A number of well-validated quality of life (QOL) measures for adults have been developed, many of which are used in routine clinical trials. The inclusion of QOL measures in clinical trials has in part come from increasing recognition that self-reports on subjective states can provide information about the consequences of treatment plans (such as behavioral or psychological outcomes) that may not be captured by traditional outcome indices 1. In the last twenty five years, a number of well-validated child instruments have been developed 2.

Given the lower cognitive and language skills of young children, the majority of child QOL instruments have been developed for children above eight years with proxy reports (usually parent) used to gain information about younger children 3. However the value of obtaining children's self-reports about their health, functioning, abilities, and emotions is increasingly recognized within both medical care and child health research 2. Several generic and disease-specific QOL measures are now available that include parallel child and parent proxy-report versions (for example, generic: the Pediatric Quality of Life Inventory™ (PedsQL™) 45, the Child Health and Illness Profile – Child Edition (CHIP-CE™) 6, and the KINDL™ 7, disease-specific: the Cystic Fibrosis Questionnaire (CFQ) 8, the Child Health Ratings Inventory (CHRIs) 9, and the How Are You? (HAY) 10).

The availability of measures with parallel child and parent versions has raised questions about the level of agreement between children's own views and those of their parents about child functioning. The literature is relatively confused, with reported of poor parent-child agreement [e.g., 1112], and of moderate to high agreement [e.g., 1314]. Parent-child agreement may be affected by a number of variables 15. In a review of the relationship between child and parent QOL ratings, Eiser and Morse 16 concluded agreement is dependent on the domain being measured, with higher agreement for physical aspects of health compared to emotional or social aspects. Eiser and Morse 16 also reported evidence of higher agreement between parents and chronically sick children compared with parents and healthy children. Some researchers have found evidence that parents of sick children tend to underestimate their child's QOL compared with children's own ratings [e.g., 17]. The reverse (i.e., overestimation) has been reported with parents of healthy children [e.g., 131819].

Agreement between child and parent proxy-ratings may also vary by the age of the child. Eiser and Morse 16 identified only two studies examining the effect of age 413. Varni et al. 4 reported that agreement was highest between children with cancer and their parents for cognitive functioning, and highest between adolescents and parents for physical functioning. Theunissen, Vogels, Koopman, Verrips, Zwinderman, and Verloove-Vanhorick 13 found that parent-child agreement was related to child's age and their positive emotions ratings. Specifically, Theunissen et al. 13 reported that older children (10–11 years) with low positive emotion scores agreed less with their parents than younger children (8–9 years), and older children with high positive emotion scores agreed more with their parents. A study by Annett, Bender, DuHamel, and Lapidus 20 with children with asthma reported parent-child agreement increased with child age. Ronen, Streiner, and Rosenbaum 21 reached similar conclusions, with younger age predicting greater differences between parents and children with epilepsy.

An additional factor for consideration here is the impact of parents' own functioning and well-being. Eiser, Eiser, and Stride 22 found that mothers who rated their own well-being as poor also rated their child's QOL as poor, suggesting that parents project their own feelings on to judgments about the child's functioning. In addition, Goldbeck and Melches 23 reported a significant interaction effect of parental QOL and patients' self-reported QOL in predicting parental proxy reports of their children's QOL.

Part of the confusion described above may relate to the statistical methods employed to compute parent-child agreement. The most frequently used statistic for examining agreement between child and parent reports has been the Pearson product-moment correlation coefficient 16. However Pearson r values provide information on the covariation among scores but do not indicate absolute agreement 24. A more appropriate statistic for examining agreement between raters is the Intra-class correlation coefficient (ICC). ICC values provide an index that reflects the ratio between subject variability and total variability 25.

It is useful to examine mean differences between children's and parents' reports, as it is possible for their scores to be correlated (i.e., linearly related) but also show statistically significant differences in mean scores 26. Analyses which include both correlation and mean difference testing are needed in order to provide more conclusive evidence regarding the relationship between parent and child ratings. We identified two studies which adopted this approach 2627. Both assessed parent-child agreement for QOL ratings in pediatric cancer populations. These researchers found moderate correlations between child and parent scores and no group differences between their mean scores 2627. It is questionable if test scores displaying moderate correlations can be considered equivalent. Correlation coefficients of at least 0.70 are usually required for a reasonable prediction of individual test scores 28.

Our goal was to extend knowledge of the factors influencing child-parent agreement in rating child QOL in healthy populations. First, we considered differences in agreement across two different statistical methods. Intra-Class correlation coefficients (ICC) were used to evaluate correlational consistency between child and parent scores on the generic core scales of the UK-English version of the Pediatric Quality of Life Inventory™ 4.0 (PedsQL™) 45, and Wilcoxon median testing to evaluate agreement between child and parent ratings on this measure. Second, we considered the effect of chronological age and domain type (i.e., physical vs. psychosocial aspects) on agreement between children's and parents' reports on the PedsQL™. We assessed parent-child agreement across three age groups (stratified by year in school), and across the Physical Health domain and Psychosocial Health domains. Third, we investigated the relationship between parents' own well-being and their ratings of their child's QOL.

Based on the findings of Eiser and Morse 16, we predicted that in this healthy sample parent-child correlations would be low to moderate. Furthermore, we expected statistically significant group differences in child and parent median scores, specifically parents' scores would be higher than children's scores due to the over-estimation effects found in previous studies with healthy child populations 131819. In relation to chronological age, we expected that parent-child agreement would increase with child age, based on the findings of previous work 42021. In relation to domain type, we expected that parent-child correlations would be higher for the physical health compared to psychosocial health domains. Finally following on from the findings of Eiser, Eiser and Stride 22 on the effect of mother's own well-being on their ratings of their children's QOL, we expected that parents' own QOL levels would be correlated to their proxy-reports of child QOL.

The results of this study show that agreement between child and parent proxy-reports of child QOL in healthy populations can be affected by the domains investigated, the age of children, and parents' own QOL.

Consistent with our predictions parent-child agreement on the PedsQL™ was low, with ICC values ranging from 0.02 to 0.23. Differences in levels of parent-child consistency were found when analyses were performed separately by age group and domain type. Statistically significant parent-child correlations were found for the oldest age group (7.5 to 8.5 years) for Total Score and Psychosocial Health, and for Physical Health scores in the youngest age group (5.5 to 6.5 years). This result is consistent with the findings of Eiser and Morse 16 who found that parent-child agreement can differ across domains investigated (i.e., higher agreement for physical aspects of health vs. emotional aspects). However our results showed domain and age differences in correlational consistency between child and parent ratings (i.e., higher agreement for younger age on physical health, compared to higher agreement for older age on psychosocial aspects of health).

We examined parent-child agreement on the PedsQL™ using median difference testing in addition to correlational consistency between child and parent ratings. Consistent with our expectations there were median differences between children's and parents scores. Parents reported better child QOL than did their children on the PedsQL™, and this finding is consistent with previous research with similar populations [e.g., 131819].

Parent-child median differences were largest for the older age groups, whereas parent-child scores were not different for the youngest age group (5.5 to 6.5 years). This result contradicts findings from other researchers who have shown agreement increasing with child age [e.g., 2021]. However we included healthy children and their parents, but previous researchers tested children with asthma 20 and children with epilepsy 21. Eiser and Morse 16 found that parent-child agreement can be affected by children's illness status, therefore the difference in populations may account for the variation in results between studies.

We also considered the relationship between parents' own QOL on the SF-36 and their ratings of their child's QOL on the PedsQL™. Although correlations give no information about the causal direction of a relationship, we found statistically significant correlations between parents' own QOL ratings and their ratings of their child's QOL. These results are consistent with the findings of Eiser, Eiser, and Stride 22 and Goldbeck and Melches 23. This correlation was only statistically significant for the middle age group (6.5 – 7.5 years). Future research needs to explore these subtle age differences in parent-child agreement in more detail.

Our results have implications for the measurement of child QOL and assessment of agreement between parents' and children's reports. We found parent-child agreement can be effected by the types of domains investigated and the ages of children in the sample. Our use of two different statistical methods allowed consideration of both the correlational consistency and the mean differences between parent proxy-report and child self-report scores to be considered. Our findings suggest that differences in the levels of parent-child agreement across previously reported studies may be either: 1) an artefact of statistical methods used; or 2) affected by the different ages of children in their sample populations.

There is a need for further research to explore whether parent-child agreement is dependent on additional factors such as relationship to child (i.e., mother vs. father), the mental health of the parent themselves, and for sick populations by different disease types (e.g., asthma vs. cancer vs. epilepsy). The optimal predictors of high or low parent-child agreement remain to be conclusively determined 26. In addition, future researchers should provide details of both correlation consistency and means difference testing when investigating parent-child agreement. Using more than one statistical method can help provide meaningful data as high correlations between scores do not necessarily indicate high agreement between raters 26. Correlations provide a criterion of relative agreement, but researchers also need an indicator of differences in group mean scores 19.

The results from this study suggest that domains assessed and ages of children can effect parent-child agreement levels. In addition correlational consistency and mean differences in scores can provide different information on levels of agreement in ratings. Our findings support previous researchers 10151623 suggestions of future research to systematically examine the predictors of agreement levels between child and parent proxy-reported child QOL (such as child age or gender, relationship to child, health status, disease type).

The author(s) declare that they have no competing interests.

JC conceived the study, collected the data, conducted the analysis, drafted and revised the manuscript. CE and MB both contributed to the design of the study; interpretation of the data and analyses; and revised the article for important intellectual content. All authors gave final approval of the version to be published.