬
nents of socioeconomic status (income and employment status).

Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Pittsburgh User on 12/06/2015

 Table 5.—T1 Clinical Characteristics of ECA Subjects With Table 6.—Rates of Comorbidity al T1 Between MD and
Persistent Depression After 1 Year (Five Sites)* Other Diagnoses and Odds Ratio of Depression Persistence
by Diagnosis (Five Sites)*
Persistence,
% Comorbid
No. of Diagnoses % T1
prior episodes with Comorbidityt Odds
1 21.1 MDatTI (n 344)
=
Ratio*
51 16.0 Substance
10.13 .02 abuse 12.8 0.81
3-9 120 18.3
Alcohol
>10 93 34.8 abuse 9.4 0.73
Longest prior Panic 8.6 1.87
episode, mo
<1 108 15.7 Obsessive-
1-3
compulsive 8.8 1.78
10.50 .01 Phobia 41.5 1.35
22.5
6-24 64 35.9 Any anxiety 43.2 1.43

No. of Dysthymia 31.1 1.63

symptoms
positive at T1
4 137 15.3
24.5
11.08 .01
62 30.7
7-8 47 36.2
No. of concurrent
T1 diagnoses
0 110 20.9
1 145 20.7
14.18 .004
65 23.1
3-5 24 54.2
*T1 indicates time point 1 ; ECA, Epidemiologie Catchment Area.
Clinical Characteristics
Table 5 indicates the rates of depression persistence according to Tl
clinical characteristics. The DIS records the number of depressed
episodes, as well as the longest prior episode. Being depressed 1 year
after an index episode is more likely for those subjects who reported a
large number of previous depressed periods (P= .02); these subjects
may well have remitted and relapsed between waves of ECA inter¬
views. Conversely, persons with a history of chronicity seem to have
maintained their pattern (P .01), perhaps remaining depressed dur¬
= with depression persistence. Table 6, column 3 compares the odds of
ing the interim. The reliability of distinguishing greater than 10 prior
episodes and chronic depression is compromised by retrospective
ECA data; however, subsequent logistic analysis demonstrated their
independent contribution to persistence. It is noteworthy that, for
these factors, only the most severe histories seem to result in an
appreciable increase in the risk of persistent depression.
Episode severity was measured as the number of symptom groups
present from the eight areas assessed (disturbances of appetite,
weight, sleep, sexual interest, physical activity, energy, or concen¬
tration and a preoccupation with thoughts of guilt or suicide). Since
subjects were selected because of their diagnosis of depression, the
symptom count ranged from 4 (just meeting DSM-III criteria) to 8 (all
symptom groups present). There was a positive association between
case persistence and the number of symptom groups reported at the
initial interview (F=.01). This relationship was more clearly a step-
wise function than that of the other clinical factors, perhaps because it
was the least retrospective and did not depend on complex decision-
rules required to assign comorbid diagnoses. Among specific symp¬
toms, a positive association was significant only between case persis¬
tence and suicidal features. In contrast to sociodemographic
characteristics, none of the preceding clinical variables showed evi¬
dence of differing effects due to sex or age.
The Tl comorbidity was assessed for eight additional diagnoses that
were current at Tl: substance abuse, alcohol abuse, panic disorder,
phobia (simple and agoraphobia), obsessive-compulsive disorder, so-
matization disorder, dysthymia, and schizophrenia. (Mania and bipo¬
lar disorder were criteria for exclusion in this sample and thus are not
represented.) Most subjects met criteria for more than one current
disorder; the maximum number of additional diagnoses was five.
Examination of comorbidity effects on depression persistence demon-
Somatization 2.1 2.58
Schizophrenia 3.8 2.08
*T1 indicates time
point 1 ; MD major depression.
fPercent of those with T1 current MD also meeting criteria for specified
(current) second disorder.
JRatio of odds of depression persistence if comorbid to odds of depression
persistence if not comorbid, for specified disorder.
strated that one or two additional diagnoses did not make an apprecia¬
ble difference in outcome. For three or more diagnoses, however,
there was a significantly higher persistence of depression (P= .004).
Patterns were similar for both sexes and all age groups.
The rates of comorbidity were evaluated for each diagnosis sepa¬
rately. In Table 6 (column 2), substance and alcohol abuse show
intermediate rates of comorbidity, with about 10% to 12% of de¬
pressed subjects simultaneously meeting criteria for these disorders
and depression at Tl. Comorbidity with phobic disorders was primari¬
ly responsible for the high rate of comorbidity between anxiety disor¬
ders and depression. Almost one third of the sample met criteria for
both major depression and dysthymia. The lowest rates of comorbi¬
dity were found with schizophrenia and somatization disorder.
Individual diagnoses were also evaluated for specific associations
persistence between those subjects who were comorbid against those
subjects who were not comorbid for each diagnosis. Odds ratios
greater than 1 reflected an increased likelihood of depression persis¬
tence for comorbid individuals. Although the likelihood of depression
persistence for those subjects who were comorbid with somatization
disorder and schizophrenia appeared to be elevated, the low frequen¬
cy of these disorders rendered this statistically insignificant. Only
dysthymia achieved marginal significance (P =. 10 level), with an odds
ratio of 1.6. Investigation of sex and age effects revealed that a
significantly positive association with depression persistence held
only for male subjects with dysthymia (P<-05).
Clinical characteristics not found to be associated with case persis¬
tence were age at onset of depression, the number of additional
diagnoses ever present (lifetime comorbidity), and the current Tl
comorbidity with specific disorders other than dysthymia.
Logistic Regression Models
Logistic regression was employed to determine whether the signifi¬
cant effects of sociodemographic and clinical factors on persistence
observed in bivariate analyses were mutually independent. For all
models the overall goodness of fit, as measured by the likelihood ratio
test, was highly significant (P<.001).
The first model employed sociodemographic terms (Table 7). Once
site was controlled (Baltimore vs others), only education maintained a
significant effect on the likelihood of depression persisting during the
course of the study period. Significant effects of age and marital
status, observed in separate intermediate models (not shown), were
probably lost in the full model because of an intercorrelation between
these factors (Pearson r= .36). However, the effect of education, as

Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Pittsburgh User on 12/06/2015

 Table 7.—Odds Ratios of Depression Persistence After 1 Year: Multivariate Logistic Models
Model 1 Model 2 Model 3 Model 4
Variables Odds Odds Odds Odds
In Model Ratio SE* Ratio SE Ratio SE Ratio SE
Sitet 6.9* 0.62 7.0* 0.63 7.8* 0.63 7.9* 0.64
Sex§ 1.3 0.36
Agell 1.3 0.35
Marital
statusH 1.3 0.30
Education* 1.5** 0.19 1.3 0.20
T1tt
comorbidity 2.1 0.52
Symptom
severity** 2.6* 0.49 2.6* 0.49 2.7* 0.49
Prior
episode
length§§ 1.61 0.19 0.19 1.7III 0.20
Prior
episode
numberHTl 2.0* 0.31 2.111 0.30 2.2III 0.30
Sex§
Dysthymia + 0.78 0.57
Dysthymia -
2.3## 0.49
Dysthymia***
Female 0.89 0.36
Male 2.6 0.68
*SE of log odds ratio.
tOther site vs Baltimore, Md.
*P<.005.
§Female vs male.
j|Age older than 30 years vs younger than 30 years.
HUnstable vs stable.
#Five-year period.
**P<.05
ffGreater than two additional diagnoses.
ttFour-symptom increase.
§§Twelve-month period.
||||P<.01.
flGreater than 10 vs less than 10.
##P<.1.
***Dysthymia (presence) vs dysthymia (absence).

well as of age, marital status, and sex, was modest; the estimated odds
ratios for each of the sociodemographic factors were less than 1.5. The
ECA site strongly affected the likelihood of persistence. Although
previous analyses suggested that the effects of sex might vary with
age, the inclusion of an age-sex interaction term resulted in no im¬
provement in this model.
In contrast, individual clinical characteristics were more likely to
remain significantly related to persistence, as shown in the second
multivariate model. Since Tl comorbidity was only moderately corre¬
lated with the prior episode number and length (Pearson r .21 and =
.15, respectively), the loss of a significant association with outcome in
this model appeared to result from a failure to contribute additional
predictive information once the other factors were controlled. In this
model, the effects of the remaining factors were more substantial than
those for sociodemographic characteristics, with odds ratios ranging
from 1.6 to 2.6. Site was again significantly associated with outcome.
When factors with significant effects from the first two models were
combined (model 3), only site and clinical characteristics retained
predictive significance. Other models that contained sociodemogra¬
phic, clinical, and (where appropriate) interactive terms did not
change this pattern of effects.
The final model included effects due to dysthymia, together with
those of previously significant factors. Since dysthymia was found to
be associated with depression persistence only for males, an interac¬
tive term was included for sex x dysthymia. As a result, the odds
ratios for dysthymia were calculated that were contingent on sex and
those for sex were contingent on the presence or absence of
dysthymia.
In this model, clinical characteristics remained significantly associ-
ated with persistence, with odds ratios similar to those of previous
models. There was also some evidence that sex effects on outcome
depended on whether dysthymia was present, after controlling for
important clinical factors. In the absence of dysthymia, females were
more than twice as likely as males to remain depressed during the
follow-up year (odds ratio, 2.28; P<.1). However, sex effects in the
dysthymic group were essentially absent (odds ratio, 0.78; not signifi¬
cant), suggesting that males were more adversely affected by dysthy¬
mia than were females. This was reflected in the pattern of odds ratios
for dysthymia that revealed no increased risk of persistent depression
among females (odds ratio, 0.89; not significant). In contrast, males
with dysthymia were 2.6 times as likely to exhibit persistence com¬
pared with those without. With relatively low numbers of males, this
effect just fell short of statistical significance (P =. 15); nevertheless,
these results suggest that for males, chronic dysphoria and mild
depressive symptoms are more important prognostically than for
females.
COMMENT
The results indicate that persistence of major depression in
the general population is high; approximately 20% of those
subjects with an index episode will be depressed 1 year later.
The diagnostic status of 18.7% ofthis sample lost to attrition is
unknown. Thus, while there is no evidence that the loss of
subjects was biased for factors related to outcome, this esti¬
mate of depression persistence remains a conservative one
because of subject attrition.

Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Pittsburgh User on 12/06/2015

 The effect of sex on outcome appears to vary with age, with
women aged older than 30 years showing significantly in¬
creased rates of depression persistence. For women, the
sociodemographic factors of less than a high school education
and an unstable marital history also were associated with poor
outcome. In these high-risk groups, persistence was approxi¬
mately 30% or higher. For men, there was no evidence that
the factors of age, education, or marital history were associ¬
ated with outcome, although low frequencies limited the reli¬
ability of these observations. Forthe entire sample, the simul¬
taneous analysis of these factors showed that only a poor
education was significantly associated with depression persis¬
tence. A 5-year decrement in education increased the likeli¬
hood of depression after 1 year by a factor of 1.5.
Clinical characteristics of depression history, increased in¬
dex episode severity, and greater index comorbidity were
significantly associated with outcome. There was no trend for
the effects of clinical characteristics to vary with age or sex.
When analyzed together, all, except comorbidity, were inde¬
pendently predictive of depression persistence. The likelihood
of depression persistence for a prior longest episode of 12
months, more than 10 prior episodes, and maximum current
severity was increased by factors of 1.6, 2.0, and 2.6,
respectively.
The comorbidity in this community sample was high, with
one quarter of subjects meeting criteria for three or more
other current diagnoses, excluding bipolar disorder. The only
individual diagnosis associated with poor outcome was dys¬
thymia, and this held only for males.
Comparison With Other Studies
The comparison of overall results with those of previous
studies is complicated by a number of factors. First, unlike
some other studies,6,8,10 the ECA design did not include de¬
tailed course and outcome measures; thus, it precludes sepa¬
rate estimates of chronicity and relapse. By obtaining only
two "snapshots" of mental functioning taken 1 year apart,
some of those with a relapsing course would be classified as
recovered in this study. The resulting bias toward underesti¬
mating poor outcome is compounded by subject attrition, as
discussed above.
Second, other studies have generally focused on depressive
episodes and did not exclude those with a bipolar illness.5,6,8,28
Since bipolar disorder may be more prone to relapse than
unipolar illness, these investigations may slightly overesti¬
mate relapse in depressive disorder.
Design issues, then, lead to an expected rate of persistence
here that is intermediate between the rates of chronicity and
the sum of chronicity plus relapse from other studies. In
general, patient-based 1-year follow-up studies have found
that chronicity ranges from 20% to 35%, and that relapse
accounts for another 20% to 30% of subjects.5,6,8 High-risk
groups in the general population give similar results,10 while
less selected individuals from the community appear to have a
better prognosis.'1 The persistence rate here (approximately
20%) is slightly less than expected. As discussed below, this
may reflect the presence of less severe illnesses in the
community.
Comparison With Other Studies:
Clinical Characteristics
Previous patient-based studies have documented the im¬
portance of clinical characteristics in determining the outcome
of both depressive episodes and chronic depression.29,30 In the
National Institute of Mental Health Collaborative Study on
the Psychobiology of Depression, for example, the duration of
index episode and the number of prior episodes were the most
Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Pittsburgh User on 12/06/2015
significant predictors of outcome.89,31,32 The principal factors
associated with outcome in this study were similar, and in this
respect, the results from this community sample are compara¬
ble with those from treated groups.
Also consistent with patient-based research was the associ¬
ation between the episode severity and outcome.8,33 Differ¬
ences in depression outcome across studies may thus reflect
sample variation in this factor. Forty percent of subjects in
this study met minimal criteria for major depression, and this
may partially explain why slightly lower than expected persis¬
tence rates were observed.
Comorbidity effects on depression outcome have rarely
been comprehensively assessed in previous studies; usually,
there has been a focus on depression and only one other
disorder. Thus, either panic disorder or depression alone have
been reported as having a better outcome than a combined
diagnosis.34 Similar results have been obtained for dysthy¬
mia.35 Our results do not indicate that individual disorders,
comorbid with depression, are significantly associated with
poor depression outcome in the community. Rather, total
comorbidity appears to be more important prognostically than
any particular diagnosis.
For males, dysthymia was an exception to this finding, and
its presence removed any trends for a difference in outcome
between sexes. Double depression has previously been found
to affect prognosis adversely,35 although others have not re¬
ported a stronger effect for males. Nevertheless, a tendency
for male underreporting of symptoms has been offered as a
partial explanation for an excess female prevalence of depres¬
sion,36 and Angst and Dobler-Mikola36 have suggested that
impairment-based diagnostic criteria offset symptom report¬
ing bias, thus equalizing sex-specific depression rates. A dif¬
ferential sex effect of dysthymia on depression outcome is
consistent with these observations, and the question is again
raised of how diagnostic classification should incorporate
symptom, impairment, and prognostic information.
It is of interest that, despite a lack-of individual significance,
almost all comorbid diagnoses were associated with an in¬
creased likelihood of depression persistence. The two excep¬
tions to this were substance and alcohol abuse, suggesting
that later depression may be less likely in the presence of these
disorders. If substantiated, this may reflect a masking of later
depressive symptoms through "self-medication". Alterna¬
tively, the Tl depressive symptoms may have been secondary
to substance abuse and not the result of depressive illness.
Further research is required to clarify the relationship among
depression, depressive symptoms, and substance abuse.
Comparison With Other Studies:
Sociodemographic Characteristics
Significant differences among the sites have been reported
previously with other ECA analyses,19,37 and distinguishing
between real site effects and subtle méthodologie discrepan¬
cies has been difficult. Site effects here primarily resulted
from a significantly lower persistence rate at one site (Balti¬
more), and its size suggests that it is an artifact. However,
there was no evidence from additional analyses that this was
due to ECA subject selection bias, interviewer characteris¬
tics, or attrition patterns, although it may be of some rele¬
vance that Baltimore also reported the second lowest rates of
depression prevalence.19 Site was retained as a control vari¬
able for logistic regression analysis, but ultimately, this result
has remained largely unexplained.
Other epidemiologie studies have typically examined rela¬
tionships between major depression prevalence and sociode¬
mographic factors (discussed further below). The recognition
that prevalence is determined jointly by incidence, duration,
and relapse rates has prompted research on the specific roles
 of clinical31,32 and sociodemographic risk factors38,38,39 in these
terms.
Sex differences in depression prevalence are among the
most robust sociodemographic findings.16,18,19 Incidental ex¬
amination of sex effects on outcome have yielded both positive6
and negative4,8,85 results. However, two studies that were
designed to investigate sex differences prospectively have
found a higher recurrence or less favorable outcome in wom-
en 11,88,38-12 jjesuits from our analysis agree that differences in
course may contribute to acknowledged sex differences in
depression prevalence. Higher persistence for women aged
older than 30 years is particularly consistent with an elevated
prevalence in women of intermediate age (approximately 30 to
50 years).18,43
Education also emerged as a significant predictor of out¬
come in this study, although it is still unclear how its effects on
depression are mediated. Analyses here suggest that clinical
factors may be unequally distributed among education
groups, with less educated persons tending to have more and
longer prior episodes of depression. Evidence that there are
important age and sex interactions with education, income,
and employment indicates that their effects are com¬
plex,16,18,19,43 reinforcing the notion that these factors may con¬
tribute to illness prevalence via the different processes of
chronicity and relapse.
Of interest in this study was an assessment of the relative
importance of sociodemographic and clinical factors that influ¬
ence outcome. Logistic models that contained both types of
variables gave no significant effects due to sociodemographic
factors. In contrast, clinical characteristics contributed signif¬
icant prognostic information, and to this extent, major depres¬
sion in this community sample conformed to a medical model of
illness. Given its prognostic significance, the clinical history of
subjects must be considered as a primary control variable in
the design and analysis of future depression outcome studies.
It is perhaps surprising that so few sociodemographic fac¬
tors were significantly associated with persistence. It is possi¬
ble that sociodemographic effects were attenuated by a reduc¬
tion in their variation in the process of sample selection,
although distributions of these variables in the initial sample
did not appear to be skewed. However, expectations of finding
important outcome effects of sociodemographic characteris¬
tics are largely based on previous studies that used samples of
mixed diagnostic composition,33,88,44,46 and the application of
their results specifically to depressive subjects who meet
clinical criteria is problematic. Moreover, the ECA data pro-
References
1. Kraepelin E; Barclay M, trans. Manic-depressive Insanity and Paranoia.
8th ed. London, England: Churchill Livingstone; 1921.
2. Cassano GB, Maggini C, Akiskal HS. Short-term, subchronic, and chronic
sequelae of affective disorders. Psychiatr Clin North Am. 1983;6:55-67.
3. Hirschfeld RMA, Goodwin FK. Mood disorders. In: Talbott JA, Hales RE,
Yudofsky SC, eds. Textbook ofPsychiatry. Washington, DC: American Psychi-
atric Press Inc; 1988;chap 13.
4. Angst J. The course of major depression, atypical bipolar disorder, and
bipolar disorder. Presented at the New Results in Depression Research Sympo-
sium; March 26-28,1985.
5. Bronisch T, Wittchen H-U, Krieg C, Rupp H-U, von Zerssen D. Depres-
sive neurosis: a long-term prospective and retrospective follow-up study of
former inpatients. Acta Psychiatr Scand. 1985;71:237-248.
6. Ceroni GB, Nevi C, Pezzoli A. Chronicity in major depression. J Affective
Disord. 1984;7:123-132.
7. Prien RF, Kupfer DJ, Mansky PA, Small JG, Tuason VB, Voss CB,
Johnston WE. Drug therapy in the prevention of recurrence in unipolar and
bipolar affective disorders. Arch Gen Psychiatry. 1984;41:1096-1104.
8. Keller MB, Shapiro RW. Major depressive disorder: initial results from a
one year prospective naturalistic follow-up study. J Nerv Ment Dis.
1981;169:761-768.
9. Keller MB, Klerman GL, Lavori PW, Coryell W, Endicott J, Taylor J.
Long term outcome in major depression. JAMA. 1984;252:788-792.
10. Surtees PG, Sashidharan SP, Dean C. Affective disorder amongst wom-
en in the general population: a longitudinal study. Br J Psychiatry.
Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Pittsburgh User on 12/06/2015
vide limited social information, and outcome was defined in
clinical terms. In contrast, social psychiatric research has
generally employed more comprehensive assessments of so¬
cial circumstances, such as life events, social supports, and
interpersonal bonds, and often, it has incorporated social
function in outcome measures.13,33,38,4"9
The analyses presented here do not deal with treatment
effects on outcome. Information on help-seeking is available
on ECA subjects, but confounding effects of illness severity
present major difficulties in analysis. How to control for this
properly is an unresolved issue at present although the results
here may be helpful in this regard.
Finally, it is worth noting that despite the number of factors
investigated and associations found, final logistic models did
not classify subjects well: sensitivity and specificity of optimal
models were approximately 17% and 97%, respectively.
Moreover, significant factors were associated with only mod¬
est increases in the likelihood of persistence. Clearly, there is
much variation in outcome not accounted for by variables that
are assumed to determine the course of illness. Improvements
in understanding outcome may require simultaneous incorpo¬
ration of additional parameters,30,50,51 such as genetic loading,
personality, and social environment.
The Epidemiologie Catchment Area Program is a series of five epidemiologie
research studies performed by independent research teams in collaboration
with staff of the Division of Biometry and Epidemiology—reorganized in 1985
with components now in the Division of Clinical Research and the Division of
Biometry and Applied Sciences—of the National Institute of Mental Health,
Rockville, Md. The National Institute of Mental Health Principal Collaborators
are Darrel A. Regier, MD, Ben Z. Locke, MSPH, and Jack D. Burke, Jr, MD;
the National Institute of Mental Health Project Officer was Carl A. Taube, PhD
(1978-1985) and is now William J. Huber (1985- ). The Principal Investigators
and Coinvestigators from the five sites are: Yale University, New Haven,
Conn, grant U01 MH 34224 to Jerome K. Myers, PhD, Myrna M. Weissman,
PhD, and Gary L. Tischler, MD; The Johns Hopkins University, Baltimore,
Md, grant U01 MH 33870 to Morton Kramer, DSc, Ernest Gruenberg, MD, and
Sam Shapiro, MS; Washington University, St Louis, Mo, grant U01 MH 33883
to Lee N. Robins, PhD, and John E. Heizer, MD; Duke University, Durham,
NC, grant U01 MH 35386 to Dan Blazer, MD, and Linda George, PhD;
University of California, Los Angeles, grant U01 MH 35865 to Marvin Karno,
MD, Richard L. Hough, PhD, Javier I. Escobar, MD, M. Audrey Burnam,
PhD, and Dianne M. Timbers, PhD.
Dr Sargeant was supported through a Clinical Fellowship from the Alberta
Heritage Fund for Medical Research Edmonton, Alberta, Canada. Support
from the New York State Psychiatric Institute, and the Research Foundation
for Mental Hygiene, New York, NY, is also acknowledged (Dr Sargeant). This
work was supported in part by grant 86-212 from the John D. and Catherine T.
MacArthur Foundation for Mental Health Research on Risk and Protective
Factors in Major Mental Disorders (Dr Weissman). Computational support was
provided by National Institute of Mental Health grant MH 30906-10 (New York
State Psychiatric Institute).
1986;148:176-186.
11. Amenson CS, Lewinsohn PM. An investigation into the observed sex
difference in prevalence of unipolar depression. J Abnorm Psychol. 1981;90:1-13.
12. Cohen P, Cohen J. The clinician's illusion. Arch Gen Psychiatry.
1984;41:1178-1182.
13.Shapiro S, Skinner EA, Kessler LG, VonKorff M, German PS, Tischler
GL, Leaf PJ, Bernham L, Cottler L, Regier DA. Utilization of health and
mental health services: three epidemiologic catchment area sites. Arch Gen
Psychiatry. 1984;41:971-976.
14. Surtees PG, Dean C, Ingham JG, Kreitman NB, Miller PMcC, Kreitman
NB, Sashidharan SP. Psychiatric disorder from an Edinburgh community:
associations with demographic factors. Br J Psychiatry. 1983;142:238-246.
15. Surtees PG, Miller PMcC, Ingham JG, Kreitman NB, Rennie D, Sashid-
haran SP. Life events and the onset of affective disorder: a longitudinal general
population study. J Affective Disord. 1986;10;37-50.
16. Hirschfeld RMA, Cross CK. Epidemiology of affective disorders. Arch
Gen Psychiatry. 1982;39:35-46.
17. Eaton WW, Kessler LG, eds. Epidemiologic Field Methods in Psychia-
try. Orlando, Fla: Academic Press Inc; 1985.
18. Leaf P, Weissman MM, Myers JK, Holzer CE, Tischler GL. Psychosocial
risks and correlates of major depression in one United States urban community.
In: Barrett JE, Rose RM, eds. Mental Disorder in the Community: Progress
and Challenge. New York, NY: Guilford Press; 1986; chap 4.
19. Weissman MM, Leaf PJ, Tischler GL, Blazer DG, Karno M, Bruce ML,
Florio MP. Affective disorders in five United States communities. Psychol
 Med. 1988;18:141-153.
20. Robins LN, Helzer JE, Croughan J, Ratcliff KS. National Institute of
Mental Health Diagnostic Interview Schedule: its history, characteristics, and
validity. Arch Gen Psychiatry. 1981;38:381-389.
21. Anthony JC, Folstein M, Romanoski AJ, Von Korff MR, Nesadt GR,
Chahal R, Merchant A, Brown CH, Shapiro S, Kramer M, Gruenberg EM.
Comparison of the lay Diagnostic Interview Schedule and a standardized psy-
chiatric diagnosis. Arch Gen Psychiatry. 1985;42:667-675.
22. Helzer JE, Robins LN, McEvoy LT, Spitznagel EL, Stoltzman RK,
Farmer A, Brockington IF. A comparison of clinical and Diagnostic Interview
Schedule diagnoses. Arch Gen Psychiatry. 1985;42:657-666.
23. Wittchen HU, Semler G, von Zerssen D. A comparison of two diagnostic
methods: clinical ICD diagnoses vs DSM-IIIand RDC using the DIS (version 2).
Arch Gen Psychiatry. 1985;42:677-684.
24. Burke JD. Diagnostic categorization by the Diagnostic Interview Sched-
ule: a comparison with other methods of assessment. In: Barrett JE, Rose RM,
eds. Mental Disorder in the Community: Progress and Challenge. New York,
NY: Guilford Press; 1986; chap 13.
25. Myers JK, Weissman MM, Tischler GL, Holzer CE III, Leaf PJ, Orvas-
chel H, Anthony JC, Boyd JH, Burke JD, Kramer M, Stoltzman R. Six-month
prevalence of psychiatric disorders in three communities. Arch Gen Psychiatry.
1984;41:959-967.
26. Fleiss JL, Williams J, Dubro AF. The logistic regression analysis of
psychiatric data. J Psychiatr Res. 1986;20:145-209.
27. Nam CB, LaRoque J, Powers MG, Holmberg J. Occupational status
scores: stability and change. In: Proceedings of the Social Statistics Section of
the American Statistical Association. Washington, DC: American Statistical
Association; 1975:570-575.
28. Akiskal HS, Bitar AH, Puzantian VR, Rosenthal TL, Walker PW. The
nosological status of neurotic depression: a prospective follow-up. Arch Gen
Psychiatry. 1978;35:756-766.
29. Predicting chronicity in depression. Lancet. 1986;2:897-898. Editorial.
30. Scott J. Chronic depression. Br J Psychiatry.
1988;153:287-297.
31. Keller MB, Shapiro RW, Lavori PW, Wolfe N. Recovery in major
depressive disorder: analysis with the life table and regression models. Arch
Gen Psychiatry. 1982;39:905-910.
32. Keller MB, Shapiro RW, Lavori PW, Wolfe N. Relapse in major depressive
disorder: analysis with the life table. Arch Gen Psychiatry. 1982;39:911-915.
33. Mann AH, Jenkins R, Belsey E. The twelve month outcome of patients
with neurotic illness in general practice. Psychol Med. 1981;11:535-550.
34. Van Valkenburg C, Akiskal HS, Puzantian V, Rosenthal T. Anxious
depressions: clinical, family history, and naturalistic outcome\p=m-\comparisons
Downloaded From: http://archpsyc.jamanetwork.com/ by a University of Pittsburgh User on 12/06/2015
with panic and major depressive disorders. J Affective Disord. 1984;6:67-82.
35. Keller MB, Shapiro RW. Double depression, superimposition of acute
depressive episodes on chronic depressive disorders. Am J Psychiatry.
1982;139:438-442.
36. Angst J, Dobler-Mikola A. Do the diagnostic criteria determine the sex
ratio in depression? J Affective Disord. 1984;7:189-198.
37. Robins LN, Helzer JE, Weissman MM, Orvaschel H, Gruenberg E,
Burke JD, Regier DA. Lifetime prevalence of specific psychiatric disorders in
three sites. Arch Gen Psychiatry. 1984;41:949-958.
38. Cooper B, Fry J, Kalton G. A longitudinal study of psychiatric morbidity
in a general practice population. Br J Prev Soc Med 1969;23:210-217.
39. Dunn G, Skuse D. The natural history of depression in general practice:
stochastic models. Psychol Med. 1981;11:755-764.
40. Angst J, Dobler-Mikola A. The Zurich Study: a prospective epidemiologi-
cal study of depressive, neurotic, and psychosomatic symptoms, IV: recurrent
and nonrecurrent depression. Eur Arch Psychiatr Neurol Sci. 1985;234:408\x=req-\
416.
41. Winokur G, Morrison J. The Iowa 500: follow-up of 225 depressives. Br J
Psychiatry. 1973;123:543-548.
42. Frank E, Carpenter LL, Kupfer DJ. Sex differences in recurrent depres-
sion: are there any that are significant? Am J Psychiatry. 1988;145:41-45.
43. Holzer CE, Shea BM, Swanson JW, Leaf PJ, Myers JK, George L,
Weissman MM, Bednarski P. The increased risk for specific psychiatric disor-
ders among persons of low socioeconomic status. Am J Soc Psychiatry.
1986;6:259-271.
44. Sims A, Prior P. The pattern of mortality in severe neuroses. Br J
Psychiatry. 1978;133:299-305.
45. Huxley PJ, Goldberg DP, Maguire GP, Kincey VA. The prediction and
course of minor psychiatric disorders. Br J Psychiatry. 1979;135:535-543.
46. Brown G, Harris T. The Social Origins of Depression. London, England:
Tavistock Publishers Ltd; 1978.
47. Hornstra RK, Klassen D. The course of depression. Compr Psychiatry.
1977;18:119-125.
48. Flaherty JA, Gaviria FM, Black EM, Altman E, Mitchell T. The role of
social support in the function of patients with unipolar depression. Am J
Psychiatry. 1983;140:473-476.
49. Parker G, Blignault I. Psychosocial predictors of me subjects
outcoin
with untreated depressive disorder. J Affective Disord. 1985;8:73-81.
50. Akiskal HS. Factors associated with incomplete recovery in primary
depressive illness. J Clin Psychiatry. 1982;43:266-271.
51. Weissman MM, Prusoff BA, Klerman GL. Personality and the long-term
outcome of depression. Am J Psychiatry. 1978;135:797-800.