Parent–child agreement on health-related quality of life (HRQOL): a longitudinal study

Over the past years, a number of self-reported instruments assessing health-related quality of life (HRQOL) have been developed for children and adolescents. A systematic review identified almost a hundred instruments, 30 of them generic and more than sixty specific questionnaires addressed to this population by the year 2008 [1]. Most of the specific instruments collect information from proxies, mainly parents.

HRQOL assessment by proxies is controversial given the subjective nature of this concept. However, parents’ ratings are often used to assess HRQOL in young children, due to age, and/or lack of cognitive and linguistic skills necessary for self-completion of HRQOL measures [2]. Proxy-respondent bias represents a limitation on the assessment of HRQOL by parents [3, 4]. Nevertheless, the parents’ perspective is frequently necessary given that they are responsible for the children and the ones who decide on their health needs and their use of healthcare services. This is an open research area where no consensus exists on when and how to collect and interpret information from parents on HRQOL [5].

Several factors have been described as associated to the degree and direction of parent–child agreement on the children’s HRQOL, mainly in cross-sectional studies. Parental well-being and child pain have been associated to disagreement in children with cerebral palsy [6]. Another study in children with chronic pain found no significant differences between self- and parents report on the PedsQL [7]. Direction of disagreement was also variable. A relatively high proportion of children with chronic conditions agreed with their parents while a low proportion of children scored lower (24%) and higher (32%) than their parents [8]. Some factors such as age [9], gender [10], parent’s age [11], or parental health [12‐14] show variable, non-consistent results, while other factors such as child’s health status present more consistent results. In general, a moderate to high level of parent–child agreement was shown in children with chronic conditions [2, 15, 16], except for children with psychiatric conditions, such as depression, for whom a low level of agreement on their HRQOL was described [17]. Levels of agreement when comparing the evaluations of more objective dimensions, such as physical domains, are higher than those for more subjective aspects of health [9, 11, 14]. Direction of disagreement was also different according to the dimension of HRQOL analyzed [18].

Few studies have evaluated changes on parent–child agreement in HRQOL over time [19, 20]. Some characteristics of these studies (such as small sample sizes or different instruments used for parents and children) call for additional longitudinal research. The magnitude and direction of this change can be clearly appreciated through longitudinal studies, since they allow identifying factors associated with changes over time and dimensions with marked changes in the degree of agreement.

The KIDSCREEN follow-up study [21] based on the Spanish sample of the European KIDSCREEN project [22], was designed to analyze changes on HRQOL over a period of 3 years of follow-up. We collected information of the children’s HRQOL both from the children themselves and from their parents at baseline and at a 3-year follow-up. The objectives of the present study were to assess parent–child agreement on HRQOL in a 3-year longitudinal study, and to analyze factors associated with these changes.

The main interest was to explore the association of age and time of follow-up on the level of parent–child agreement. According to the literature review, a higher level of agreement in those more “observable” dimensions was expected. It was also expected that the health status of both children and parents among other factors would be associated with the level of agreement at the follow-up.

The response rate at follow-up was 54% (n = 454). A total of 418 parent–child pairs with complete information were included in the analyses. Table  1 shows the sample characteristics of those included in the study; mean age at baseline was 12.6 (standard deviation, SD= 4.8); 55.5% of the sample was in the 10 to 13 years-old group at baseline, and the 49.3% reported intermediate FAS category. Mental health was stable in 75.4% of children, and perceived health at baseline and follow-up was very good for approximately 70% of the sample.

Table  2 shows self- and parent-administered responses on children’s HRQOL both at baseline and follow-up as well as child- parent differences and 95% CI of the lower and upper limits of agreement. Statistically significant differences between self and parent reports were found on the dimensions of Physical well-being and Autonomy and relations with parents of the KIDSCREEN-27 at baseline (p<0.05). Mean child- parents differences ranged from 0.03 for Parent relationships and Autonomy at baseline to 1.76 on Physical well-being at baseline. Lower and upper limits of agreement showed great variability.

The level of agreement for Physical well-being varied according to the HRQOL level (Figure  1). Agreement was better for worst values of the Physical well-being, and slightly diminished at follow-up even in these worst values of HRQOL. Psychological well-being presented a similar distribution (Figure  2).

In general, ICC showed a level of agreement between low and moderate in both baseline and follow-up administrations. Approximately 20% of agreement was very low in both administrations (Table  3). Physical well-being showed the highest level of agreement at baseline (CCI=0.59; 0.53-0.65); ICC decreased in girls at the follow-up (ICC in 2003 for girls = 0.61; 0.52-0.69; and ICC in 2006 = 0.37; 0.24-0.48), and in the older age group (ICC in 2003 = 0.69; 0.58-0.77; and ICC in 2006= 0.42; 0.24-0.56), as the 95% CI did not overlap. The ICC for Psychological well-being was 0.46 (0.38-0.53) in 2003 and 0.39 (0.30-0.47) at follow-up (2006). No differences were found in the level of agreement neither in the total sample nor stratifying the sample in the dimensions of Parent relationships and Autonomy (baseline ICC = 0.42; follow-up ICC = 0.41) and the KIDSCREEN-10 (baseline ICC = 0.59; follow-up ICC= 0.44).

Table  4 shows the results of GEE models. The main factors associated to parent–child disagreement were age, rater and perceived health. Interaction was found between age and rater: with each year of increasing child’s age, children scored lower than parents on Parent relationships and Autonomy (−0.47; -0.71 / -0.23), and on the KIDSCREEN-10 (−0.49; -0.73 /-0.25). An interaction was found between the rater and year of assessment: children scored higher than parents at follow-up (in 2006) on Parent relationships and Autonomy dimension (2.92; 1.38 /4.46).

The present study is one of the few that analyze changes over time in the level of parent–child agreement on the kid’s HRQOL, based on a general population sample of children and adolescents. In general, the level of agreement was low to moderate in both the baseline and follow-up assessments, but tended to be lower at the follow-up. Child’s age and parent’s self-perceived health were the main factors associated to parent–child differences over time.

Few studies have analyzed parent–child agreement in a longitudinal design. One of these studies was carried out on 83 children with diabetes type I and analyzed HRQOL using the Child Health Questionnaire for parents (CHQ PF-50) and the self-reported version for adolescents (CHQ CF-80) from 12 to 18 years old [19]. Similarly, another study was carried out in a sample of 31 children with attention deficit hyperactivity, and HRQOL was evaluated through the Child Health and Illness Profile (CHIP-CE) which includes 44 items in the self-reported version and 75 items in the parent-reported version [20]. None of them carried out a formal comparison of longitudinal changes in the level of parent–child agreement. Our study included a sample of 418 parent–child pairs. The study allowed to evaluate changes over time in the level of agreement in a general population sample of children and adolescents, and to analyze factors with potential influence in these changes using an instrument with identical content for each respondent. Moreover, the HRQOL instrument used in the present study (KIDSCREEN) has demonstrated its ability to detect changes over time [35], which is a necessary psychometric property to determine the level of agreement between parent and child in a longitudinal design.

The results confirm those of previous studies that reported a higher level of agreement in the more observable dimensions [16]. Moreover, agreement on Physical well-being presented a decrement over time, and especially in girls and in the older age group. These results are consistent with one of the few longitudinal studies conducted in a population with diabetes type I in which agreement between adolescents and their parents also diminished over time in observable dimensions [19]. Other studies analyzing factors with potential influence on the level and direction of agreement showed a slightly higher level of parent–child agreement in children with cerebral palsy, compared with the results from the present study [6]. A general population study of parent–child agreement showed similar results to the present study, compared with our baseline results [18].

It is worth noting that the main factors associated to parent–child disagreement over time in the present study were age and to a lesser extent self-perceived health of both children and their parents. We found scarce or even no influence of other factors such as gender or chronic conditions in the level of agreement. The latter may be due to the healthy characteristics of the population sample analyzed in the present study. Different results could be found in children with serious chronic conditions and worse health because parents spend more time with the children helping with their care [6]. In this case, HRQOL could be assessed from both parent and child, or even just from the parent. This fact deserves attention for future studies.

The statistical method used to analyze agreement could be a factor with potential influence on the results. The most frequently used statistic for examining agreement between child and parent reports has been the Pearson product–moment correlation coefficient [2]. However, Pearson coefficients provide information on the covariation among scores but do not indicate absolute agreement [36]. A more appropriate statistic tool for examining agreement between raters is the ICC. ICC values provide an index that reflects the ratio between subject variability and total variability [37]. Moreover, the use of Bland and Altman method allowed graphical display of differences in the level of agreement and determine the degree of its variability [32]. For example lower and upper 95% CI for agreement on Physical well-being at baseline ranged from −20.89 to 22.27, a fact that discourages the use of parents as proxies on HRQOL assessment at individual level.

Some limitations of our study deserve comment. Response rates at follow-up (54%) could have biased the assessment of changes on HRQOL, and consequently, the level of agreement. Nevertheless, the response rate was similar to that in other longitudinal population-based studies [38], and although those followed up were slightly younger and from more educated families than non-participants, there were no significant differences in their baseline HRQOL scores [21]. Secondly, the sample size allowed us to detect a difference in the ICC of 0.13 (a relatively high difference). However, when we analyzed the agreement at the baseline contact including all parent–child pairs (n = 840), 95% CI were similar to those obtained at follow-up (data not shown). These data support the idea that the results are more related to variability in the level of agreement than to the small sample size. On the other hand, the present study included a higher sample size than other published studies [19, 20]. Thirdly, some factors with potential influence on the level of agreement such as changes on some health-related behaviors (starting to smoke, drinking, etc.) and special needs were not included and should be taken into account in future studies. It is worth noting that response shift could have influenced the results on the level of agreement over time, although probabilities of this fact are low given that the sample characteristics and the time passed between baseline and follow-up could have made this less likely [39]. Finally, adolescents 19–21 years old would be expected to show a lower level of agreement than the rest of age groups. Nevertheless, very few differences were found in the results on this age group. This could be associated to the fact that most adolescents from the present study were still living with their families and very few of them declared to live independently from the nuclear family at follow-up.