The PedsQL™ in Type 1 and Type 2 Diabetes

RESEARCH DESIGN AND METHODS

Diabetes sample

Participants were children aged 5–18 years (n = 300) and parents of children aged 2–18 years (n = 308) diagnosed with type 1 or type 2 diabetes, with 331 families accrued overall. For 279 children aged 5–18 years, both child self-report and parent proxy-report were available. Participant characteristics are shown in Table 1.

Healthy sample

Participants were healthy children and their parents from the PedsQL 4.0 field test (8). The demographics are described in Varni et al. (8). This healthy sample was younger (mean age 12.19 vs. 14.19 years) and represented fewer African-American (2.9 vs. 5.2%) and Asian/Pacific Islander (1.2 vs. 4.1%) children and more Hispanic children (27.3 vs. 12.3%) than the diabetes sample.

Measures

The PedsQL 4.0 Generic Core Scales.

The 23-item PedsQL 4.0 Generic Core Scales encompass: 1) physical functioning (8 items), 2) emotional functioning (5 items), 3) social functioning (5 items), and 4) school functioning (5 items). They were developed through focus groups, cognitive interviews, pretesting, and field testing measurement development protocols (7,8).

Child self-report includes ages 5–18 years, and parent proxy-report includes ages 2–18 years. The items for each form are essentially identical. The instructions ask how much of a problem each item has been during the past 1 month. A five-point response scale is used (0 = never a problem, 4 = almost always a problem). Items are reverse-scored and linearly transformed to a 0–100 scale (0 = 100, 1 = 75, 2 = 50, 3 = 25, and 4 = 0), so that higher scores indicate better HRQOL. Scale scores are computed as the sum of the items divided by the number of items answered. If >50% of the items in the scale are missing, the scale score is not computed (11). After imputing missing values, 99% of child respondents and 98% of parent respondents were included in the scale score analyses. The physical health summary score (eight items) is the same as the physical functioning subscale. The psychosocial health summary score (15 items) is computed as the sum of the items divided by the number of items answered in the emotional, social, and school functioning subscales.

The PedsQL 3.0 Type 1 Diabetes Module.

The 28-item multidimensional PedsQL 3.0 Diabetes Module encompasses five scales: 1) diabetes symptoms (11 items), 2) treatment barriers (4 items), 3) treatment adherence (7 items), 4) worry (3 items), and 5) communication (3 items). The format, instructions, Likert-type response scale, and scoring method are identical to the PedsQL 4.0, with higher scores indicating fewer symptoms or problems. The PedsQL 3.0 Diabetes Module development consisted of a review of the literature, patient and parent focus groups and individual focus interviews, item generation, cognitive interviewing, pretesting, and subsequent field testing (12–14).

PedsQL Family Information Form.

The Family Information Form contains the demographic information required to calculate the Hollingshead socioeconomic status (SES) index (15).

Procedure

Clinic assessment.

Subjects were identified at the two sites when they presented for clinic. Written parental informed consent and child assent were obtained. Parents and children completed the PedsQL separately. A research assistant administered the PedsQL for children aged 5–7 years and was available to assist the self-administered instrument for ages 8–18 years.

Telephone assessment.

Patients identified by their physician as meeting inclusion requirements were sent a letter from their physician. The letter included an 800 number that the family could call if they did not want to participate. A week after the letter was sent, a research assistant called the family and obtained verbal consent from the primary caregiver. Another research assistant on the telephone line witnessed the verbal consent. The questionnaires were read individually to the parent and child verbatim, and the answers were recorded. These research protocols were approved by the institutional review boards at each site.

Statistical analysis

Feasibility was determined from the percentage of missing values (16). Internal consistency reliability was determined by Cronbach’s α-coefficient (17). Scales with reliabilities of ≥0.70 are recommended for comparing patient groups, whereas a criterion of 0.90 is recommended for analyzing individual patient scale scores (18,19).

Construct validity was determined using the known-groups method. The known-groups method compares scale scores across groups known to differ in the health construct being investigated. PedsQL 4.0 Generic Core Scales scores in groups differing in known health condition (healthy children and children with type 1 or type 2 diabetes) were computed (20,21) using t tests and one-way ANOVA. We hypothesized that: 1) healthy children would report higher PedsQL 4.0 scores (better HRQOL) than patients with diabetes based on previous PedsQL 4.0 findings with other pediatric chronic health conditions (8–10,22). ANOVAs were conducted to examine whether there were differences in PedsQL 4.0 scores among children with type 1 or type 2 diabetes and healthy children.

Construct validity was further examined through an analysis of the intercorrelations among the PedsQL 4.0 Generic Core Total Scale score with the PedsQL 3.0 Diabetes Module scales scores. Computing the intercorrelations among scales provides initial information on the construct validity of an instrument (19). We hypothesized that: 2) higher scores on the diabetes symptom scale (fewer symptoms) would be correlated with higher Generic Core Total Scale scores, based on the conceptualization of disease-specific symptoms as causal indicators of HRQOL, (1) and previous PedsQL 4.0 findings (9,22); 3) higher scores on the treatment barriers and treatment adherence scales (fewer problems with barriers and adherence) would be correlated with higher Generic Core Total Scale scores, based on the conceptualization that treatment adherence is associated with better symptom control and, consequently, better HRQOL (23); 4) higher scores on the treatment barriers and treatment adherence scales would be correlated with higher scores on the diabetes symptom scale based on the treatment adherence literature (24,25); 5) higher scores on the worry and communication scales (less worry and better communication, respectively) would be correlated with higher Generic Core Total Scale scores, based on previous PedsQL disease-specific module studies (9,22). Intercorrelations were expected to demonstrate medium to large effect sizes (1). Correlation effect sizes are designated as small (0.10), medium (0.30), and large (0.50) (26). Finally, we explored whether the PedsQL 4.0 scores would be different in children with type 1 or type 2 diabetes, and whether HbA_1c was related to HRQOL. Parent-child intercorrelations were computed to examine cross-informant variance. Response equivalence has been demonstrated across language and mode of administration (8); therefore, responses were pooled.

RESULTS

Missing item responses

For PedsQL 4.0 child self-report and parent proxy-report, the percentage of missing item responses was 0.1 and 1.4%, respectively. For the Diabetes Module, the percentage of missing item responses was 1.4% for child self-report and 3.6% for parent proxy-report.

Means and standard deviations

Table 2 presents the means and SDs of the PedsQL Generic Core Scales for children with diabetes and healthy children (8) and for the type 1 Diabetes Module.

Internal consistency reliability

α-Coefficients for the PedsQL across ages 2–18 years are presented in Table 2. Most child self-report scales and parent proxy-report scales exceeded the reliability standard of 0.70 (18). The PedsQL 4.0 total score across the ages approached the reliability criterion of 0.90 recommended for analyzing individual patient scores (18,19). Child self-report for the Diabetes Module exceeded the reliability standard of 0.70 for the diabetes symptoms scale and the communication scale, and it was in the 0.63–0.66 range for the other scales. All parent proxy-report scales except one exceeded the 0.70 standard.

Construct validity

Table 2 demonstrates the comparisons between the PedsQL Generic Core Scales for healthy children and children with diabetes as a group across ages 2–18 years. For child self-report, there was a significant difference between healthy children and children with diabetes for all scales except physical functioning and social functioning. For parent proxy-report, there was a significant difference between healthy children and children with diabetes on all scales. Tables 3 and 4 display the one-way ANOVAs comparing healthy children with children with type 1 and type 2 diabetes using the PedsQL Generic Core Scales for ages 8–18 years (matched to the age range for patients with type 2 diabetes). For all scales except physical functioning and social functioning, children with type 1 diabetes reported lower HRQOL than healthy children. For all scales except physical functioning, children with type 2 diabetes reported lower HRQOL than healthy children. Children with type 2 diabetes also reported lower generic HRQOL scores than children with type 1 diabetes on the total scale score, psychosocial health, and school functioning. For all parent proxy-report scales, children with type 1 and type 2 diabetes were reported as manifesting lower generic HRQOL than healthy children. Parents did not report any generic HRQOL differences between children with type 1 and type 2 diabetes.

The intercorrelations between the PedsQL Generic Core Scales total score and the Diabetes Module were in the medium-to-large effect size range, with the largest intercorrelations between the diabetes symptoms scale and the Generic Core total score (0.66 child-report, 0.54 parent-report). Intercorrelations among the other scales were consistent with the a priori hypotheses and ranged from 0.35 to 0.66. Parent-child intercorrelations for the PedsQL Generic Core Scales and Diabetes Module ranged from 0.28 to 0.47, with most in the medium effect size range.

HbA_1c

For a subset of children with type 1 and type 2 diabetes, HbA_1c values were available. For the type 1 sample (n = 211), correlations between the child self-report generic core and diabetes scales and HbA_1c levels revealed small-to-medium correlation effect sizes for Generic Core total score (−0.17, P < 0.05), psychosocial health (−0.20, P < 0.01), school functioning (−0.29, P < 0.001), treatment barriers (−0.27, P < 0.01), and treatment adherence (−0.20, P < 0.05). Correlations between parent proxy-report Generic Core and diabetes scales and HbA_1c levels also revealed small-to-medium correlation effect sizes for Generic Core total score (−0.22, P < 0.01), psychosocial health (−0.28, P < 0.001), emotional functioning (−0.18, P < 0.05), social functioning (−0.16, P < 0.05), school functioning (−0.34, P < 0.001), treatment barriers (−0.23, P < 0.01), and treatment adherence (−0.19, P < 0.05). There were no significant correlations between HbA_1c and PedsQL child self-report and parent proxy-report scales for the type 2 diabetes sample (n = 70).

CONCLUSIONS

This study presents the measurement properties for the PedsQL in type 1 and type 2 diabetes. The analyses support the reliability and validity of the PedsQL as a child self-report and parent proxy-report HRQOL measurement instrument for diabetes. The PedsQL is the only empirically validated pediatric HRQOL instrument to span this broad age range for child self-report and parent proxy-report while maintaining item and scale construct consistency.

Items on the PedsQL had minimal missing responses, suggesting that children and parents are able to provide good-quality data regarding the child’s HRQOL. The PedsQL self-report and proxy-report internal consistency reliabilities generally exceeded the recommended minimum α-coefficient standard of 0.70 for group comparisons. Across the age ranges, the PedsQL 4.0 Generic Core Scales total score for both child self-report and parent proxy-report approached or exceeded an α of 0.90, recommended for individual patient analysis (18), making the total scale score suitable as a summary score for the primary analysis of HRQOL outcome in clinical trials and other group comparisons. The only exception was child self-report for 5–7 years of age, where the α was acceptable for group comparisons only. The physical health and psychosocial health summary scores are recommended for secondary analyses. The emotional, social, and school functioning subscales may be used to examine specific domains of functioning, with the caveat that until further testing is conducted, scales not achieving an α ≥0.70 should be used only for descriptive or exploratory analyses.

The PedsQL 3.0 Diabetes Module scales internal consistency reliabilities generally exceeded the recommended minimum α-coefficient standard of 0.70 for group comparisons for child self-report for ages 8–18 years and parent proxy-report for ages 2–18 years. For young-child self-report ages 5–7 years, only the treatment adherence scale met the 0.70 standard for group comparisons, whereas several of the other scales were in the 0.60 range. Although Cronbach’s α internal consistency coefficients represent the lower bound of the actual reliability of a measurement instrument, and thus are a conservative estimate of actual reliability (27), until further testing is conducted, scales that did not achieve the standard of 0.70 should be used only for descriptive or exploratory analyses. For the purposes of a clinical trial, the PedsQL 3.0 Diabetes Module diabetes symptom scale in combination with the PedsQL 4.0 Generic Core Scales would provide an integrative HRQOL measurement model, with the advantages of both generic and diabetes-specific scales and with known reliability and validity across a wide age range.

The cross-informant variance observed in the parent/child-report supports the need to measure the perspectives of child and parent informants in evaluating pediatric HRQOL. The use of parent proxy-report to estimate child HRQOL may be necessary when the child is either unable or unwilling to complete the HRQOL measure, or when young child self-report scale reliabilities do not achieve the 0.70 standard. The present findings have several potential limitations. Information on nonparticipants was not available, which may limit the generalizability of the findings. Test-retest reliability was not conducted; however, test-retest reliability may be less useful than internal consistency reliability in HRQOL instrument development, given that short-term fluctuations are highly likely in a health condition in which external factors, such as disease and treatment variables, are expected to influence functioning. The method for testing construct validity used the known-groups approach. Additional methods for testing construct validity include correlating the instrument with other standardized measures of functioning. There were higher PedsQL 4.0 scores for the telephone versus clinic mode of administration in type 2 diabetes, indicating social desirability responding, consistent with the survey research literature (28). Although including these higher values for the 20 patients with type 2 diabetes is a conservative test of our known-groups hypothesis, this finding suggests that the telephone mode of administration for type 2 diabetes may result in higher HRQOL scores. In the type 1 diabetes sample, there was no mode of administration differences, consistent with the field test (8). Finally, the healthy sample was somewhat younger than the diabetes sample, with several race/ethnic differences. The differences in generic HRQOL between type 1 and type 2 diabetes may be attributed to differences in participant characteristics. In type 2 diabetes, patients were more likely to be overweight, had lower SES, and were from ethnic groups in which health disparities have been documented. Other causes of these HRQOL differences should be a focus of future research.

Finally, the small-to-medium correlation effect sizes between perceived HRQOL and HbA_1c are consistent with the broader literature across diseases, as succinctly summarized by McHorney (29, p. III58): “QOL scores correlate modestly at best with clinical outcomes. This finding suggests that clinical and human function are relatively independent. It does not imply that one or the other is inherently superior or correct. They simply measure different things, and using both will likely yield more information than any set alone.”