Addressing Disparities in Mother’s Milk for VLBW Infants Through Statewide Quality Improvement

Our Massachusetts PQC, the Neonatal Quality Improvement Collaborative (NeoQIC), includes all 10 level 3 NICUs in Massachusetts. All NICUs participated in the Human Milk QI Collaborative from January 2015 to December 2017. Hospital characteristics are shown in Table 1. Other stakeholders included the Massachusetts Department of Public Health, Mother’s Milk Bank Northeast, and Massachusetts Breastfeeding Coalition. Hospital team participation was voluntary with in-kind effort. Teams comprised multidisciplinary staff as well as patient family members.

For this report, we included only the 9 level 3 birth centers and excluded 1 outborn-only level 4 NICU because virtually all infants cared for there were included in the data collection from other birth centers. From our centralized database, we included 1900 VLBW infants or infants <30 weeks’ gestation and excluded those who died (n = 118; 6%), infants who were ineligible to receive mother’s milk (ie, presumed milk protein allergy) (n = 64; 3%), infants with mothers who were ineligible to provide their own milk (ie, maternal substance abuse, HIV infection, and/or expected adoption or foster care placement) (n = 47; 2%), and when month of birth was missing (n = 1; 0.1%), leaving 1670 (88%) infants for this report (Table 2). This project was determined to be QI and not human subjects research by institutional review boards at NeoQIC leadership team hospitals. All participating hospitals completed a data use agreement with the coordinating center before submitting data.

We used the Institute for Healthcare Improvement Breakthrough Series framework.³³ A key driver diagram, including outcome, process, and balancing measures, was decided on collectively in early 2015, in-person meetings were held biannually, and webinars were held quarterly for the purposes of team sharing and delivery of QI education by QI experts. NeoQIC leaders conducted site visits approximately twice for each participating team. Site visits in years 1 to 2 were focused on QI education and propelling local QI work forward. Visits in years 2 to 3 were focused on increasing hospital leadership buy in and fostering collaboration between hospitals. Hospital teams collected data using standard definitions, submitted data to a centralized database, and received quarterly data reports of statewide and local team progress. Hospital names were reported openly at in-person meetings and webinars but were coded by letter for written reports.

Teams conducted interventions focused on key drivers (Table 3). Interventions were tracked by NeoQIC leaders in the form of completed monthly “plan-do-study-act (PDSA) homework” forms and team sharing at in-person meetings and webinars. Teams received detailed feedback on PDSA homework forms and live team sharing events by NeoQIC leaders. Interventions were focused on increasing availability of supplies (eg, colostrum collection kits, breast pumps, and skin-to-skin chairs), updating practice guidelines related to breastfeeding support (eg, skin-to-skin care), increasing staff awareness through education campaigns in which breastfeeding support practices were tracked closely (eg, “kangaroo-a-thons”), and use of automatic electronic medical record orders or visual cues to remind staff of breastfeeding support practices.

Regarding racial and ethnic disparities, given limited racial and ethnic differences in process and outcome measures in year 1, we sought to determine more targeted interventions in years 2 and 3. We conducted in-depth qualitative interviews with non-Hispanic black and Hispanic mothers at 2 hospitals to better understand facilitators and barriers to prolonged lactation.³⁴ Relevant facilitators included perceptions of strong practical and emotional support by NICU staff when delivered in a racially and ethnically unbiased manner in the mother’s primary language. Follow-up interventions included examination of variation in NICU family and breastfeeding support groups across hospitals and creation or enhancement of such groups. Relevant barriers included difficulties with NICU visitation due to transportation issues, sibling care, return to work, and exhaustion with pumping. Follow-up interventions included examination of variation in sibling visitation policies, parking costs, difficulty with hospital-grade pump acquisition, and meals for mothers. In response, teams increased lactation education and support (by phone or in person) and provided meal vouchers for breastfeeding mothers. The NeoQIC leadership team also created multicultural, multilingual family education materials that were available for team members (https://www.neoqicma.org/educational-materials).

Infant feeding information was abstracted from the medical record every 1 to 2 weeks and the 24 hours before discharge or transfer by using the following standard definitions: (1) exclusive mother’s milk (100% of “base” milk was from the mother, with or without a fortifier [bovine or human]), (2) any mother’s milk (any amount of mother’s milk, with or without the addition of donor milk, formula, and/or any fortifier), and (3) no mother’s milk (donor milk, formula, and/or bovine fortifier only). We considered provision of mother’s milk to occur when the infant was fed milk from the infant’s own mother, not donor milk. Our main outcomes were any and exclusive mother’s milk in the 24 hours before the initial disposition, which was either discharge from the hospital or transfer to another hospital for ongoing care. This measure mimics the human milk measure used by the Vermont Oxford Network, which has been widely used in QI and research publications.^{10,18,26,35,–37} 

Process measures included the following: (1) prenatal consultation with human milk education (which was mention of “human milk,” “breast milk,” “mother’s milk,” “mother’s own milk,” or “breastfeeding”) benefits in a written prenatal consultation note, by physicians or nurse practitioners only, among mothers who received prenatal consultations (yes or no); (2) first milk expression within 6 hours of birth, either by hand expression or pumping (yes or no); and (3) any skin-to-skin care with the mother on days of life 7, 14, 21, or 28 (yes or no).

Other outcomes were presence of medical or surgical necrotizing enterocolitis (yes or no) or late-onset bacterial or fungal sepsis (yes or no) per Vermont Oxford Network definitions³⁸ and length of stay. Balancing measures were percent small for gestational age (<10th percentile) at discharge or transfer and infant growth over the course of the hospitalization, defined as birth weight for gestational age z score change, by using the sex-specific Fenton and Kim³² growth curves. Maternal race and ethnicity was categorized as non-Hispanic white, non-Hispanic black, and Hispanic of any race (which we hereafter refer to as white, black, and Hispanic, respectively), as well as other race and ethnicity, and unknown race and ethnicity.

Statistical process control charts were used to analyze monthly performance over time for the main outcome and process measures as well as additional outcome and balancing measures of length of stay and growth. Run charts were used to analyze quarterly performance over time for those measures with small sample sizes or rare outcomes, including the main outcomes and process measures by racial and ethnic subgroups and necrotizing enterocolitis and late-onset sepsis for the overall population. In a post hoc analysis, we used control charts to examine the secondary outcome measure of mother’s milk provision at 28 days after birth.

Control charts were created and analyzed by using established methods; initial limits were created after 20 data points and were then fixed or adjusted on the basis of the presence or absence of special cause variation and expected persistence of system change.³⁹ Rules for determining special cause included the following: (1) a single point outside the control limits, (2) a run of 8 or more points in a row above or below the centerline, (3) 6 consecutive points increasing (trend up) or decreasing (trend down), (4) 2 out of 3 consecutive points near (outer one-third) a control limit, and (5) 15 consecutive points close (inner one-third of the chart) to the centerline.³⁹ Because run charts had only 12 data points, centerlines were not adjusted on the basis of the presence of signal, given limited ability to assess persistence of change. We examined rates of any mother’s milk according to racial and ethnic subgroups at 1- to 2-week intervals during the hospitalization and hospital variation in racial and ethnic disparities using χ² tests. We determined the coefficient of variation of race and ethnicity across hospitals in any mother’s milk at discharge. QI charts in Excel and SAS 9.4 (SAS Institute, Inc, Cary, NC) were used to conduct this analysis.

Our key driver diagram was updated throughout the collaborative, and the final version is presented in Supplemental Fig 5. Attendance at collaborative-wide webinars and in-person meetings was high, with ≥8 of 10 teams participating at each webinar and in-person meeting. One hospital collected data in year 1 only, and the rest collected data for all 3 years. Overall, 69 interventions were conducted. Team sharing was focused on prenatal education for year 1, first milk expression in all 3 years, and skin-to-skin care mostly in years 2 and 3 (Fig 1).

In our analysis of 1670 infants, 1637 were ≤1500 g and 33 were >1500 g and <30 weeks. We found improvements in all 3 of our main process measures (Fig 1). Prenatal human milk education increased from 57.6% to 76.6%, first milk expression within 6 hours increased from 36.6% to 57.1%, and any skin-to-skin care in the first month increased from 31.2% to 39.0%. We found no change in any or exclusive mother’s milk at discharge or transfer (63.7% and 46.4%, respectively) (Fig 2) nor did we find changes in any necrotizing enterocolitis, late-onset sepsis, weight for gestational age z score change (growth), or length of stay (Supplemental Fig 6).

Improvement ofprocess measures (Fig 1) and any mother’s milk for the first 3 weeks of hospitalization (Fig 3) did not differ among racial and ethnic subgroups. Disparities in mother’s milk emerged after week 3, such that, at the point of discharge or transfer, 69% of white mothers, 53% of black mothers, and 48% of Hispanic mothers provided any mother’s milk (P < .001) (Fig 3). Regarding the disparity between black and white mothers, white mothers had higher rates of mother’s milk provision compared with black mothers at discharge among all hospitals, but the degree of disparity between black and white mothers varied dramatically by hospital (2%–39% difference, median 16%) (Fig 4). Regarding the disparity between Hispanic and white mothers, white mothers had higher rates of mother’s milk provision compared with Hispanic mothers at all hospitals except hospital C. The degree of disparity between Hispanic and white mothers, among the 8 hospitals where white mothers had higher rates of mother’s milk provision compared with Hispanic mothers, varied from 9% to 48%, with a median of 15% (Fig 4). The coefficient of variation of race and ethnicity performance within hospitals was 26.6 among black mothers, 28.3 among Hispanic mothers, and 21.4 among white mothers, indicating substantial variation in racial and ethnic subgroups between hospitals.

We performed a post hoc analysis of any and exclusive mother’s milk provision at day 28 to determine if improvements in mother’s milk provision occurred earlier in the hospitalization than discharge; we did not find any change at day 28.

In a population-level Massachusetts PQC composed of all 10 level 3 NICUs, we tracked individual-level data on 1670 mother–VLBW infant dyads over 3 years. We found significant improvements in key hospital-based breastfeeding support practices focused on the first month of life that were similar among racial and ethnic subgroups. However, racial and ethnic disparities in provision of mother’s milk emerged after the first 3 weeks of hospitalization in our state. Although we did not find improvements in our main outcome or provision of mother’s milk at discharge or transfer, numerous aspects of our initiative can inform other hospitals and PQCs seeking to address similar goals.

We were pleased to find strong and sustained engagement from Massachusetts NICUs during the project time period, as evidenced by ongoing interventions, data submission, and ≥80% team participation in all webinars and in-person conferences, despite ongoing disrupting events such as changes in leadership, team lead absences, and electronic medical record rollouts. Numerous factors contributed to strong engagement, including the recognized importance of mother’s milk for VLBW infants and structure of PQC initiatives. Teams greatly appreciated the open sharing of data and ideas and receipt of basic QI education, which propelled their local work. The multidisciplinary teams also contributed to engagement, in which different disciplines championed different aspects of the work.

Across the state, significant improvements were noted in key process measures, prenatal human milk education, first milk expression within 6 hours after birth, and skin-to-skin care in the first month of life. Although these are important improvements in and of themselves, the lack of improvement in our main outcome measure of mother’s milk provision at discharge is notable and important. Several factors likely contributed to this inconsistency. First, it is possible that the magnitude of the improvements in our process measures were not great enough to yield improvement in mother’s milk provision at discharge. Next, we chose prenatal education,¹¹ first milk expression within 6 hours,^13,14 and skin-to-skin care⁴⁰ as process measures because they are recommended in national nursing guidelines⁴¹ and have been associated with increased duration of mother’s milk provision until hospital discharge; however, many of the studies used to inform these guidelines are observational and/or from single centers, raising concerns of generalizability to NICUs in Massachusetts. We speculate that our process measures (focused on the early part of the hospitalization) may not have led to sustained improvements in mother’s milk provision until discharge in our state. Additionally, teams prioritized their interventions on key drivers differently; this variability may have limited the greater success of the collaborative as a whole. Finally, the baseline rate of any mother’s milk provision at discharge in Massachusetts was much higher than national rates,¹⁰ which may have made it more difficult to achieve further improvement. A previous California human milk collaborative of 11 hospitals started from a baseline rate of 55% any mother’s milk at discharge and improved to 64% in the postintervention period, using many of the same improvement strategies as our Massachusetts collaborative.¹⁹ California’s postintervention rate matches our Massachusetts baseline. Factors that impact mother’s milk for the duration of the NICU course are complex; processes driving improvement among hospitals that are below the national average may be different from those starting above average.

Reflecting on the efficacy of interventions our teams trialed, we have many “lessons learned.” Regarding prenatal education, prenatal consultation checklists in the electronic medical record were highly effective, especially in settings where trainees primarily perform prenatal consultations. Regarding early initiation, increasing accessibility of supplies (pumps and colostrum expression kits) and incorporating early milk expression support into the labor and delivery and postpartum nursing workflow were highly effective. Regarding continuation of milk production and skin-to-skin care, staff and family education campaigns with close tracking of skin-to-skin care and use of visual reminders for skin-to-skin eligibility, like crib cards, were successful in the short-term. Focusing on nursing staffing barriers and attitudes toward parent engagement in skin-to-skin care was helpful, whereas focusing on nursing knowledge of skin-to-skin care benefits was less helpful. Other successful interventions included pumping at the bedside (versus separate room), breastfeeding support groups, and meal vouchers.

To our knowledge, no previous human milk PQC focused on VLBW infants has made reducing racial and ethnic disparities a main aim of their project. We felt that clearly defining this goal and tracking all metrics by maternal race and ethnicity was critical because it greatly increased awareness of disparities in health outcomes within the Massachusetts PQC community. Among our teams, the open labeling of hospital names was well received and sparked conversation about the possible etiologies of the variation in racial and ethnic disparities between hospitals. When we conceived the project, we hypothesized that our key drivers would vary by maternal race and ethnicity and that focusing on reducing disparities in these key drivers would lead to reduction in racial and ethnic disparities in mother’s milk provision at discharge or transfer. This did not prove to be true. There were minimal differences by race and ethnicity in our key drivers, other than reduced skin-to-skin care among Hispanic mothers, compared with non-Hispanic white mothers.²² We examined our data in more depth in the beginning of year 2 of the project and realized that racial and ethnic disparities emerged after the first 3 weeks of hospitalization, which we have described in detail elsewhere,²² and that there was wide variation in disparities between hospitals. We consequently conducted qualitative interviews with black and Hispanic mothers at 2 hospitals and found that transportation and parking costs, family visitation policies, and returning to work were identified as significant barriers for maternal NICU visitation and ongoing pumping.³⁴ Spanish-speaking mothers also emphasized the importance of communication in the primary language.³⁴ We hypothesized that variation in NICU family engagement practices may explain, at least in part, the variation in disparities. We therefore performed a practice survey in year 2, in which we identified wide variation in the use of scheduled family or breastfeeding support groups, provision of hot meals for breastfeeding mothers, transportation vouchers, family visitation rules, and ease of use of in-person versus phone interpreters. The leadership team created multilingual family education materials that were released in 2016 and videos that were released in 2018. Many teams began to focus on improving family engagement practices in the latter half of our collaborative, which has continued into 2018.

Strengths of our statewide QI collaborative include engagement of all 10 NICUs and robust data collection on 1670 VLBW infants eligible to receive mother’s milk over several time points during the hospitalization. We differentiated mother’s milk from donor milk. Limitations include lack of many other variables that may potentially impact mother’s milk provision, such as maternal medical morbidities, mode of delivery, maternal intent to breastfeed, type of pump or pumping frequency, and sociodemographic variables. We did not track individual-level family engagement measures, such as presence at the bedside or participation in lactation consultations or breastfeeding support groups. We captured maternal education during the prenatal period only, whereas families likely received important education throughout the hospital stay, and our measure of skin-to-skin care was limited to 4 chart audit days in the first month only. A limitation is that we did not focus on etiologies of racial and ethnic disparities beyond the prenatal period and first month of hospitalization from the inception of the project. Instead, we focused on interventions targeted on racial and ethnic disparities that occurred later in the hospitalization, mainly in years 2 and 3, which may have provided less time to achieve reduction in racial and ethnic disparities in mother’s milk provision at discharge or transfer.

As disparities in quality of neonatal care have become more well described,^26,28 PQCs focusing on reducing these disparities represent a promising approach to address this critical problem. In our Massachusetts PQC focused on reducing racial and ethnic disparities in provision of mother’s milk, we found wide hospital variation. Tracking metrics according to race and ethnicity and sharing transparently was well received and sparked interventions related to reduction in racial and ethnic disparities in provision of mother’s milk for VLBW infants at the local and statewide level.

We acknowledge the hospitals that participated: Baystate Medical Center, Beth Israel Deaconess Medical Center, Boston Children's Hospital, Boston Medical Center, Brigham and Women’s Hospital, Massachusetts General Hospital, St. Elizabeth’s Medical Center, Tufts Medical Center, South Shore Hospital, and UMass Memorial Hospital.

NeoQIC

Neonatal Quality Improvement Collaborative

PDSA

plan-do-study-act

PQC

perinatal quality collaborative

QI

quality improvement

VLBW

very low birth weight