Childhood Depression Subscales Using Repeated Sessions on Children's Depression Rating Scale – Revised (CDRS-R) Scores

Abstract

Background: Although acute treatments have been shown to be effective in treating early-onset depression, only one-third or thereabouts reach a remission within 3 months. Unfortunately, delayed time to remission in early-onset depression leads to poorer therapeutic outcomes. Clearly, there is a need to identify, diagnose, and provide effective treatment of a depressed patient quickly. A sophisticated understanding of depression subscales and their change over time with treatment could enhance pathways to individualized treatment approaches for childhood depression. Objective: Previous studies have found that the clinician-measured instrument, Children's Depression Rating Scale-Revised (CDRS-R) measures multiple subscales (or components) of depression. The aim of this study was to see how these subscales may change over the course of a 12-week study. This knowledge will help determine if dimensions/subscales of childhood depression (paralleling the adult literature) using the subscales derived from factor analysis procedure is useful. Methods: We examined two clinical trials in which youth (n = 234) with major depressive disorder (MDD) were treated openly with fluoxetine for eight sessions spread over 12 weeks. The CDRS-R was completed based on clinician interviews with parent and child at each session. Classical test theory and component analysis with associated parallel analysis (oblique rotation) were conducted on each week's scores. Results: Although more factors were needed for the baseline and first two therapy sessions, a two-factor solution sufficed thereafter. Depressed facial affect, listless speech, and hypoactivity best defined Factor I, whereas sleep problems, appetite disturbance, physical symptoms, irritability, guilt, and weeping best defined Factor II. All other symptoms cross-loaded almost equally on the two factors. The scale's reliability (internal consistency) improved from baseline to exit sessions (a = 0.65-0.91). As a result, the clinicians' assessments of the various symptoms became more highly related to one another. This caused the first eigenvalue to increase from 3.24 to 7.38 and the variance explained to increase (%) from 19% to 43% over sessions. These two factors may reflect 1) clinician-observed signs and 2) reported symptoms of depression. Conclusions: Factor analysis of CDRS-R data in a single session consistently generates a complex and difficult to interpret structure of at least three factors. This makes it very difficult to understand what these factors measure. However, when gathered over additional sessions, the CDRS-R structure tends to simplify to two factors. The reasons for this simplification are as yet unclear and in need of further study.

Estimating patient-specific treatment advantages in the ‘Treatment for Adolescents with Depression Study’

“…Second, commonly-used depression rating scales have been criticized for not being psychometrically valid (Bech, 2010;Fried and Nesse, 2015;Fried et al, 2016) and this issue has been raised with the CDRS-R (Isa et al, 2014). If the outcome measure used is unreliable, true effect heterogeneity might be clouded by measurement error, especially in case of smaller effect modifications.…”

The proportion of non-depressed subjects in a study sample strongly affects the results of psychometric analyses of depression symptoms

“…Indeed, it was noted that conditioning on a proxy of a collider has the same consequence as conditioning on the collider itself [31]. In agreement with this, previous studies indicate that correlations between different depression scales increased substantially during clinical trials [4,5], mirroring the increase of symptom correlations on the scale-level [2,3]. However, additional studies are needed to establish the conjecture empirically.…”

“…This provides an explanation for the lack of reproducibility of previous psychometric studies [1]. In addition, these findings provide a potential explanation why psychometric results differed to such an extent when the baseline versus end-of-trial data of one and the same patient sample were compared in previous studies [2,3]. Obviously, over the course of any trial, an increasing number of patients tend to remit.…”

“…In American and Dutch depression patients, Fried et al found that while the average depression severity in the samples decreased over time, the average symptom correlation increased and the factor structure of the correlations became simpler [2]. Meanwhile, in American adolescent depression patients, Isa et al found that while the average depression severity decreased over time, symptom correlations became stronger, the first eigenvalue and the total variance explained in the symptom correlations increased, the factor structure became simpler, and the measurement scale's internal consistency grew stronger [3]. Other authors found that correlations between different depression scales increased substantially during clinical trials of depressed patients, while average depression severity on the same scales decreased [4,5].…”

“…If these hypotheses hold, the implication is that dimensional, item response, and network models yield substantially different results, depending on sample composition. This would be a serious issue, since such analyses were based on rather different sample compositions [1][2][3][6][7][8][9][10][11][12][13][14][15][16]. Hence, these studies might actually be incomparable.…”