Depressive Symptoms in Parents of Children With Chronic Health Conditions

A Meta-analysis

Martin Pinquart, PHD


J Pediatr Psychol. 2019;44(2):139-149. 

In This Article



When searching through electronic databases (PSYCINFO, MEDLINE, Google Scholar, CINAHL, PSYNDEX [an electronic data base of psychological literature from German-speaking countries]), we used the following search terms: (mothers or fathers or parents) AND (illness OR disease OR disability OR chronic condition) and depress*. In addition, we checked the references sections of the identified papers for additional studies.

Given the very large numbers of chronic physical diseases (World Health Organization, 1992), we could not include the complete list of these diseases in our search terms. Therefore, we used the broader search terms illness OR disease OR disability OR chronic condition, and decided for each identified condition whether it fulfills the criteria of a chronic disease by Thompson and Gustafson (1996): A chronic disease is associated with functional impairment, persists for >3 months in a year, and/or necessitates a period of continuous hospitalization for >1 month. We had learned from electronic searches for previous meta-analyses that the use of the search terms chronic disease and chronic illness does not identify larger numbers of relevant studies because this phrase is not used in many studies on cancer, asthma, and other chronic diseases. Therefore, we did not use the phrase chronic illness or chronic disease in our search.

Criteria for inclusion of studies in the present meta-analysis were:

  1. they assessed depressive symptoms or depressive disorders in parents who have a child with a chronic physical disease or sensory disability or physical disability;

  2. as we were interested in psychological consequences of caring for a child with a chronic condition, depression had to be assessed after the onset of the chronic condition;

  3. the studies provided sufficient information for a comparison of levels of parental depression with established normative data or a similar group of families with healthy/nondisabled children from the same county;

  4. the mean age of children at being diagnosed was <18 years—the legal age of majority in most countries (; and

  5. the studies were published or presented before April, 2018.

The literature search was completed on March 29, 2018.

Unpublished studies (e.g., dissertations and master theses) were identified as part of the systematic search with the electronic data bases PSYCINFO, CINAHL, Google Scholar, and PSYNDEX, as well as cross-referencing, and were included if they met the criteria listed above. We identified 1,634 papers. After screening and assessing for eligibility, we were able to include 460 studies in the meta-analysis. A flow chart of the search for studies is provided in Figure 1, and the studies included are listed in Appendix A1 and A2 (see Supplementary Online Material).

Figure 1.

PRISMA flow diagram.

If between-group differences were provided for several subgroups within the same publication (e.g., for different chronic conditions), we entered them separately in our analysis instead of entering the global association. If more than one depression scale was applied in the same sample, the effect sizes were averaged. Data from intervention studies were only used if pretest scores were provided and participants had not been selected based on elevated depression scores or elevated psychological symptoms in general.

Information from the IMF (2018) was used for coding countries as developed (countries with advanced economy) or developing (emerging market and developing economies). Study quality was assessed by the sum of four criteria from the Modified Quality Index that has been previously used in pediatric psychology (Ferro & Speechley, 2009): (a) external validity (whether the parents were representative of the entire population from which they were recruited; 1 = yes, 0 = no), (b) internal validity (use of valid and reliable depression measures; 1 = yes, 0 = no), (c) whether parents of children with and without chronic conditions did not differ in third variables or whether the analysis adequately adjusted for confounding effects of third variables (1 = yes, 0 = no), and (d) sufficient statistical power (for detecting the mean effect size from the meta-analysis by Easter et al., 2015; 1 = yes, 0 = no).


After assessing studies for eligibility, all studies were coded by the author, and a random sample of 120 studies was also coded by a psychologist with experience in meta-analyses. Differences between the two coders were resolved by discussion. We coded the number of parents of individuals with pediatric chronic conditions (intraclass correlation coefficient [ICC] = 1.0), number of persons in the control group (ICC = 0.91), mean parental age at assessment (ICC = 0.96), percentage of mothers (ICC = 0.93), percentage of married parents (ICC = 0.97), mean age of the child (ICC = 0.98), time since diagnosis (ICC = 0.93), type of chronic condition (inter-rater agreement 100%), whether all young people have a progressive disease (e.g., AIDS, cystic fibrosis, neuromuscular disorders; 2 = yes, 1 = no; ICC = 0.90), representativeness of the sample (1 = yes, 0 = no/not reported; ICC = 0.92), comparison group (1 = parents of healthy/nondisabled children from same recruitment area, 0 = test norms/other reference groups; ICC = 1.0), equivalence of patient and control group (1 = yes, 0 = not tested/no; inter-rater agreement 94%), available support for the validity and reliability of the depression measure (1 = yes, 0 = limited/no; ICC = 0.94), method for assessing depressive symptoms (inter-rater agreement 98%), publication status (2 = published, 1 = unpublished, ICC = 1.0), standardized size of between-group differences in depressive symptoms (ICC = 0.94), and percentage of parents meeting the criteria of a depressive disorder based on structured clinical interviews (ICC = 1.0). The sample sizes were used to determine whether the study had a sufficient test power for identifying moderate effect sizes (1 = yes, 0 = no). A review protocol can be accessed from the author.

Statistical Integration of the Findings

Calculations for the meta-analysis were performed in seven steps with Comprehensive Meta-Analysis (Borenstein, Higgins, Hedges, & Rothstein, 2005) using random-effects models and the method of moments.

  1. For comparisons of parents of children with and without a chronic condition, we computed effect sizes d as the difference in parental depression between the sample with a chronic condition and the control sample divided by the pooled SD. If the authors only provided test scores for parents of children with a chronic condition, we used the norms from the test manuals or from national comparative samples (with similar age and gender distributions) for comparison. Outliers that were >2 SD from the mean of the effect sizes were recoded to the value at 2 SD, based on Lipsey and Wilson (2001).

  2. The effect sizes d were transformed to Hedges' g to correct for bias because of overestimation of the population effect size in small samples.

  3. Weighted mean effect sizes and 95% confidence intervals [CIs] were computed. The significance of the mean was tested with Z-tests by dividing the weighted mean effect size by the standard error of the mean. To interpret the practical significance of the results, we used Cohen's criteria (Cohen, 1992): Effect sizes of g = .20 are interpreted as small, g = .50 as medium, and g = .8 as large.

  4. Homogeneity of effect sizes was computed by use of the Q statistic.

  5. To test the influence of categorical moderator variables, we used an analogue of an analysis of variance. A significant Q score indicates heterogeneity of the effect sizes between the compared conditions. If more than two chronic conditions were compared, differences between individual conditions were interpreted as significant if the 95% CIs of effect sizes do not overlap (Lipsey & Wilson, 2001). For analyzing effects of continuous moderator variables, we used a weighted regression analysis (meta-regression).

  6. Egger's test and the trim-and-fill algorithm by Duval and Tweedie (2000) were used to check whether the results may have been influenced by a publication bias.

  7. For computing the weighted mean prevalence of clinical depression, prevalence rates from the individual studies were weighted by their sample size.