Fax: (901) 495–3121
Assessment of health-related quality of life in children with cancer
Consistency and agreement between parent and child reports
Version of Record online: 10 APR 2006
Copyright © 2006 American Cancer Society
Volume 106, Issue 10, pages 2267–2274, 15 May 2006
How to Cite
Russell, K. M. W., Hudson, M., Long, A. and Phipps, S. (2006), Assessment of health-related quality of life in children with cancer. Cancer, 106: 2267–2274. doi: 10.1002/cncr.21871
- Issue online: 27 APR 2006
- Version of Record online: 10 APR 2006
- Manuscript Accepted: 15 DEC 2005
- Manuscript Revised: 23 NOV 2005
- Manuscript Received: 1 SEP 2005
- National Cancer Institute. Grant Numbers: R01 CA60616, R01 CA 82378
- American Lebanese-Syrian Associated Charities (ALSAC)
- health-related quality of life;
- proxy measures
In pediatric settings, measures of health-related quality of life (HRQL) are often obtained by proxy, especially from parents, but the correlation between parental report and patient report has not been clearly established. In the current study, the authors examined the agreement between child and parent reports of HRQL in children with cancer (both those receiving treatment and those off treatment) and healthy controls.
Two groups of children with cancer who differed with regard to treatment status (n = 199) and a group of healthy control children (n = 108) were assessed using a standardized measure of HRQL. Both children and parents reported on the child's HRQL using parallel forms of the Children's Health Questionnaire.
Significant parent-child correlations were found for all 10 HRQL scales, and these correlations were higher in the cancer groups compared with controls. Parents in the cancer groups tended to underestimate the HRQL of their children compared with the child report, although there were no significant parent-child mean differences in the group of patients who were off treatment and only 2 significant differences in the group of children receiving treatment. In contrast, parents in the control group tended to overestimate the HRQL of their children, and the parent-child differences were found to be larger, achieving statistical significance on 8 of 10 scales.
Parents and children tend to report comparable child HRQL outcomes, and this is particularly true in oncology populations. In cases in which the child is either too young or too ill to provide a self-report, parent-reported HRQL can be viewed as a reliable substitute. Cancer 2006. © 2006 American Cancer Society.
As survival rates for pediatric cancers have increased, the endpoint criteria for judging the success of clinical trials has expanded beyond measures of survival to include additional outcomes, particularly health-related quality of life (HRQL).1, 2 In equivalence, or noninferiority trials, HRQL may emerge as the primary outcome measure, highlighting the need for accurate measures of this construct.3
In contrast to adult measures of HRQL, which rely on self-report, measures of HRQL in children may use either self-report or proxy report.2 Proxy reporters most often are parents, but occasionally others such as physician, nurse, or teacher informants may be used. There are many reasons for the use of parental or other proxy reports for the assessment of children's HRQL, including cases in which the child may be too young or too ill to accurately report on their HRQL.4–7 In addition, parents are the principle decision makers regarding the child's treatment, at least until the child reaches the age of majority.7, 8 Therefore, understanding parents' perceptions of their children's HRQL and comparing them with the children's perceptions will have important implications for treatment decisions.
When using parent-reported HRQL, either in addition to or in place of child-reported HRQL, it is important to understand the areas of agreement and disagreement between the two. Several studies have reported low to moderate correlations between parent and child reports of HRQL,5, 6, 9–13 with generally stronger concordance noted on physical health domains.11–13 However, strong correlations do not necessarily indicate high agreement. For example, if parents consistently underestimate or overestimate HRQL compared with the child's report, the 2 measures may be highly correlated despite large mean differences. Therefore, correlations provide a criterion of relative agreement that is a necessary, but not sufficient, indicator of source agreement.
A few studies to date have examined differences between parent and child reports of HRQL in clinical populations. Although not shown in all studies,5 research has generally demonstrated a tendency for the parents of children with serious illness to underestimate the HRQL of their children (ie, to report a poorer QOL for their child than the child reports for himself) on various domains including mental health, general QOL, role limitations, and physical functioning.4, 7, 8 This parental tendency to report lower levels of HRQL exists for both children and adolescents, although there is some evidence that it is more pronounced among parents of older patients.7 This trend in parent-child differences appears to apply only to clinical samples. In fact, a few studies have reported that parents of a sample of healthy children tended to overestimate in their report of HRQL compared with their children.4, 6
Although these studies have suggested trends in parental reporting of child HRQL, significant gaps remain in our understanding of parent-child agreement. First, although both correlational and mean difference information are needed to adequately assess parent-child agreement, to our knowledge only a few studies to date have included both.5, 6, 12 Second, most studies examining parent-child agreement of HRQL in clinical samples have not included both illness and healthy comparison groups, or have done so in the context of very small samples.5, 7, 11–14 Finally, studies addressing cancer samples have not distinguished between patients who are receiving treatment and those who are off active treatment.4, 5, 7
The goal of the current study was to extend our understanding of parent-child agreement in reports of HRQL in the pediatric oncology setting. We compared reports of HRQL among 3 different groups: 1) children with cancer who were receiving active treatment, 2) children with cancer who had completed therapy, and 3) healthy control children. Healthy children were predicted to display the best HRQL according to both parent and child reports, and cancer children receiving active treatment were predicted to display the worst HRQL. Beyond the examination of group differences, the primary objective of the current study was to examine relative agreement between parents and children for the overall sample and within groups using both correlational and descriptive comparisons.
MATERIALS AND METHODS
Children with cancer were recruited from outpatient clinics at St. Jude Children's Research Hospital. Children who were ages 7 years to 18 years, who spoke English, and who had a primary diagnosis of malignancy and no known cognitive deficits were eligible. Patients who were at least 1 month from diagnosis were eligible, with no upper limit to time elapsed since diagnosis. Sampling procedures were designed to obtain a broad cross-section of patients in terms of diagnosis, and the time elapsed since diagnosis. Of the 339 patients we approached, 249 (73%) were enrolled initially. Of these enrolled participants, 35 returned incomplete information and 15 withdrew before completing all materials, leaving a sample of 199 cancer patients. Of these 199 patients, 71 (36%) were still receiving active treatment and 128 (64%) were off treatment at the time of their participation in the study.
Children in the control group were recruited through an “acquaintance control” methodology. Cancer patients recruited to the study were asked to identify up to 3 friends from their home neighborhood or school who were within ± 2 years of the age of the patient. Data from those patients who agreed to participate were obtained through the mail, with telephone assistance when necessary. Using this procedure, 367 potential control participants were identified. Of these, 8 refused participation, 5 proved ineligible, and 57 had insufficient or incorrect contact information. Survey packets were sent to the remaining 297 potential control participants and complete information was received back from 108 (36%).
Table 1 displays the medical and demographic information for both cancer groups and controls. The diagnostic breakdown of the cancer groups generally was representative of the population served by the institution. Although the acquaintance control method was used in an attempt to obtain a control group similar to the cancer group with regard to demographic characteristics, a few significant differences between the cancer and control groups emerged. The groups did not differ significantly by gender or age, but were found to differ by race (chi-square test = 11.72; P = .003), with significantly more African American children in the cancer group than in the control group. The groups also differed significantly with regard to socioeconomic status (SES) (chi-square test = 13.65; P = .001), with more SES Group I and II children (i.e., higher SES, using the Hollingshead index15) and significantly fewer SES Group IV and Group V children in the control group compared with the cancer group.
|Control (n = 108)||Cancer Group, Off Treatment (n = 128)||Cancer Group, On Treatment (n = 71)|
|Mean age, y (SD)||12.38 (3.0)||12.95 (3.2)||11.17 (3.5)|
|Gender, no. (%)|
|Male||44 (40.7)||62 (48.4)||42 (59.2)|
|Female||64 (59.3)||66 (51.6)||29 (40.8)|
|Race, no. (%)|
|White||99 (91.7)||100 (78.1)||52 (73.2)|
|Black||6 (5.6)||22 (17.2)||17 (23.9)|
|Other||3 (2.7)||6 (4.7)||2 (2.8)|
|SES group (%)*|
|I and II||84 (77.8)||76 (59.4)||39 (54.9)|
|III||18 (16.7)||30 (23.4)||15 (21.1)|
|IV and V||3 (2.8)||16 (12.5)||11 (15.5)|
|Parent participant, no. (%)†|
|Mother||98 (90.7)||100 (78.1)||60 (84.5)|
|Father||9 (8.3)||25 (19.5)||8 (11.3)|
|Other‡||1 (0.9)||3 (2.3)||1(1.4)|
|Diagnostic category, no. (%)|
|ALL||25 (19.5)||22 (31.0)|
|Other leukemia||5 (3.9)||6 (8.5)|
|HD/NHL||20 (15.6)||10 (14.1)|
|Solid tumor||46 (35.9)||21 (29.6)|
|Brain tumor||32 (25.0)||12 (16.9)|
|Time since diagnosis, no. (%)|
|≥ 1 mo to 6 mo||12 (9.4)||40 (56.3)|
|> 6 mo to 18 mo||31 (24.2)||16 (22.5)|
|> 18 mo to < 5 y||32 (25.0)||15 (21.1)|
|> 5 y||53 (41.4)||—|
Children's Health Questionnaire
All participants were assessed using the Children's Health Questionnaire (CHQ),14 a generic, health-related QOL measure that is available in a child report format and several parent report formats of differing length. We used the intermediate-length parent report version, which contains 50 items covering 10 different domains, including physical functioning, role/social limitations-physical, general health perceptions, body pain, change in health, impact on the family, role/social limitations-behavioral, self-esteem, mental health, and general behavior. All subscales are scored on a range of 0 to 100 so that higher numbers reflect a better HRQL.
The child report version of the CHQ contains items identical to those in the parent report version, with the identifier “you” or “your” in place of “your child,” and contains an additional 37 items not found in the parent form. Because the goal of the current study was to compare child and parent reports of HRQL, only the 44 overlapping items were used. The items used in the child report cover the same 10 domains as the parent report. The CHQ is one of the most widely used measures of HRQL for children, and the reliability and validity for both the parent report and child report forms have been well established.14 Child and parent completed their questionnaires independently, and were requested not to consult with each other regarding their responses.
Demographics and HRQL
Before comparing child and parent reports of HRQL, we examined the effects of demographic variables on each report. There was no effect of gender found on either the child or parent reports on any of the 10 subscales of HRQL, but a few differences were found by age, ethnicity, and SES. The age of the child was found to be correlated with 5 scales reported by the child and 3 scales reported by the parent. For children, age was found to be negatively correlated with change in health and self-esteem and positively correlated with role/social limitations-behavioral, role/social limitations-physical, and impact on the family. For parents, child age was found to be positively correlated with role/social limitations-physical, general behavior, and impact on the family. However, all these correlations were weak (correlation coefficient [r] range of 0.12–0.22). Differences by ethnicity also were observed in child-reported HRQL on 3 scales: pain, mental health, and the impact on family. White children reported better HRQL on these scales than African American or other minority children. There were no other differences by ethnicity noted on the remaining 7 subscales by child report or any of the 10 subscales by parent report. Differences with regard to SES were observed in child-reported HRQL on 2 scales (physical functioning and general health perceptions) and by parent report on 3 scales (change in health, physical functioning, and role/social limitations-behavioral). Higher SES groups reported better HRQL than lower SES groups on these scales. With regard to medical diagnosis, using the 5-group diagnostic breakdown from Table 1, there were no consistent differences noted between diagnostic groups with regard to HRQL, by either child or parent report.
HRQL Differences by Group
The mean values on the 10 HRQL subscales for the 2 cancer groups and the control group are presented in Table 2. Group differences were assessed using analysis of covariance (ANCOVA), controlling statistically for age, ethnicity, and SES differences. There were significant group differences noted on child-reported HRQL on 7 of the 10 scales. Generally, and as expected, children in the control group reported the best HRQL, and the cancer group that was receiving treatment reported the worst HRQL, with the cancer group that was off treatment found to be intermediate. The significance of post hoc tests between the 3 groups differed across scales as follows: With regard to the change in health, physical functioning, and impact on the family scales, all 3 groups differed significantly from each other. With regard to the role/social limitations-physical scale and the general health perceptions scale, children in the control group reported a better HRQL than children in the 2 cancer groups, who did not differ from each other. With regard to the body pain scale, children in the cancer group receiving treatment reported significantly lower HRQL compared with children in the off treatment group or the control group, who did not differ from each other.
|Control Mean (SD)||Cancer Group, Off Treatment Mean (SD)||Cancer Group, On Treatment Mean (SD)||Group Difference.|
|Child||94.50†‡ (10.7)||79.48† (22.7)||66.22†‡ (28.0)||30.15||< .01|
|Parent||98.07†‡ (5.9)||75.24† (27.9)||62.91†‡ (27.9)||45.01||< .01|
|Child||96.24†‡ (11.3)||80.95† (27.9)||73.67†‡ (28.8)||14.77||< .01|
|Parent||99.49†‡ (2.9)||75.44† (35.3)||57.55†‡ (37.8)||34.65||< .01|
|General health perceptions|
|Child||74.81†‡ (11.6)||61.82†‡ (16.8)||58.80† (14.1)||25.46||< .01|
|Parent||81.21†‡ (13.2)||54.56†‡ (17.1)||54.45† (19.7)||81.57||< .01|
|Child||76.88†‡ (22.0)||72.14† (26.0)||63.60† (25.5)||7.76||< .01|
|Parent||82.55†‡ (18.2)||73.60† (25.4)||56.04† (22.5)||24.86||< .01|
|Change in health|
|Child||85.97†‡ (16.4)||75.41† (23.3)||67.30† (24.5)||16.11||< .01|
|Parent||91.33†‡ (13.7)||72.41† (22.2)||66.32† (23.9)||38.03||< .01|
|Impact on the family|
|Child||87.86† (16.3)||76.15† (24.4)||59.92† (26.7)||22.05||< .01|
|Parent||89.63† (12.9)||74.74† (24.4)||56.45† (24.2)||40.68||< .01|
|Child||90.68‡ (20.7)||85.77 (23.3)||82.44 (23.4)||–||NS|
|Parent||94.67†‡ (13.7)||82.75† (29.0)||75.05† (35.5)||6.65||< .01|
|Child||78.67† (17.7)||75.22† (17.6)||72.92† (18.5)||3.73||.03|
|Parent||81.17† (19.5)||72.95† (19.3)||70.50† (18.5)||6.72||< .01|
|Child||73.06‡ (18.1)||74.12 (20.9)||70.40 (18.2)||–||NS|
|Parent||76.84†‡ (12.3)||73.61† (15.4)||70.57† (16.3)||3.19||.04|
|Child||76.34‡ (13.4)||75.96 (13.4)||76.34 (17.5)||–||NS|
|Parent||80.71‡ (13.5)||76.45 (17.6)||79.56 (15.5)||–||NS|
There were significant group differences noted with regard to parent–reported HRQL on 9 of the 10 scales, with only the general behavior scale failing to differentiate between the groups. The pattern of group differences was similar to that observed by child report. On the physical functioning, role/social limitations-physical, body pain, and impact on the family scales, all 3 groups differed significantly from each other. With regard to the general health perceptions, change in health, role/social limitations-behavioral, and self-esteem scales, parents of children in the control group reported better HRQL than the parents of children in the 2 cancer groups, who did not differ from each other. With regard to the mental health scale, parents of children in the control group reported better HRQL than parents of children in the cancer group who were receiving treatment, whereas neither group differed significantly from the cancer group of children who were off treatment.
Agreement between Parent/Child Report
Pearson correlation coefficients between child and parent reports on the 10 HRQL subscales are presented in Table 3. The agreement between child and parent reports was found to be moderate to high for the entire sample, with correlations ranging from 0.397 (mental health) to 0.660 (physical functioning). When correlations were calculated separately for the 3 groups; the correlations generally were higher for the 2 cancer groups (0.344–0.592 for those receiving treatment and 0.384–0.677 for those cancer patients who were off treatment) compared with the control group (0.204–0.483). Closer examination revealed that correlations in the control group were lower in part because of a restriction in range. When many parents and children in this group reported the maximum value (100) for a scale, it weakened the linear correlation between the 2 reports.
|Control||Cancer Group, Off Treatment||Cancer Group, On Treatment||Total|
|General health perceptions||0.366*||0.384*||0.252‡||0.508*|
|Change in health||0.269‡||0.677*||0.344‡||0.567*|
|Impact on the family||0.204‡||0.532*||0.544*||0.588*|
Agreement in mean scores
Agreement in parent and child reports of HRQL on the 10 scales were assessed using a series of Student t tests for paired data samples within each of the 3 groups. In the group of cancer patients who were off treatment, there was only 1 significant child-parent difference in the 10 scales. Parents reported a worse HRQL than their children on the general health perceptions scale (t  = 4.08; P <.01). In the group of cancer patients receiving treatment, there were only 2 significant child-parent differences. Parents reported a worse HRQL than their children on the physical functioning scale (t = 2.09; P = .04) and role/social limitations-physical scale (t = 3.20; P <.01). In contrast, in the control group, significant child-parent differences were observed on 8 of 10 scales. Parents reported a significantly better HRQL for their child than the children did by self-report on the physical functioning (t = −2.26; P = .03), role/social limitations-physical (t = −2.62; P = .01), general health perceptions (t = −4.98; P <.01), body pain (t = −2.41; P = 0.02), change in health (t = −3.20; P <.01), role/social limitations-behavioral (t = −2.13; P = .04), mental health (t = −2.04; P = .04), and general behavior (t = −3.06; P < .01) scales. These outcomes are displayed graphically in Figure 1 for 6 of the scales that were most relevant to the pediatric oncology setting.
The major conclusion of the current study is that reports of a child's HRQL from parents and children are highly consistent, both in terms of correlation and absolute levels of HRQL, and this was found to be particularly true in populations of children with cancer. Not only were there moderate parent-child correlations found on the majority of CHQ scales, but parents and patients reported very similar mean levels of HRQL. This level of parent-child consistency is considerably higher than that typically reported with other measures of child affect or behavior.16–18 It is not clear why parent-child correspondence should be so much better with a QOL measure than more traditional measures of anxiety or depression. Although some of the scales assess observable aspects of health and behavior, others assess less observable internal functioning (mental health, self-esteem) and parent-child agreement was found to be equivalent on both types of outcomes. Although it would be preferable to obtain data from both informants, the primary implication of these findings is that data can be obtained with reasonable confidence from either parent or child if only 1 informant is available.
Past research using clinical samples has suggested that parents tend to underestimate their child's HRQL when compared with the child's report.4, 7, 8 Although this trend toward parental underestimating was present in our cancer sample on the majority of scales, the difference was seldom found to be significant. Parents and their children had moderate to high agreement and no significant differences on 7 subscales for the group of patients receiving treatment and on 9 of 10 subscales for the group of patients who were off treatment. The only nonsignificant correlation found in the cancer group that was receiving treatment was for the role/social limitations-physical scale. Results for this group also demonstrated a significant parent-child difference on this scale, with parents reporting more physical limitations than their children. Although parent and child reports were found to be moderately to strongly correlated, a significant parent-child difference also was found in the cancer group that was receiving treatment on the physical functioning scale. Parents were consistently reporting worse physical functioning than their children, resulting in both a significant correlation and difference between the reports.
The opposite trend in healthy samples, in which parents overestimate their child's HRQL compared with their child's report,5, 7 was clearly observed in the current study. In the healthy control group, significant parent-child correlations and nonsignificant parent-child differences indicating good agreement were found on only 2 scales: self-esteem and impact on the family. Six other scales demonstrated low to moderate parent-child correlations and small but significant parent-child differences. In each case, the mean of parent–reported HRQL was better than the mean of the child report, but both reporters typically rated HRQL very highly. Therefore, the low correlations appear to partly be the result of a restricted range, given that a high percentage of both parent and child reports were at the ceiling level. The parent-child differences for these scales typically were small, raising the possibility that although the discrepancies were statistically significant, they may not be clinically significant.
The solid finding of child-parent agreement regarding HRQL in cancer patient samples has important implications. Foremost, the observed agreement lends confidence to researchers and clinicians in situations in which only 1 reporting source is available. In cases in which the child is either too young or to ill to complete the CHQ, parent-reported HRQL can be viewed as a reliable substitute. There is sufficient evidence to suggest that both parent and child reports provide the same information, allowing the researcher to choose the most cost-effective and time-effective source.
Even with good overall parent-child agreement, the parents of cancer patients tended to underestimate the HRQL of their child, a trend that was found to be significant on 2 scales. In the cancer group that was receiving treatment, the lowest parent-child agreement was with regard to physical functioning and physical limitations. For both of these scales, parents overestimated the challenges faced compared with the child's evaluation. Therefore, when using parent-reported HRQL as an outcome measure, these 2 scales should be interpreted carefully, especially for patients who are receiving treatment.
A limitation of the current study is the differential rate of participation in the cancer and control groups, thereby resulting in a possible selection bias. Although 73% of cancer patients who were approached agreed to enroll in the current study, only 36% of the control children enrolled. This differential selection would be a greater concern if our interest was primarily absolute cancer control differences. However, because our primary focus was the consistency of child and parent reports, between-group selection is less of a concern and not a major limitation for the generalization of the findings. Another potential limitation involves demographic differences between cancer and control groups with regard to SES and ethnicity. To address this, group differences in HRQL were examined after statistically controlling for these variables. Finally, only 1 parent, typically the mother (84.6% of the total sample), reported on their child's HRQL. Obtaining reports from 2 parents is desirable, but often impractical. Although our reliance on reports predominantly received from the mother may limit the generalizability of the findings, it also is reflective of typical clinical assessment practices in pediatric oncology settings. A post hoc examination revealed no significant difference in HRQL outcomes or in parent-child differences between maternal and paternal informants.
There is sufficient evidence to suggest that parents and children are in agreement when reporting on the child's HRQL. Furthermore, this agreement was found to be greatest in the cancer sample, supporting its use as a meaningful outcome measure even when only 1 informant is available.
- 14The Child Health Questionnaire (CHQ): a users manual. Boston: HealthAct; 1999., , .
- 15Four factor index of social status. New Haven, CT: Yale University Press; 1975..