The relationship between physical and mental health multimorbidity and children’s health-related quality of life

Data were pooled from eight waves of the Longitudinal Study of Australian Children (LSAC); a nationally representative sample of Australian children followed every 2 years between 2004 and 2018 [26]. The LSAC study was approved by the Australian Institute of Family Studies Ethics Committee and families provided written consent to participate. The current study was approved by the data custodians and no further ethics approval was required.

Participants were children aged ≥ 4 years and < 18 years at the time of data collection, and their primary caregiver. This age range was chosen to match the recommended age range for the measures described below, and data availability within the LSAC study.

LSAC cross-sectional population weights were applied to each wave to account for the unequal probability of participant selection and sample attrition so that the individually weighted waves are population representative [35]. This approach of utilising the cross-sectional characteristics of LSAC survey allows us to investigate a population-representative sample with individual respondents that each have different representation across each wave based on the weights applied. Complete case analysis was undertaken. Due to the large sample size, a conservative significance level of p < 0.0001 was used. All data were analysed using StataSE 16 (Statacorp, Texas, US).

Sample characteristics were explored using descriptive statistics for the total sample, and for children who had (1) physical health problems and SHCN; (2) borderline or clinical mental health symptoms and (3) physical–mental multimorbidity. Group differences were explored using t-tests for continuous variables and χ² for categorical variables comparing each of these problem groups to children not classified into these problem groups.

For regression analyses, we explored the distribution of the error terms using a Modified Park Test, and the results suggested an Ordinary Least Squares (OLS) linear regression model performed well. Firstly, we conducted a regression of the dependent variable (HRQoL) on independent variables (presence/not of borderline or mental health symptoms; or any physical health problem) including the interaction terms between borderline/clinical mental health symptoms and physical health problems. Secondly, we repeated the analysis adjusting for all covariates described above, examining the individual associations between HRQoL and each variable holding all others constant. Thirdly, we repeated the adjusted regression analysis within subgroups of child age and sex.

We conducted a range of sensitivity analyses to examine the robustness of the relationships under alternate definitions of physical health problems, mental health symptoms and HRQoL, in a series of adjusted regressions. See Supplementary Table 2 for full details of the definitions tested and measures used. Of note, as per previous research and recommendations regarding HRQoL measurement in mental health populations [10, 36, 37], our sensitivity analyses tested whether any observed relationship between mental health symptoms and HRQoL was due to item overlap between the two measures. Additionally, due to the known differences and challenges associated with using parent proxy-report versus child self-report of HRQoL [8, 10, 38] we conducted a sensitivity analysis using child self-reported HRQoL measured via the Child Health Utility instrument (CHU9D) [39]. This was included as a sensitivity analysis rather than the main analysis as the child’s self-report was only available in Waves 6–8.

The total number of observations that met inclusion criteria for age ≥ 4 years and < 18 years was 51,501 across the eight waves. Observations from each wave were dropped if they had insufficient PedsQL (n = 3645) or SDQ data (n = 1943); or if they were missing SHCN information (n = 610). Total excluded cases were n = 3934, leaving 47,567 observations (n = 9270 unique children, 92.4% of original sample) included in the final analyses. We present the participant flowchart in Supplementary Fig. 1. A larger proportion of excluded observations were for children who were aged 4–7 years (65.8% vs 32.6%); spoke a language other than English (20.0% vs 8.9%); and met the clinical cut-off on the SDQ (11.1% vs 7.5%) and fewer excluded observations had a physical health problem (64.0% vs 70.8%). Missingness of covariates was < 1.0%, except for income (6.7%).

Table 1 shows the sample characteristics overall, and for each problem group. Overall, the mean age was 10.1 years (SD 3.7) and 51.2% were males. The majority of the sample (72.7%) were observations not classified into one of the problem groups, whereas 15.2% had at least one physical health problem and SHCN; 17.4% had borderline or clinical mental health symptoms and 5.4% of observations were children with physical–mental multimorbidity. Significant differences were observed between each problem group and those not otherwise classified for nearly all sample characteristics except for parent sex (97.9% were mothers), and proportion residing in a major city (64.1%).

Table 2 shows results for unadjusted and adjusted regression models. In both analyses, physical and mental health problems were each associated with significantly lower ratings of HRQoL, however, only mental health problems were associated with a clinically meaningful difference in HRQoL. In adjusted analyses, though slightly attenuated, the associations between borderline and clinical mental health symptoms and HRQoL remained equal to more than two-times (10.5 points) and more than three-times (16.8 points) the clinically meaningful difference, respectively. A statistically significant interaction effect was observed for children with physical–mental multimorbidity, where children with clinical (but not borderline) levels of mental health symptoms had lower ratings of HRQoL than the additive effect of mental and physical health problems. Of note, when the primary carer screened positive for their own ‘probable serious mental illness’, children showed an additional significant and clinically meaningful difference in HRQoL (4.5 points lower), after accounting for all other factors Fig. 1 shows results of adjusted regression analyses alongside results of each subgroup and sensitivity analysis (see Supplementary Table 3 for data tables for Fig. 1).

The same pattern of results was observed for subgroup analyses by child age and sex (Fig. 1, and Supplementary Table 4), with differences in HRQoL associated with borderline and clinical symptoms ranging from 7.6–12.5 points lower and 12.1–19.6 points lower, respectively. Lower HRQoL was generally observed for older (8–12 and 13–17 years) compared to younger children (4–7 years). Physical health problems were only associated with a significant and clinically meaningful difference in HRQoL in the oldest age band (13–17 years; 5.6 points lower) and for girls (4.8 points lower). The interaction of physical health and clinical mental health problems was significant for boys, but not girls (see Supplementary Fig. 2), and was not significant in each age band individually.

Relationships between physical and mental health problems and HRQoL were similar to the main analysis when examining relationships within individual domains of the PedsQL (Supplementary Table 5). As would be expected, there was a significant and clinically meaningful relationship between mental health problems (both borderline and clinical levels) and the emotional functioning domain. Importantly, however, this relationship was additionally observed in the domains of social functioning and physical health. Clinical levels of mental health symptoms were additionally associated with a clinically meaningful difference in school functioning. Reflecting the main analysis, physical health problems were associated with a significant, but not clinically meaningful, difference in HRQoL across all domains. The interaction of physical health and clinical mental health problems was significant within the domains of emotional functioning and social functioning.

When using child-reported HRQoL (CHU9D) as the outcome, we found a similar pattern of results as with parent-report regarding the relative relationships between clinical and borderline mental health problems and physical health problems. However, in contrast to parent-report, there was no significant interaction effect of physical and mental health problems. Additionally, the model explained less variance in child self-reported HRQoL scores (7%) compared to parent-reported HRQoL (26%).

Results of further sensitivity analyses regarding alternative definitions of physical health, mental health and HRQoL are presented in Fig. 1 (see also Supplementary Tables 3 and 6). Findings were similar to the main analysis, with the exception that the interaction of physical health and clinical mental health problems was no longer significant when mental health was defined as ‘parent report of anxiety and/or depression’ or ‘parent report of ADHD’.

Children with borderline or clinical levels of mental health symptoms each had lower ratings of HRQoL than those with physical health problems, whilst children with physical–mental multimorbidity showed even lower ratings over-and-above the additive effect of both problems. Mental health problems were associated with poorer HRQoL for older compared to younger children, though were still associated with a significant and clinically meaningful difference in children aged 4–7 years. In subgroup analyses, the interaction effect was significant for boys only, meaning that boys with physical–mental multimorbidity are at particular risk of disproportionately poorer HRQoL than boys with either mental or physical health problems alone.

In line with previous research we found children with clinical levels of mental health symptoms have poorer HRQoL than children with physical health problems [9, 12‐14], and that greater mental health problems are associated with poorer HRQoL [18‐20]. Our study provides novel evidence that quantifies the relative relationships between HRQoL and borderline mental health symptoms versus physical health problems, and showed that borderline mental health symptoms are also associated with a clinically important difference in children’s HRQoL that is larger than that of physical health problems. This is important in determining groups of children in greater need for support.

Our finding that mental health problems were associated with lower ratings of HRQoL in older compared to younger children has been noted previously for children with depression (as rated by their parents) [13], though we extended this knowledge to include borderline symptoms. This finding may be due to greater disease severity [14] or greater psychiatric comorbidities [13] in adolescents. Yet others have noted no difference by child age [9, 40]. Longitudinal studies examining the onset and progression of mental health problems are warranted to disentangle these mixed findings.

Our study extends a previous finding [12] of a significant interaction effect in children with physical–mental multimorbidity by examining a larger multimorbid sample across a broader range of mental health symptoms, and adjusting for more child, parent, family and social factors. Whilst we found a significant interaction effect, as per Sawyer et al. [12], our analyses revealed that this does not extend to borderline mental health symptoms, possibly due to the smaller span of scores within the borderline range compared with the clinical range of symptoms.

Our finding that the interaction effect of physical–mental multimorbidity was significant for boys but not girls is novel. This difference may arise due to our finding that the associations between physical and mental health problems and HRQoL were each larger and more significant for girls than boys; i.e. girls with physical health problems show significantly lower ratings of HRQoL compared to their healthy peers (regardless of their mental health status), however, for boys this was only true if they also had mental health problems. The differing relationships observed by sex may also relate to the type of mental health problems experienced by boys and girls, with internalising problems previously found to be related to poorer HRQoL in girls and externalising problems more-so for boys [20]. This warrants further exploration in the context of the interaction of physical–mental multimorbidity, to determine whether a particular combination of problems accounts for this difference.

Our sensitivity analysis found no interaction effect when using child self-reported HRQoL, which may be related to a number of factors. Firstly, parent ratings of children’s HRQoL tend to be lower than the child’s self-report when the child has chronic health conditions [8, 14, 40, 41], which may overestimate the deficit in HRQoL and lead to a stronger interaction effect of having multiple chronic conditions by parent-report. Secondly, in the current study, children and parents rated HRQoL using different measures. Whilst the PedsQL and CHU9D are moderately correlated (r = 0.63; when children self-report on both measures [42]), the use of different measures that comprise different domains of HRQoL, with different respondents makes it difficult to compare the findings. Similarly, the higher degree of variance in HRQoL scores that is explained using parent-report would likely be related to having the same respondent (parents) rate both the predictor (child’s health status) and outcome (HRQoL); i.e. part of this observed relationship reflects the parent’s underlying perception of the child’s health and wellbeing. Parent’s perception of their child’s health status can be further augmented by parents’ own mental health problems [43], which may also lead to the observed differences in child versus parent-report of HRQoL.

The overlap between measures of mental health and HRQoL is important to consider when assessing associations between the two [10, 37], though many studies have not addressed this [36]. Our finding that mental health problems were associated with larger differences in psychosocial domains of HRQoL is in line with a previous systematic review [11], however, this differs from others that have found stronger relationships in school functioning domains [14, 44]. Importantly, the relationship between mental health and HRQoL extended beyond the expected emotional health domain into all other domains examined. This broad relationship has previously been noted in studies using the PedsQL [9, 44], but is less consistently found using other measures [12‐14]. Our findings add to a growing literature that mental health and HRQoL are related but distinct concepts [8].

In our adjusted model, only parental mental illness was associated with a clinically meaningful difference in the child’s HRQoL, holding all other factors constant. As the respondent in our study was the child’s mother in 98% of instances, our findings are similar to previous work by Bastiaansen et al. [20], which found that mother’s psychopathology was individually associated with poorer child HRQoL, though this was no longer significant in their final adjusted model. In contrast, recent evidence suggests parental mental illness may only be related to poorer HRQoL in the domain of autonomy & parent relation for adolescent girls, and may not be associated with any deficits in HRQoL for adolescent boys [45]. Whilst parental mental health may be an important, potentially modifiable, focus for intervention that may improve children’s HRQoL [20], further exploration is warranted particularly regarding the persistence of parental mental illness and links to children’s mental health problems as well as the implications for child self-report and parent-report of the child’s HRQoL over time.

This study presents the first evidence of the relative relationships between physical health, borderline and clinical levels of mental health symptoms and children’s HRQoL in a large, population-based sample. Using a population sample enabled us to detect a greater number of children within the borderline mental health range than might be possible with a clinical sample. The size of the sample enabled us to examine relationships within smaller subgroups of children, and relationships with a range of child, parent, family and social factors that have not been previously examined in this context.

Limitations include the use of parent-reported data on children’s physical and mental health problems rather than diagnosed conditions; in the absence of a gold-standard diagnostic interview, and the inconsistent availability of parent-reported mental health conditions across waves, we used the SDQ as a validated tool available at every time point. We recognise that use of the SDQ to measure mental health problems means not all children who screen positive for mental health problems would meet clinical criteria for a psychiatric diagnosis and, vice versa [46], some children with diagnosed conditions may be well-managed with treatment and may have low scores on the SDQ. This misalignment in identifying children with mental health problems can be seen within the current study regarding the differing results between SDQ and parent-report of children’s mental health diagnoses and speaks to the wider difficulty in accurately identifying this population. On the other hand, a strength of using the SDQ for all children meant we could identify children with borderline symptoms that may have gone undetected using formal clinical criteria. Using the same respondent for child mental health and HRQoL is a limitation which we have made efforts to address by including a range of parent factors as covariates in adjusted analyses, and conducted a sensitivity analysis using child self-reported HRQoL. An additional limitation arises from the pooled cross-sectional design of the study, such that we cannot comment on the direction of effect, and the analysis cannot control for unobserved factors (e.g. parent factors) that may impact reports of child HRQoL and child health. Whilst it seems likely that physical health problems would cause poorer HRQoL, it is not necessarily the case that the relationship between mental health problems and HRQoL is unidirectional. A longitudinal analysis is planned to examine this interplay of symptoms, as well as the potentially changing influence of child, parent, family and social factors, on children’s HRQoL throughout childhood and adolescence.

Given the relationship between greater mental health symptoms and poorer HRQoL in children of all ages, clinicians should seek to identify mental health symptoms in children as young as 4–7 years old, even if these problems would not meet formal clinical criteria for a mental health disorder. Particular attention should be paid to the mental health and HRQoL of children, especially boys, with physical–mental multimorbidity. Maternal mental illness may represent a potentially modifiable area for intervention alongside treatment of the child’s symptoms. Siloing of child and adult mental health services may mean that this does not happen in practice, though this has been trialled in the United Kingdom [47].

Children with physical–mental multimorbidity are at risk of disproportionately poorer HRQoL than the additive effect of both problems. Our results suggest that these children are a group with particularly high needs who would benefit from health policy and clinical decision making to address these needs. Similarly, if we are to improve children’s HRQoL, monitoring and addressing borderline mental health symptoms in children as young as 4–7 years old is important but not happening in Australia [48].