Research in Social and Administrative Pharmacy 8 (2012) 309320

Original Research

Identifying iatrogenic depression using conrmatory factor analysis of the Center for Epidemiologic Studies Depression Scale in patients prescribed a verapamil-sustained-release-led or atenolol-led hypertension treatment strategy
Debbie L. Wilson, Ph.D.a, L. Douglas Ried, Ph.D.b,*
Department of Pharmaceutical Outcomes and Policy, College of Pharmacy, PO Box 100496, University of Florida, Gainesville, FL 32610-0496, USA b Department of Pharmacotherapeutics and Clinical Research, College of Pharmacy, University of South Florida, 12901 Bruce B. Downs Blvd., MDC 30, Tampa, FL 33612, USA
a

Abstract Background: b-blockers and calcium channel blockers are highly eective medications indicated for treatment and prevention of hypertension. However, the literature regarding the potential depressive eects of b-blockers and calcium channel blockers is equivocal regarding whether one or both are associated with depression. Objectives: To determine whether self-reported depressive symptoms of older persons with hypertension and coronary artery disease and who were randomly assigned to a verapamil-sustained-release-led (Ve-led) or atenolol-led (At-led) hypertension treatment strategy were similar using conrmatory factor analytical models of the Center for Epidemiologic Studies Depression Scale (CES-D). Methods: This study used a mail survey of patients enrolled in a substudy of an international randomized controlled clinical trial. Complete data on the CES-D after 1 year of treatment were obtained from 1019 study subjects. Multiple group conrmatory factor analysis (CFA) procedures were used to test for dierences in the t of the data to the initial 4-factor CES-D model among patients assigned to the 2 treatment groups after 12 months of therapy. A test of congural invariance was conducted by sequentially constraining various matrices to be equal across groups. The convergent validity of the model was tested by examining the standard errors of the lambda-X parameter estimates of the congural model. The factor loadings for like items were investigated across the 2 groups using a test of strong factorial invariance. Finally, the 2 treatment groups were compared on the 4 factors to detect dierences in the models parameters. Results: Overall, the data t the CFA models across the 2 treatment groups based on the 4-factor model. However, 3 items diered slightly, including appetite, depressed, and crying. The data suggested signicant
At the time of this work, Debbie L. Wilson, Ph.D., was at the Rehabilitation Outcomes Research Center, North Florida South Georgia Veterans Health System, 1601 SW Archer Road (151B), Gainesville, FL 32608-1197, and Dr Ried was Professor, Department of Pharmaceutical Outcomes and Policy, College of Pharmacy, University of Florida, Gainesville, FL. * Corresponding author. 12901 Bruce B. Downs Blvd., MDC30, Tampa, FL 33612. Tel.: 1 813 974 1309; fax: 1 813 905 9890. E-mail address: lried@health.usf.edu (L.D. Ried). 1551-7411/$ - see front matter 2012 Elsevier Inc. All rights reserved. doi:10.1016/j.sapharm.2011.08.002

310

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320

dierences across groups on the positive aect, interpersonal relations, and somatic and retarded activity latent variables. Conclusions: The domains indicating less happiness and more depressive symptoms were most likely to be unfavorably impacted by the At-led treatment strategy. Given a choice between these equally eective high blood pressure treatment strategies, it may be prudent to use the Ve-led strategy. This is especially true if the risk of the occurrence of a mood-related side eect of the b-blocker outweighs its other benets in comparison. 2012 Elsevier Inc. All rights reserved.
Keywords: Hypertension; Treatment outcome; Depression; Factor analysis; Statistical; Measurement equivalence; Depressive symptoms; Iatrogenic; Coronary artery disease

Introduction Treatment and prevention of high blood pressure and coronary artery disease (CAD) are largely lifestyle based and pharmacologic based. b-blockers and calcium channel blockers are highly eective medications indicated for treatment and prevention of hypertension.1 At least some of the medications from these 2 pharmacologic categories (ie, atenolol, a b-blocker, and verapamil-sustained-release, a calcium channel blocker) appear to be equally eective when included in treatment strategies for hypertension among patients with CAD.2 Medications from dierent pharmacologic classes used for the same indication sometimes have dierent side effects and impact patients health-related quality of life (HRQoL) in dierent ways. However, the literature regarding the potential depressive eects of b-blockers and calcium channel blockers is equivocal regarding whether one or both are associated with depression.3,4 Rates of depression are purportedly higher for b-blockers than calcium channel blockers5,6 based on ndings that antidepressant prescriptions were more likely to be prescribed after b-blocker treatment.7,8 In other instances, the ndings have not been replicated.9,10 With respect to calcium channel blockers, there are reports of an association with depression.11 Consequently, the choice of antihypertensive treatment may inuence risk for depression and subsequent clinical outcomes. A resolution to this issue is important because older persons with cardiovascular diseases already have a higher prevalence of depression and higher risk of subsequent depression-related morbidities.12-17 The relationship between b-blockers and depressive symptoms may be dependent on the measurement of the depression outcome or confounded in patients with concomitant diseases.4

Consequently, elimination of these plausible reasons for misclassication error and confounding is essential to elucidating the relationship. The Center for Epidemiologic Studies Depression Scale (CES-D) has been used in epidemiologic research into depressive symptoms for nearly 3 decades. In these studies, patients with major chronic medical conditions are most likely to score higher on the somatic symptoms domain of the CES-D. If this occurs, chronically ill persons are more likely to exceed the depressed threshold because of their responses to the somatic/retarded activity domain and they may be misclassied as depressed. Although the associations do not support the conclusion that physical inrmities inate the CES-D scores exclusively through the somatic items among those with chronic illness,18,19 b-blockers and calcium channel blockers have not been compared. Consequently, this study will be the rst to examine the predominance of mood or somatic symptoms in iatrogenic depression associated with these medications. Second, the predominance of the mood or somatic subscales is critical to the hypothesis of iatrogenic depression because mood disturbance is 1 of the 2 symptoms required by the Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition) as evidence of depression. The somatic and retarded activity domain evaluates the somatic symptoms associated with depression, including tiredness, lethargy, concentration, and other central nervous system side eects. However, these symptoms may be confounded in patients with concomitant physical diseases, including CAD. This study was conducted in a large sample of persons with documented CAD whose hypertension treatment was led with medications from these 2 pharmacologic classes. Because all the patients randomly assigned to both pharmacologic

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320

311

strategies have documented CAD, classication of depression using the CES-D should not be overly inuenced by CAD somatic symptoms. Given the conuence of these 2 ndings, a depression diagnosis mistakenly assigned to iatrogenic central nervous system side eects that mimic the fatigue and lethargy of major depression is of concern. The misclassication of the patient as depressed may be because of somatic-related adverse eects mimicking depression associated with both of these pharmacologic categories but not the adverse mood eects.20,21 Consequently, the impact of the CAD somatic symptoms has been eectively controlled by studying only patients with CAD in both treatment strategies. However, the central nervous system side eects of the b-blockers may remain confounded with the CAD somatic symptoms important to a depression diagnosis (tiredness, lethargy). As such, it is essential that patients responses to important mood-related subscales should be dierent if persons taking 1 of these 2 medications are to be appropriately classied as more depressed because of the medication. The classication of depression using the CES-D should be related to which of the 2 medication strategies the patients are taking. However, a necessary precondition is that any measure of depression must be valid and measure the mood and somatic symptoms similarly for patients assigned to both groups. In other words, both the somatic and mood domain scores should dier if patients assigned to 1 treatment group are more depressed. However, the model conguration of the depression measure should t both treatment groups equally well. The primary goal of this study was to clarify whether patients responses to the mood or somatic domains dier depending on whether they were assigned to a b-blocker or calcium channel blocker. The objective was to determine whether self-reported depressive symptoms of older persons with hypertension and CAD and who were randomly assigned to a verapamilsustained-release-led (Ve-led) or atenolol-led (Atled) hypertension treatment strategy were similar using conrmatory factor analytical models of the CES-D.

the INternational VErapamil-Trandolapril STudy (INVEST).2 Subjects were older than 50 years with documented CAD and hypertension according to Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure guidelines. Documented CAD was dened as any of the following: history of myocardial infarction (MI), abnormal coronary angiogram (O50% narrowing of at least 1 major coronary artery), abnormalities on 2 dierent types of stress tests, or diagnosis of classical angina pectoris. Exclusion criteria included unstable angina, angioplasty, coronary bypass, or stroke within the previous month; b-blocker use within the previous 2 weeks or previous year for post-MI patients; sinus bradycardia, sick sinus syndrome, or atrioventricular block of more than rst degree in the absence of an implanted pacemaker; severe (New York Heart Association class IV) heart failure; severe renal (creatinine R 4.0) or hepatic failure; or contraindication to verapamil. Subjects were treated with either a Ve-led or At-led strategy. The verapamil-based strategy was found equivalent to an atenolol-based strategy with respect to clinical outcomes.2 SADD-Sx enrolled 2317 ambulatory subjects residing in the United States who were consecutively randomized to INVEST between April 1, 1999, and October 31, 1999.22 Surveys designed to collect information on sociodemographic characteristics, HRQoL, and depressive symptoms were mailed to participants at baseline and again after 1 year. Subjects were mailed a second survey if they did not return the initial survey within 10 working days. If subjects did not respond to this second survey, no further attempt was made to contact them. Subjects responses to questions about their depressive symptoms after 1 year of hypertension treatment were analyzed in this study. SADD-Sx was conducted according to the principles of the Declaration of Helsinki. The University of Florida Institutional Review Board approved the study protocol. Data collection and measure of depressive symptoms The studys primary measure of depressive symptoms was the CES-D. The CES-D is a 20item rating scale designed for respondents to selfreport their current level of depressive symptoms. The items stem reads Below is a list of the ways you might have felt or behaved. Please tell me how

Methods Study subjects The Study of Antihypertensive Drugs and Depressive Symptoms (SADD-Sx)22 is a substudy of

312

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320

often you have felt this way during the past week. Items are measured on a scale of 0 to 3 with the following anchors: 0, Rarely or none of the time (less than 1 day); 1, Some or a little of the time (12 days); 2, Occasionally or a moderate amount of time (3-4 days); and 3, Most or all of the time (5-7 days).23 Following Radlo,23 the data were treated as being continuous. The CES-D is widely used and has been extensively validated.23,24 Scores range from 0 to 60, with higher scores indicating more depressive symptoms. It is a reliable (Cronbach a range, 0.83-0.90)19,23 and valid instrument.23,24 Radlo22 described the model as having 4 factors, including (1) depressed aect, (2) positive aect, (3) somatic and retarded activity, and (4) interpersonal feelings. Others have replicated Radlos model using dierent groups of subjects.25,26 Although the specic item loadings and their magnitudes dier somewhat among the models, the basic somatic/retarded activity and mood-related domains have been replicated. Liang et al27 compared the structures of the CES-D from 4 dierent studies. The models 4 factors and items were (1) somatic and retarded activity (bothered by things that dont usually bother me, my appetite was poor, everything was an eort, restless sleep, could not get going, trouble keeping my mind on what I was doing, talked less than usual, could not shake o the blues); (2) depressed aect (felt depressed, felt lonely, crying spells, felt sad, felt fearful, thought my life was a failure); (3) positive aect (as good as other people, felt hopeful, was happy, enjoyed life); and (4) interpersonal feelings (people were unfriendly, people dislike me).27 This model structure was chosen for the present study because of its generalizability and because it was based on multiple samples in multiple reports and it is theoretically relevant to the purposes of this study.28 The structure of the theoretical model is represented in Fig. 1.28

concomitant medications. Study medications were sent to the patient directly from INVEST. Statistical analyses The Pearson chi-square statistic was used to assess the equivalency of the 2 blood pressure (BP) treatment groups at baseline.27,28 Next, multiple group conrmatory factor analysis (CFA) procedures were used to test for dierences in the t of the 2 treatment groups data to the initial CES-D model after 12 months of therapy.27,29 The CESD items were constructed using the same units of measurement for the 2 groups. Thus we performed the CFA using the 2 groups covariance matrixes and by setting 1 factor loading to 1 for each factor using a multiple step process. Multiple group CFA requires the testing of a series of nested models. CFA was used rst to investigate whether our model adequately t the data to ensure the construct validity of our analyses. CFA was then used to investigate how the groups diered on the constructs within the model.12 By conducting the test of the constructs within the model using latent means, the researchers were able to correct for attenuation because of unreliability or measurement error.29 This would not have been possible using traditional methods.29 Initially, a series of tests were conducted to investigate the usefulness of the Liang conguration and whether the specied 4-factor model t the data from the 2 groups. A test of congural invariance (a test of a weak factorial invariance null hypothesis) was conducted by specifying the same conguration of factor loadings for both groups. One factor loading was set to 1 for each factor, and the remaining factor loadings were free. The t was measured using various indicators of the goodness of t.28 The ratio of the chi-square to the degrees of freedom (c2:df) was set as %4.30 The Standardized Root Mean Square Residual (SRMR) was set as %0.09 a priori for adequate t.31 The Non Normed Fit Index (NNFI) and Comparative Fit Index (CFI) were set a priori as R0.95.31 For the test of poor t or Root Mean Square Error of Approximation (RMSEA) values of R0.1 suggest poor t; values between 0.05 and 0.08 suggest reasonable error of approximation and values of %0.05 suggest close approximate t.31,32 The P-values for the RMSEA and the Minimum Fit Function Chi-Square were set as P % .05.31 Multiple tests were used because the Minimum Fit Function Chi-Square often is very sensitive to minor dierences in the predicted

Hypertension treatment strategies INVEST subjects were randomly assigned to an At-led (the b-blocker) or Ve-led (the calcium channel blocker) hypertension treatment strategy. Both treatment strategies permitted addition of hydrochlorothiazide and/or trandolapril if blood pressure was not controlled by monotherapy. Additional medications (including over-the-counter drugs) were not permitted without prior consultation with their primary care physician. INVEST documented all changes or discontinuation of

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320

313

bothered by things that dont usually bother me my appetite was poor everything was an effort Somatic and Retarded Activity restless sleep could not get going trouble keeping my mind on what I was doing talked less than usual could not shake off the blues

e e e e e e e e

felt depressed Depressed Affect felt lonely crying spells felt sad felt fearful thought my life was a failure as good as other people Positive Affect felt hopeful was happy enjoyed life people were unfriendly people dislike me

e e e e e e e e e e

Interpersonal Feelings

e e

Fig. 1. Unmodied theoretical model targeted in CFA.

and observed covariance matrices. LISREL (LISREL 8.7 for Windows [Computer software]. Lincolnwood, IL, Scientic Software International, Inc.) reports the 2 groups combined indices for the SRMR, NNFI, CFI, RMSEA, and Minimum Fit Function Chi-Square. Except where described, the dierence chi-square test was used for the remaining tests with a signicant value (P % .05), suggesting worsening t. Measurement invariance testing involves the comparison of increasingly more restricted models by sequentially constraining

various matrices to be equal across groups. A signicant decline in t between models indicates dierences between groups in the constrained matrix.33 The convergent validity of the model was tested by examining the standard errors of the lambda-X parameter estimates of the congural model.34 If the parameter estimate is greater than 2 times the standard estimate, convergent validity may exist.34 It was not possible to calculate this for parameters that had been set to 1 to set the scale for the analysis.

314

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320

After the congural invariance test, additional CFA models were tested. The factor loadings for like items were investigated across the 2 groups using a test of strong factorial invariance (Ho: 0 Lg Lg ). A factor loading is an estimate of the factors eect on the item.35 If a factors loadings are equal for the 2 groups, the unit of the measurement of the underlying factor is identical.36 Next, the hypothesis that the unique variancesd measurement errors or disturbancesdof like items 0 are invariant across groups (Ho: Qg Qg ) was tested. At least partial unique invariance was necessary to proceed, so the unique variances (QD) of items were allowed to vary across groups for those items whose unique variances had a greater than 0.1 variance between groups. The unique variances were forced to be the same for all other like items in the remaining tests. Finally, the 2 treatment groups were compared on the 4 factors to detect dierences in the models parameters. To do this, dierences in the latent variables mean scores were tested for the 2 treatment groups for the 4 factors simultaneously using a dierence chi-square test and then individually using the z-test (critical z 1.64) of the latent mean dierences (k) across groups.37 A 1-tailed test was chosen because the hypotheses were 1 tailed and required ndings of a specic pattern to be supported. If the physiological/ mood hypotheses were to be supported, the At-led somatic symptom scores needed to be greater than the Ve-led somatic symptom scores on the CES-D and the Ve-led strategy should have higher on the positive aect scores. If neither medication causes depressed mood (eg, the side eect is the somatic symptoms instead), then the CES-D negative mood scale scores should be similar for patients taking either of the 2 pharmacologic categories of medications. Covariance matrices and standard deviations (SDs) for the items were calculated using SPSS v12.0 SPSS for Windows, version 12.0, Chicago, IL. The CFAs were performed using the covariance matrices and SDs using Jo reskog, K.G. & So rbom, D. (2004). LISREL 8.7 for Windows [Computer software]. Lincolnwood, IL: Scientic Software International, Inc. LISREL v8.70. Methods and data are reported following Schreiber.28

had complete data on the CES-D after 1 year of treatment. Of these, 522 were included in the Ve-led treatment group and 497 were included in the At-led treatment group (Table 1). The absolute value of the mean dierence between the baseline CES-D item correlations for those who responded only to the baseline data with those who completed both surveys was small (Pearson r 0.047, SD 0.03, n 189 correlations). The actual values of the mean dierences were normally distributed. Sex (c2 7.19, P .007) and race (c2 92.57, P ! .001) were the only signicant predictors of nonresponse. Within sex and race, response and nonresponse were similar for the 2 BP treatment strategies (females: c2 1.92, P .17; males: c2 0.08, P .78; whites: c2 1.36, P .24; and nonwhites: c2 0.15, P .70). Conrmatory factor analysis Congural invariance CFA was used to investigate ts of the model across the 2 treatment groups. Factor loadings were constrained to be equal to test for congural invariance. The SRMR was 0.06, which is less than the a priori level (%0.09), suggesting that the 4-factor model adequately ts both treatment groups. The Minimum Fit Function Chi-Square (c2 1182.04, df 328, P ! .01) for both groups measured simultaneously suggested that the factor loadings weakly vary across groups. The c2:df ratio was 3.6, suggesting adequate t. The error of approximation test was RMSEA 0.07, P ! .01 (90% condence interval 0.0690.078), suggesting a reasonable error of approximation. The CFI of 0.97 and NNFI or TLI of 0.96 also suggested that the model adequately t the data of the 2 groups simultaneously. Of the 6 a priori goodness of t indices, 5 suggested that the 4-factor model reasonably or adequately t the data from both hypertension treatment groups. The test of convergent validity tests (parameter estimates/standard errors) were all greater than 8 suggesting convergent validity was achieved. Full metric invariance The factor loadings were freed for the 2 groups in the next model to evaluate whether at least 1 items factor loadings (LX) varied between the 2 treatment groups. The 2 groups models are invariant or equal for the 2 groups (metric invariance c2congural invariance c2 16.76, df 16, P .40). The 4-factor model t the 2

Results Slightly less than half (n 1019, 43.9%) of the subjects originally enrolled (n 2317) in the study

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320 Table 1 Baseline demographic comparisons of subjects in analysis sample Demographic and control variables Age category, n (%) %65 66-74 75-84 R85 Age (y) (SD) Sex (%) Female Male Race (%) White African American Hispanic and others High school graduate (%) No Yes Living status (%) Alone With someone Baseline history of depression (%) No Yes When (%) !6 mo ago O6 mo ago Doctor-diagnosed depression since study (%) No Yes When (%) !1 mo 1-2 mo 3-6 mo ago O6 mo ago At-led 205 191 89 12 (41.2) (38.4) (17.9) (2.4) Ve-led 219 190 103 10 (42.0) (36.4) (19.7) (1.9)

315

Test value (P-value) c2 1.06 (.79)

66.7 (9.3) 199 (40.0) 298 (60.0) 399 (80.3) 73 (14.7) 25 (5.0) 149 (34.1) 288 (65.9) 110 (25.1) 328 (74.9) 345 (82.3) 74 (17.7) 33 (45.2) 40 (54.8) 432 (88.5) 56 (11.5) 12 8 9 26 (21.8) (14.5) (16.4) (47.3)

66.5 (9.2) 238 (45.6) 284 (54.4) 419 (80.1) 80 (15.3) 24 (4.6) 144 (30.9) 322 (69.1) 118 (25.0) 354 (75.0) 377 (82.9) 78 (17.1) 32 (41.6) 45 (58.4) 458 (88.4) 60 (11.6) 7 11 9 30 (12.3) (19.3) (15.8) (52.6)

t 0.28 (.78) c2 2.98 (.08)

c2 0.17 (.92)

c2 0.91 (.34)

c2 0.00 (1.00)

c2 0.02 (.91)

c2 0.82 (.76)

c2 0.00 (1.00)

c2 2.04 (.56)

groups data as shown in the previous step. Allowing the factor loadings to vary across groups does not signicantly improve the t, suggesting that the unit of the measurement of the underlying factors is identical across the 2 groups. Uniqueness The invariance of the unique variance or error variance of the factor loadings (QD) across groups was investigated next. The uniqueness c2metric invariance c2 dierence chi-square was c2 85.23 (df 30, P ! .001) suggesting that the items unique variance varies across groups for at least 1 of the items and that the like items error terms dier between groups.

Partial uniqueness To impose partial uniqueness, the QD was freed for like items where the dierence between the QD of the At-led group and the Ve-led groups was O0.1 in the full metric invariance model. This dierence was greater than 0.1 for 3 items: appetite, depressed, and crying. The partial uniqueness c2metric invariance c2 dierence chi-square was 26.37 (df 27, P .50), and the QD was freed in the remaining steps for those 3 items. Latent mean dierences Next, the mean dierences of the 4 latent factors were tested across treatment groups. The

316

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320

Table 2 Correlations of the factor means within groups Pharmacologic treatment Ve-led Latent factors At-led Somatic Somatic Negative Positive Interpersonal 0.89 0.42 0.65 Negative 0.92 0.43 0.79 Positive 0.53 0.52 0.44 Interpersonal 0.68 0.80 0.36

4 factor means (k) were xed for 1 group, and the others were allowed to vary. The latent mean c2partial uniqueness c2 dierence chi-square was 12.69 (df 16, P .70), suggesting that we do not have invariant factor mean scores across treatment groups. In other words, the factor mean score for at least 1 factor was dierent between the 2 groups. The correlations for these means were also similar (Table 2). Because the means of the 4 latent factors (k) varied, we determined those that signicantly differed across groups. The z-tests for 3 of the latent mean dierences were greater than the critical value (1.64, P % .05) and are probably nonchance occurrences. The At-led treatment group had larger estimated latent factor mean scores than the Ve-led group on the somatic and retarded activity (dierence in k 0.09, SD 0.05, z 1.82) and interpersonal feelings factors (dierence in k 0.06, SD 0.04, z 1.72) (Table 3). On the other hand, the At-led group had a smaller estimated latent factor mean score than the Ve-led group on the positive aect factor (dierence in k 0.08, SD 0.04, z 2.06). The dierence between the estimated latent factor mean score of the At-led group and that of the Ve-led group on the depressed aect factor was not signicant (dierence in k 0.06, SD 0.05, z 1.24).

for those assigned to the atenolol strategy. However, it is still dicult to ascertain whether these are symptoms of depression or central nervous system side eects known to be associated with atenolol. The depression-like, central nervous system side eects of fatigue (!2%), confusion (!2%), lethargy/drowsiness (!1%), and insomnia (!2%) are listed in the patient and professional labeling (eg, patient package insert, Food and Drug Administration and manufacturer labeling) for verapamil.38 The eects of depression (0.6-12%), drowsiness (0.6-2.0%), fatigue (3.0-6.0%), lethargy (1.0-3.0%), and tiredness (0.6-26%) are listed for atenolol.39 The somatic and retarded activity items of the CES-D that may reect and be confused with these side eects include trouble concentrating, cant get going, everything was an eort, and restless sleep. An earlier study investigated the psychometric properties of the CES-D using item response theory methodologies with a dierent sample of older persons. The study found that items indicative of more severe depressive symptoms were those contained in the interpersonal feelings and positive aect domains.40 At the same time, the depressed aect and somatic-retarded activity domains were less dicult to endorse and were less

Discussion The CFA undertaken in this study demonstrated that patients with CAD responded to the somatic and retarded activity, positive aect, and interpersonal feelings domains of the CES-D dierently depending on whether they were taking a Ve-led or At-led hypertension treatment strategy. Specically, their average scores on the positive aect domain were lower and higher on the interpersonal feelings domain, both dierences indicative of more severe depressive symptoms. As expected, patients responses to the somatic and retarded activity domain of the CES-D were higher

Table 3 Latent mean dierences Latent factor Mean dierence in latent factor mean scores for At-led minus Ve-led 0.09 0.06 0.08 0.06 SD z-Score

Somatic and retarded activity Interpersonal feelings factors Positive aect factor Depressed aect factor
a

0.05 0.04 0.04 0.05

1.82a 1.72a 2.06a 1.24

Signicantly dierent at 0.05.

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320

317

useful in distinguishing between those who were depressed and those who were not depressed. The authors also found that the positive aect items did not t well into a unidimensional continuum of depression, and a multidimensional solution (depressed aect, interpersonal feelings, positive aect, and somatic and retarded activity) actually t the data better. Similar to the ndings in this study, they also found that an exploratory 4-factor solution t the data better than either the 1- or 2-factor models and the positive aect items also were not highly correlated with the somatic and retarded activity, negative aect, and interpersonal feelings factors. Stansbury et al were unable to compare 2 groups of patients taking different pharmacologic hypertension treatment strategies with dierent side eect proles as in the present study. In the present study, the 2 hypertension treatment strategies diered within the same domains that were found to be the most dicult to endorse and most indicative of more severe depressive symptoms in the Stansbury et al study, namely higher scores on the interpersonal feelings and lower scores on the positive aect domains. Therefore, the dierences in the somatic and retarded activity domain could be associated with the diering side eect proles; however, the mood-related symptoms indicative of more severe depression also were clearly dierent, as well. Second, the data t the preconditions deemed necessary to conrm mood-related dierences, namely that the higher total CES-D scores were not just found in the somatic and retarded activity portion of the scale, which might be expected among persons with multiple medical conditions and no depression. First, the 4-factor model of Liang adequately t the data. This nding is important because each of the 4 factors have dierent implications for ascertaining the degree of depression caused by the medications. Second, patients assigned to the 2 treatment groups responded dierently to only 3 items, namely the depressed, crying, (depressed aect domain) and appetite (somatic and retarded activity domain) items. Despite the dierences on these 3 items, it was not sucient to change the interpretation that the 4-factor model adequately t the data. The conrmation of the 4-factor model was an important precondition to interpreting the next set of ndings regarding the development of iatrogenic depression. The subjects reports on the items of the depressed aect domain, namely, sadness, tiredness, lethargy, and fatigue, were not

signicantly dierent across the 2 groups. These items are often used by the lay public and patients to describe the symptoms of depression. Conversely, items from the 2 domains that are most useful to describe the more severe mood-related symptoms and social impairment associated with major depressive disorder40 were signicantly different between the 2 treatment groups. This nding indicates that CES-D score dierences because of dierences in the positive aect and interpersonal feelings domains may be indicative of iatrogenic depression. These ndings are an important addition to the information available regarding the question of whether b-blockers cause depression because (1) they support earlier ndings that happiness and depression are not part of a unidimensional construct and (2) the social disablement in normal roles by depression is an indication of more severe depressive symptoms. In both cases, the mean scores of the 2 factors indicating a more serious mood-related disorder were higher in the At-led treatment group. The literature regarding whether b-blockers cause depression is mixed.3,4 These ndings do not put an end to the discussion, but they do indicate that atenolol may have a more specic impact on mood-related and the somatic items that indicate more severe depressive symptoms compared with the Ve-led strategy. Our ndings seem to support those of Berkman et al18 who found that the positive aect and interpersonal feelings domains, such as I felt as happy as most people and people were unfriendly to me, were more likely to be endorsed among those with higher depressive symptom scores. Limitations Interpretation and clinical actions based on the ndings of this study must occur in the context of its limitations. First, only half of the study subjects enrolled in the SADD-Sx study responded to both the baseline and follow-up survey. We calculated the absolute value of the mean dierence between the baseline CES-D item correlations for those who responded only to the baseline data with those who completed both surveys to get a sense of their similarity in magnitude. The actual values of the mean dierences were normally distributed and unlikely to impact the factor structure. Only the mean dierence was likely to have been impacted. Demographic dierences among those who responded to both surveys and those who did not were statistically signicant

318

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320

although none by treatment assignment. This study also does not indicate clinical signicance of the dierence; the analytic strategy only points out the items and domains where the treatment groups diered. On the other hand, this study has multiple benets over previous work investigating the mood-related impact of pharmacologic treatment for hypertension. First, this study ameliorates some of the methodological concerns related with seminal work in this area, such as use of single-item indicators of depression,21,41 observational study designs,42,43 and retrospective database analysis with proxy indicators of depression, such as antidepressant dispensings.7,10 In contrast, the SADD-Sx uses a prospective multisite randomized study design using a validated scale with multiple items that eectively distinguish between the moodrelated and somatic and retarded activity symptoms of the depressive syndrome in a large ethnically diverse subsample of INVEST subjects who were randomized to Ve-led or At-led treatment at geographic locations throughout the United States.

a mood-related side eect of the b-blocker outweighs its other benets in comparison. This study also suggests that further work needs to be done with items that ask patients about happiness and functional indicators associated with worsening depression (eg, social and interpersonal limitations/disabilities). Acknowledgments SADD-Sx was conducted at the Department of Pharmaceutical Outcomes and Policy, College of Pharmacy, University of Florida, Gainesville, Florida. The study was supported in part by a grant from Knoll Laboratories (now Abbott Laboratories). The measurement invariance analyses were conducted under the guidance of Robert Vandenberg, Ph.D., Department of Management, University of Georgia, Athens, Georgia. References
1. Chobanian AV, Bakris GL, Black HR, et al. Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure. Hypertension 2003;42: 12061252. 2. Pepine CJ, Handberg EM, Cooper-DeHo RM, et al. A calcium antagonist vs a non-calcium antagonist hypertension treatment strategy for patients with coronary artery disease. The International Verapamil-Trandolapril Study (INVEST): a randomized controlled trial. JAMA 2003;290:28052816. 3. Ko DT, Hebert PR, Coey CS, Sedrakyan A, Curtis JP, Krumholz HM. Beta-blocker therapy and symptoms of depression, fatigue, and sexual dysfunction. JAMA 2002;288:351357. 4. Ried LD, McFarland BH, Johnson RE, Brody KK. Beta-blockers and depression: the more the murkier? Ann Pharmacother 1998;32:699708. 5. Lacy CF, Armstrong LL, Goldman MP, Lance LL. Drug Information Handbook 2000-2001. 8th ed. Hudson, OH: Lexi-Comp, Inc; 2000. 6. Thompson PDR. Physicians Desk Reference. 61st ed. Montvale, NJ: Thomson PDR; 2007. 7. Avorn J, Everitt DE, Weiss S. Increased antidepressant use in patients prescribed beta-blockers. JAMA 1986;255:357360. 8. Thiessen BQ, Wallace SM, Blackburn JL, Wilson TW, Bergman U. Increased prescribing of antidepressants subsequent to beta-blocker therapy. Arch Intern Med 1990;150:22862290. 9. Bright RA, Everitt DE. Beta-blockers and depression. Evidence against an association. JAMA 1992; 267:17831787. 10. Gerstman BB, Jolson HM, Bauer M, Cho P, Livingston JM, Platt R. The incidence of depression

Conclusions In summary, the ndings of this CFA support the notion that the higher summated CES-D scores earlier found in SADD-Sx were consistent with greater mood-related impact versus higher scores on the somatic items alone. These ndings, together with earlier ndings22,40 that the patients assigned to the Ve-led high blood pressure treatment strategy consistently had lower average CES-D scores than study subjects assigned to the At-led strategy, lends some credence to the proposition that b-blockers may have mood-related consequences. In this study, as predicted, patients responses to the mood-related subscales reecting higher levels of depressive symptoms were signicantly dierent and somatic and depressive aect were nearly evenly distributed among the 2 high blood pressure treatment groups. Moreover, it lends additional support to the notion that depression and happiness may not be polar opposites of the same unidimensional continuum. The factor means of the domains indicating less happiness and more depressive symptoms were most likely to be unfavorably impacted by the At-led treatment strategy. Given these ndings, if a practitioner has a choice between these equally eective high blood pressure treatment strategies,2 it may be prudent to use the Ve-led strategy if the risk of the occurrence of

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320 in new users of beta-blockers and selected antihypertensives. J Clin Epidemiol 1996;49:809815. Lindberg G, Bingefors K, Ranstam J, Rastam L, Melander A. Use of calcium channel blockers and risk of suicide: ecological ndings conrmed in population based cohort study. BMJ 1998;316:741745. Horn JL, McArdle JJ. A practical guide to measurement invariance in aging research. Exp Aging Res 1992;18:117144. Jiang W, Alexander J, Christopher E, et al. Relationship of depression to increased risk of mortality and rehospitalization in patients with congestive heart failure. Arch Intern Med 2001; 161:18491856. Turvey CL, Schultz K, Arndt S, Wallace RB, Herzog R. Prevalence and correlates of depressive symptoms in a community sample of people suering from heart failure. J Am Geriatr Soc 2002;50: 20032008. Mallik S, Krumholz HM, Lin ZQ, et al. Patients with depressive symptoms have lower health status benets after coronary artery bypass surgery. Circulation 2005;111:271277. Barth J, Schumacher M, Herrmann-Lingen C. Depression as a risk factor for mortality in patients with coronary heart disease: a meta-analysis. Psychosom Med 2004;66:802813. van Melle JP, de Jonge P, Spijkerman TA, et al. Prognostic association of depression following myocardial infarction with mortality and cardiovascular events: a meta-analysis. Psychosom Med 2004;66: 814822. Berkman LF, Berkman CS, Kasl S, et al. Depressive symptoms in relation to physical health and functioning in the elderly. Am J Epidemiol 1986;124:372388. Turner RJ, Noh S. Physical disability and depression: a longitudinal analysis. J Health Soc Behav 1998;29: 2337. Ried LD, Johnson RE, McFarland BH, Brody KK. Antihypertensive drug use and the risk of depression among older hypertensives in an HMO. J Pharmacoepidemiol 2000;8:128. Davis BR, Furberg CD, Williams CB. Survival analysis of adverse eects data in the Beta-Blocker Heart Attack Trial. Clin Pharmacol Ther 1987;41:611615. Ried LD, Tueth MJ, Handberg E, Kupfer S, Pepine CJ. A Study of Antihypertensive Drugs and Depressive Symptoms (SADD-Sx) in patients treated with a calcium antagonist versus an atenolol hypertension Treatment Strategy in the International Verapamil SR-Trandolapril Study (INVEST). Psychosom Med 2005;67:398406. Radlo LS. The CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Meas 1977;1:385401. Weissman MM, Sholomskas D, Pottenger M, Pruso BA, Locke BZ. Assessing depressive symptoms in ve psychiatric populations: a validation study. Am J Epidemiol 1977;106:203214.

319

11.

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25. Clark CH, Mahoney JS, Clark DJ, Eriksen LR. Screening for depression in a hepatitis C population: the reliability and validity of the Center for Epidemiologic Studies Depression Scale (CES-D). J Adv Nurs 2002;40:361369. 26. Takeuchi H, Hiroe T, Kanai T, et al. Childhood parental separation experiences and depressive symptomatology in acute major depression. Psychiatry Clin Neurosci 2003;57:215219. 27. Liang J, Van TT, Krause N, Markides KS. Generational dierences in the structure of the CES-D scale in Mexican Americans. J Gerontol 1989;44: S110S120. 28. Schreiber J. Core reporting practices in structural equation modeling. Res Social Adm Pharm 2008;4: 8397. 29. Vandenberg RJ, Lance CE. A review and synthesis of the measurement invariance literature: suggestions, practices, and recommendations for organizational research. Organ Res Methods 2000;3:470. 30. Kline RB. Principles and Practice of Structural Equation Modeling. 2nd ed. New York, NY: Guilford Press; 2005. 31. Hu L, Bentler PM. Cuto criteria for t indexes in covariance structure analysis: conventional criteria versus new alternatives. Struct Equ Modeling 1999; 6:155. 32. Browne MW, Cudeck R. Alternative ways of assessing model t. In: Bollen KA, Long JS, eds. Testing Structural Equation Models. Beverly Hills, CA: Sage; 1993. p. 136162. 33. French BF, Finch WH. Conrmatory factor analytic procedures for the determination of measurement invariance. Struct Equ Modeling 2006;13: 378402. 34. Anderson JC, Gerbing DW. Structural equation modeling in practice: a review and recommended two-step approach. Psychol Bull 1988;103: 411423. 35. Kline RB. Principles and Practices of Structural Equation Modeling. New York, NY: Guilford; 1998. 36. Chen F, Sousa KH, West SG. Testing measurement invariance of a second-order factor model: an illustration using a measure of quality of life. Struct Equ Modeling 2005;12:471492. 37. Hoyle RH, ed. Structural Equation Modeling. Thousand Oaks, CA: SAGE Publications, Inc.; 1995. 38. Gold Standard, Incorporated. (2007). Verapamil. Clinical Pharmacology [database online]. Available at: http://www.clinicalpharmacology.com. Accessed 18.09.07. 8-6-0007. 39. Gold Standard, Incorporated. (2007). Atenolol. Clinical Pharmacology [database online]. Available at: http://www.clinicalpharmacology.com. Accessed 18. 09.07. 6-17-0007. 40. Stansbury JP, Ried LD, Velozo CA. Unidimensionality and bandwidth in the Center for Epidemiologic Studies Depression (CES-D) Scale. J Pers Assess 2006;86:1022.

320

Wilson & Ried / Research in Social and Administrative Pharmacy 8 (2012) 309320 42. Carney RM, Rich MW, Tevelde A, Saini J, Clark K, Jae AS. Major depressive disorder in coronary artery disease. Am J Cardiol 1987;60:12731275. 43. Prisant LM, Spruill WJ, Fincham JE, Wade WE, Carr AA, Adams MA. Depression associated with antihypertensive drugs. J Fam Pract 1991;33:481485.

41. Hjalmarson AC, Herlitz J, Hjalmarson A. Use of beta blockers in postinfarct prophylaxis: aspects on quality of life, occurrence of anterior ST depression in inferior myocardial infarction and relation to clinical outcome. Am Heart J 1987;114: 245250.