Center for Epidemiological Studies-Depression Scale: Evaluation of a New Cutoff Score for Women Living with HIV

To the Editor:

Major Depressive Disorder (MDD) is one of the most burdensome mental health conditions among adults in the United States, with a lifetime prevalence of 7%, accounting for 3.7% of all US disability-adjusted life years.¹ Among persons living with HIV (PLWH), the lifetime prevalence rates of depression are two to three times higher, at estimates ≥20%.² In addition, PLWH diagnosed with depression are less likely to be adherent to antiretroviral therapy (ART)³ and thus are at a significantly higher risk of death.⁴

The Center for Epidemiological Studies-Depression (CES-D) scale is a well-established 20-item measure that was developed for the screening of cognitive, affective, and somatic symptoms of depression involving women with and without breast cancer.⁵ Cutoff scores are used to indicate clinically relevant symptoms and suggest a need for further assessment using a diagnostic tool. Optimal cutoff scores should be established to maximize detecting true illness and minimize false positives. The established clinical cutoff indicating clinical depression when using the CES-D is ≥16.⁶ However, higher cutoff scores have shown better balance between sensitivity and specificity among chronically ill, female, and vulnerable populations.⁷

Among low-income women in primary care, a 16-point cutoff score had high sensitivity (0.95) and specificity (0.70); however, two-thirds of those screened positive for depression did not meet criteria for MDD [positive predictive value (PPV) = 0.28].⁸ A cutoff score of 34 had high specificity (0.95) and a higher predictive value for MDD (0.53), but much lower sensitivity (0.45).

To our knowledge, although there are established CES-D cutoff scores for chronic illnesses other than HIV disease, a CES-D cutoff score for women living with HIV (WLWH) has never been formally ascertained. Determining an appropriate cutoff score for WLWH will aid both HIV care providers and researchers whose focus is on individuals who meet DSM criteria for depression. This study provides preliminary data on the assessment of the CES-D to screen for MDD using data from a randomized controlled trial of African American low-income WLWH in rural Alabama.

Participants included in this analysis were African American WLWH (≥19 years of age) enrolled in the Tele-C.A.R.E. study, a randomized controlled trial comparing cognitive behavioral therapy designed to increase medication adherence and reduce depression in WLWH to a supportive psychotherapy control intervention.⁹ Participants were recruited at HIV clinics across rural Alabama. Informed consent was obtained from all participants included in the study.

This study used data from the baseline assessment of the Tele-C.A.R.E. study. MDD was indicated by diagnosis using the Mini-International Neuropsychiatric Interview. We compared CES-D scores and participant characteristics between the MDD (n = 29) and non-MDD (n = 18) groups using measures of effect size: Cohen's d (standardized mean difference) for continuous variables, and d-equivalent for categorical variables.¹⁰ We estimated the CES-D score cutoff that minimized the distance between the receiver operating characteristic (ROC) curve and the point of perfect prediction of MDD diagnosis (specificity = 0, sensitivity = 1) in the sample. Nonparametric Bootstrap (2000 resamples, stratified by MDD group) was used to estimate uncertainty about the cutoff value and accuracy measures, in the form of 95% confidence intervals (CIs) for area under the ROC curve (AUC), number of false positives and false negatives, sensitivity, specificity, PPV, and negative predictive value (NPV). The Bootstrap approach was used to avoid making any distributional assumptions for these quantities; however, the resulting CIs are still based on the sample and, therefore, should be interpreted with respect to the specific population from where the sample was drawn: African American WLWH. Analyses were conducted using R system software v.3.3.2.

On average, participants were 47 years of age, received an HIV diagnosis 9 years previously, and reported taking ART for about 7.5 years. About 11% of participants were immunosuppressed (CD4 count <200 cells/μL) and 13% had a viral load of HIV RNA >200 copies/mL. About 8% participants were college graduates, 38% attended college, 32% completed high school, and 21% had less than high school education. Thirty-eight percent of participants were employed, with 57% reporting an income of <$10,000. Less than 5% of participants reported current substance abuse and 27% reported depression medication use.

Relevant differences between the MDD and non-MDD groups were observed in CES-D scores, marital status, and history of depression. Participants with MDD diagnosis were more likely to be separated, divorced, or widowed (48% vs. 22%), more likely to have a prior history of depression (72% vs. 50%) and to use an antidepressant medication (69% vs. 39%) than participants without MDD. The distributions of CES-D scores by MDD diagnosis group are shown in Fig. 1. The CES-D scores ranged from 21–52 and 11–35 in the MDD and non-MDD groups, respectively. The mean difference in scores was 10.9 (standard error = 2.6, p < 0.0001).

FIG. 1.

Distributions of CES-D scores by depression diagnosis. Depression diagnosis ascertained through the Mini-International Neuropsychiatric Interview. CES-D, Center for Epidemiological Studies-Depression.

In the non-MDD group, 83% of participants had CES-D scores ≥16. A cutoff score of ≥16 resulted in zero false negatives, but 15 false positives (sensitivity = 1 but specificity = 0.17). The CES-D score that minimized the distance from the ROC curve to the point of perfect prediction was 30 (95% CI: 28–38). In terms of accuracy of the classification of MDD diagnosis this cutoff score, AUC = 0.81 (95% CI: 0.68–92), indicated good in-sample prediction. The number of false positives was 6 (95% CI: 0–7) and false negatives was 6 (95% CI: 3–14). Sensitivity was 0.79 (95% CI: 0.52–0.89) and specificity = 0.67 (95 CI: 0.61–1). PPV was 0.79 (95% CI: 0.75–1) and NPV was 0.67 (95% CI: 0.53–0.82).

In our sample, the recommended CES-D cutoff score of ≥16 resulted in a large number of false positives; a higher cutoff of 30 decreased the number of classification errors to 12, reducing the false positives to 6 at the cost of 6 false negatives. Reducing the number of false positives has important implications in the design and implementation of research studies, impacting power calculations. Gender differences in optimal CES-D score cutoffs have been suggested by others, and may be attributed to differences in response patterns between women and men.⁷ In addition, neurocognitive disorders, which commonly occur among HIV-positive populations, could affect depression screening scores in this population.¹¹

A cutoff of ≥16 in this study greatly overestimated the true number of persons diagnosed with depression; 83% of persons without depression were incorrectly screened in. By screening out persons without depression at recruitment, raising the cutoff would allow for optimal allocation of resources and case ascertainment. It is important to note that although these participants should not have screened into the study because they did not meet DSM criteria for depression, their elevated depression symptom score may warrant further clinical assessessment.³

Based on the current findings, and support from previous research suggesting a higher optimal CES-D cutoff among persons with chronic diseases, vulnerable women, and African Americans, we conclude that a higher cutoff should be considered for the use among WLWH in research settings; in clinical settings, it may be prudent to further explore lower levels of reported distress for the sake of triaging patients to appropriate clinical resources. This study is limited by a relatively small and regionally and racially specific sample; therefore, larger studies among more demographically diverse samples powered to validate a more accurate CES-D cutoff score are needed to provide a more definite answer regarding the right cutoff to be used among PLWH.

Author Disclosure Statement

S.S. receives royalties from Oxford University Press and Guilford Publications, and anticipates royalties from Humana Press for being an author on books. All other authors report no conflicts of interests.