Determinants of health-related quality of life proxy rating disagreement between caregivers of children with cancer

Data of this study are derived from the “Amsterdam Parent Project”, a cross-sectional multicenter study on parental sleep, QoL and psychosocial functioning in childhood cancer. A detailed description of the Amsterdam Parent Project is given elsewhere [14]. For the purposes of the current study, the questionnaires on parent-rated child HRQoL as well as those on parental QoL and distress were included. In the Amsterdam Parent Project, both parents were invited through mail to complete validated questionnaires independently from each other, either on paper or through an online assessment. If parents were not living at the same address, the parent that received the invitation was asked to transfer it to the other parent. Inclusion criteria were having a child diagnosed with cancer or any type of brain tumor between January 2010 and January 2015 in the Emma Children’s Hospital, Amsterdam University Medical Centers (location VUmc or AMC) and receiving follow-up care in one of these centers at time of study. Parents of deceased children or children receiving palliative care were excluded, as well as parents who were insufficiently fluent in Dutch to complete questionnaires independently.

Ultimately, 352 parents (202 mothers and 150 fathers, response rate 48%) were included, comprising 121 parental couples. One couple was excluded from the current study because they had a child younger than 2 years of age and therefore could not complete the proxy HRQoL questionnaire (see below). The data of the remaining 120 couples are reported here.

The 120 parental couples in this study consisted of 120 mothers and 120 fathers; there were no same-sex couples. Three respondents were step parents (two stepfathers and one stepmother). Mean parental age was 43.6 (± 7.6) years (45.1 (± 7.6) in fathers and 42.2 (± 7.4) in mothers). Their children (n = 120) were on average 3.3 (± 1.4) years from diagnosis and the majority (87%) had finished treatment. No significant differences were found in child characteristics between participants of the study and other children of the Amsterdam Parent Project plus non-responders, except for the percentage of children in active treatment (13% in participants vs 7%, p = 0.04; data not shown). Parent and child characteristics are summarized in Table 1.

Compared to proxy reports of healthy children, fathers and mothers of children with cancer both reported significantly lower child HRQoL on all domains (Table 2).

Parent’s own QoL scores were available from 110 couples. Mothers reported lower mental QoL (MCS) scores than fathers (mean difference 3.9 ± 12.2, p = 0.001), but similar physical QoL (PCS) scores (Table 1). Additionally, mothers reported higher distress than fathers, although their average distress levels were not in the clinical range (mean thermometer score 3.4 vs 2.6, p = 0.005).

Table 3 shows the correlations between proxy-rated child HRQoL and parental QoL. In both fathers and mothers, small to moderate positive correlations (± 0.5) were found between parental MCS and parent-rated child’s total HRQoL score and psychosocial score (PSHS). Additionally, there was a moderate correlation between MCS in fathers and child’s emotional functioning. Correlations between parental MCS and other child domains were small (0.3–0.4). There were no significant correlations between maternal PCS and parent-rated child HRQoL. In fathers, own physical QoL (PCS) had a significant small correlation (0.3) with parent-rated child’s emotional functioning. Regarding distress, for both mothers and fathers, significant moderate negative correlations (0.4–0.6) were found between parental distress and all proxy scales.

Table 2 summarizes parental agreement on different PedsQL subscales. Differences in proxy ratings between fathers and mothers were not significant and effect sizes were small (0.03–0.10). Intra-class correlation coefficients were good on the PSHS, social functioning and school functioning scales, and excellent on the physical functioning, emotional functioning and total HRQoL scales.

Bland–Altman plots (Fig. 1) did not reveal any systematic patterns in disagreement, but there was a very wide range in the magnitude of differences (up to 50 points in some couples). These differences went in both directions; i.e. both higher ratings by fathers compared to mothers and vice versa.

Table 4 shows the quartiles of the proxy difference irrespective of the direction, with corresponding means. The mean difference in the highest quartile ranged from 18.6 (total HRQoL) to 29.9 points (school functioning). Univariate regression models are displayed in Table 5, and the final multiple regression models in Table 6. For the total score and PSHS score, couples with the largest difference were more likely to have a child in active treatment (OR 3.1 [0.8–11.2], p = 0.091, and OR 4.2 [1.2–15.2], p = 0.030, respectively). For the physical functioning and school functioning score, couples were more likely to diverge in their mental QoL (OR 1.1 [1.0–1.1], p = 0.023, and OR 1.1 [1.0–1.1], p = 0.06, respectively). For the emotional functioning score, couples were more likely to diverge when they had a child with retinoblastoma (OR 9.1 [1.5–54.9], p = 0.02) or a child with a relapse or second tumor (OR 6.5 [1.8–22.9], p = 0.004).

The aims of this study were to assess potential differences in proxy HRQoL ratings between fathers and mothers of children with cancer, and to explore determinants of caregiver disagreement. On average, we found good agreement in paternal and maternal proxy ratings, and no significant differences between parents. This implies that if only one report is available, clinicians and researchers can reasonably assume that, in general, paternal and maternal proxy HRQoL ratings of their child with cancer are interchangeable. However, when comparing the highest rating parent to the lowest rating parent (instead of concentrating on the difference between fathers and mothers), proxy reports in the highest quartile differed on average 20 points. This means that approximately 25% of parents report quite differently about their child’s HRQoL. These couples were more likely to have a child in active treatment or with a relapse, have a child with retinoblastoma, and to diverge in their own mental QoL—compared to the parental couples with a proxy difference in the lower three quartiles (a difference between 0 and 10 points on most scales). Yet for some scales (e.g. social functioning) we could not specify determinants of this discrepancy.

This study adds important information regarding the interpretation of parent-reported outcomes in pediatric oncology.

Previous research in healthy children showed that paternal and maternal proxy HRQoL reports were interchangeable [12]. Moreover, in a sample of adolescent burn survivors as well as in children and adolescents in outpatient psychotherapy treatment, moderate to high agreement was found in proxy HRQoL reports of parent-dyads [30, 31]. However, to our knowledge, no previous studies have been performed on this subject in pediatric oncology. Furthermore, previous studies only focused on differences in paternal and maternal reports, and perhaps overlooked differences that would have been found if they had compared the highest and lowest proxy ratings. Also, determinants of caregiver disagreement have not been previously studied yet.

It is known that mothers are most often the primary caregivers. Especially in pediatric illness populations, mothers report higher distress and lower QoL, factors that can both influence proxy ratings [15‐17]. Also in our study population, mothers reported lower QoL and higher distress for themselves than fathers. Therefore, if parents would rate their child’s HRQoL differently, we anticipated mothers to give lower ratings than fathers. This hypothesis is further strengthened by findings of previous studies in primary caregivers of healthy children, which showed that the relationship between parent’s own (mental) health and their proxy ratings was specifically present in mothers, more than in fathers [15, 18].

However, results of our study show that differences in proxy ratings were bidirectional, and in both parents equally correlated with own distress and mental QoL. The latter could explain why specifically the difference in mental QoL within a couple was a significant predictor of finding a large proxy HRQoL difference (for the physical functioning and school functioning scales). Furthermore, these findings might indicate that the situation of healthy children cannot be directly applied to pediatric illness populations, but future research is warranted.

We found that parents were more likely to differ in their proxy-rated total HRQoL and PSHS scores if their child was still in active treatment. This finding should be interpreted with caution, since we only had few children in our sample that were still in active treatment. Yet a previous study also found differences in agreement on and off treatment, although this study investigated child proxy agreement instead of parental agreement [19]. Especially during the child’s cancer treatment the primary caregiver is likely to be more closely involved with the child than the second caregiver [32, 33]. It might be that this could either enhance or relieve the worries of the second parent, considering the bidirectional differences that we found in the proxy ratings, but this warrants further research.

Additionally, for the emotional functioning score, we found larger differences if parents had a child with retinoblastoma or a relapse. This might be explained by the larger uncertainty that these characteristics entail; i.e. the hereditary nature of retinoblastoma or fear of blindness, and the worse prognosis of relapsed cancer.

Based on previous studies, we would have expected that a discrepancy in distress level between parents would be a significant predictor as well [15, 16]. However, we did not find this. An explanation might be that our sample did not have clinical distress scores on average.

With regards to sociodemographic factors, we did not find a relationship between having a proxy difference in the highest quartile and differences between parents in educational level or cultural background, although we know that these factors can influence proxy ratings [3, 4]. This is probably explained by the little variance that we had in our sample regarding these variables. The majority of parents in our sample had a Dutch background and middle or high educational level, which is not representative for the entire Dutch population.

Finally, age and sex of the child did not predict a larger proxy difference in our study. Previous studies found conflicting results regarding these factors [3]. It might be that parents with a very young child diverge more in their scores, since it is more difficult to rate for example emotional functioning if a child cannot well express his feelings [4]. However, the children in our study were on average 11 years old, and we had few toddlers in our sample (i.e. 17 children under the age of 4 years).

Several implications can be derived from our study. First, in some hospitals, electronic PROs are already systematically implemented in clinical practice. An example of this is the KLIK-portal in pediatric oncology [34, 35]. Before each outpatient visit—both during and after cancer treatment—parents and children from 8 years of age complete an online HRQoL assessment through this portal. A ten points decline in child’s HRQoL over time is flagged as clinically relevant. Since only one parent completes the proxy report and this parent may differ at various time points, our study findings are directly important to clinical care. About 25% of the parents in our study diverged more than 10 points, especially during treatment. In this case, if the respondent across both time points would be different, a decline in child’s HRQoL might reflect a proxy difference rather than a real decline. It should therefore be part of health-care providers’ training in using PROs in clinical practice that caregivers might differ in their proxy reports. This could be specifically important in divorced parents. Unfortunately, we had too few divorced parents in our sample to draw any conclusions on this group.

Furthermore, an important conclusion of our study is that parent gender is not really important in the interpretation of proxy reports; if in doubt about the interpretation, clinicians should be aware of the family context (i.e. child’s clinical characteristics and parents’ own functioning). Future studies should further explore the characteristics of the parents with a clinically relevant difference in proxy ratings. Additionally, it would be interesting to investigate parental agreement in other pediatric illness populations.

Our study has several limitations. First, the cross-sectional design only provides information on one specific moment in time, and the relatively low response rate might indicate some participation bias. Second, the far majority of the parent-dyads in our sample had a middle or high educational level, were born in the Netherlands, and were living together, which is not comparable to the general population. Especially in divorced parents, agreement in parents might be lower, but this should be confirmed in future research. Additionally, since there are no Dutch reference values available of the PedsQL proxy reports from 8 years of age, we used the reference values of Varni et al.; yet Dutch values might be slightly different. Furthermore, we did not have information on which parent was the primary caregiver; perhaps we would have found that this parent’s rating was consequently higher or lower than the second parent’s rating. However, we do not really expect this, since the mother is still the primary caregiver in the majority of the Dutch families [14]. Finally, it would have been interesting to include child self-reports as well, and to take these into account in the comparison of parental proxy ratings.