Association of a Family Integrated Care Model With Paternal Mental Health Outcomes During Neonatal Hospitalization

Key Points Question Is there an association between the neonatal care setting—a family integrated care (FICare) model in single family rooms with complete couplet-care for the mother-newborn dyad vs standard neonatal care in open bay units—and mental health and participation outcomes among fathers of preterm newborns? Findings In this cohort study of 263 fathers, fathers in the FICare model perceived less stress and participated more in caring for their newborns compared with those in standard care. Participation mediated the beneficial association of the FICare model on fathers’ depressive symptoms and parent-newborn bonding. Meaning These findings suggest that supporting fathers to actively participate in all aspects of care of preterm newborns should be encouraged regardless of the neonatal unit’s architectural design.


eAppendix 1. Supplemental Methods Scales Used in the Study
To measure stress, the Parental Stress Scale: Neonatal Intensive Care Unit (PSS-NICU) was used. This scale was developed by Miles in 1993 and measures the parental perception of stressors due to the physical, psychosocial and emotional environment of the neonatal intensive care unit 1 . This scale consists of 26 questions and was filled in both at admission and at discharge.. The PSS-NICU scale has 3 subdomains and measures the degree of stress experienced by parents during hospitalization related to alterations in their parental role, the appearance and behavior of their infant, and sights and sounds of the unit. Parents rate their experiences on a 5-point rating scale ranging from "not at all stressful" to "extremely stressful." In an update of the tool in 2007, sights and sounds of the environment (5 items) were combined with infant's appearance subscale (14 items) and scored as one subscale and Parental Role Alteration as the second subscale (7 items). 2 If fathers did not experience the stressor, we transformed the score to "0". 3 The Hospital Anxiety and Depression Scale was developed in 1983 and is used to measure the possible presence of anxiety and depressive complaints (Zigmond & Snaith, 1983). This scale has also been converted to Dutch, so that it is also available in this language (Spinhoven et al, 1997). Parents were asked to fill in this 14item self-report at admission and at discharge.

Parent-and infant bonding
The Postpartum Bonding Questionnaire (PBQ), was devised by Brockington et al. (2001) as a screening instrument to detect bonding problems in obstetric and primary care services 4,5 . The PBQ is a 25-item scale reflecting a mother's feelings or attitudes towards her baby (e.g. ''I feel close to my baby'', ''My baby irritates me''). Participants rate how often they agree with these statements on a 6-point Likert scale ranging from always (score=0) to never (score=5) with low scores denoting good bonding. The PBQ has four subscales which reflect impaired bonding (Scale 1) (12 items, ranging from 0 to 60), rejection and anger (Scale 2) (7 items, scores ranging from 0 to 35), anxiety about care (Scale 3) (4 items, scores ranging from 0 to 20) and risk of abuse (Scale 4) (2 items, scores ranging from 0 to 10). Scale 1 (impaired bonding) has a sensitivity of 0.93 and a specificity of 0.85 in detecting mothers with a bonding disorder.

Parental empowerment and satisfaction
Parent satisfaction was measured using the EMpowerment of PArents in THe Intensive Care -Neonatology questionnaire 6 . This scale was developed and tested in a single center in the Netherlands, and available in Dutch. The domains covered are: Information (14 statements); Care and Treatment (20 statements); Parental Participation (nine statements); Organization (11 statements); and Professional Attitude (13 statements). The 57 statements divided in five domains provide a conceptualization of parent satisfaction within the neonatal ward from a family-centred care perspective 6 .

Parental Self Efficacy
The Perceived (Maternal) Parenting Self-Efficacy (PMP-SE) tool, was used to measure perceived parental selfconfidence when caring for the infant admitted to the Neonatal Ward 7 . The internal consistency reliability of the Perceived Maternal Parenting Self-Efficacy tool is 0.91, external/test-retest reliability is 0.96. A total of four conceptually unique subscales of parenting are: "Care taking procedures" (parents' perceptions of their ability to perform the activities and tasks related to the baby's basic needs like feeding). "Evoking behaviour(s)" (perceptions in their ability to elicit a change in the baby's behaviour, for example, soothing the baby). "Reading behaviour(s) or signalling" (perceptions in their ability to understand and identify changes in their baby's behaviour, for example, 'I can tell when my baby is sick'). "Situational beliefs"(parents' beliefs about their ability to judge their overall interaction with the baby). Responses to each item were recorded on a four point Likert scale ranging from 'strongly disagree' (score 1) to 'strongly agree' (score 4). A low score on this scale indicates a low parental self-efficacy.
We previously validated a measurement tool on parent participation in neonatal care (the CO-PARTNER tool), which includes 6 domains: 1) daily care 2) medical care 3) information gathering 4) advocacy and leadership 5) time spent with infant 6) closeness and comforting the infant. 8 Total scores per domain were obtained by summing scores. For Domain 1, 2 and 6 we calculated 0 for 'the nurse does this', 1 for 'the nurse and I do this together' and 2 for 'I do this independently' (minimum scores 0 to 22, 8 and 14 respectively). For domain 3 and 4 'yes' was scored as 1, and 'no' as 0 (minimum scores 0 to 3). Non-applicable items were transformed to 0 (no participation in this item). For the domain Time Spent with Infant (3 items) quartiles were calculated resulting in 0 to 4 score (total score in domain 5 minimum 0 maximum 12). A total participation score was obtained by summing all domain scores. Minimum total scores were 0 and maximum 62. Additionally, we aimed to analyze in multiple mediation models the extent to which domain of participation contributes to ameliorated outcomes in the FICare model.
Inclusion and exclusion criteria of the AMICA study.
In order to be eligible to participate in this study, a subject had to meet all of the following criteria: -Born between 24 and 36 6/7 weeks gestational age, with a postconceptional age ≤ 44 weeks on admission -At least 1 week of hospital stay on the Neonatal Ward. Infants <35 weeks gestational age were recruited in the first days after birth, as their hospital stay is usually more than 7 days. Infants with a gestational age between 35 and 37 weeks were recruited after consultation with the attending physician, and were recruited when a hospital stay of >7 days was expected.
-Would visit the Outpatient Clinic of the OLVG East, OLVG West or NWZA after discharge or consents to gather required information through infant health centers -Written informed consent from the parents A potential subject who met any of the following criteria was excluded from participation in this study: -Metabolic or chromosomal/syndromal diseases -Therapeutic hypothermia for perinatal asphyxia -Severe psychiatric or psychosocial problems i.e. parents under supervision of youth care -Transfer to another hospital before discharge -Parents unable to answer the questionnaires in Dutch/English -Death of an infant

Sample size calculation
The primary outcome of the AMICA study is neurodevelopment in infants at 2 years of age. We pre-stratified the study population towards infants born <32 weeks of gestation with a previous admission to a level 3 NICU and infants that were born >32 weeks of gestation. Within each gestational age group, we did a power calculation for the primary outcome of neurodevelopment. We calculated to have 64 experimental subjects and 128 control subjects with power 0.90 (1-) at a significance level of 0.05 () with a true difference in the outcome of neurodevelopment at the age of 2 years of ½ SD. To allow for 30% withdrawal we aimed to include 91 patients in group A and 182 patients in group B per risk group (post-intensive care versus inborn infants). A total of 546 infants who were hospitalised and their parents were expected to be included in this study.
For this study we calculated the power for the outcome on stress in fathers post-hoc. 11 The group sample sizes of 89 and 93 achieved 84% power to detect a difference of -8.6 between the null hypothesis that both group means are 40.8 and the alternative hypothesis that the mean of group 2 is 49.4 with known group standard deviations of 20.3 and 18.9 and with a significance level (alpha) of 0.05 using a two-sided two-sample t-test.

Variables used in multiple imputation model
For the imputation model we used the following strategy 12 1. Include all variables that are part of the analysis model, including the dependent (outcome) variable.
2. Include additional (auxiliary) variables that are related to nonresponse 3. Include additional variables that are related to variables with missing values.
The following variables were used in the multiple imputation model: -Gestational age in days -Inborn -Work hours per week -FICare model -Singleton status -Work hours per week -Identifies with Dutch cultural background -University degree -HADS at admission/discharge* -PBQ at admission/discharge* -PMP at admission/discharge* -PSS-NICU at admission/discharge* -Participation at admission/discharge* -Smoking -Use of recreational drugs -Alcohol use Collinear variables -Gestational age < 32 weeks -Length of stay in NICU (with total gestational age) -Length of stay total (with total gestational age) -Paid work (with work hours per week) -GA admission (with total gestational age) *depending on parcel mean summary score and which database was imputed.

Confounders and effect modifiers
Potential confounders and effect modifiers were identified from the literature and assessed using statistical analyses. We considered socio-economic class (including education and employment status), family composition (single parent vs co-parenting), older/other infants in the family, stress at birth, gestational age of infant, singleton status and mode of delivery. If the beta-regression coefficient differed at least 10% in regression analyses, this was used as an indication of statistical confounding, and the variable was included in the adjusted model. If collinearity was present, the strongest confounder (largest change in crude betacoefficient) was used to adjust for.

. Supplemental Results
We included fathers of infants who filled out any questionnaires during infant hospital stay in the analyses. Fathers who did not fill out any questionnaires during infant hospital stay were not analyzed, as we did not have any data on their baseline characteristics, such as work status, identification with Dutch cultural background, (see eTable 3, n= 81) and therefore multiple imputation was not deemed fit to perform on this subsample, with too much missingness in characteristics. Fathers who were non-responders, and thus not included, were not different in their baseline characteristics from responders (eTable 3), yet were still deemed unfit to be included in multiple imputation processes. Similar non-response rates were found in FICare (48.9% vs. 45.7%, p=0.727), the gestational age of their infants was similar ( 33 +1 (30 +6 -35 +0 ) vs 33 +3 (31 +3 -34 +6 ), p=0.682) and length of stay in the neonatal unit. Thus 81 fathers were not analyzed for this study. Of those fathers who filled out any questionnaires during hospital stay (n=182), 156 fathers filled out questionnaires at discharge (eTable 4). Thus, 26 fathers filled out questionnaires solely at admission. As we did have data on these fathers, we deemed it fit to include them in analysis and use their data for multiple imputation (see Methods) to generate imputed datasets.

No.: number
Baseline characteristics difference between FICare model and two standard care groups.
After review, we tested for differences between the FICare model and two standard care groups. For comparison of multiple groups for normally distributed data we used the one-way analyses of variance (ANOVA) and for non-normally distributed data the Kruskall-Wallis test. If differences were significant within the normally distributed data we computed Tukey HSD (Tukey Honest Significant Differences) for performing multiple pairwise-comparison between the means of groups. For non-parametric tests, we post-hoc performed multiple pairwise U-tests, with a Holm-Bonferroni correction, to test for differences in non-normally distributed data.
Chi-square (χ2) tests were used to test for differences in binary outcomes. If expected cellcounts were 5 or lower, we calculated differences with the Fisher's exact test. If differences between all groups were significant we performed a pairwise Z-test with Bonferroni correction between proportions of groups.