Psychosocial health comparing NICU and healthy children
The unadjusted mean MCS score for parents of NICU children did not differ from parents of healthy children (48.2 versus 48.8; p = .305). We also compared MCS scores after adjusting for the three sample characteristics that differed between the two groups (i.e., proportion of biological parents; gender of subject; and those with lower household income), and no differences were found in the outcome variable.
Psychosocial health by child health status problem
On the HSCS-PS, 55.2% of healthy children had no health problems in any area, compared with 39.8% of NICU children (p < .001 on Chi-square). Table shows the joint distribution of health status problems across the four categories for the NICU and healthy sample. These results show that the NICU sample had a higher proportion of children with more health status problems, as well as a higher proportion with moderate/severe versus mild problems.
Distribution of children with health status problems across the 4 health status categories for NICU and healthy children
For parents of NICU children, for all 4 health status categories, parental MCS scores decreased as severity of the child health problem increased (see Table ). Effect sizes comparing parents of children with no health status problems with parents of children with a moderate or severe health status problem were all moderate to large indicating important differences in parental mental health according to Cohen's benchmarks. The results for parents of healthy children show similar trends, with mainly moderate to large effect sizes.
Parental mental health summary score, 95% confidence intervals, number of subjects, p-value and effect size for child health status category
Psychosocial health by child behavior problem
Child behavior was strongly related to parental psychosocial health in both groups of parents (see Table ). Parents whose child scored in the clinical range for internalizing and externalizing symptoms and the total problem score on the CBCL/1.5–5 had the lowest mean (i.e., poorest) MCS scores. The differences between this group and the group with children scoring in the normal range resulted in large effect sizes, indicative of clinically important differences in parental psychosocial health.
Mean score, p-value and effect size for SF-36 psychosocial summary score comparing CBCL/1.5–5 normal with borderline and clinical groups
Parental psychosocial health by birth-related risk factors
Within the NICU sample, MCS score did not vary by any birth-related risk factor (i.e., gestational age; small for gestational age; apgar score; multiple birth; the presence of a major morbidity; and neonatal illness severity score), with the exception of the presence of a congenital anomaly. For this variable, MCS scores were significantly lower in parents of children with versus without a congenital anomaly (mean difference = -3.8; p = .017; effect size = -.37). Children with a congenital anomaly (n = 87) had proportionally more mild and moderate/severe health status problems in all 4 categories (see Table ).
Number (%) of NICU children with and without a congenital anomaly to report a problem for each health status category and p-value for Chi-square test of significance
Correlates of psychosocial health in general
In general, variables significantly associated with the MSC score in bivariate analysis were as follows: any health status problems (mean difference = -3.8; p < .001); neurosensory problems (mean difference = -3.7; p = 0.04); motor development problems (mean difference = -4.4; p < .001); learning/remembering problems (mean difference = -2.9; p < .001); poorer quality of life (mean difference = -4.8; p < .001); more internalizing behaviour symptoms (mean difference = -8.3; p < .001); more externalizing behavior symptoms (mean difference = -9.9; p < .001); household income below $30,000 per year (mean difference = -2.6; p < .001); female gender (mean difference = -2.6; p < .001);not living as common-law or married (mean difference = -3; p = .03); more caregiver strain (r = .41; p < .001); and lower family function (r = -.44; p < .001). Borderline significance was also found for less than high school education (mean difference = -2; p = .08).
We examined a pooled model (both groups together) for a direct comparison of the NICU and healthy groups after adjustment for other variables. Due to the low number of male respondents in the healthy group, we restricted the pooled multivariable analysis to only female respondents. Predictors significantly associated with the outcome were the following: parental age (Beta = 0.15; p = 0.001); internalizing behavior (Beta = -2.06; p = 0.017); externalizing behavior (Beta = -3.24; p = 0.004); parental strain (Beta = 0.15; p < 0.001); and family function (Beta = -0.53; p < 0.001). The pooled model also showed an interaction effect between NICU admission and education (less than high school) (Beta-education = -5.94 with p = 0.009; Beta-interaction = 7.28 with p = 0.005)(see Table .) For the NICU group, education did not show any effect in terms of difference in outcome, but for the healthy group, lower education was associated with a significantly lower mean MCS score. More specifically, for respondents with less than high school education, the healthy group reported lower MCS scores than did the NICU group. The results were not affected by exclusion of multiple births and cases of congenital anomalies from the analysis.
Beta coefficients, 95% confidence intervals, standardized beta coefficients and p-values for predictor variables in the multiple regression models for pooled model
Although other interaction terms with NICU status did not add any more significant results in the pooled model (non-significant partial F-test), we examined separate models for the NICU and the healthy baby group to further explore the association between gender and MCS score, and to evaluate the potential influence of congenital anomalies in NICU group.
Correlates of psychosocial health for NICU sample
Variables that were significantly associated with lower MCS scores at the bivariate level include the following: female caregivers (mean difference = -3.2; p = .037); household income below $30,000 per year (mean difference = -3.3 and p < .001); not living as common-law or married (mean difference = -5.1; p < .001); neurosensory problems (mean difference = -6.44; p = .011); motor development problems (mean difference = -7.1; p < .001); learning/remembering problems (mean difference = -5.9; p < .001); poorer quality of life (mean difference = -10.4; p < .001); more internalizing behaviour symptoms (mean difference = -9.03; p < .001); more externalizing behavior symptoms (mean difference = -13.9; p < .001); the presence of a congenital anomaly (mean difference = -3.8; p = .017); more caregiver strain (r = .411; p < .001); and lower family function (r = -.441; p < .001).
Predictors that were significant in the final regression model appear in Table . Female gender was an independent risk factor for lower MCS score: females scored on average 5.3 points (CI interval 2.5 to 8.0) lower, which represents a moderate effect size of 0.51 (when overall NICU parents group standard deviation (SD) 10.4 for MCS was used as the denominator). Scoring outside the normal range for internalizing and externalizing child behavior symptoms independently contributed to lower MCS scores (-1.9 and -2.8, both with wide confidence intervals), with the change representing small effect sizes of 0.18 and 0.27. More caregiver strain (i.e., lower PTT) was related with poorer MCS scores. A one point change in PTT corresponded to a 0.15 (CI: 0.11–0.19) change in MCS score. In NICU parents, the mean PTT was 86.9 and SD was 18.5. Therefore, 2 SD on the PTT would represent 5.5 points on the MCS, or an effect size of 0.53. The mean score for family function (FAD) was 8.1 and the SD was 6.4. A one point change in FAD corresponded to a 0.5 (CI: 0.62; 0.42) change in MCS. Therefore a 2 SD increase in family function score (i.e., poorer family functioning) would result in a 6.4 decrease (worsening) in MCS, representing a moderate effect size of .62. Overall, the adjusted R2 was .2884 (F = 73.96; df = 5; p = < .0001), with 5 out of 15 predictors included in the full model.
Beta coefficients, 95% confidence intervals, standardized beta coefficients and p-values for predictor variables in the multiple regression models for both samples
Correlates of psychosocial health for healthy baby sample
Variables that were significantly associated with poorer SF-36 MCS scores at the bivariate level include the following: younger parental age (r = .19; p < .001); household income below $30,000 per year (mean difference = -4.6; p = .005); less than high school education (mean difference = -6.22; p = 0.065); not living as common-law or married (mean difference = -6.1; p = .005); motor development problems (mean difference = -11.3; p = .043); learning/remembering problems (mean difference for any problems versus none = -2.68, p = 0.021); poorer quality of life (mean difference = -13.9; p < .032); more internalizing behavior symptoms (mean difference = -9.5; p < .001); more externalizing behavior symptoms (mean difference = -15.2; p < .018); more caregiver strain (r = .385; p < .001); and lower family function (r = -.438; p < .001).
Predictors that were significant in the final regression model appear in Table . The model for parents of healthy children did not include female gender (because of low numbers) and externalizing behavior symptoms, and included several variables not predictive in the NICU model (i.e., parental age; education; quality of life). Both models included internalizing child behaviors, caregiver strain and family function.
In the healthy baby sample, younger parental age was related to poorer MCS score, with a one year change in age resulting in a 0.26 (CI: 0.08; 0.45) change in MCS. A ten year difference in age would correspond to a 2.6 difference in MCS, which would represent a small effect size of 0.27 (when the overall healthy baby parent group SD for MCS (9.6) was used as a denominator). Education was also associated with MCS. Compared with high school graduates, the MCS score for parents with less than a high school education were on average 5.0 lower (CI: 0.84; 9.1), which represents a moderate effect size of 0.52, although the effect could range from minimal to large due to lower precision of the beta estimate. Child internalizing symptoms, family function and caregiver strain were associated with parental MCS in a similar way as for NICU parents. However, due to lower numbers and resulting low precision in beta estimates, the effects ranged from minimal to large. Lower parent-reported child quality of life was also associated with a lower parental MCS. Parents who reported a problem with their child's quality of life had MCS scores that were 6.9 (CI: 0.37; 13.4) lower than parents who reported at least one quality of life problem compared with those who reported at least one problem. Again, due to the small numbers, the effect could range from minimal to large. In the final regression model, the adjusted R2 was .3046 (F = 25.97; df = 6; p < .0001), with 6 out of 16 predictors included in the full model.