Health Psychology Copyright 2001 by the American Psychological Association, Inc.

2001, Vol. 20, No. 2, 112-119 0278-6133/OW5.00 DOI: 10.103W0278-6133.20.2.U2

Psychometric Evaluation of the Beck Depression Inventory-II

With Primary Care Medical Patients
Randolph C. Arnau, Mary W. Meagher, Margaret P. Norris, and Rachel Bramson
Texas A&M University

This study evaluated the psychometric characteristics of the Beck Depression Inventory-II (BDI-II; A. T.
Beck, R. A. Steer, & G. K. Brown, 1996) in a primary care medical setting. A principal-components
analysis with Promax rotation indicated the presence of 2 correlated factors, Somatic-Affective and
Cognitive, which explained 53.5% of the variance. A hierarchical, second-order analysis indicated that
all items tap into a second-order construct of depression. Evidence for convergent validity was provided
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

by predicted relationships with subscales from the Short-Form General Health Survey (SF-20; A. L.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Stewart, R. D. Hayes, & J. E. Ware, 1988). A receiver operating characteristic analysis demonstrated
criterion-related validity: BDI-II scores predicted a diagnosis of major depressive disorder (MOD), as
determined by the Primary Care Evaluation of Mental Disorders (PRIME-MD) Patient Health Question-
naire (PHQ). This study demonstrated that the BDI-II yields reliable, internally consistent, and valid
scores in a primary care medical setting, suggesting that use of the BDI-II in this setting may improve
detection and treatment of depression in these medical patients.

Key words: Beck Depression Inventory-II, psychometrics, primary care, validity,

reliability, depression

Depression is one of the most common psychological disorders ever, despite its widespread use, relatively little is known about the
seen in primary care medical patients, with prevalence rates as psychometric properties of the BDI-n in primary care patients.
high as 23% (Coyne, Fechner-Bates, & Schwenk, 1994; Regier et The BDI-II is a revised version of the 21-item Beck Depression
al., 1993; Zung, 1990). The detection of depression in primary care Inventory (BDI; Beck & Steer, 1993), which assesses the severity
patients is especially important because 50-75% of individuals of depression in adults and adolescents. The BDI-II is an update of
seeking treatment for a depressive disorder present to their physi- the original BDI, which was altered to correspond to criteria from
cian, whereas only 16-23% present to a mental health practitioner the Diagnostic and Statistical Manual of Mental Disorders (DSM-
(Depression Guideline Panel, 1993; Munoz, Hollon, McGrath, IV; American Psychiatric Association, 1994) for major depressive
Rehm, & VandenBos, 1994). Unfortunately, because depression disorder and to improve the content validity of the instrument. For
often is not detected in primary care settings, over half of the example, the revised instrument includes modifications to 17 re-
patients in these settings with major depressive disorder (MDD) sponses, including options for both increases and decreases in
remain untreated (Coyne, Fechner-Bates, & Schwenk, 1995; Mon- appetite, weight, and sleep. In addition, four items were dropped (body
tono, 1994; Vasquez-Barquero, Herran, & Altai, 1997). Patients image change, work difficulty, weight loss, and somatic preoccu-
with untreated depression exhibit significant functional impair- pation) and replaced by four new items (agitation, worthlessness,
ments and higher morbidity and mortality than average, and tend loss of energy, and concentration difficulty). The time frame for
to be high utilizers of medical services (Eisenberg, 1992; Wells et responses was lengthened from 1 week to 2 weeks to be consistent
al., 1989). To increase rates of detection and treatment of depres- with the DSM-IV temporal criterion for a major depressive episode.
sion in primary care settings, primary care providers are turning to The BDI-n has been validated with college students (Beck et al.,
screening instruments (Katon et al., 1997) such as the Beck De- 1996; Dozois, Dobson, & Ahnberg, 1998; Osman et al., 1997),
pression Inventory-H (BDI-H; Beck, Steer, & Brown, 1996). How- adult psychiatric outpatients (Beck et al., 1996; Steer, Ball,
Ranieri, & Beck, 1999), and adolescent psychiatric outpatients
(Steer, Geetha, Ranieri, & Beck, 1998). Although the BDI-n
Randolph C. Arnau, Mary W. Meagher, and Margaret P. Norris, De- demonstrated excellent test-retest reliability, high internal consis-
partment of Psychology, Texas A&M University; Rachel Bramson, Col- tency, and moderate to high convergent validity, the factor struc-
lege of Medicine, Texas A&M University. ture varied across studies. Using exploratory factor analysis on a
This research was funded by grants from Scott & White Hospital, college-student sample, Beck et al. (1996) obtained a two-factor
Temple, Texas to Mary W. Meagher and from the Bush School of Gov- solution involving a Cognitive-Affective factor and a Somatic
ernment and Public Service, College Station, Texas. Portions of this article
factor. The same two-factor solution was found for another
were presented at the 107th Annual Convention of the American Psycho-
logical Association, Boston, Massachusetts, August 24, 1999. We express
college-student sample using both exploratory and confirmatory
our appreciation to Jack Bodden for his help in facilitating this project. factor analyses (Dozois, Dobson, & Ahnberg, 1998). In contrast,
Correspondence concerning this article should be addressed to Mary W. Osman et al. (1997) reported a three-factor model corresponding to
Meagher, Department of Psychology, Texas A&M University, College Negative Attitude, Performance Difficulty, and Somatic clusters in
Station, Texas 77843-4235. Electronic mail may be sent to mwm@ college students. In adult psychiatric outpatients, a two-factor
psyc.tamu.edu. structure corresponding to a Somatic-Affective factor and Cogni-
112
 BDI-H PRIMARY CARE 113

live factor has been consistently obtained with both exploratory involving screening for "stress-related problems," and they were given a
and confirmatory factor analyses (Beck et al., 1996; Steer et al., brief, standardized statement of the purpose and procedures. In the second
1999). Moreover, confirmatory factor analyses of clinically de- stage of recruitment, patients who previously agreed to participate were
pressed adult outpatients suggest that the BDI-II comprises one greeted by a researcher when they arrived for their appointment. The
second-order Depression factor that can be further divided into two researcher provided complete details, answered questions, and distributed
the informed consent form and questionnaire packet.
first-order factors (Somatic-Affective and Cognitive factors; Steer
A total of 1,486 patients were invited to participate; 759 (51%) declined.
et al., 1999). Because these data were collected to address several distinct investigations,
Similarly, Steer et al. (1998) found a second-order factor of the BDI-II was administered to a subset of only 340 patients.
Depression in a sample of adolescent psychiatric outpatients, but
the structure differed from adult outpatients in that three first-order Procedure
factors emerged, rather than the two that have been found in adult
outpatient samples. However, two of the three factors were com- While waiting for their appointments with their physicians, patients gave
parable to the Somatic-Affective and Cognitive factors found in written informed consent and completed the BDI-n, the Primary Care
Evaluation of Mental Disorders (PRIME-MD) Patient Health Question-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

adult outpatient samples; the third factor was composed of only

naire (PHQ; Spitzer, Kroenke, & Williams, 1999), and the SF-20 as part of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

three items and was deemed ungeneralizable.

a larger battery of instruments. Following their appointments, patients
Because no studies, to our knowledge, have investigated the
returned completed questionnaires to a masters-level graduate student who
psychometric characteristics of the BDI-n in primary care medical scored the screening instruments and provided written feedback to the
patients, such an investigation was the purpose of the present patients and their physicians.
study. To assess reliability, we evaluated the internal consistency
of item scores and item-total correlations. Validity was assessed by Participants
examining: (a) the first- and second-order factor structure using
exploratory and hierarchical factor analysis, (b) convergent and The sample for the present study consisted of 340 primary care patients
discriminant validity with the Medical Outcomes Study Short- (68.8% female), 7 of whom were excluded because of missing responses.
Form General Health Survey (SF-20; Stewart, Hayes, & Ware, Participants' ages ranged from 18 to 54 years (M = 36.5, SD = 10.1).
Other available sample demographics are summarized in Table 1. Data
1988) and diagnoses of MDD and (c) receiver operating charac-
teristics analyses to determine criterion-related validity for predict-
ing MDD.
This study also examined the validity of somatic symptoms Table 1
(e.g., disturbed eating and sleeping) as indicators of depression in Sample Demographics
primary care patients. In a sample of psychiatric outpatients, loss
Variable
of energy, sleep disturbance, and appetite disturbance were the
best predictors of MDD (Buchwald & Rudick-Davis, 1993). How- Sex
ever, some have proposed that somatic symptoms of depression Female 229 68.8
may lose diagnostic utility in a medical setting because somatic Male 104 31.2
Race/ethnicity
complaints may be confounded with physical illness (see Clark,
Caucasian 231 69.4
Cavanaugh, & Gibbons, 1983). Empirical studies have consistently African American 23 6.9
demonstrated that for elderly medical patients somatic symptoms Hispanic 22 6.6
are, indeed, good indicators of depression (Koenig, Cohen, Blazer, Asian American/Pacific Islander 4 1.2
Krishnan, & Sibert, 1993; Norris, Snow-Turek, & Blankenship, Other 7 2.1
Unreported 46 13.8
1995; Norris & Woehr, 1998). Therefore, we expected that the Education
diagnostic utility of somatic symptoms would also be demon- Some high school 12 3.6
strated in younger adult primary care patients. High school or GED test 62 16.6
Some college 82 24.6
College degree 50 15.0
Method Some graduate school 19 5.7
Graduate degree 60 18.0
Study Setting Unreported 48 14.4
Household income
The setting for this study was a large staff-model health maintenance Under $10,000 21 6.3
organization (HMO) in a suburban community in southeast Texas. Partic- $10-20,000 36 10.8
ipants were recruited from the family medicine clinic, which is staffed $21-30,000 60 18.0
by 15 board-certified family physicians. Approximately 60% of the pa-
$31-50,000 80 24.0
$50-80,000 61 18.3
tients presenting to this clinic subscribed to the health plan of the staff- Over $80,000 21 6.3
model HMO, and the remaining approximately 40% either subscribed to Unreported 54 16.2
other health plans or were self-pay patients. Marital status
Single 56 16.8
Married 208 62.5
Recruitment of Participants Separated 3 0.9
Divorced 20 6.0
All patients over age 18 who presented to the clinic for an appointment Widowed 2 0.6
with their primary care physician were eligible. Recruitment was two- Unreported 44 13.2
staged. First, eligible patients were contacted by telephone the day before
their appointment. They were informed that a study was being conducted Note. GED = general education development.
 114 ARNAU, MEAGHER, NORRIS, AND BRAMSON

collected for older adults (from 54 to 86 years of age) were not included in Item-total correlations. The item Ms, SDs, and corrected item-
the present study.1 total correlations are presented in Table 2. The item-total correla-
Because of the number of patients who declined to participate, we tions ranged from .54 to .74, indicating good internal consistency.
evaluated whether participants and decliners differed on several dimen-
sions. Some demographic data (gender and age) and health and medical Factorial Validity
utilization data (total number of diagnoses received in the past year and
total number of visits to a doctor in the past year) were available for 210 First-order analysis. Because little factor-analytic research on
of the decliners. Analyses of these data indicated that there were no the original BDI has been conducted in a medical setting and none
statistically significant differences between participants and decliners for has been conducted on the BDI-II in a medical setting, the factor
any of these variables. structure of BDI-II scores was examined using exploratory factor
analysis. Although the factor structure was expected to be similar
Measures to that of the BDI and the BDI-II in nonmedical settings, no
BDI-ll. The BDI-n2 is a 21-item self-report measure of the severity of
explicit structure was hypothesized a priori for this sample.
depressive symptomatology. Each of the 21 items is rated on a 4-point A principal-components analysis of the BDI-n item correla-
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

scale ranging from 0-3. The ratings are summed, yielding a total score that tions was conducted. We based the decision about the number of
This document is copyrighted by the American Psychological Association or one of its allied publishers.

can range from 0-63. factors to retain on (a) parallel analysis (Horn, 1965) and (b) the
SF-20. We assessed self-reported health and functioning using the interpretability and theoretical salience of the rotated factors. Par-
SF-20, which is a 20-item self-report instrument that measures the effects allel analysis has consistently been shown to be superior to other
of health on physical and role functioning, bodily pain, general health factor retention rules in terms of extracting the correct number of
perceptions, and mental health. Lower scores reflect poorer perceived factors in Monte Carlo studies (Crawford & Koopman, 1973;
health and functioning. Zwick & Velicer, 1986). In the present study, parallel analysis
PHQ. The PHQ is a self-report instrument designed to be used in
indicated that two factors should be retained. Furthermore, when
primary care settings to diagnose a number of DSM-IV disorders, including
MDD, panic disorder, and bulimia nervosa. For the purposes of the present
the two factors were rotated, they were theoretically salient and
study, the depression module of the PHQ was used to identify participants accounted for a total of 53.5% of the variance.
with MDD. Because both factors are theorized to tap into depression, we
The PHQ is a revised version of the PRIME-MD Patient Problem hypothesized that the factors would be correlated. Therefore, we
Questionnaire (PPQ), which was developed as a structured interview. The subjected the two factors to a Promax rotation (Hendrickson &
original PPQ was validated in a study involving 1,000 primary care patients White, 1964), which allows correlated factors and generally yields
and was subsequently revised into the self-report version called the PHQ, good simple structure (Gorsuch, 1983).
which was intended to increase the clinical and research utility of the As expected, the rotated factors were correlated (r = .70). The
original PPQ. factor-pattern coefficients are presented in Table 3, along with the
The diagnostic accuracy of the PHQ was recently assessed with a sample
communalities (h2) of the measured variables. Factor pattern co-
of 550 primary care patients. Within 48 hr of administration of the PHQ,
a mental health professional, who had not been informed of the PHQ
efficients of .40 or greater were considered salient (Stevens, 1996,
results, contacted patients and determined diagnoses using a structured p. 372). The first factor was labeled Somatic-Affective, given the
interview. This analysis demonstrated a 93% agreement between a PHQ salient pattern and structure coefficients for measured variables
diagnosis of MDD and diagnosis by the mental health professional making up that component, such as tiredness, sleep problems,
(Spitzer, Kroenke, & Williams, 1999). problems with appetite, sadness, and loss of pleasure. The second
factor was labeled Cognitive, given the salient pattern and struc-
Results ture coefficients for the measured variables making up that com-
ponent, such as suicidality, pessimism, worthlessness, and guilt.
Prevalence and Severity of Depression Symptoms One item, self-criticalness, did not make a substantial contribution
to either factor, with a pattern coefficient of .37 for the Somatic-
The average BDI-n score was 8.74 (SD = 9.7). Scores of 14 or Affective component, and .32 for the Cognitive component.
higher, suggesting at least a mild level of depression (Beck et al., Whenever factors are correlated, structure coefficients (correla-
1996), were observed in 23.2% of the sample. tions of the measured variables with the extracted components) are
also important aids to interpretation (Thompson, 1997; Thompson
Socioeconomic Status and Gender Differences & Borrello, 1985). Therefore, the structure coefficients are pre-
sented in Table 4. The large structure coefficients for all measured
As expected, there was a small relationship between BDI-II variables on both components are consistent with the high corre-
scores and socioeconomic status variables. Specifically, BDI-II
lation between the rotated components. It is also important to note
scores were inversely related to level of education (r = — .24, p <
that the structure coefficients for the self-criticalness item were
.01) and household income (r = —.20, p < .01). In addition, there
large, indicating this item correlates with both first-order factors
was an expected gender difference in BDI-II scores. The mean
and thus is a relevant symptom of depression, despite the low
BDI-II score for women, M = 9.88, SD = 10.33, was higher than pattern coefficients.
the mean score for men, M = 5.82, SD = 6.91; f(338) = 3.70,
p < .001.
1
These data were analyzed and reported separately to examine different
Reliability hypotheses than those tested in the present study (see Morris, Meagher, &
Arnau, 1999).
Internal consistency. The internal consistency of BDI-II re- 2
The BDI-II can be ordered from The Psychological Corporation, P.O.
sponses was excellent, with an alpha coefficient of .94. Box 839954, San Antonio, TX 78282-3954.
 BDI-II PRIMARY CARE 115

Table 2 Table 4
Means, Standard Deviations, and Corrected Item-Total Structure Matrix for the BDI-H
Correlations of the BDl-Il
Factor
Item M SD r
to.
Item 1 2
Sadness .28 .54 .68
Pessimism .35 .61 .66 Sadness .699 .653
Past Failure .41 .70 .69 Pessimism .538 .779
Loss of Pleasure .44 .68 .69 Past Failure .584 .799
Guilty Feelings .40 .59 .62 Loss of Pleasure .701 .664
Punishment Feelings .23 .65 .54 Guilty Feelings .561 .623
Self-Dislike .43 .74 .74 Punishment Feelings .418 .705
Self-Criticalness .41 .69 .62 Self-Dislike .691 .721
Suicidal Thoughts .08 .31 .54 Self-Criticalness .594 .576
Crying .32 .72 .67 Suicidal Thoughts .352 .728
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Agitation .41 .61 .58 Crying .594 .679

This document is copyrighted by the American Psychological Association or one of its allied publishers.

Loss of Interest .40 .67 .73 Agitation .671 .467

Indecisiveness .31 .68 .73 Loss of Interest .740 .680
Worthlessness .22 .56 .68 Indecisiveness .740 .627
Loss of Energy .70 .67 .64 Worthlessness .578 .784
Changes in Sleeping .77 .95 .58 Loss of Energy .753 .444
Irritability .48 .78 .67 Changes in Sleeping .671 .412
Changes in Appetite .53 .83 .60 Irritability .733 .572
Concentration Difficulty .46 .67 .68 Changes in Appetite .655 .453
Tiredness .65 .72 .67 Concentration Difficulty .747 .559
Loss of Interest in Sex .48 .80 .57 Tiredness .794 .443
Loss of Interest in Sex .666 .451
Note. BDI-II = Beck Depression Inventory-H.
Note. BDI-II = Beck Depression Inventory-II. Factor 1 = Somatic-
Affective, Factor 2 = Cognitive.
Second-order analysis. The correlation of the rotated factors
(r = .70) implied a higher level of conceptualization, which should
correspond to a second-order factor of depression. For the second-
order analysis, a principal-components analysis of the first-order was conducted. One second-order factor was extracted, yielding
factor correlation matrix (in this case, a single correlation of .70) pattern coefficients of .92 for both of the first-order factors loading
onto the second-order factor.
As an aid to interpretation of the second-order analysis, a
Table 3 Schmid-Leiman orthogonalized solution (Schmid & Leiman,
Rotated Factor Pattern Matrix for the BDI-II 1957) was computed, as recommended by Gorsuch (1983). The
Factor Schmid-Leiman transformation yields a solution in which (a) the
second-order factors are expressed in terms of the measured vari-
Item 1 2 h2 ables (rather than in terms of the first-order factors) and (b) the
common variance accounted for by the second-order factors is
Sadness .474 .323 54.3
Pessimism -.007 .784 60.7 residualized from the first-order factors. In other words, the pattern
Past Failure .055 .761 64.0 coefficients of the variables for the first-order factors represent the
Loss of Pleasure .464 .341 55.2 unique variance that is accounted for by the first-order factors and
Guilty Feelings .247 .451 41.9
-.140 .802 50.7
is not accounted for by the second-order factors (see Arnau, 1998;
Punishment Feelings
Self-Dislike .367 .466 59.0 Gorsuch, 1983; Thompson, 1990; Thompson & Borrello, 1992).
Self-Criticalness .373 .317 40.4 The Schmid-Leiman solution for the present analysis was com-
Suicidal Thoughts -.299 .936 57.6 puted using Interactive Matrix Language (IML; SAS Institute,
Crying .235 .516 49.0
Agitation .670 .001 45.0 1990) and is presented in Table 5. All of the measured variables
Loss of Interest .517 .321 60.0 from the BDI-II contributed a noteworthy amount of variance to
Indecisiveness .589 .217 57.1 the second-order depression factor, with pattern coefficients rang-
Worthlessness .062 .741 61.7
ing from .59 to .77.
Loss of Energy .861 -.155 58.0
Changes in Sleeping .744 -.105 45.6 It is also apparent from Table 5 that very little variance in the
Irritability .650 .119 54.5 first-order factors is not accounted for by the second-order factor.
Changes in Appetite .658 -.005 42.9 This is evidenced from the magnitudes of the pattern coefficients
Concentration Difficulty .694 .076 56.0
.942 -.212 65.4 in the third and fourth columns of Table 5, which represent the
Tiredness
Loss of Interest in Sex .683 -.024 44.4 unique variance still accounted for by the first-order factors after
being residualized of the variance captured by the second-order
Note. Pattern coefficients with values of .40 or greater are in boldface. factor. None of the first-order coefficients retained salience after
Factor 1 = Somatic-Affective, Factor 2 = Cognitive, h2 = Communalities
of the measured variables; figures in this column represent %s. BDI-II = being residualized of the variance accounted for by the second-
Beck Depression Inventory-II. order factor.
 116 ARNAU, MEAGHER, MORRIS, AND BRAMSON

Table 5 depressive diagnosis group averaging 6.7 (SD = 7.1) and the
Schmid-Leiman Orthogonalized Solution depressive diagnosis group averaging 28.0 (SD = 9.7), a differ-
ence that is noteworthy and statistically significant; f(333) = 15.3,
Factor p < .001.
Item Depression 1 2
Receiver Operating Characteristics (ROC) Analysis
Sadness .734 .185 .126
Pessimism .715 -.003 .306 Given the mean BDI-II score difference between patients with and
Past Failure .751 .021 .297 without MDD, it was deemed useful to evaluate how well BDI-II
Loss of Pleasure .741 .181 .133 scores predict the presence of MDD. When used as a screening
Guilty Feelings .643 .096 .176
Punishment Feelings .610 -.055 .313 measure, the BDI-II could alert primary care providers that a
Self-Dislike .767 .143 .182 patient should receive further evaluation for a depressive disorder.
Self-Criticalness .635 .145 .124 ROC analysis is specifically designed to provide an index of the
Suicidal Thoughts .587 -.117 .365 diagnostic accuracy of a screening instrument (Rey, Morris-Yates,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Crying .692 .092 .201

This document is copyrighted by the American Psychological Association or one of its allied publishers.

& Stanislaw, 1992). The results of the analysis can also be used to
Agitation .618 .261 .000
Loss of Interest .772 .202 .125 determine an optimal cutoff score. For any given cutoff score,
Indecisiveness .742 .230 .085 there are several indices of diagnostic efficacy. The true positive
Worthlessness .739 .024 .289 rate (sensitivity) refers to the proportion of individuals with the
Loss of Energy .650 .336 -.060 disorder who are identified as such by the screen (i.e., score above
Changes in Sleeping .588 .290 -.041
Irritability .708 .253 .046 the cutoff), whereas the false positive rate is the proportion of
Changes in Appetite .602 .257 -.002 those scoring above the cutoff who actually do not have the
Concentration Difficulty .709 .271 .029 disorder. The true negative rate (specificity) is the proportion of
Tiredness .672 .367 -.083 individuals without the disorder who are identified as such by the
Loss of Interest in Sex .607 .266 -.009
screen (i.e., score below the cutoff), and the false negative rate is
Note. The Depression column represents the second-order factor. The the proportion of those scoring below the cutoff who actually have
Factor 1 and Factor 2 columns represent the first-order solution, based on the disorder.
variance orthogonal to the second-order factor (Gorsuch, 1983). Coeffi- Two other important indices derived from an ROC analysis are
cients with an absolute value of .40 or greater are in boldface. Factor 1 = positive predictive power (PPP) and negative predictive power
Somatic-Affective, Factor 2 = Cognitive. Communalities (7i2) for the
second-order solution are the same as those of the first-order solution (NPP). PPP is the ratio of individuals screening positive who
presented in Table 2. actually do have the disorder to the total number of individuals
who screen positive (true positives + false positives). NPP is the
ratio of individuals who screen negative who do not have the
Convergent Validity disorder to the total number of individuals who screen negative
(true negatives + false negatives).
The convergent validity of the BDI-II scores was examined
In an ROC analysis, the sensitivity and specificity associated
through correlations of BDI-JI total scores and factor scores with
with every possible cutoff score are calculated and plotted, with
SF-20 subscale scores, including the Mental Health, Perception of
sensitivity on the x axis and specificity on the y axis. The area
Overall Health, Pain, Physical Functioning, and Role Functioning
under the curve (AUC) is an index of the amount of diagnostic
subscales. The correlations between BDI-II total scores and factor
information provided by an instrument (see Hanley & McNeil,
scores with the SF-20 subscales are presented in Table 6. When
1982; Swets, 1988) and can range from 0.0 (no diagnostic infor-
interpreting these results, it is important to note that higher scores
mation) to 1.0 (perfect diagnostic accuracy). The diagonal along
on the SF-20 indicate greater functioning and positive health
the middle of the curve represents the diagnostic information that
perceptions, so negative relationships between these scales and
BDI-II scores indicate higher depression severity associated with
lower functioning and lower perceptions of health. The correla-
Table 6
tions between the total BDI-II scores and the SF-20 subscales
ranged from —.19 for the Physical Functioning subscale to —.65 Convergent Validity Coefficients of the BDI-II
for the Mental Health subscale. As expected, the BDI-II total and With SF-20 Subscales
factor scores correlated more strongly with the Mental Health BDI-II scores
subscale than with any of the other SF-20 subscales.
Total
Criterion-Related Validity SF-20 subscale score Factor 1 Factor 2

Because the BDI-II purportedly measures severity of depressive Mental Health -.65 -.61 -.59
Health Perceptions -.42 -.43 -.34
symptomatology, there should be a difference in mean BDI-II Physical Pain -.24 -.24 -.21
scores between a group of patients with a diagnosis of MOD Physical Functioning -.19 -.18 -.18
versus a group with no depressive disorder diagnosis. As men- Role Functioning -.35 -.35 -.31
tioned previously, DSM-IV diagnostic criteria were assessed with
the PHQ. Using this criterion, 31 patients obtained a DSM-IV Note. Higher scores on SF-20 subscales indicate better health and greater
functioning. BDI-II = Beck Depression Inventory-II. Factor 1 = Somatic-
diagnosis of MOD and 304 patients did not. The mean BDI-II Affective, Factor 2 = Cognitive. All correlations are statistically signifi-
scores differed substantially across the two groups, with the non- cant at the p < .01 level.
 BDI-II PRIMARY CARE 117

could be expected by chance, and represents an AUC of .50. An the factor structure we found for primary care medical patients
AUC of .80 or higher indicates a useful screening instrument differed from that of college students, which have typically yielded
(Holmes, 1998). a Cognitive-Affective factor and a Somatic factor (Beck et al.,
The computer program Simstat (Prevails Research, 1996) was 1996; Dozois et al., 1998). Therefore, it seems that the structure of
used to conduct an ROC analysis with BDI-n scores predicting the depression in primary care medical patients is similar to that of
diagnosis of MDD (determined by the PHQ, as described previ- both adult and adolescent psychiatric outpatients.
ously). The AUC was .96, indicating that the BDI-II exhibits As mentioned previously, a few items in our sample demon-
excellent performance as a screen for MDD. strated different factor patterns than those found for a psychiatric
The ROC analysis indicated that using a BDI-II cutoff score outpatient sample reported by Beck et al. (1996). In our sample,
of 18 yielded the best balance between sensitivity and specificity. sadness loaded most highly on the Somatic-Affective factor,
Table 7 depicts the percentages of true positives (TP), true nega- whereas this item loaded on the Cognitive factor in the psychiatric
tives (TN), false positives (FP), and false negatives (FN) resulting outpatient sample. However, sadness is probably best conceptual-
from this cutoff score in the present sample. This classification ized as consisting of affective and cognitive components, and thus
distribution yielded a sensitivity of 94% and a specificity of 92%,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

might be expected to contribute to an affective or a cognitive factor.

This document is copyrighted by the American Psychological Association or one of its allied publishers.

with an overall correct classification rate of 92%. In addition, this Consistent with our findings, crying contributed to the Cognitive
yielded a PPP of 54% and an NPP of 99%. These values illustrate factor in the psychiatric sample studied by Steer et al. (1999),
the fact that although false positive results were produced, almost although crying was mainly associated with the Somatic-Affective
all of those who actually had MDD were identified as such (i.e., factor in the Beck et al. (1996) sample. These differences are con-
scored above the cutoff). There were very few FN (i.e., patients sistent with the hypothesis of Beck et al. (1996) that affective
with MDD who were not identified as such). symptoms may be located in different dimensions depending on the
Although 18 is the cutoff recommended for the best balance background and diagnostic composition of the sample under study.
between sensitivity and specificity, one might opt for either higher
The present study by presenting results of a second-order factor
sensitivity or higher specificity. The ROC analyses indicated that
analysis in a primary care medical setting, expands the literature on
if higher sensitivity is desired (fewer FN), a cutoff of 10 is recom-
the BDI-II. The Schmid-Leiman orthogonalized solution indicated
mended, which yields a sensitivity of 100% and specificity of 70%.
one second-order factor made up of strong contributions by every
If higher specificity is desired (fewer FP), a cutoff of 25 is recom-
item on the BDI-II. This gave stronger evidence for the construct
mended, which yields a sensitivity of 58% and specificity of 97%.
validity of the BDI-II than could be provided by a first-order
analysis alone because the second-order analysis produced a more
Discussion accurate model of what the BDI-II is intended to measure (i.e., a
This study evaluated the reliability and validity of BDI-II scores single construct of depression). Specifically, the second-order
in a primary care medical setting. The findings indicate that the analysis indicated that the instrument measures distinct, albeit
BDI-II yields reliable and valid scores for assessing depression in related, factors of depression that both tap into a higher-order
this setting. With regard to reliability, BDI-II scores demonstrated factor that can be called depression. Furthermore, the first-order
excellent internal consistency. In addition, corrected item-total factors account for only a small amount of variance after being
correlations ranged from .54 to .74, which are somewhat higher residualized of the variance accounted for by the higher-order
than those reported by Osman et al. (1997; range = .44 to .65) and factor of depression.
Dozois et al. (1998; range = .41 to .62). The strong correlations of BDI-n scores with the Mental Health
This study also yielded strong evidence for the factorial validity subscale of the SF-20 provided evidence for the convergent valid-
of the BDI-II in a primary care setting. Principal-components ity of BDI-II scores. In addition, BDI-H scores were moderately
analysis suggested that two factors (Somatic-Affective and Cog- correlated with SF-20 subscales corresponding to general percep-
nitive) summarized the data parsimoniously. Although a few in- tions of health and social-role functioning. Consistent with other
dividual items loaded differently, the two extracted factors were studies, depression was associated with physical functioning and
the same factors found for adult psychiatric outpatients (Beck et physical pain (Romano & Turner, 1985; Williamson & Schulz,
al., 1996), and comparable to the first two factors found using 1992). In addition, noteworthy and statistically significant differ-
adolescent psychiatric outpatients (Steer et al., 1998). Conversely, ences in total BDI-U scores across groups of patients who were
and were not diagnosed with MDD demonstrated discriminant
validity for the BDI-II.
Table 7 The ROC analysis indicated that a cutoff score of 18 yielded the
Contingency Table Obtained When Using a Cutoff Score best balance between sensitivity and specificity for predicting
of 18 on the BDI-II MDD in this setting, which led to an overall rate of correct
classification of 92%. These results are particularly positive given
BDI-II score the widespread call for psychological screening in primary care
Diagnosis Score > 18 Score < 18
medical settings due to the underdetection of depression in such
settings (Katon et al., 1997; Montono, 1994). Results from the
MDD n = 29 n =2 ROC analysis indicate that the BDI-II is a very sensitive and
(true positives) (false negatives) moderately specific screen for depression in a primary care med-
No MDD n = 25 n = 279
(false positives) (true negatives)
ical setting. Because the BDI-U takes only 5-10 min to complete
and is easily scored, it could be integrated into a primary care
Note. MDD = major depressive disorder. setting without slowing patient flow. Patients scoring above a
 118 ARNAU, MEAGHER, NORRIS, AND BRAMSON

cutoff score could be flagged for follow-up by the physician or items in medical settings (Karanci, 1988). An instrument has
mental health provider. actually been developed for use in medical settings that is based on
It is important to note that the PPP (.54) indicated that the false the BDI but excludes somatic items (see Beck, Steer, Ball, Ciervo,
positive rate is almost as high as the true positive rate. In this sample, & Kabat, 1997). Further research has suggested that some somatic
for patients scoring at or above the cutoff of 18, there was a 54% symptoms (e.g., diminished energy, appetite and sleep disturbance,
chance that the patient had MDD. However, a lower PPP than NPP and health worries) are accurate indices of depression, whereas
is not inconsistent with the main goal of screening in primary care other somatic symptoms (e.g., work inhibition, weight loss, loss of
settings, which is to increase the rate of detection. The results of interest in sex) are not consistent with overall depression (Morris &
this analysis indicate that the BDI-II missed only 1% of patients Woehr, 1998). These latter items were deleted in the BDI revision.
with MDD; in contrast, previous studies have shown that without Several findings in the present study challenge the argument that
screening, sometimes over half the patients with MDD in primary somatic items should be excluded from measures of depression
care settings go untreated (Coyne et al., 1995; Montono, 1994; used in medical settings. First, if scores from somatic items are less
Vasquez-Barquero et al., 1997). The practical implication of rel- valid indicators of depression for a medical patient sample, then
atively lower PPP is that patients scoring positive on screens corrected item-total correlations should be lower for somatic items,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

should be flagged for further evaluation by a physician or mental which was not the case. Second, the predicted relationships be-
health practitioner to determine if they should be treated for tween SF-20 subscales and BDI-II factor scores did not vary across
depression. The present study suggests that the BDI-II is an excellent factors including or not including somatic items. Third, all items
screen for identifying patients who may need treatment for MDD. from the BDI-II (including somatic items) made salient contribu-
tions to the second-order factor of depression. Therefore, regard-
Study Limitations and Future Research less of whether somatic items on the BDI-n are sometimes con-
founded with physical illness, they still make salient contributions
One limitation of the present study is its use of a self-report
to a second-order factor of depression and are covarying with other
diagnostic measure (the PHQ) as the gold standard against which
cognitive and affective symptoms of depression.
to assess the diagnostic utility of the BDI-II. Although the PHQ
Fourth, results from the ROC analysis indicated that the BDI-H
has been found to have very strong agreement with independent
total score, which includes somatic items, is quite efficacious for
clinician-rated diagnoses (Spitzer et al., 1999), two such scores
predicting the presence of MDD. Our finding of 92% overall
that are obtained by the same method (in this case, self-report)
accuracy in detecting MDD is strong evidence that the inclusion of
almost always have some degree of correlation. Therefore, the
somatic items in the total score did not detract from the predictive
relationships between BDI scores and MDD diagnoses reported by
utility of the BDI-H scores. In fact, in comparison to the nonso-
the present study may be somewhat inflated due to shared method
matic item BDI evaluated by Beck et al. (1997), the specificity was
variance. Future studies should address this issue by using a
nearly identical, and the sensitivity was higher in the present study.
structured interview, such as the Structured Clinical Interview for
Finally, correlations between the SF-20 subscales and BDI-II
DSM-TV (SCID-IV; Spitzer, Williams, Gibbon, & First, 1994) or
factor scores revealed that the Somatic-Affective factor, as well as
Schedule for Affective Disorders and Schizophrenia (SADS; En-
the Cognitive factor, correlated least with the SF-20 Physical Pain
dicott & Spitzer, 1978).
and Physical Functioning subscales, suggesting neither factor is
Future research should also study the BDI-II's responsiveness to
confounded with physical illness.
change in depressive symptomatology. Although the differences in
These are five strong lines of evidence that somatic items should
BDI-II scores between people with and without MDD suggest that
be included in measures of depression in medical settings.
lower scores would follow treatment or spontaneous recovery, the
cross-sectional data of the present study do not allow for a defin-
itive conclusion. Therefore, longitudinal research is needed to Conclusion
study the validity of the BDI-II as a treatment outcome measure.
Some potential drawbacks to using the BDI-n should be men- The present study suggests that the BDI-II yields reliable and
tioned. One is the cost of the BDI-II relative to some other valid scores when used in a primary care setting. More extensive
measures of depression. Because the BDI-n must be purchased, it use of the BDI-H may improve detection, treatment, and under-
could be quite expensive to use for large-scale screening. Another standing of depression in primary care medical patients.
potential drawback could arise if phone administration were de-
sired, such as for treatment follow-up. The response format of the References
BDI-II requires respondents to choose among four statements, a
format that may not fit well with phone interviews, although the American Psychiatric Association. (1994). Diagnostic and statistical man-
ual of mental disorders (4th ed.). Washington, DC: Author.
specific phrases associated with each response option probably add
Arnau, R. C. (1998, April). Second-order factor analysis: Methods and
to the information obtained. interpretation. Paper presented at the annual meeting of the Southwest-
ern Psychological Association, New Orleans, LA. (ERIC Document
Use of Somatic Items in a Medical Setting Reproduction Service ED 417 216).
Beck, A. T., & Steer, R. A. (1993). Manual for the Beck Depression
A possible criticism of using the BDI-II in a medical setting is
Inventory. San Antonio, TX: Psychological Corporation.
that the BDI-n includes somatic items, which may be confounded Beck, A. T., Steer, R. A., Ball, R., Ciervo, C. A., & Kabat, M. (1997). Use
with physical illness rather than tapping into a depressive syn- of the Beck Anxiety and Depression Inventories for primary care with
drome (Clark et al., 1983). In another study, a somatic-vegetative medical outpatients. Assessment, 4, 211-219.
factor extracted from the BDI was correlated with illness severity Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the Beck
and age, which was interpreted as support for excluding somatic Depression Inventory-II. San Antonio, TX: Psychological Corporation.
 BDI-II PRIMARY CARE 119

Buchwald, A. M., & Rudick-Davis, D. (1983). The symptoms of major & Chiros, C. E. (1997). Factor structure and psychometric characteristics
depression. Journal of Abnormal Psychology, 102, 197-205. of the Beck Depression Inventory-n. Journal of Psychopathology &
Clark, D. C., Cavanaugh, S. V., & Gibbons, R. D. (1983). The core Behavioral Assessment, 19, 359-376.
symptoms of depression in medical and psychiatric patients. The Journal Provalis Research. (1996). Simstat for Windows [Computer Software].
of Nervous and Mental Disease, 171, 705-713. Montreal, Quebec, Canada: Author.
Coyne, J. C., Fechner-Bates, S., & Schwenk, T. L. (1994). The prevalence, Regier, D. A., Narrow, W. E., Rae, D. S., Manderscheid, R. W., Locke,
nature, and co-morbidity of depressive disorders in primary care. Gen- B. Z., & Goodwin, F. K. (1993). The de facto U.S. Mental and Addictive
eral Hospital Psychiatry, 16, 267-276. Disorders Service System: Epidemiologica! Catchment Area prospective
Coyne, J. C., Fechner-Bates, S., & Schwenk, T. L. (1995). Nondetection of 1-year prevalence rates of disorders and services. Archives of General
depression by primary care physicians reconsidered. General Hospital Psychiatry, 50, 85-94.
Psychiatry, 16, 267-276. Rey, J. M., Morris-Yates, A., & Stanislaw, H. (1992). Measuring the accuracy
Crawford, C. B., & Koopman, P. (1973). A note on Horn's test for the of diagnostic tests using receiver operating characteristics (ROC) analysis.
number of factors in factor analysis. Multivariate Behavioral Re- International Journal of Methods in Psychiatric Research, 2, 39-50.
search, 8, 117-125. Romano, J. M., & Turner, J. A. (1985). Chronic pain and depression: Does
Depression Guideline Panel. (1993). Depression in primary care (Vol. 1: the evidence support a relationship? Psychological Bulletin, 97, 18-34.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Detection and Diagnosis, Clinical Practice Guideline, No. 5). Rockville, SAS Institute. (1990). SAS/IML software: Usage and reference (Version 6,
This document is copyrighted by the American Psychological Association or one of its allied publishers.

MD: U.S. Department of Health and Human Services, Public Health First ed.). Cary, NC: Author.
Service, Agency for Health Care Policy and Research. (AHCPR Publi- Schmid, J., & Leiman, J. M. (1957). The development of hierarchical factor
cation No. 93-0550). solutions. Psychometrika, 22, 53-61.
Dozois, D. J. A., Dobson, K. S., & Ahnberg, J. L. (1988). A psychometric Spitzer, R. L., Kroenke, K., & Williams, J. B. W. (1999). Validation and
evaluation of the Beck Depression Inventory-II. Psychological Assess- utility of a self-report version of the PRIME-MD: The PHQ primary care
ment, 10, 83-89. study. JAMA, 282, 1737-1744.
Eisenberg, L. (1992). Treating depression and anxiety in primary care: Spitzer, R. L., Williams, J. B., Gibbon, M., & First, M. B. (1994).
Closing the gap between knowledge and practice. New England Journal Structured Clinical Interview for DSM-JV Axis I Disorders—Patient
of Medicine, 326, 1080-1084. Edition. New York: New York State Psychiatric Institute.
Endicott, J., & Spitzer, R. L. (1978). A diagnostic interview: The Schedule Steer, R. A., Ball, R., Ranieri, W. F., & Beck, A. T. (1999). Dimensions of
for Affective Disorders and Schizophrenia. Archives of General Psychi- the Beck Depression Inventory-n in clinically depressed outpatients.
atry, 35, 837-844. Journal of Clinical Psychology, 55, 117-128.
Gorsuch, R. L. (1983). Factor analysis (2nd ed.). Hillsdale, NJ: Erlbaum. Steer, R. A., Geetha, K., Ranieri, W. F., & Beck, A. T. (1998). Use of the
Hanley, J. A., & McNeil, B. J. (1982). The meaning and use of the area under Beck Depression Inventory-n with adolescent psychiatric outpatients.
a receiver operating characteristic (ROC) curve. Radiology, 143, 29-36. Journal of Psychopathology and Behavioral Assessment, 20, 127-137.
Hendrickson, A. E., & White, P. O. (1964). Promax: A quick method for Stevens, J. (1996). Applied multivariate statistics for the social sciences
rotation to oblique simple solution. British Journal of Statistical Psy- (3rd ed.). Mahwah, NJ: Erlbaum.
chology, 17, 65-70. Stewart, A. L., Hayes, R. D., & Ware, J. E. (1988). The MOS short-form
Holmes, W. C. (1998). A short, psychiatric, case-finding measure for HIV general health survey: Reliability and validity in a patient population.
seropositive outpatients. Medical Care, 36, 237-243. Medical Care, 26, 724-735.
Horn, J. L. (1965). A rationale and test for the number of factors in factor Swets, J. E. (1988). Measuring the accuracy of diagnostic systems. Science,
analysis. Psychometrika, 30, 179-185. 240, 1285-1293.
Karanci, N. A. (1988). Patterns of depression in medical patients and their Thompson, B. (1990). SECONDOR: A program that computes a second-
relationship with causal attributions for illness. Psychotherapy & Psy- order principal components analysis and various interpretation aids.
chosomatics, 50, 207-215. Educational and Psychological Measurement, 50, 575-580.
Katon, W., Von Korff, M., Lin, E., Unutzer, J., Simon, G., Walker, E., Thompson, B. (1997). The importance of structure coefficients in structural
Ludman, E., & Bush, T. (1997). Population-based care of depression: equation modeling confirmatory factor analysis. Educational and Psy-
Effective disease management strategies to decrease prevalence. Gen- chological Measurement, 57, 5—19.
eral Hospital Psychiatry, 19, 169-178. Thompson, B., & Borrello, G. M. (1985). The importance of structure
Koenig, H. G., Cohen, H. J., Blazer, D. G., Krishnan, K. R. R., & Sibert, coefficients in regression research. Educational and Psychological Mea-
T. E. (1993). Profile of depressive symptoms in younger and older surement, 45, 203-209.
medical inpatients with major depression. Journal of the American Thompson, B., & Borrello, G. M. (1992). Measuring second-order factors
Geriatrics Society, 41, 1169-1176. using confirmatory methods: An illustration with the Hendrick-Hendrick
Montono, C. B. (1994). Recognition and treatment of depression in a Love Instrument Educational and Psychological Measurement, 52,69-77.
primary care setting. Journal of Clinical Psychiatry, 55, 18—34. Vasquez-Barquero, J. L., Herran, A., & Altai, J. (1997). Epidemiological
Mufloz, R. F., Hollon, S. D., McGrath, E., Rehm, L. P., & VandenBos, research in primary care. European Journal of Psychiatry, 11, 133-143.
G. R. (1994). On the AHCPR Depression in Primary Care guidelines: Wells, K. B., Stewart, A., Hays, R. D., Bumam, A., Rogers, W., Daniels,
Further considerations for practitioners. American Psychologist, 49, 42-61. M., Berry, S., Greenfield, S., & Ware, J. (1989). The functioning and
Norris, M. P., Meagher, M. W., & Arnau, R. C. (1999, August). Structure well-being of depressed patients. Journal of the American Medical
of depression in older primary care patients. Poster session presented at Association, 262, 914-919.
the 107th Annual Convention of the American Psychological Associa- Williamson, G. M., & Schulz, R. (1992). Pain, activity restriction, and
tion, Boston, MA. symptoms of depression among community-residing elderly adults.
Norris, M. P., Snow-Turek, A. L., & Blankenship, L. (1995). Somatic Journals of Gerontology: Psychological Sciences, 47, 367-372.
depressive symptoms in the elderly: Contribution or confound? Journal Zung, W. W. (1990). The role of ratings scales in the identification and
of Clinical Geropsychology, 1, 5-17. management of the depressed patient in the primary care setting. Journal
Norris, M. P., & Woehr, M. D. (1998). Differentiating somatic symptoms of Clinical Psychiatry, 51, 72-76.
in the elderly: Depression, aging, or illness? Journal of Mental Health Zwick, W. R., & Velicer, W. F. (1986). Comparison of five rules for
and Aging, 4, 347-359. determining the number of components to retain. Psychological Bulle-
Osman, A., Downs, W. R., Barrios, F. X., Kopper, B. A., Gutierrez, P. M., tin, 99, 432-442.