Journal of Abnormal Psychology 1987, Vol. 96, No. 3,179-183

Copyright 1987 by the American Psychological Association, Inc. 0021-843X/87/S00.75

Differentiating Anxiety and Depression: A Test of the Cognitive Content-Specificity Hypothesis Aaron T. Beck and Gary Brown

Center for Cognitive Therapy Department of Psychiatry University of Pennsylvania Medical School

Robert A. Steer Department of Psychiatry University of Medicine and Dentistry of New JerseySchool of Osteopathic Medicine

Judy I. Eidelson Philadelphia, Pennsylvania

John H. Riskind George Mason University

The development and initial psychometric properties of the Cognition Checklist (CCL), a scale to measure the frequency of automatic thoughts relevant to anxiety and depression, are described in this article. Item analyses of the responses of 618 psychiatric outpatients identified a 14-item depression and a 12-item anxiety subscale that were significantly related, respectively, to the revised Hamilton Rating Scales for Depression and Anxiety. Patients diagnosed according to the Diagnostic and Statistical Manual of Mental Disorders (DSM-III; American Psychiatric Association, 1980) with anxiety disorders had higher mean CCL anxiety scores than patients with DSM-III depression disorders who, in turn, had higher mean CCL depression scores. The validity of the CCL supports the content-specificity hypothesis of the cognitive model of psychopathology (Beck, 1976).

The cognitive model of psychopathology stipulates that each neurotic disorder can be characterized by a cognitive content specific to that disorder (Beck, 1976). Thus the transient automatic thoughts, interpretations, and imagery of the depressed patient center around the theme of self-depreciation and negative attitudes toward the past and future. Anxiety disorders, on the other hand, are characterized by the theme of danger; anxious patients tend to misread their experiences as constituting either a physical or psychosocial threat and to overestimate both the probability and intensity of anticipated harm in future situations (Beck & Emery, 1985). Automatic thoughts are denned as nonvolitional, stream-of-consciousness cognitions (Beck, 1976). This study describes the development and psychometric properties of the Cognition Checklist (CCL), which was designed to measure the frequency of automatic thoughts. Previous scales of this type, such as the Automatic Thoughts Questionnaire (Hollon & Kendall, 1980) and the Crandell Cognitions Inventory (Crandell & Chambless, 1981), have concentrated exclusively on the loss and failure cognitions characteristic of depression. In contrast, the CCL includes cognitions related to danger, which are said to be characteristic of anxiety disorders (Beck & Emery, 1985), in addition to cognitions related to depression. Consequently, the CCL can be used to test explicitly the content-specificity hypothesis of the cognitive model (Beck, 1967,1976).

This study was supported by National Institute of Mental Health Grant MH38843 to Aaron T. Beck. We gratefully acknowledge the contributions of Paul McDermott and Auke Tellegen. Correspondence concerning this article should be addressed to Aaron T. Beck, Center for Cognitive Therapy, Room 602, 133 South 36th Street, Philadelphia, Pennsylvania 19104.

179

Method Patients A total of 618 consecutive outpatients received the CCL prior to admission to the Center for Cognitive Therapy, Philadelphia, as part of a standard pretreatment evaluation. A subsample of 210 patients was set aside for cross-validation analyses, leaving 408 patients in the index sample. The overall sample consisted of 278 (45%) men and 340 (55%) women. The mean age was 36.20 years (SD = 12.16 years).

Instruments Diagnostic interview. A clinician diagnosed each patient according to the Diagnostic and Statistical Manual of Mental Disorders (DSMIII; American Psychiatric Association, 1980). For a significant portion of these patients (approximately 415, or 68%) the Structured Clinical Interview for DSM-III (SCID; Spitzer & Williams, 1985) was used. All of the patients in the cross-validation sample were diagnosed with the SCID. The SCID provides a standardized format for questioning patients about their symptoms, and the sequence of questions approximates the DSM-III decision rules. DSM-III criteria are embedded directly in the SCID interview, thus ensuring adequate coverage of the relevant criteria. Evidence for the reliability of SCID-based diagnoses on a portion of the present sample is provided by Riskind, Beck, Berchick, Brown, and Steer (in press) who found kappa coefficients of .72 for major depression and .79 for generalized anxiety disorder. In the index sample (N = 408), 99 patients (24%) were given a primary diagnosis of a DSM-III anxiety disorder, 207 patients (51%) were given a primary diagnosis of a depression disorder, and 102 (25%) were not given a depression or an anxiety disorder as a diagnosis. In the crossvalidation sample (N = 210), 75 (35.7%) of the patients were given primary anxiety diagnoses, 96 (45.7%) were given primary depression diagnoses, and the remaining 39 patients (18.6%) received diagnoses other than anxiety or depression. Hamilton psychiatric rating scales for depression and anxiety. The clinician also rated each patient on the Hamilton Psychiatric Rating Scales for Anxiety (Hamilton, 1959) and Depression (Hamilton, 1960). Because the standard Hamilton scales overlap substantially in content,

180

BECK, BROWN, STEER, EIDELSON, RISKIND

they were rescored as suggested by Riskind, Beck, Brown, and Steer (in press) to enhance discrimination of anxious and depressive disorders. Cognition checklist. The initial pool of items for the CCL was compiled from the verbatim reports of automatic thoughts provided by patients during the course of treatment with cognitive therapy. These cognitions are routinely recorded by cognitive therapy patients on the Daily Record of Dysfunctional Thoughts (Beck, Rush, Shaw, & Emery, 1979). A preliminary 43-item version of the scale was extracted from a pool of nearly 100 potential items by including only cognitions that were judged to be most typical of those encountered either in anxious or depressed patients. In addition, cognitions that were confounded with symptomatology (e.g., "I have become unable to manage on my own") or that reflected a disability typical of a variety of disorders (e.g., "I can't cope") were excluded, as were redundant and overlapping cognitions. Respondents rated how often each thought typically occurred to them on a 5point scale ranging from 0 (never) to 4 (always) in the context of one of four specific situations (attending a social occasion, with a friend, working on a project, and experiencing pain or physical discomfort) and regardless of the situation. In addition to rating the frequency of each cognition, the first 212 patients were asked to label the predominant affect they experienced while thinking each of the thoughts contained in the CCL.

Results Cognition-Affect Consistency Analysis Labelings of predominant affect supplied by the respondent that were associated with nonzero ratings were categorized as depressed ("depressed," "sad," etc.), anxious ("anxious," "nervous," etc.), angry ("angry," "mad," etc.), or other. Thirty-five of the items were labeled as expected: Items reflecting themes of hopelessness and loss were most often labeled as depressed, whereas items describing themes of danger and threat of loss were most often labeled as anxious. Only two items were not labeled as expected, and six items were labeled as neither anxious nor depressed (see Table 1).

Item Selection For the following item-selection analyses, the anxious group was denned as those patients with revised Hamilton Anxiety Rating Scale (HARS-R) scores greater than or equal to 0.5 standard deviations above their revised Hamilton Rating Scale for Depression (HRSD-R) scores. Similarly, the depressed group was defined as those patients with HRSD-R scores greater than or equal to 0.5 standard deviations above their HARS-R scores. The total number of patients meeting either of the aforementioned criteria in the index sample was 202, 106 (52.5%) in the anxious group and 96 (47.5%) in the depressed group; 206 met neither of the criteria and so were not included in the initial set of item-selection analyses. The 43 initial CCL items were entered into the SPSS" Discriminant program (SPSS, 1983). The classification variable was coded 1 for anxiety and 2 for depression. The resulting discriminant function loadings are shown in Table 1. Sixteen (76.2%) of the 21 highest loading positive items had depressed content and 12 (92.3%) of the 13 items with negative loadings had anxious content. Items were next assigned tentatively to anxiety and depression subscales if the direction of the discriminant loading was

Table 1 Cognition Checklist (CCL) Items by Discriminant Function Loading and Labeling of Affect CCL item There's no one left to help me. I'm worse off than they are. I'll never be as good as other people are. I'm falling behind. Life isn't worth living. There's no point in trying, I'm sure to fail. I don't deserve to be loved. He(she) won't want to see me again. Nothing ever works out for me anymore. I won't know what to say. People don't respect me anymore. I'll never be as capable as I should

be. I'm not worthy of other people's attention or affection. I will never overcome my problems. I won't have enough time to do a good job. I have become physically unattractive. I'm worthless. I'm a social failure. I've lost the only friends I've had. Other things might get in the way. No one cares whether I live or die. I will hurt someone I care about. What if I fail? He(she) will reject me. People will keep me from getting what I want. He(she) will be irritated with me. I'm losing my mind. They won't be there when I need them. I might make a mistake. I will make a fool out of myself. I am a defective human being. People will laugh at me. Something might happen that will ruin my appearance. There's something very wrong with

me. I'm going to have an accident. Something awful is going to happen. I am going to be injured. Something will happen to someone I care about. I might be trapped. I am not a healthy person. What if no one reaches me in time to help? What if I get sick and become an invalid? I am going to have a heart attack.

Loading

Labeling

Situation

.28 .27

Depressed" Depressed11

5 5

.23 .23 .20

Depressed" Anxious Depressed"

3 4 2

.20 .18 .18

Anxious Depressed"

—

3 5 5

.15 .15 .15

Depressed" Anxious Depressed"

2 3 5

.15

—

.14 .14

Depressed" Depressed"

5 2

.13

Anxious

4

.12 .12 .11 .11 .09 .09 .08 .07 .06

Depressed" Depressed" Depressed" Depressed" Anxious Depressed"

2 3 3 5 4 5 2 4 5

.05 .04 .04

—

Anxious

—

4

—

2 5 2

—

-.01 -.02

Anxious Anxious Depressed Anxious"

5 4 3 1 3

-.04

Anxious"

2

-.05 -.05

Anxious" Anxious"

1 1

-.08 -.09

Anxious" Anxious"

2 1

-.10 -.12 -.14

Anxious" Anxious" Anxious"

2 1 1

-.14

Anxious*

1

-.20 -.30

Anxious" Anxious"

1 1

.03 .03 .01

Anxious Anxious"

Note. For situations; 1 = feeling pain or physical discomfort; 2 = regardless of the situation; 3 = attending a social occasion; 4 = working on a project; 5 = with a friend. TV = 202 for discriminant analysis; N = 212 for labeling of affect. Values have been rounded to the nearest hundredth. " Included on tentative Anxiety scale. " Included on tentative Depression scale.

DIFFERENTIATING

181

Table 2 Varimax- Rotated Principal-Factor Loadings for Cognition Checklist (CCL) Items CCL item

Depression

I'm worthless. I'm not worthy of other people's attention or affection. I'll never be as good as other people are. I'm a social failure. I don't deserve to be loved. People don't respect me anymore. I will never overcome my problems. I've lost the only friends I've had. Life isn't worth living. I'm worse off than they are. There's no one left to help me. No one cares whether I live or die. Nothing ever works out for me anymore. I have become physically unattractive. What if I get sick and become an invalid? I am going to be injured. What if no one reaches me in time to help? I might be trapped. I am not a healthy person. I'm going to have an accident. There's something very wrong with me. Something might happen that will ruin my appearance. I am going to have a heart attack. Something awful is going to happen. Something will happen to someone I care about. I'm losing my mind.

.77 .73 .73 .71 .69 .66 .63 .62 .61 .61 .60 .60 .56 .50

% total variance % common variance

Anxiety

.36 .31 .32

.36

.35 .35 .71 .70 .70 .66 .65 .63 .62 .56 .56 .53 .51 .51

.80 .55

.20 .45

.34 .36

Communality

Situation

.63 .57 .56 .51 .49 .48 .52 .40 .47 .45 .46 .41 .44 .37 .49 .51 .50 .48 .49 .44 .50 .35 .32 .42 .35 .39

1 2 3 3 2 2 1 2 1 2 1 2 1 1 4 4 4 4 4 1 1 4 1 1 1 1

Note. For situations, 1 = regardless of the situation; 2 = with a friend; 3 = attending a social occasion; 4 = feeling pain or physical discomfort. N = 408. Values have been rounded to the nearest hundredth. Loadings less than .30 are not shown.

consistent with the labeling of affect. The disposition of each item is shown in Table 1.

Treatment of Marginal Items We constructed tentative subscales by summing the unitweighted 14 depression and 12 anxiety items that met both the labeling-of-affect and discriminant-loading criteria. Partial correlations were calculated between each of the unassigned items that had an absolute discriminant function loading less than .10 and each subscale controlling for the correlation of the remaining subscale. One item, "I'm losing my mind," which loaded in the depressed direction on the discriminant function but was labeled as anxious in the cognition-affect consistency analysis, had a higher partial correlation with the tentative anxiety scale than with the depression scale; it was therefore added to the anxiety scale. To confirm that dimensions of anxious and depressed cognition were underlying the 13 anxiety and 14 depression items, we conducted a principal-factor analysis with iterations to establish communalities on the entire index sample (N = 408), and two factors were retained for rotation. After a varimax rotation, one item ("People will laugh at me") that had been assigned to the anxiety subscale now loaded on the depression factor and was therefore removed from the anxiety scale. The analysis was repeated without this item. The resulting factor pattern of the final 12 anxiety subscale (CCL-A) and 14 depression subscale (CCL-D) items is shown in Table 2.

Reliability Cronbach coefficient alphas were calculated in the cross-validation sample for both CCL subscales to estimate their internal consistency. The alpha coefficient for CCL-A (Anxiety) was .90, and the average corrected item-total correlation was .62. The overall alpha coefficient for CCL-D (Depression) was .92, and the average corrected item-total correlation was .65. To estimate test-retest reliability, the CCL was readministered to a subsample of 66 patients during their sixth week of treatment. The correlation between intake and 6-week CCL-A scores was .79 (p< .001); the correlation between intake and 6week CCL-D scores was .76 (p < .001). The test-retest reliabilities of the subscales did not differ.

Discriminant Validity We calculated simple and partial correlations between each CCL subscale and each revised Hamilton scale, controlling, respectively, for the correlation with the remaining Hamilton scale. Because we had used the Hamilton scales in the itemselection process, this analysis was cross-validated in a separate sample (N= 210). The intercorrelation of the CCL subscales was .58 in the index sample and .57 in the cross-validation sample. Although both the CCL-D and the CCL-A correlated significantly with both revised Hamilton scales, the relationship was stronger between the CCL subscale and the same-affect Hamilton scale

182

BECK, BROWN, STEER, EIDELSON, RISKIND

Table 3 Simple and Partial Correlations of Cognition Checklist (CCL) Subscales With Revised Hamilton Scales Index sample (JV= 408) CCL scale Depression r Partial r Anxiety r Partial r

HRSD-R HARS-R

Cross-validation sample (N =210)

t

HRSD-R

HARS-R

t

.56** .45**

.38** .08

4.79**

.62** .54**

.37** .03

6.89**

.43** .16*

.55** .41**

3.20*

.36* .08

.54** .44**

3.29**

had higher mean CCL-A scores than did depressed patients in both the index study and cross-validation study, whereas depressed patients had higher mean CCL-D scores than did anxious patients. To determine the accuracy with which patients could be assigned to their correct diagnostic group on the basis of CCL subscale scores, we performed a discriminant classification analysis. By applying the discriminant function derived on the index sample to the cross-validation sample, 30 of 38 (79%, or 59% above chance) anxious patients and 34 of 41 (83%, or 65% above chance) depressed patients were correctly classified.

Discussion

Note. HRSD-R = Hamilton Depression Scale-Revised. HARS-R = Hamilton Anxiety Scale-Revised. *p<.01. **/><.001.

(Table 3). The differences in magnitude of the correlations between same- and different-affect scales were all significant beyond the .05 level using Hotelling's t test. When partial correlations were calculated between each CCL and each revised Hamilton scale controlling, in turn, for the remaining revised Hamilton scale, all of the correlations between each CCL subscale and the same-affect Hamilton scale in both samples remained significant. All opposite-affect correlations were not significant, with the exception of the partial correlation of the CCL-A with the HRSD-R in the index sample, controlling for HARS-R (partial r = . 16, p< .05). Next, each sample was regrouped according to both DSMIII diagnosis and salience of affect. Patients were included in the second anxiety grouping if they had a primary diagnosis of a DSM-III anxiety disorder (generalized anxiety disorder, panic disorder, social phobia, etc.) and their HARS-R standard score (z score) was at least 0.5 standard deviations higher than their HRSD-R standard score. Likewise, patients were included in the depression grouping if they had a primary diagnosis of a DSMIII depression disorder (major depression, dysthymic disorder, etc.) and their HRSD-R standard score (z score) was at least 0.5 higher than their HARS-R standard score. The mean CCL subscale scores for the criterion groups are shown in Table 4. rtests indicated that the CCL subscale scores differentiated the groups in both the index and the cross-validation studies (all one-tailed ps < .025). Thus anxious patients

The present set of results supports the content-specificity hypothesis of the cognitive model that anxious and depressed groups could be distinguished by the types of cognitive content intrinsic to the two conditions. It is clear that the items retained on the anxiety and depression subscales of the CCL are consistent in content with the cognitive themes ascribed to them by the cognitive model of psychopathology (Beck, 1976). The content of the subscales also conforms to a broader framework of affective thought processes recently proposed by Tellegen (1985). Tellegen proposed that depressive states are characterized by affective disengagement and that the associated cognitions are indicative of an "oriented" or "knowing" mode; in contrast, affectively engaged states such as anxiety reflect an "orienting" or "asking" mode. Thus the anxiety cognitions on the CCL embody a greater degree of uncertainty and an orientation toward the future, whereas depressive cognitions are either oriented toward the past or reflect a more absolute negative attitude toward the future. The CCL items were subjected to a variety of statistical procedures. Multivariate analyses using both internal and external criteria yielded a 12-item subscale of anxious cognitions and a 14-item subscale of depressed cognitions. Evidence for the discriminant and convergent validity of the two measures was demonstrated by (a) correlations with a set of independent ratings of anxiety (the HARS-R) and depression (HRSD-R); (b) the mean scores of the two scales, which differentiated samples of patients diagnosed with anxious and depressed DSM-III disorders; and (c) a good classification rate of patients into their correct DSM-III diagnostic category on the basis of their CCL subscale scores.

Table 4 Means of Cognition Checklist (CCL) Subscales for DSM-IIIAnxiety and Depression Diagnostic Groups CCL-Depression

CCL-Anxiety Group Index sample Anxiety Depression Cross-validation sample Anxiety Depression

N

M

SD

41

71

52.28 49.14

7.84 8.54

38 41

52.97 47.56

11.21 6.50

Note. Scores have been converted to /"scores [(z score X 10) + 50]. *p< .025. **p< .005. ***/>< .001.

M

SD

t

. „„

47.93 53.46

9.72 8.50

3.14**

2 52,».

45.02 54.78

7.71 8.98

5.05***

/

3

^

DIFFERENTIATING Although the correlation between the subscales was substantial, they afforded moderate discrimination between the criterion groups on the basis of mean scores and good above-chance classification rates. The discrimination achieved with the CCL is comparable to the best results that have been obtained with symptom-based psychometric measures of anxiety and depression, in which a high degree of overlap is commonly found (see Dobson, 1985, for a review). It is hoped that the CCL will be used in conjunction with symptom-based measures of psychopathology to afford enhanced discrimination of the two syndromes. In addition, the CCL would have utility in a variety of studies relating cognitive factors to diagnostic groups as well as in process studies of psychotherapy. References American Psychiatric Association. (1980). Diagnostic and statistical manual of mental disorders (3rd ed.). Washington, DC: Author. Beck, A. T. (1967). Depression: Clinical, experimental, and theoretical aspects. New 'York: Hoeber. Beck, A. T. (1976). Cognitive therapy and the emotional disorders. New York: International Universities Press. Beck, A. T, & Emery, G. (1985). Anxiety disorders and phobias: A cognitive perspective. New York: Basic Books. Beck, A. T, Rush, A. J., Shaw, B. E, & Emery, G. (1979). Cognitive therapy of depression. New \brk: Guilford Press. Crandell, C. J., & Chambless, D. L. (1981, November). The validation of an inventory for measuring depressive thoughts: The Crandell Cog-

183

nitions Inventory. Paper presented at the meeting of the Association for Advancement of Behavioral Therapy, Toronto, Ontario, Canada. Dobson, K. S. (1985). Relationship between anxiety and depression. Clinical Psychology Review, 5, 307-324. Hamilton, M. (1959). The assessment of anxiety states by rating. British Journal of Medical Psychology, 32, 50-55. Hamilton, M. (1960). A rating scale for depression. Journal of Neurology, Neurosurgery, and Psychiatry, 23, 56-61. Hollon, S. D., & Kendall, P. C. (1980). Cognitive self-statements in depression: Development of an automatic thoughts questionnaire. Cognitive Therapy and Research, 4, 383-395. Riskind, J. H., Beck, A. T., Berchick, R. J., Brown, G., & Steer, R. A. (in press). Interrater reliability of the Structured Clinical Interview for DSM-III (SCID) for major depression and generalized anxiety disorder. Archives of General Psychiatry. Riskind, J. H,, Beck, A. T, Brown, G., & Steer, R. A. (in press). Taking the measure of anxiety and depression: Validity of reconstructed Hamilton Scales. Journal of Nervous and Mental Disease. Spitzer, R. L., & Williams, J. B. W. (1985). Instruction manual for the Structured Clinical Interview for DSM-III (SCID). New York: Biometrics Research Department, New \fork State Psychiatric Institute. SPSS, Inc. (1983). SPSS* user's guide. New York: McGraw-Hill. Tellegen, A. (1985). Structures of mood and personality and their relevance to assessing anxiety, with an emphasis on self-report. In A. H. Tuma & J. D. Maser, (Eds.), Anxiety and the anxiety disorders (pp. 681-706). Hillsdale, NJ: Erlbaum. Received August 18,1986 Revision received February 16,1987 Accepted February 27,1987 •