Factor structure of the Children’s Depression Inventory in a multisite sample of children and adolescents with chronic pain

Abstract

This study examined the factor structure of the Children’s Depression Inventory (CDI) among children and adolescents with chronic pain using exploratory and confirmatory factor analysis in a large, multisite sample of treatment-seeking youth. Participants included 1043 children and adolescents (ages 8–18) with a range of chronic pain complaints who presented for initial evaluation at one of three tertiary care pediatric chronic pain clinics across the U.S. They completed the CDI and reported on pain intensity and functional disability. Factor analysis was conducted using a two-step (exploratory and confirmatory) approach. Results supported a 5-factor model for the CDI with good fit to the data. The distribution and item-total correlations of the somatic items (e.g. pain complaints, fatigue) were explored in this sample. Results indicate that the CDI is a useful tool for assessing depressive symptoms in youth with chronic pain, but some caution is warranted in interpreting the clinical significance of scores in light of the overlap of specific symptoms common to both pain and depression.

Assessment of children’s emotional functioning is an important component in the comprehensive assessment of youth with chronic pain. In particular, the assessment of depressive symptoms allows researchers and clinicians to characterize affect/mood, identify youth at risk for clinical depression, and assess response to intervention. One of the most widely used measures of children’s depressive symptoms is the Children’s Depression Inventory (CDI).^25,26 Recently, the PedIMMPACT consensus statement recommended the CDI as the preferred instrument for the assessment of depressive symptoms in youth with chronic pain.³²

The CDI is a 27-item self-report measure of childhood depression and a downward extension of the Beck Depression Inventory (BDI).²⁷ The CDI yields a total score and five subscales: Negative Mood, Ineffectiveness, Anhedonia, Negative Self-Esteem and Interpersonal Problems, derived from an exploratory factor analysis conducted in a community child and adolescent sample.²⁵ Previous studies investigating the factor structure of the CDI with both psychiatric and community samples^{5,15, 37,47,48} have typically yielded two to five factor solutions accounting for relatively small percentages of the measure’s variance. The low variance accounted for and differing factor structures across samples indicate that the CDI may not measure a uniform experience of childhood depression. It is unknown if the factor structure found in community and psychiatric studies is consistent in a pediatric pain population. The construct validity of the CDI in pediatric pain populations is an important topic as it may affect interpretation of research data as well as clinical applications that involve comparisons of children with pain to physically healthy children.

To our knowledge, despite the measure’s wide use in these contexts, few studies have examined the factor structure of the CDI in children and adolescents with chronic illness⁴⁴ or chronic pain. Because youth with chronic pain have higher scores on the CDI even in comparison to children with other chronic illnesses³⁵, consideration of the validity and structure of the CDI in this population is particularly important. Specifically, in the current study we aimed to apply both exploratory factor analysis and confirmatory factor analysis to allow for estimation of the CDI measurement model in a large sample of youth with chronic pain. Several studies in adult pain samples have used factor analysis of similar measures of depression to examine the specific influence of somatic symptoms on total and subscale scores.^{8,14,33,36,49} Overall, these studies have concluded that caution is needed when using traditional self-report depression measures to assess depression in the context of chronic pain due to a high level of endorsement of somatic symptoms. It is unclear whether these somatic symptoms truly reflect depression or are more closely linked to the chronic pain experience itself. Similar questions are relevant to the measurement of child and adolescent depressive symptoms in the context of pediatric chronic pain.

The primary aim of the present study was to examine the factor structure of the CDI in a large, multisite sample of treatment-seeking youth with chronic pain using exploratory and confirmatory factor analysis. To further our understanding of the assessment of depressive symptoms in youth with chronic pain, secondary aims were to explore (1) associations among CDI total and subscale scores with reports of pain and functional disability, and (2) the frequency of endorsement of somatic items relative to other CDI items in this population.

METHODS

Participants

Participants were 1043 children and adolescents presenting to three large multidisciplinary pediatric pain treatment centers between January 2000 and February 2010. Inclusion criteria were: 1) primary presenting complaint of chronic or recurrent pain (persistent pain ≥ 3 months), 2) 8–18 years of age, and 3) patient and parent ability to read and comprehend questionnaires in English. Patients were excluded if they had significant developmental delays or impairments that affected their ability to respond to self-report measures. Patients were typically referred from a variety of medical subspecialties (e.g., orthopedics, rheumatology, neurology, gastroenterology) and primary care pediatricians, usually after other treatment attempts failed to substantially reduce symptoms or pain was more severe than expected given the underlying medical condition. The data included from all study sites were collected as part of site-specific Institutional Review Board (IRB) approved protocols. Further, IRB permission was obtained for combining the de-identified databases from all contributing sites (Boston Children’s Hospital, Children’s Hospital of Philadelphia, and Cincinnati Children’s Hospital) for the purposes of this study.

Measures

Demographic Information

Parents completed a demographic information form including the child or adolescents’ age, sex, race and ethnicity, as well as information regarding the onset, duration, and frequency of pain.

Children’s Depression Inventory (CDI)

The CDI is comprised of 27 items assessing self-reported symptoms of depression in children and adolescents 7–17 years of age.^25,26 The CDI is well-validated³⁹ and is frequently used to measure depressive symptoms in youth with chronic pain.^7,18,21 The measure scoring system yields five subscales that were developed based on normative data from 592 boys and 674 girls ages 7–16 in the Florida public school system. The normative sample was 77% Caucasian and described as “largely middle class²⁵”. The subscales are Negative Mood, Interpersonal Difficulties, Ineffectiveness, Anhedonia, and Negative Self Esteem. Although a few patients in this sample (< 1%) were older than the normative age limit of 17, the CDI is clinically used in 18-year-olds to maintain consistency of measures in pediatric pain clinics and has also been used in 18-year-old adolescents in prior pediatric pain research.^16,18,19,29 CDI raw scores and corresponding norm-based T-scores were calculated. In the current sample, internal consistency (Cronbach α) reliabilities for the original factors specified in the CDI manual were as follows: CDI Total Score α = 0.88; Negative Mood subscale α = 0.71; Interpersonal Difficulties subscale α = 0.54; Ineffectiveness subscale α = 0.60, Anhedonia subscale α = 0.67; Negative Self-Esteem subscale α = 0.62. These subscale internal consistency values are somewhat low but are consistent with subscale internal consistencies reported for the CDI’s standardization sample (ranging from 0.59–0.68).²⁵

Functional Disability Inventory (FDI)

The FDI⁴⁶ is a 15-item self-report inventory of pain related disability in daily activities including home, school, recreational, and social domains such as “doing chores at home,” “being at school all day,” or “walking the length of a football field.” Items are rated on a 5-point Likert scale, ranging from 0 to 4 (“No Trouble” to “Impossible”) and summed to create a total score (range 0–60) with higher scores indicating greater pain-related disability. The FDI was initially created for use in children and adolescents 8–17 years old with recurrent abdominal pain⁴⁶ and has subsequently been used with a wide range of pediatric pain conditions in children and adolescents 8–18 years of age.^{12,17, 28,30,31} The FDI has been reported to have high internal consistency, moderate to high test-retest reliability, moderate cross-informant (parent-child) reliability, and good predictive validity.^6,46 For the current sample, internal consistency alpha was 0.90.

Pain Intensity

Children and adolescents provided pain intensity ratings of average or usual pain intensity using 10 cm Visual Analog Scales (VAS) or Numeric Rating Scales where 0 = “no pain” and 10 = “worst pain imaginable.” Both methods of rating pain intensity have been validated for this age group.^32,42,45 Average/usual pain intensity was assessed over the past two weeks or, in the case of one participating site, the immediate time period preceding the clinic visit..

Procedures

A working group of pediatric psychologists was established representing four large pediatric pain treatment centers that utilized similar assessment tools in their clinical and research protocols. The current study includes data from three of the four sites as the fourth site did not use the CDI in their assessment battery. The working group engaged in multiple electronic communications and face-to-face meetings at professional conferences to achieve consensus on: 1) classification of pain locations and pain diagnoses, 2) measures to be included in the combined database, 3) data cleaning and transmission of de-identified data, and 4) the data analysis plan. At each study site, the CDI and other measures described above were administered to children as a standard part of an initial clinical evaluation, prior to their receiving treatment.

Statistical Analysis Plan

Initial data entry and data cleaning were conducted using PASW 17.0 software.⁴⁰ Statistical analyses were conducted using R 2.15.0⁴³ and Stata 12.⁴¹

Missing data

Of the 1152 consented participants, 98 did not answer any item on the CDI measure and an additional 11 participants had 18 or more items missing on the questionnaire. These participants were excluded from analyses leaving 1043 participants. Of these 1043, 956 (91.7%) had complete data on all CDI items. Fifty six participants (5%) had one missing response, 15 (1.4%) had two missing responses, 9 (< 1%) had three missing responses, and 7 (< 1%) had missing responses ranging from 5 to 10 items. Participants with missing items did not differ significantly from those with complete responses in terms of age, race, pain frequency or pain intensity, although males tended to have more missing items than females (11.8% vs. 7.4%, p=0.03). To account for missing data a nearest neighbor hot-deck imputation of missing items was conducted¹, with all subsequent analyses performed on the imputed data. Specifically, for each participant (recipient) with missing CDI item(s), a “donor” or a “donor group” was identified through minimizing a distance metric (L1 norm in our case) between the recipient and the donor. If a donor was identified, the observed item value from the donor was then used for the missing item from the recipient. If multiple donors were identified, a value was randomly sampled from the observed item values as the imputed value. For example, if a recipient is missing response on item 2 only, a donor is found who has complete items and matches closest to this recipient on all other observed items, and the missing value is replaced with the observed value from the donor. The hot deck imputation method is used extensively in practice; for a recent detailed review see Andridge and Little¹.

Approach to factor analysis

Factor analysis was conducted using a two-step (exploratory and confirmatory) approach,^22,23 with a stratified randomization algorithm to split the sample. Within each site, half of the sample was randomly assigned to exploratory factor analysis (EFA) and the other half to confirmatory factor analysis (CFA). The factor structure identified via EFA was subsequently validated on the remaining half of the sample. The random split was done using Stata 12.⁴¹ Of the 1043 subjects with CDI data, 521 were in the estimation sample, 522 were in the validation sample. Demographics and CDI items were compared between the two halves using a series of t-tests and chi-squared tests. No significant differences were found.

First, EFA was conducted on the CDI items to explore 1) whether an item was strongly correlated with one or more latent subscales (factors) and 2) how items were differentially correlated with the factors. For these reasons, no attempt was made to drop items even when the factor loadings were below the commonly used threshold 0.3.⁹ Two methods of EFA were used, iterated principal factor (IPF) extraction and maximum likelihood extraction (MLE), in parallel using the statistical software Stata version 12.⁴¹ Number of factors was determined using eigenvalues ≥ 1.0 and at least 60% of variance explained. We also took into account additional information such as the interpretability of the factors, Bayesian Information Criterion (BIC) and Aikake Information Criterion (AIC) from the MLE approach, and factor structures when deciding the number of factors. The resultant loading matrices from EFA were rotated using two methods, orthogonal rotation and oblique rotation, to determine the factor structure. Items were classified to the factor on which they had the highest loading.

Using results from the above EFA, multiple-factor measurement models, i.e., confirmatory factor analyses (CFA) were conducted on the validation sample in order to test the adequacy of fit of the hypothesized models from the EFA. The CFA was conducted in a structural equation modeling (SEM) framework²⁴ and coefficients were estimated using maximum likelihood estimation. The CFA was conducted using Stata 12.0.⁴¹

RESULTS

Descriptive statistics

Participants’ demographics and pain characteristics are summarized in Table 1. The majority of the cohort was white (89.6%), female (78.1%), and suffering from daily pain (94.5%). The mean rating on average pain intensity was 5.5 (SD = 2.1) on a 0–10 NRS. Average pain chronicity was 120 weeks (SD = 136), or over two years; median pain chronicity was 68 weeks. Interquartile ranges for pain chronicity were: 25^th percentile = 32 weeks; 50^th percentile = 68 weeks; 75^th percentile = 148 weeks The most common pain locations included widespread pain (32%), abdominal pain (14.5%), back pain (12%), and head pain (11%).

Table 1.

Participants’ Characteristics

Variable	N	Statistics
Age (yrs), mean(SD)	1042	14.5 (2.3)
Gender, N(%)	1042
Male		228 (21.9%)
Female		814 (78.1%)
Race, N(%)	1031
White		924 (89.6%)
Black		59 (5.7%)
Asian		9 (0.9%)
American Indian		3 (0.3%)
Biracial		13 (1.3%)
Hispanic		23 (2.2%)
Site, N(%)	1042
CCHMC		374 (35.9%)
CHOP		414 (39.7%)
Boston		255 (24.5%)
Time since onset of pain (wks), mean(SD)	989	119.8 (135.6)
Pain frequency, N(%):	578
Daily, nearly everyday		546 (94.5%)
4–5 days/wk		16 (2.8%)
1–3 days/wk		10 (1.7%)
Several days/month (>=5)		5 (0.9%)
A couple of days/month(<5)		1 (0.2%)
Pain score, mean(SD)	758	5.52 (2.1)
Pain location, N(%)	1043
Head pain		111 (10.6%)
Face pain		8 (0.8%)
Neck or shoulder pain		21 (2.0%)
Back pain		126 (12.1%)
Hip pain		11 (1.1%)
Arm pain		27 (2.6%)
Leg pain		72 (6.9%)
Joint pain		86 (8.3%)
Abdominal, pelvic, or flank pain		151 (14.5%)
Widespread pain		334 (32.0%)
Foot/ankle		66 (6.3%)
Chest		19 (1.8%)
Hand/wrist		11 (1.1%)

Factor Analysis

EFA

EFA with IPF extraction method resulted in 3 factors with eigenvalues above 1.0 and 58% variation explained, a 4 factor had eigenvalue 0.83 and explained an additional 5% variation, leading to a total of 63% variation explained. EFA with MLE extraction method first retained 10 factors by doing a likelihood ratio test against the saturated model (with all 27 items), then estimated 5 factors with eigenvalues above 1.0, which accounted for 81% of variation in the 10-factor model. These results suggest both 4-factor and 5-factor models should be examined. The factor loading matrices from orthogonal rotations are shown in Table 2.

Table 2.

EFA Factor Loading Matrices for CDI Items

Item (name)	IPF (orthogonal rotation)				MLE (orthogonal rotation)
	1	2	3	4	1	2	3	4	5
1 (Sadness)	0.67	0.18	0.22	0.17	0.71	0.19	0.15	0.18	0.09
2 (Pessimism)	0.44	0.25	0.24	0.08	0.41	0.26	0.08	0.22	0.13
3 (Self-deprecation)	0.29	0.45	0.14	0.12	0.27	0.34	0.09	0.14	0.32
4 (Anhedonia)	0.30	0.09	0.50	0.20	0.31	0.10	0.18	0.52	0.07
5 (Misbehavior)	0.10	0.33	0.18	0.28	0.12	0.20	0.22	0.20	0.30
6 (Pessimistic worry)	0.52	0.28	0.05	0.07	0.46	0.26	0.08	0.06	0.16
7 (Self-hate)	0.29	0.53	0.40	0.07	0.24	0.61	0.10	0.30	0.19
8 (Self-blame)	0.24	0.57	0.04	0.10	0.25	0.36	0.05	0.05	0.44
9 (Suicidal ideation)	0.38	0.35	0.18	0.05	0.39	0.33	0.05	0.13	0.18
10 (Crying spells)	0.62	0.03	0.13	0.11	0.66	0.05	0.10	0.11	0.01
11 (Irritability)	0.60	0.17	0.14	0.27	0.58	0.18	0.26	0.14	0.11
12 (Less social interest)	0.24	0.17	0.62	0.10	0.24	0.20	0.07	0.65	0.07
13 (Indecisiveness)	0.33	0.17	0.12	0.27	0.29	0.16	0.27	0.13	0.12
14 (Neg self-image)	0.26	0.37	0.18	0.05	0.17	0.64	0.12	0.02	−0.01
15 (Schoolwork probs)	0.20	0.15	0.07	0.63	0.23	0.05	0.59	0.09	0.21
16 (Sleep probs)	0.21	−0.11	0.25	0.32	0.18	0.00	0.33	0.25	−0.10
17 (Fatigue)	0.25	−0.08	0.17	0.43	0.21	0.09	0.47	0.13	−0.13
18 (Reduced appetite)	0.24	0.05	0.19	0.31	0.23	0.10	0.31	0.17	0.03
19 (Somatic concerns)	0.38	0.02	0.09	0.14	0.35	0.06	0.14	0.10	0.01
20 (Loneliness)	0.54	0.33	0.35	0.03	0.54	0.39	0.04	0.28	0.12
21 (School dislike)	0.26	0.11	0.41	0.30	0.27	0.12	0.28	0.40	0.09
22 (Few friends)	0.13	0.33	0.37	−0.07	0.11	0.33	−0.08	0.33	0.17
23 (Academic decline)	0.15	0.17	0.08	0.51	0.12	0.17	0.54	0.06	0.12
24 (Neg peer comparison)	0.28	0.33	0.18	0.19	0.20	0.40	0.22	0.13	0.12
25 (Feels unloved)	0.17	0.38	0.23	−0.07	0.17	0.35	−0.09	0.18	0.20
26 (Disobedience)	0.02	0.43	−0.01	0.19	0.06	0.07	0.10	0.04	0.58
27 (Fighting)	0.10	0.46	0.13	0.24	0.13	0.20	0.15	0.17	0.47

IPF = iterated principal factor (IPF) extraction; MLE = maximum likelihood extraction.

Bolded numbers are loadings >0.3

CFA

The EFA process yielded 4- and 5-factor models that were then tested on the remaining half of the data (validation sample). To assess the goodness of fit of the specified multiple-factor measurement models, we examined both the comparative fit index and the Tucker-Lewis index.² Values of these two indices close to 1 indicate good fit. In addition, we examined the root mean squared error of approximation (RMSEA) and coefficient of determination (CD). A RMSEA of 0.05 or less is often considered good fit while RMSEA above 0.10 are often considered poorly fit.³ Like R-squared for linear models, CD for SEM measures the amount of variation accounted for in the endogenous constructs by the exogenous constructs. A perfect fit corresponds to a CD of 1. We fitted models with 4 correlated latent factors and 5 correlated factors. The CFA model with 5 correlated latent factors is depicted in the path diagram shown in Figure 1. Between the two models tested, the model with 5 latent factors had better fit (Comparative fit index = 0.89, Tucker-Lewis index = 0.88, RMSEA = 0.05, coefficient of determination = 0.99) than the 4-factor variant (Comparative fit index = 0.86, Tucker Lewis index = 0.85, RMSEA = 0.05, coefficient of determination = 0.98). All coefficient estimates were significant at the 5% level. The mean, standard deviation, score range, and internal consistency for each of the newly derived 5 factors obtained through the EFA/CFA procedure are reported in Table 3.

Figure 1.

Path diagram for 5 correlated factors model of the CDI in pediatric chronic pain

Table 3.

Descriptive statistics for CDI total and derived subscales

Factor	# items	# subjects	Mean	SD	Median	IQR	Range	Alpha coeff.
Derived Factor 1: Negative mood	9	1043	4.65	3.46	4	(2, 7)	0 – 18	.81
Derived Factor 2: Ineffectiveness	5	1043	3.73	2.30	4	(2,5)	0 – 10	.62
Derived Factor 3: Interpersonal problems	4	1043	0.56	1.03	0	(0, 1)	0 – 7	.61
Derived Factor 4: Anhedonia	4	1043	1.40	1.49	1	(0, 2)	0 – 8	.66
Derived Factor 5: Negative Self-esteem	5	1043	1.24	1.65	1	(0, 2)	0 – 10	.70
CDI total	27	1043	11.58	7.71	10	(6, 16)	0 – 49	.88

Associations among CDI subscales and reports of pain and disability

This analysis and subsequent analyses are based on the entire dataset. Correlations among the derived CDI subscale scores and total score, FDI scores, and pain intensity are shown in Table 4. The derived CDI Interpersonal Problems scale was not significantly associated with average pain intensity. All other derived subscales and total CDI score were significantly correlated with FDI and pain intensity.

Table 4.

Correlations between independent and dependent variables (N=780)

Independent variables	Total FDI score (P) ρ (95% CI)	Average pain intensity (P) ρ (95% CI)
CDI total score	0.26 (0.19, 0.32)***	0.25 (0.19, 0.32)***
Negative mood (derived)	0.21 (0.14, 0.27)***	0.26 (0.19, 0.33)***
Ineffectiveness (derived)	0.35 (0.29, 0.41)***	0.24 (0.17, 0.30)***
Interpersonal Problems (derived)	0.01 (−0.06, 0.08)ns	0.04 (−0.03, 0.11)ns
Anhedonia (derived)	0.21 (0.14, 0.28)***	0.20 (0.13, 0.26)***
Negative self-esteem (derived)	0.08 (0.01, 0.15)*	0.09 (0.02, 0.16)*

^*<.05;

^**<.01;

^***<.001;

^nsNot significant

Exploratory analysis of individual CDI items

To explore the role of specific items, particularly those items with a somatic focus, to overall CDI scores among children with chronic pain, we examined the distribution of scores for each item and the correlations between each item and the aggregate score of the remaining items of the original scale on which the items are classified. The items most frequently endorsed on the CDI are from the original Subscale D (Anhedonia), reflecting somatic-related concerns. See Table 5.

Table 5.

Results of exploratory analysis of individual CDI items.

Item # (name)	Original subscale	Derived subscale	Distribution of scores (N=1043)			Rrest	Item mean
			0 (No symptom)	1 (Mild symptom)	2 (Definite symptom)
1 (Sadness)	A	A	701 (67%)	290 (28%)	52 (5%)	0.60	.38
2 (Pessimism)	E	A	547 (52%)	451 (43%)	45 (4%)	0.53	.52
3 (Self-deprecation)	C	E	911 (87%)	114 (11%)	18 (2%)	0.51	.14
4 (Anhedonia)	D	D	601 (58%)	419 (40%)	23 (2%)	0.51	.45
5 (Misbehavior)	B	C	939 (90%)	88 (8%)	16 (2%)	0.35	.12
6 (Pessimistic worry)	A	A	681 (65%)	333 (32%)	29 (3%)	0.48	.38
7 (Self-hate)	E	E	865 (83%)	157 (15%)	21 (2%)	0.57	.19
8 (Self-blame)	A	C	861 (83%)	167 (16%)	15 (1%)	0.44	.19
9 (Suicidal ideation)	E	A	859 (82%)	174 (17%)	10 (1%)	0.46	.19
10 (Crying spells)	A	A	695 (67%)	234 (22%)	114 (11%)	0.52	.44
11 (Irritability)	A	A	515 (49%)	369 (35%)	159 (15%)	0.61	.66
12 (Less social interest)	B	D	817 (78%)	199 (19%)	27 (3%)	0.47	.24
13 (Indecisiveness)	A	A	426 (41%)	490 (47%)	127 (12%)	0.43	.71
14 (Neg self-image)	E	E	658 (63%)	328 (31%)	57 (5%)	0.46	.42
15 (Schoolwork probs)	C	B	484 (46%)	332 (32%)	227 (22%)	0.43	.76
16 (Sleep probs)	D	B	292 (28%)	387 (37%)	364 (35%)	0.31	1.07
17 (Fatigue)	D	B	265 (25%)	412 (40%)	366 (35%)	0.38	1.10
18 (Reduced appetite)	D	B	726 (70%)	209 (20%)	108 (10%)	0.34	.41
19 (Somatic concerns)	D	A	255 (24%)	508 (49%)	280 (27%)	0.37	1.03
20 (Loneliness)	D	A	726 (70%)	270 (26%)	47 (5%)	0.59	.35
21 (School dislike)	D	D	599 (57%)	372 (36%)	72 (7%)	0.46	.49
22 (Few friends)	D	D	830 (80%)	195 (19%)	18 (2%)	0.35	.22
23 (Academic decline)	C	B	701 (67%)	266 (26%)	76 (7%)	0.42	.40
24 (Neg peer comparison)	C	E	701 (67%)	251 (24%)	91 (9%)	0.52	.42
25 (Feels unloved)	E	E	985 (94%)	46 (4%)	12 (1%)	0.31	.07
26 (Disobedience)	B	C	879 (84%)	154 (15%)	10 (1%)	0.29	.17
27 (Fighting)	B	C	962 (92%)	72 (7%)	9 (1%)	0.39	.09

Notes: “Original Scale” indicates the subscale that each item represented according to the original scoring system reported in the CDI Manual. A = Depressed Mood; B = Ineffectiveness; C = Interpersonal problems; D = Anhedonia; E = Negative self-esteem.

R_rest = The correlations between each item and the aggregate score of the remaining items of the original scale

DISCUSSION

To our knowledge, this study represents the first effort to describe the underlying factor structure of the CDI in a clinical population of children and adolescents with chronic pain. Given that the CDI is the most widely used measure for assessing depressive symptoms in pediatric pain, a clear understanding of the construct validity of this instrument is crucial to the accurate interpretation of total and subscale scores. Results of the exploratory and confirmatory factor analyses revealed that a 5-factor model provided a very good fit to the data. While the five-factor structure is similar to what has been found in previous studies, our CFA indicated a closer fit and higher percent of variation accounted for (i.e., coefficient of determination) than has been found in previous studies.^10,48

The newly derived factors that emerged in the current study correspond broadly to the original 5-factor structure of the CDI²⁶, although there were some differences in the specific items that comprised each of the subscales in the current model. In particular, the largest number of items (9 of 27) loaded onto the newly derived Negative Mood subscale (including items relating to sadness, pessimism, hopelessness, irritability and isolation) in contrast to the smaller 6-item Negative Mood subscale of the original CDI. This 9-item Negative Mood scale was found to correlate highly with the remaining 4 derived factors (coefficients ranging from .63–.85) with the strongest association with the Negative Self-Esteem factor. Another notable difference from the original factor structure was that half of the items on the original Anhedonia subscale (4 items primarily relating to somatic symptoms), were now subsumed under other factors (primarily the Ineffectiveness factor). As a result, the items remaining on the Anhedonia (derived) factor more narrowly reflected “withdrawal” type symptoms (loss of interest, reduced social engagement, school dislike and few friends). The composition of the derived Ineffectiveness subscale included items with a somatic focus (relating to problems with sleep, fatigue and appetite), which most likely represents the direct impact of pain on the daily functioning of children suffering from chronic pain.

Although the internal consistency of the full CDI was very good in our sample, internal consistency estimates for most of the original subscales were low, ranging from .54 to .71. This is perhaps further indication that the original factor structure of the measure did not fit well for this sample. Alpha coefficients for the derived factors ranged from questionable (.62 for Interpersonal Problems and Ineffectiveness subscales) to good (.81 for Negative Mood). Low numbers of items loading onto some factors may contribute to the lower internal consistency of those subscales. This pattern lends support to the primacy of the derived Negative Mood factor in this sample; among children with chronic pain, this subscale may represent a distillation of the affective and cognitive symptoms that most purely represent depression in comparison to other CDI subscales.

Upon closer examination of the performance of the CDI measure at an item level, it was found that all items were endorsed at some level by participants in the study. The items least likely to be endorsed (>85% scoring 0 on the item) were those relating to misbehavior, feeling unloved and self-deprecation. On the other hand, the three items relating to sleep difficulties, fatigue and somatic concerns (which fell under the Anhedonia scale in the original CDI) were the most highly endorsed items in this clinical pain population, with 70% of patients reporting at least some problems in these areas. While this is not a surprising finding, it should be kept in mind when interpreting Anhedonia scores using the original scoring method. In fact, given the high likelihood of endorsement of these items among children and adolescents with chronic pain in general, it is not unexpected that overall CDI scores in this and other studies of pediatric pain^4,12,13,21 fall in the slightly elevated range when compared to population- based norms. Based on the contribution of the somatic items on the CDI alone, it is likely that total raw scores in the 11–12 range on the CDI are in fact “normative” for children with chronic pain and do not necessarily reflect elevated depressive symptoms or signify risk of a depressive disorder.

Aside from the above findings unique to pediatric chronic pain, the full CDI was well supported in this study as a useful measure of global depressive symptoms, and there did not appear to be strong justification for the elimination of any items or necessity for proposing an entirely new scoring system. Nevertheless, the findings of this study have several important implications for the interpretation of the CDI in clinical and research use with this population. As noted earlier, caution should be exercised when interpreting total CDI scores clinically, as slight elevations might be expected in children and adolescents with pain based upon the somatic symptoms that often accompany a pain diagnosis. Similarly, as these somatic items originally loaded on the Anhedonia scale, an interpretation of the Anhedonia score should give greater weight to the symptoms of lack of interest and social withdrawal as being diagnostic of depressive symptomatology as opposed to the subscale score per se. Additionally, an interpretation of the original Ineffectiveness domain should take into account the disruptive impact that pain might have on normal day-to-day life including school, academic and peer-related functioning. Admittedly, it is particularly difficult in pediatric populations to draw distinct boundaries between symptoms that characterize depression and those that characterize chronic pain. Factor analyses of the Beck Depression Inventory in adult chronic pain samples have concluded that interpretation of depressive symptoms may be different from a psychiatric model of depression in that cognitive and somatic symptoms seem to be endorsed by pain patients more frequently than affective disturbance.³³.

Overall, the findings of this study lead us to endorse the continued use of the CDI for the broad or global assessment of psychological functioning in children with chronic pain. To gain a comprehensive picture of pain-specific distress in these children, we recommend that assessments also incorporate measures of pain-specific emotional distress such as pain catastrophizing¹¹ and pain-related fear³⁸ that may better characterize children’s cognitions.

In addition to enriching our knowledge about clinical interpretation of the CDI, the findings of this study have implications for research use of this measure. Most clinical trials in pediatric pain have not found significant reductions in total CDI scores after cognitive-behavioral treatment (CBT)³⁴, with the exception of one trial in adolescents with juvenile fibromyalgia that did find significant reduction in total CDI scores after CBT.²⁰ The latter finding might be due to the fact that adolescents with widespread chronic pain are likely to have significantly higher levels of depressive symptoms than those with other pain conditions, and the measure might be more sensitive to change at higher levels of symptomatology. An alternative to using total CDI scores in such trials may be to use a narrower focus on the newly derived Negative Mood subscale, which may be more sensitive to changes in the cognitive and affective symptoms of depression. This recommendation is bolstered by the derived Negative Mood subscale’s relatively strong psychometric performance, e.g. its good internal consistency and the variance in scores obtained on this subscale.

The aim of the current study was to examine CDI scores among youth presenting for assessment and treatment of chronic pain conditions. Given this, the generalizability of the findings beyond youth seeking treatment in the tertiary care setting is unclear. A limitation of the study is the absence of a contemporaneous comparison group of healthy children and adolescents. A further limitation is that we did not have information from clinical diagnostic interviews for psychological conditions, without which it was not possible to elaborate on the diagnostic utility of the CDI in pediatric pain populations or to suggest potential CDI cut-off scores that would be optimal for identifying clinically significant depressive symptoms. This is an important next step for future research efforts in this area. An additional focus for future research is to advance our understanding of how optimally to integrate data obtained from general measures of emotional distress such as the CDI with pain-specific measures of emotional distress (e.g., catastrophizing).

In summary, the Children’s Depression Inventory appears to be a valid measure of depressive symptoms in youth with chronic pain complaints. It is a useful tool for quantifying the extent of negative mood and related symptoms that are commonly experienced among children and adolescents experiencing prolonged pain conditions, and may be best used when augmented by other measures aimed at eliciting types of emotional distress that may be specific to the chronic pain experience. In combination with these more pain-specific scales, the CDI can be useful in comprehensive assessment of the spectrum of affective symptoms and how they may change over time in a pediatric chronic pain population in both clinical and research settings.

Perspective.

The Children’s Depression Inventory can be considered a valid tool for assessing mood symptoms in children with chronic pain. Caution is encouraged when interpreting the clinical significance of scores due to symptom overlap between chronic pain and depression.