Study Design and Data Source
This study was a secondary analysis of publicly available, cross-sectional data that was combined from the 2016 and 2017 National Survey of Children’s Health (NSCH). The data analyzed for the current study are available through the U.S. Census Bureau at https://www.census.gov/programs-surveys/nsch/data.html. The NSCH is a parent-reported survey about healthcare access and quality, educational experiences, parent and family health, and child health for a nationally-representative sample of children ages 0-17 years. The NSCH is sponsored by the Maternal and Child Health Bureau of the Health Resources and Services Administration, part of the U.S. Department of Health and Human Services. The 2016 and 2017 NSCH were conducted by the U.S. Census Bureau using web- or mail-based survey administration, with a telephone questionnaire assistance option. Questionnaires were available in English or Spanish. The overall weighted response rates were as follows: 40.7% for the 2016 NSCH and 37.4% for the 2017 NSCH.(35,36) Additional details about the NSCH methodology are available from the U.S. Census Bureau.(37,38)
Two parent advisors were continuously and regularly involved in the study’s conceptualization, design, and interpretation of results. Each parent advisor had a young child who was 2 to 3 years old that was born prematurely, and each advisor was involved on a family advisory committee for a neonatal intensive care unit (NICU) at a large academic medical center following their child’s discharge. The Institutional Review Board at Massachusetts General Hospital determined that this study was not human research and it was exempt from review.
Participants
The full study sample included 19,482 U.S. children ages 0-5 years. We limited the study sample to children ages 0-5 years, because early childhood is a critical period for development and when children born prematurely and their families may experience the greatest adverse impact.(13,14,28) In the sample, 242 children were born very low birthweight (< 1,500 grams), 1,236 children were born low birthweight (1,500 to 2,499 grams), 969 children were born preterm but not low birthweight or very low birthweight, and 17,035 other children were not born very low birthweight, low birthweight, or preterm. Because children born preterm but not with low birthweight may be similarly prone to experience health risks as children born low birthweight (not very low birthweight) (4,39), we combined children born low birthweight and children born preterm not low birthweight or very low birthweight (n = 2,205) into one group (LBW/PTB) that was mutually exclusive from children born very low birthweight (VLBW) or other children. In both the 2016 and 2017 NSCH, parents were asked the following question to determine if children were born prematurely: “Was this child born more than 3 weeks before his or her due date?” To establish each child’s birthweight, parents were also asked: “How much did he or she weigh when born?” In alignment with the Centers for Disease Control and Prevention’s case definition (1), very low birthweight was defined as < 1,500 grams and low birthweight was defined as 1,500 to 2,499 grams for this study.
Measures
Healthcare access. Per past research about healthcare access and quality for child subgroups at high risk of health disparities (e.g., children with special health care needs, children with autism spectrum disorder) (29,40–42), we used the following three healthcare access measures: adequate health insurance, access to medical home, and developmental screening receipt. Adequate health insurance was a composite measure only assessed among children who were insured during the past 12-months. In the study sample, 635 children were uninsured. Adequate health insurance was determined by the following three subcomponents: health insurance benefits met the child’s needs (usually or always versus sometimes or never), coverage allowed the child to see needed providers (usually or always versus sometimes or never), and the child’s out-of-pocket health care expenses were reasonable (usually or always versus sometimes or never). To qualify as having adequate health insurance, children had usually or always on all three subcomponents. Access to medical home was also a composite measure based on 16 items about the following five subcomponents of care in the past 12-months: child had a personal doctor or nurse, usual source for sick care, family-centered care (e.g., doctors spent enough time with the child, doctors showed sensitivity to family values and customs), no problems getting needed referrals, and effective care coordination when needed (e.g., got all needed help with care coordination, satisfaction with communication among child’s doctor and other health care providers). To qualify as having a medical home, children needed to have had a personal doctor or nurse, usual source for sick care, and family-centered care. To have been considered as having a medical home, children additionally must have had no problems getting needed referrals and effective care coordination (if they reported needing these services). Additional documentation about this medical home measure is provided elsewhere.(43) Developmental screening receipt was assessed with a 3-item measure previously validated using NSCH data.(44) The developmental screening measure was only assessed for children who were ages 9 to 35 months, in alignment with national screening guidelines.(45) Children were considered to have had developmental screening if their parent indicated a doctor or other health care provider had given them or another caregiver a questionnaire about specific concerns or observations they had about their child’s development, communication, or social behaviors and if this questionnaire had two age-specific content areas regarding language development and social behavior in the past 12-months.
Adverse family impact. We used five adverse family impact measures, which have been commonly used in relevant, past research.(18,20,29) Two of these measures were related to family financial and/or employment impacts including if the family spent $1,000 or more on out-of-pocket medical expenses for the child during the past 12-months and if a parent or other family member cut down on hours working or stopped working because of the child’s health or health condition(s) during the past 12-months. Parental aggravation was a previously used composite measure derived from the following three items: parent felt the child is difficult to care for, parent felt that the child does things that bother them, and parent felt angry with the child.(18) All of the parental aggravation items were assessed for the past month and included a five-point response scale (never, rarely, sometimes, usually, always). Parents were defined as having often experienced parental aggravation during the past month if they indicated usually or always for any of the three measure items. Overall maternal and paternal health status not being excellent were similarly measured using two items: one item about the mother’s or father’s overall physical health status and one item about the mother’s or father’s overall mental health status. Each item was rated on a five-point scale (poor, fair, good, very good, excellent). Maternal and paternal health were both considered to be not excellent, if either physical or mental health status was reported to be poor, fair, good, or very good.
Covariates. We selected child and family characteristics as covariates that have established linkages with prematurity status, healthcare access, and/or adverse family impact and were available in the 2016 and 2017 NSCH.(25,27,46,47) Covariates included the child’s age (years), sex (male or female), race and ethnicity (white and non-Hispanic, Hispanic, black and non-Hispanic, other race and non-Hispanic), parent’s nativity (born in the U.S. or not born in the U.S.), primary household language (English or Spanish/other language), highest parent education level (high school or less versus more than high school), family structure (two married parents, two unmarried parents, single mother, other family structure), household income level defined according to the family poverty ratio, health insurance coverage (private only, public only, private and public, uninsured or unspecified), and region of residence (Northeast, Midwest, South, West). In addition, the child’s special health care needs status was assessed by the Children with Special Health Care Needs (CSHCN) Screener.(48) Other covariates included current presence of one or more of 27 chronic conditions (e.g., asthma, developmental delay, speech and language disorder), number of adverse childhood experiences (e.g., parent divorced or separated, parent died), and family resiliency (i.e., family talks together about what to do when facing a problem, works together to solve a problem, knows the family has strengths to draw on when the family faces a problem, and stays hopeful even in difficult times when the family faces problems).
Statistical Analysis
We first compared characteristics of U.S. children ages 0-5 years by prematurity status using chi-square tests, as well as by using multinomial logistic regression for categorical variables and linear regression for continuous age. Both unadjusted and adjusted differences in healthcare access and adverse family impact by prematurity status were examined by estimating relative risk. All covariates that differed by prematurity status at a p < .10 level were included in the multivariable regression models used to compute adjusted differences in healthcare access and adverse family impact.
Given differences in healthcare access and adverse family impact by prematurity status and the study’s focus, we examined associations of healthcare access with adverse family impact only among children born prematurely (VLBW and PTB/LBW combined). Propensity score weighting was used to estimate the average treatment effect of each healthcare access indicator in relationship to each adverse family impact. We employed the propensity score weighting with subclassification approach recommended by DuGoff and colleagues when applying propensity score methods in using complex survey data such as that from the NSCH.(49) To compute propensity score weights, we initially included the following variables that were associated with 1 of the adverse family impact variables: age, sex of child, race/ethnicity, family structure, insurance status/type, region, VLBW, CSHCN status, comorbid condition(s), ACE(s), family resilience, and the survey weights that NCHS specified. We then assessed propensity score balance by evaluating the standardized differences of each covariate for each of the three healthcare access variables (adequate health insurance, medical home, and developmental screening). Covariates were removed if the absolute value of the standardized difference was 0.10, and propensity scores were re-estimated with the remaining covariates. Different covariates were removed for models with each of the three healthcare access variables. Family structure, insurance status/type, CSHCN status, chronic condition(s), and family resilience were removed for adequate health insurance. Race/ethnicity, family structure, insurance status/type, CSHCN status, chronic condition(s), ACE(s), and family resilience were removed for medical home. Sex of child, race/ethnicity, family structure, insurance status/type, region, VLBW, CSHCN status, and chronic condition(s) were removed for developmental screening. Doubly-robust estimators of causal effects and inverse probability of treatment weighting were used to weight the treatment (e.g., adequate health insurance) and comparison (e.g., no adequate health insurance) samples by the propensity scores for each adverse family impact variable. Standardized differences were again evaluated in the weighted samples, and the propensity score weights were multiplied by the survey weight to create a new weight used in fitting the weighted multivariable regression models. These relative risk models, with adverse family impact as the dependent variable and healthcare access as the main independent variable of interest, included the set of covariates that were initially considered for each propensity score and also adjusted for parent nativity, household language, and household income level (i.e., doubly-robust estimation). Family structure was omitted from the maternal and paternal health models due to possible collinearity with the dependent variable.
To better understand the healthcare access subcomponents contributing most to statistically significant associations with certain adverse family impacts, we additionally performed post-hoc bivariate and multivariable analyses to examine associations between adequate health insurance and medical home subcomponents and three adverse family impacts (out-of-pocket costs, parent cut-back or stopped work, and parental aggravation) among children born prematurely. For these analyses, relative risk and 95% confidence intervals were estimated. Multivariable regression models included the same set of covariates initially used to examine differences in healthcare access and adverse family impact by prematurity status.
All analyses incorporated strata and weighting to produce nationally representative estimates.(38) Weights were adjusted for multi-year analysis.(50) Family poverty ratio was analyzed in a multiple imputation framework.(51) We used a conventional alpha level of .05 to determine statistical significance. Given potential bias due to multiple comparisons made in the multivariable models, we additionally provided a Bonferroni-adjusted significance threshold to compare p-values against in relevant results table. All analyses were performed in Stata version 15.(52)