High-risk infant follow-up programs (HRIFs) are a recommended standard of care for all extremely low birth weight (ELBW) infants to help mitigate known risks to long-term health and development. However, participation is variable, with known racial and ethnic inequities, though hospital-level drivers of inequity remain unknown. We conducted a study using a large, multicenter cohort of ELBW infants to explore within- and between-hospital inequities in HRIF participation.
Vermont Oxford Network collected data on 19 503 ELBW infants born between 2006 and 2017 at 58 US hospitals participating in the ELBW Follow-up Project. Primary outcome was evaluation in HRIF at 18 to 24 months’ corrected age. The primary predictor was infant race and ethnicity, defined as maternal race (non-Hispanic white, non-Hispanic Black, Hispanic, Asian American, Native American, other). We used generalized linear mixed models to test within- and between-hospital variation and inequities in HRIF participation.
Among the 19 503 infants, 44.7% (interquartile range 31.1–63.3) were seen in HRIF. Twenty six percent of the total variation in HRIF participation rates was due to between-hospital variation. In adjusted models, Black infants had significantly lower odds of HRIF participation compared with white infants (adjusted odds ratio, 0.73; 95% confidence interval, 0.64–0.83). The within-hospital effect of race varied significantly between hospitals.
There are significant racial inequities in HRIF participation, with notable variation within and between hospitals. Further study is needed to identify potential hospital-level targets for interventions to reduce this inequity.
High-risk infant follow-up is a recommended component of postdischarge care of the preterm infant and may reduce the long-term impact of prematurity. However, only 50% of eligible infants participate and studies to identify factors associated with participation are limited.
Using a national sample and employing generalized linear mixed effects modeling, we identified significant between- and within-hospital variation in the effect of race/ethnicity on high-risk infant follow-up program participation. This study highlights how racism affects neonatal health service utilization.
Extremely low birth weight (ELBW) infants and those born extremely preterm have an increased risk of developing neurodevelopmental delays, chronic medical problems, and functional impairments later in life.1–3 Because these impairments are often not evident at NICU discharge, the American Academy of Pediatrics recommends that all high-risk infants participate in high-risk infant follow-up (HRIF) programs after NICU discharge.4,5 Although their composition varies, most HRIF programs are multidisciplinary clinics that provide enhanced neurodevelopmental and functional screening, assessment, and diagnosis as complement to standard care provided by primary pediatricians.6,7 One role of HRIF is early referral to appropriate developmental services, such as Early Intervention, which is associated with improved outcomes.4,8–10 Although limited by small sample sizes and non-random loss to follow-up, some studies suggest that early and consistent participation in HRIF is associated with improved outcomes for high-risk infants.11,12 A new conceptualization of HRIF prioritizing a “follow-through” paradigm examines both medical and social factors to further optimize pediatric health.13
Despite the importance of HRIF, only ∼50% of eligible infants participate.6,14,15 Medical factors such as younger gestational age, lower birth weight, and medical sequelae of prematurity such as bronchopulmonary dysplasia (BPD) and intraventricular hemorrhage (IVH) have been associated with increased HRIF participation, as have social factors such as maternal age, socioeconomic status, and post-NICU discharge neighborhood opportunity.6,14–18 Furthermore, racial and ethnic inequities exist in HRIF, which may contribute to long-term health inequity across the pediatric life-course, though the drivers of these inequities are unknown.6,15–17,19
Race is a social construct and inequities in health and health care utilization are the result of institutional, structural, interpersonal, and internalized racism.20–22 Inequities can originate and be propagated at multiple levels, and hospital variation in health care quality has been shown to be an important driver of racial and ethnic health inequities.23–27 Using a large, multicenter cohort of ELBW infants, we tested the hypothesis that the effect of race and ethnicity on HRIF participation varies significantly both within and between hospitals, and that hospital-level factors contribute to racial and ethnic inequities in HRIF participation.
Methods
Data Source
Vermont Oxford Network (VON) is a voluntary worldwide community of practice dedicated to improving the quality, safety, and value of newborn care through a coordinated program of data-driven quality improvement, education, and research. Of the 819 centers in the United States participating in VON from 2006 to 2017, 58 centers participated in the VON ELBW Follow-up Project and contributed data for this study (Supplemental Table 4) (N = 29 421).
Population
Participating centers reported demographic information and neurodevelopmental outcomes at 18 to 24 months’ corrected age among infants who were 401 to 1000 g or 22 to 27 completed weeks at birth and seen for follow-up from 2007 to 2019.28
We limited our study to infants born at US VON participating centers. We excluded infants who died before NICU discharge (N = 6960) or HRIF program visit (N = 211), or who had unknown hospital disposition or outcome data (N = 38). We also excluded infants who had incomplete race and ethnicity data (N = 59). The University of Vermont and Beth Israel Deaconess Medical Center institutional review boards determined that use of data from the VON research repository for this analysis is human subjects research that meets the criteria for exempt status.
Race and Ethnicity
Race is a social construct reflecting hierarchies of power. Health differences between racial and ethnic populations are not reflective of biology, genetics, or individual behavior. Rather, these differences represent the effects of racism and we therefore include race as a proxy for racism in our analysis.29,30 Participating centers submitted data on race (Black or African American, white, Asian American, American Indian or Alaskan native, native Hawaiian or other Pacific Islander, other) and ethnicity (Hispanic, not Hispanic) on the basis of the 2010 US Census definitions.31 Abstractors were instructed to obtain the information by personal interview with the mother, review of the birth certificate, or medical record, in that order.32 We combined race and ethnicity to yield non-Hispanic white, non-Hispanic Black, non-Hispanic Asian American or Pacific Islander, non-Hispanic American Indian or Alaskan native, Hispanic, or other, which we refer to as white, Black, Asian American, Native American, Hispanic, and other, respectively.
Variables
The primary outcome for the study was participation in HRIF at age 18 to 24 months corrected for gestational age, which is the standard time point for evaluation in HRIF per VON ELBW Follow-up Project protocol. Participating institutions coordinated follow-up, and informed consent for inclusion in the ELBW Follow-up Project was obtained according to each institution's institutional review board specifications. Centers attempted to schedule follow-up with the families or caregivers of all infants who were alive at hospital discharge.
Infant-level demographic and NICU comorbidity information were linked in the VON very low birth weight database. Covariables of interest included gestational age, sex, multiple gestation, inborn status, grade 3 or 4 IVH, presence of periventricular leukomalacia, BPD, necrotizing enterocolitis, necrotizing enterocolitis requiring surgical intervention, focal intestinal perforation, severe retinopathy of prematurity, and patent ductus arteriosus (PDA) requiring surgical ligation. Additionally, we included the need for durable medical equipment after NICU discharge in the analysis. Need for durable medical equipment included (1) gastrostomy or jejunostomy tube, (2) home oxygen, (3) tracheostomy or tracheotomy, or (4) discharge with apnea/cardiorespiratory monitoring. Because of the deidentified nature of the data, no measures of residential racism or opportunity could be included. Data were collected by participating center staff using uniform definitions until neonates were discharged from the hospital, died, or reached 1 year of age in the hospital.32 Measures were reported using standardized data collection tools and each data item was defined in the ELBW Infant Follow-Up Manual of Operations.28
Hospital-level covariables of interest obtained from the VON Annual Member Survey included hospital type (government-owned, nonprofit, private, other) and academic hospital (presence of residents or fellows). Average number of ELBW admissions annually and proportion of white NICU ELBW admissions were obtained from infant data.
Analysis
We performed bivariate analyses using Student’s t test for continuous variables and χ2 test for categorical variables between all infant-level variables and the primary outcome. We used generalized linear models to determine the association between hospital-level variables and the primary outcome accounting for hospital-level clustering. We derived the list of candidate variables for our multivariable model from bivariate analyses with threshold for inclusion of P < .10.
The data were structured in 2 levels, with individual infants at level 1 and hospitals at level 2. The clustering of infants within hospitals informed the analytic approach, affording the opportunity to quantify the relative contributions of covariables at each level.
We used generalized linear mixed effect models to test within- and between-hospital variation in HRIF participation. We first created a null model with random hospital-level intercepts to calculate the interclass correlation coefficient (ICC). The ICC is a ratio of the between-hospital variance to the total variance, to determine the proportion of the total variance attributable to hospital only. We then added race/ethnicity to the model to estimate the effect of racism on HRIF participation. Because racism may vary between different hospitals, we included a random term for race/ethnicity to allow the effect of race (ie, racism) to vary between hospitals. Because we were interested in inequity of at-risk infants compared with white infants, we coded white as the reference category. We then included all infant-level variables from the candidate list. To identify hospital-level factors associated with HRIF participation, we added hospital-level variables one at a time. Infants with missing data were excluded from the model.
All statistical analyses were performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC).
Results
Our study sample included 19 503 infants. Of those, 7218 infants were excluded because of unknown disposition or death, and 2711 were excluded because they were born in a non-US center (N = 2654), had unknown race (N = 59), or had unknown follow-up status (N = 1) (Supplemental Fig 3). Of the included 19 503 infants, the mean gestational age was 26 weeks (SD 1.87) and mean birth weight was 829.3 g (SD 191 g). There were 9477 (48.6%) white, 6346 (32.5%) Black, 2675 (13.7%) Hispanic, 639 (3.3%) Asian American, and 147 (0.8%) Native American infants. A total of 219 (1.1%) were identified as other. The complete characteristics of the study population are shown in Table 1.
Of the 58 hospitals included in the analysis, 18 (31.0%) contributed between 1 and 3 years of data, 11 hospitals (19.0%) contributed 4 to 7 years of data, 9 hospitals (15.5%) contributed 8 to 10 years of data, and 20 hospitals (34.5%) contributed 11 or more years of data. Eighteen hospitals (31.0%) contributed to all 12 years of data. The overall follow-up rate did not vary by year over the study period (Supplemental Table 5). The majority of birth hospitals for eligible infants were classified as not-for-profit (47, 81.0%) and academic (42, 72.4%). The average ELBW admissions annually were 63 (SD 32) and the percentage of white ELBW admissions was 48.2% (SD 21.6%).
Bivariate Analyses
Infants who had an HRIF visit at 18 to 24 months’ corrected age were more likely to be of white race (P < .001), younger gestational age (26.0 weeks [SD 1.8] vs 26.3 weeks [SD 1.9], P < .001), a product of multiple gestation (27.0% vs 23.2%, P < .001), and born in the hospital they were cared for (inborn) (84.5% vs 78.2%, P < .001). They were more likely to have BPD (47.2% vs 44.0%, P = .001), severe retinopathy of prematurity (15.1% vs 13.1%, P < .001), and a PDA requiring surgical ligation (17.5% vs 15.0%, P < .001). After discharge, infants who participated in HRIF were more likely to have been discharged with an apnea/cardiorespiratory monitor (43.6% vs 39.0%, P < .001). Infants in hospitals with a higher proportion of white ELBW admissions were also more likely to participate in HRIF (Table 2).
Generalized Linear Mixed Models
In our null model with random hospital-level intercepts, the ICC was 0.26, indicating that 26% of the variability in the outcome was due to hospital alone (Fig 1). After including race/ethnicity with a random term to the model, we found that, compared with white infants, Black infants had an adjusted odds ratio (aOR) of participating in HRIF of 0.72 (95% confidence interval [CI], 0.63–0.81; P < .0001) and Hispanic infants had an aOR of 0.83 (95% CI, 0.72–0.96; P = .01). The covariance parameter for the random race/ethnicity term was statistically significant, indicating that the effect of race/ethnicity significantly varied between hospitals (Fig 2).
Upon adding the infant-level covariables, race/ethnicity continued to be a significant predictor of HRIF participation (Table 3). In addition, infants of younger gestational ages and those who were a product of a multiple gestation, were inborn compared with outborn, and those who had a PDA requiring surgical ligation were more likely to participate in HRIF. Infants discharged from the NICU with apnea/cardiorespiratory monitoring were also more likely to participate in HRIF.
No hospital-level covariables were significantly associated with the outcome in our multivariable model. However, upon adding a term for the proportion of white ELBW admissions to the model, both the hospital-level variance and the ICC decreased, indicating that a proportion of hospital-level variability was because of the percentage of white infants in the NICU. (Results of the complete model-building process can be found in Supplemental Table 6.)
Subgroup Analyses
We pursued exploratory subgroup analyses limited to infants identified as white and Black because this was the largest inequity in the population. There were no significant changes to the infant-level factors predicting HIRF participation. Upon adding the proportion of white ELBW infants into the model in quintiles, we identified a potential dose-response relationship, and thus included the variable in the model. Upon inclusion as a continuous variable, we found that increasing proportion of white ELBW was associated with a higher aOR of 3.91 (95% CI, 1.06–14.47) of HRIF participation. This suggests an effect of the NICU racial composition on follow-up rates. We further explored whether there was a differential role of proportion of white infants in the NICU on white versus Black infants by including an interaction term (proportion white ELBW admissions × Black). This interaction term was not significant, meaning there was no difference in the effect of the proportion of white ELBWs on follow-up for Black compared to white infants (Supplemental Tables 7 and 8).
Discussion
We identified significant variation in HRIF participation among infants cared for in 58 NICUs in the United States from 2006 to 2017. The proportion of eligible infants with an HRIF visit ranged from 0% to 80% by hospital. Hospital alone accounted for 26% of the variability in outcome. In multivariable models, we did not identify specific hospital characteristics associated with HRIF participation. In addition to low HRIF participation overall that has been previously reported in the literature, we identified significant racial and ethnic inequity in HRIF both between and within hospitals. Importantly, the effect of race significantly varied between hospitals.
Previous studies have identified individual health-related factors associated with HRIF participation with varied results. Although Mercier et al identified that inborn infants, multiple gestations, and younger gestational ages were associated with an increased likelihood of participation in HRIF, Nehra et al did not find similar patterns for infants of multiple gestations.14,33 Furthermore, although Hintz and others have identified that NICU comorbidities such as IVH or BPD are associated with an increased likelihood of HRIF participation, Nehra et al did not.15,16,33 When accounting for the clustering effect of hospital, we identified few NICU comorbidities that predict individual-level HRIF participation. Those comorbidities that continue to be associated with HRIF, younger gestational age, a PDA requiring surgical ligation, and infants discharged from the NICU with an apnea/cardiorespiratory monitor may represent a subset of patients who are more ill, or perceived by caregivers to be more ill, at discharge, and thus caregivers may place a higher priority on attending HRIF for developmental support than infants without those comorbidities. Further research is needed to understand observed relationships, particularly caregiver perceptions of the role and value of HRIF, and to develop strategies to further increase HRIF attendance for infants without comorbidities.
Our findings corroborate previous findings of inequities in HRIF participation.6,15–17 Though race has been conflated with poverty and other indicators of material deprivation, it is now being acknowledged that the social disadvantage associated with race is not truly synonymous with, nor a proxy for, race. Rather, those associations are the result of inequitable programs and policies favoring one group over another; that is, long-standing structural racism. Our unique methodologic approach employing multilevel mixed effects modeling enabled us to partition the many types of racism at multiple ecologic levels that drive inequities in HRIF participation in the United States.
The fixed effect of race, consistent across individuals in our sample, may be conceptually understood to represent the impact of structural racism on health care access utilization. Structural racism refers to the ways in racial discrimination is perpetuated through mutually reinforcing systems such as health care, housing, and income inequality.22 Here, we imagine that such structural forces lead to an overall decreased odds of HRIF participation for Black ELBW infants in the United States.
In subgroup analyses, we saw that hospitals with more white infants had higher overall HRIF participation rates. This is consistent with previous data highlighting that hospital segregation is associated with care quality.23,25,26 In our study, the association between hospital demographics and HRIF participation may be considered evidence of institutional racism, the differential and racialized access to resources, where hospitals with more white patients deliver higher-quality care.21 Exploring hospital demographics as a driver of racial and ethnic inequities may prove useful in exploring the underpinnings of inequities in other health outcomes and health service utilization.
Our results indicate that the effect of race significantly differed, not just between hospitals, but also within hospitals. Within-hospital inequity may represent local structural and institutional racism that may lead to differential HRIF participation because of in-hospital lack of social supports, diminished trust between providers and patients and families, and postdischarge decreased clinic accessibility, lack of transportation, or limited time off from work. Additionally, it may be that the within-hospital inequities are the effect of interpersonal racism. Interpersonal racism refers to personally mediated prejudice, assumptions, and differential behaviors and actions on the basis of race.21 Quantitative and qualitative research has found that, within NICUs, there are differential quality, treatment, and experiences of Black compared with white families.23,27,34 Interpersonal racism, experienced as either explicit or implicit racism, may explain how infants within a single NICU have different likelihoods of HRIF participation on the basis of race.
Inequities in HRIF participation are important, not only because they may lead to unmet needs and poorer quality care in the short term, but also because of potential effects on long-term pediatric health. There are racial and ethnic inequities in NICU comorbidities like IVH that are associated with poorer long-term outcomes.34–37 HRIF participation is designed for early detection of prematurity-associated neurodevelopmental and functional impairments, and to facilitate referral to needed services. Inequities in HRIF participation may therefore serve to further widen the health equity gap that already exists between white and nonwhite infants at NICU discharge and potentially magnify the effects of inequitable perinatal health throughout the life-course.19
This work has 3 methodologic limitations. First, infant race and ethnicity were based on the classification of maternal race and ethnicity. Not only may there be misclassification by data abstractors, but identifying infant race as maternal race erases the social construction of race, as well as the important role of nonbirthing parents in a child’s identity formation. Furthermore, categorical, mutually exclusive definitions of race and ethnicity do not acknowledge individuals with multiple intersectional identities and obscures the important granularity of within-race subpopulation analysis. Second, our study lacked important social information, such as indicators of socioeconomic status, that may represent important mediators that lie downstream from structural, residential, and institutional racism.38 In line with investment and focus on equitable follow-through after NICU discharge, efforts are currently underway at VON to improve the collection of social and demographic data. Finally, we have measured follow-up visits at age 18 to 24 months as our outcome and we are unable to know if nonparticipation resulted from lack of referral from the NICU, missed HRIF appointments after a referral was made, or follow-up in a clinic that did not participate in the VON ELBW Follow-up Project.
Despite these limitations, our study has several strengths. We used a multilevel, analytic approach to capitalize on the data structure and answer questions about the roles of individuals and hospitals in follow-up. Our work is both hypothesis-driven and grounded in social science theory. We used the social ecological model39 and the public health critical race praxis40 (PHCR) to develop our hypothesis, inform methodologic approaches, and aid in the interpretation of results. The social ecological model views ecological levels, in this case, the individual and the hospital, as each having distinct influences on health.39 PHCR is a “semistructured process for conducting research that remains attentive to issues of both racial equity and methodologic rigor.”40 PHCR is based on the understanding that race is a social construct and that racism is ordinary and embedded in society. It orients research to exploring the underlying drivers of inequity and encourages researchers to consider research as a pathway to health equity and justice. Using these frameworks, we identified multiple levels at which racial and ethnic inequity is upheld and potential targets, at multiple ecological levels, for interventions to improve health equity. Improving HRIF participation may be critical in preventing the worsening of health equity between white and non-white, high-risk, preterm infants across the life-course.
Conclusions
We identified significant within- and between-hospital variation in the effect of race on HRIF participation among a national, multi-institutional cohort of ELBW infants. There are multiple mechanisms by which racism–interpersonal, institutional, and structural–and its downstream effects, may impact HRIF participation among high-risk infants. Further study is needed to identify actionable targets for interventions to address these inequities in care to close the equity gap.
Acknowledgments
We thank our colleagues who submit data to VON on behalf of infants and their families. Participating centers are listed in Supplemental Table 4. We thank Lucy Greenberg for her statistical support.
Dr Fraiman conceptualized and designed the study, conducted all analyses, drafted the initial manuscript, and reviewed and revised the manuscript; Dr Edwards designed the data collection instruments, ensured data integrity, supported the analysis and interpretation of data, and reviewed and critically revised the manuscript; Drs Soll, Horbar, and Mercier designed the data collection instruments, coordinated and supervised data collection, and critically reviewed the manuscript; Dr Litt conceptualized and designed the study, supervised all aspects of the analyses and interpretation, and critically reviewed and revised the manuscript; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
FUNDING: Dr Fraiman was supported by the Agency for Healthcare and Research, grant T32HS000063, as part of the Harvard-wide Pediatric Health Services Research Fellowship Program and by the American Academy of Pediatrics, Section on Neonatal Perinatal Medicine, Marshall Klaus Perinatal Research Award. All other authors received no additional funding.
CONFLICT OF INTEREST DISCLAIMER: Dr Edwards receives salary support from the Vermont Oxford Network. Dr Horbar is chief executive officer, president, and chief scientific officer of Vermont Oxford Network, and an unpaid member of the Vermont Oxford Network board of directors. Dr Soll is vice president, director of clinical trials and follow-up, and an unpaid member of the Vermont Oxford Network board of directors. All other authors have indicated they have no relevant financial conflicts of interest to relevant to this article to disclose.