STRENGTHS AND LIMITATIONS OF THIS STUDY

The nurse survey data are a unique strength of the study, providing information about hospital organisational factors that are not readily available.

The survey approach, in which nurses are surveyed directly and not via their employer, is also a strength which lessens bias.

The four states in the sample represent about a quarter of the births in the USA in 2016 and three different regions of the country, but generalisations should be made with caution.

Parity was not available in the administrative discharge abstracts, so the target based on the Healthy People 2030 guidelines is especially conservative (ie, the target could be set higher than it is).

The data is from 2016; however, there is reason to believe that the associations found here still exist.

Introduction

Birth is the most common reason for hospitalisation in the USA.1 The hospital where a woman gives birth is a statistically significant predictor of her birth outcome.2–4 Hospital structural characteristics (eg, teaching status, bed size) and maternal risk factors do not fully account for hospital variation in birth outcomes.5

There are three possible birth outcomes: spontaneous vaginal birth (SVB), operative vaginal and caesarean. SVB, a vaginal birth without forceps or vacuum, is the optimal, safest, birth outcome for most mothers and babies.6 7 SVB is associated with a lower risk of haemorrhage, improved recovery time, higher rates of breastfeeding and decreased postpartum pain, compared with caesarean birth.6 8 9 Likewise, SVB is associated with improved neonatal outcomes, including improved respiratory status and optimally colonised microbiomes, which, in turn, are associated with improved childhood health, including lower rates of asthma and coeliac disease.10 11 SVB is the least expensive birth outcome (both in cost and length of stay) and the one most women desire.12 In other words, SVB is high-value healthcare. SVB rates vary across hospitals in a statistically and clinically significant manner.13 Efforts to increase the number of SVBs by decreasing unnecessary caesareans have neither statistically nor clinically significantly moved the needle nationally.14 While some individual states have successfully increased the SVB rate among low-risk women,15 16 the racial disparity in the outcome remains stark.17 18

SVB is marked by racial disparities.19 Black women who are low risk for a caesarean birth, that is, have no predisposing obstetric or medical risk factors that would increase their risk of having a caesarean, are less likely to have SVBs than low-risk White women.19 20 There are stakeholder calls to increase research on SVB (as opposed to caesarean birth) and equity therein.6 21–23

Nurses provide most inpatient maternity care and are critical for ensuring patient safety.24 Nursing care, which includes labour support and patient education, is associated with higher rates of SVB.6 22 25 26 Safe, high-value maternity care, therefore, is likely the fruit of the organisational context in which care is provided. Importantly, we can change these organisational factors to improve patient care and outcomes. The hospital’s nursing organisational factors, which include the work environment and staffing, are linked with patient outcomes and racial disparities in outcomes in other patient populations.27–29 Organisational factors, especially those pertinent to nursing such as the work environment and nurse staffing, have only been minimally explored as a mechanism explaining hospital variation in birth outcomes.30 The National Academy of Medicine recommended transforming the nurse work environment to keep patients safe and decrease disparities two decades ago.26 31There is a need for research that focuses on hospital nursing organisational factors as potential system-level opportunities for intervention to improve birth outcomes and realise equity in the USA. The purpose of this study was to describe the organisational factors of hospitals that achieved (1) the SVB rate target for Black women at low risk for caesarean birth,32 (2) the SVB rate target for Black women at low risk for caesarean birth and equivalent (not statistically significantly different)28 SVB rates between Black and White women at low risk for caesarean birth, or (3) neither of these metrics. Hospitals in the first and second category were considered ‘high performing’, while those in the third category were considered ‘low performing’.28

Methods

We conducted a cross-sectional descriptive study using linked data from three different sources from 2015 to 2016: nurse surveys,33 patient administrative discharge abstracts29 and the American Hospital Association (AHA) Annual Survey.34 The nurse survey data were made available from the parent study which conducted a cross-sectional survey of nurses in four states in 2015/2016.27 33 35 Further methodologic information is published elsewhere.33 A random sample of registered nurses, taken from the state licensure list, was sent a survey at their home address. The response rate was 26%; non-respondents were also queried and found to not be statistically significantly different from respondents on key variables.33 Registered nurses identified the hospital where they worked and served as informants about hospital organisational factors, including the work environment, staffing, safety, quality and missed care. The AHA Annual Survey provided hospital structural characteristics, such as bed size and teaching status. Hospital organisational factors derived from the nurse survey were linked to AHA hospital data and the patient-level clinical data. This resulted in a patient-level merged file that could be collapsed to the hospital level to examine hospital structural and organisational characteristics and the patient outcomes in those hospitals.

Data sourcesPatient administrative discharge abstracts

The Office of Statewide Health Planning and Development in California, the Healthcare Cost and Utilization Project’s State Inpatient Database for Florida, the New Jersey Department of Health and Senior Services and the Pennsylvania Health Care Cost Containment Council provided the patient discharge abstracts.33 Birth hospitalisations were identified using a published algorithm36 crosswalked to the International Classification of Diseases, Tenth Revision (ICD-10) codes.13 Women aged less than 12 or greater than 55 were excluded from the sample, as were those with lengths of stay greater than 120 days.37 Hospitals were included if Black women gave birth in them. Hospitals with less than 100 births per year were excluded, following prior research,3 13 resulting in the exclusion of 31 hospitals and the 409 births that occurred therein, leaving a sample of 494 hospitals with 879 7444 births. One hospital was dropped for missing more than 20% on the patient race variable, following prior research.38

Nurse survey

In the nursing data, hospitals were kept if there were at least three maternity registered nurses39 reporting on the hospital’s organisational characteristics.35 Maternity nurses were those identifying labour and delivery, postpartum or labour/delivery/recovery/postpartum units as their primary workplace.

American Hospital Association (AHA)

We used data from the 2015 AHA Annual Survey which collect data on structural factors such as ownership, bed size, teaching status and rurality on all hospitals in the USA.

Measures

We identified SVB using the International Classification of Disease-10 codes and diagnosis-related group codes O80, 767, 768, 774, 775, 796, 797, 798, 805, 806, 807 and 10E0XZZ. To evaluate potentially misclassified cases, we then identified caesarean births (765, 766, O82, 10D00Z0, 10D00Z1, 10D00Z)29 and operative vaginal deliveries (10D07Z, O81) and cross-tabulated them with SVB. Almost 3000 (2994) women were listed as having both an operative and an SVB and were recategorised as having an operative vaginal birth. The 718 women listed as having both a caesarean and an SVB were recategorised as having a caesarean. Low-risk status for having a caesarean birth was defined according to the Society for Maternal-Fetal Medicine (SMFM) algorithm.40

How hospital categories were created

Hospitals were placed in one of three categories: (1) Black women at low risk for caesarean birth had an SVB rate that was greater than or equal to 73.4%; (2) SVB rates for Black women in the hospital met the target, and their rate was not statistically significantly different from that of White women in the same hospital; or (3) neither (ie, hospital neither met the target SVB rate or had equivalent rates). Categories one and two were considered high performing and category three low performing. The SVB target of 73.4% represents all births (100%) minus the Healthy People 2030 target for low-risk caesareans (23.6%),32 which equals 76.4%, minus the operative vaginal births in 2016 (3%).41 It should be noted that the Healthy People 2030 goal focuses on the prevention of caesareans in women who are at low risk and have not given birth before. The caesarean rate in women who have given birth previously is significantly lower than in women who are giving birth for the first time.42 Therefore, using the Healthy People 2030 goal is especially conservative for women who have given birth previously. We subtracted operative vaginal births because this is a category of birth outcome distinct from SVB.

Patient race

Patient race and ethnicity are available in the hospital discharge abstracts.

Work environment

The nurse survey included the National Quality Forum-endorsed Practice Environment Scale of the Nursing Work Index (PES-NWI), a tool to measure the clinical work environment.43 The PES-NWI includes five domains comprised of traits associated with professional nursing practice: (1) nurse participation in hospital affairs; (2) nursing foundations for quality of care; (3) nurse manager leadership, ability and support of nurses; (4) staffing and resource adequacy; and (5) collegial nurse-physician relations. The five subscales had an overall Cronbach’s α of 0.85 with the individual subscales ranging from 0.79 to 0.89.43 Nurses respond whether the trait is present at their primary job via a 4-point Likert scale ranging from 1 (strongly agree) to 4 (strongly disagree). The intraclass correlation coefficient (1,κ) was 0.62, demonstrating sufficient reliability to aggregate to the unit level.44

Staffing

Nurses reported the number of patients and nurses on their unit during their last shift. We created an aggregate staffing measure by dividing the average number of patients reported on the unit by the average number of nurses reported on the unit during the last shift. This measure has demonstrated predictive validity for examining the relationships between nurse workloads and patient outcomes in other patient populations.45 46

Quality

Nurses answered the question ‘In general, how would you describe the quality of nursing care delivered to patients in your work setting?’ with four possible Likert-type responses including ‘excellent’, ‘good’, ‘fair’ and ‘poor’ where ‘excellent’ equalled 1 and ‘poor’ equalled 4. These responses were averaged for each hospital to reflect the quality of maternity nursing care in the hospital. Prior research has demonstrated that nurses are objective observers of quality and safety.47

Safety

Nurses rated the safety of their work setting via the following survey item: ‘Please give your current practice setting an overall grade on patient safety’. This was a Likert-type item with five responses—‘excellent’, ‘good’, ‘acceptable’, ‘poor’ and ‘failing’—where ‘excellent’ equalled 1 and ‘failing’ equalled 5.

Safety culture

The nurse survey included items from the Agency for Healthcare Research and Quality’s Hospital Survey on Patient Safety Culture.48 49 Each item was measured via a 5-point Likert scale ranging from 1 (strongly agree) to 5 (strongly disagree). We included two of these items: (1) staff feel like mistakes are held against them and (2) staff feel free to question decision of those in authority.32 41 42

Statistical analysis

We used descriptive statistics and appropriate tests of significance, including χ2, to describe the sample. We then conducted analysis of variances (ANOVAs) and followed the ANOVAs with a pairwise comparison of means with equal variances with Tukey’s honest significant different test to account for multiple comparisons. We assessed missingness in terms of our outcome and the exposure variable. Hospitals with >20% missingness were removed the study (described above). Analyses were performed in Stata 17.050 using two-sided statistical tests with an α level of 0.05 based on complete case analysis.

Patient and public involvement

Patients and the public were not involved in the design, conduct, reporting or dissemination plans of our research.

Ethical considerations

The University of Pennsylvania Institutional Review Board deemed the current study ‘not human subjects research’ as the data were entirely deidentified. Informed consent was not required for this study as it used secondary data that were deidentified. The hospital and patient data are publicly available for research. The parent study obtained informed consent from the nurse survey participants. We used the STROBE cross-sectional checklist when writing our report.47

Results

Of the 257 hospitals in the sample (see figure 1), one-third of the hospitals (33.5%) had SVB rates for Black women that met the target, and 38.5% of hospitals had SVB rates that both met the target and were equivalent for Black and White women. Less than a third (28%) of hospitals were low performing. The proportion of women giving birth who were Black varied significantly across the three hospital categories, with low performing having the lowest average (8.1%, SD 9.6%) and the two high-performing hospital categories having substantially more (12.8%, SD 12.5%; 16%, SD 16.9%) for meeting the target and meeting the target and equivalent, respectively (p=0.001). The average percent of women at low risk for caesarean who had an SVB varied across the three categories, from 75.17% (SD 43.21) in the low-performing hospitals to 79.18% (SD 40.6) and 78.92% (SD 40.79) in the high-performing hospitals that met the target and those that also had equivalent rates, respectively (p<0.001). While the overall low-risk SVB frequency among Black (77.7%, SD 41.62) and White (78.8%, SD 40.86) women was similar, these varied significantly across the hospital categories for both groups of women and were higher in high-performing hospitals. The states in which hospitals were located across the three categories varied in a statistically significant manner as well (see table 1). Other structural characteristics, including technology, teaching, bed size, rurality and birth volume, did not vary statistically significantly across the hospital categories. Further details are presented in table 1.

Figure 1

Sample inclusion flow diagram for hospitals included in our sample.

Table 1

Summary statistics for hospitals over and across performance categories

The nursing work environment varied in a statistically significant manner across all the three hospital categories from 2.7 to 2.84 (p=0.04) as seen in table 2. Of the five work environment subscales, nurse-physician collaborative relationships also varied in a statistically significant manner across the three hospital categories (2.7 to 2.77, p =0.02). When comparing the categories against each other, high-performing hospitals that met the SVB target for Black women had significantly better rated nurse-physician collegiality than low-performing hospitals (contrast between means =0.08, 95% CI 0.01 to 0.15). Nurse manager ability, leadership and support of nurses were scored worst in low-performing hospitals compared with hospitals that met the national target (contrast between means: 0.21, 95% CI 0.008 to 0.4). The average number of nurse practitioners (NPs) (including midwives) employed across the hospital categories varied significantly (2.76 to 3.51, p =0.003), with low performing having fewer than high-performing hospitals.

Table 2

Variation of organisational factors across hospital performance categories

Quality and safety ratings did not vary in a statistically significant manner between high- and low-performing hospitals. The same held true for nurses’ assessments that their mistakes were held against them or whether they were free to question the decisions of those in authority. See table 2 for further details.

We conducted a sensitivity analysis in which hospitals were grouped into two categories based on whether they (1) met the SVB target for Black women at low risk for caesarean (high performing, 173 hospitals) or (2) did not met the target (low performing, 85 hospitals). Most of the output in this analysis did not vary from what has been presented with two exceptions. First, the nursing work environment did not vary significantly between the two categories. Second, nurses in high-performing hospitals, compared with low performing, were more likely to strongly disagree that mistakes were held against them (2.75 vs 2.57, p =0.0459) or that they were not able to question those in authority (2.86 vs 3.01, p =0.01).

Discussion

We were interested in examining hospital organisational factors associated with improved birth outcomes and equity for Black women at low risk for caesarean birth. We were motivated to do this to identify hospital organisational factors that might suggest system-level targets to improve birth outcomes and equity. Overall, we found that the nursing work environment, and the collegial nurse-physician relationship subscale, varied statistically significantly across the three hospital categories and was better rated in high-performing hospitals. When comparing the categories against each other, we found that the hospitals where SVB rates among Black women met the target not only had a significantly better overall nursing work environment but also higher rated nurse manager and collegial nurse-physician relationship subscales. Though the change in work environment scores seems relatively small, other researchers have found that similarly sized differences are clinically meaningful. Researchers have reported that improvements in the work environment score over time are associated with significantly less missed care51 and decreased odds of having poorly rated quality and safety on nursing units.27 This suggests that the small improvement in the average work environment score we report in high-performing hospitals in this sample may be clinically significant.

High-performing hospitals were statistically significantly more likely to employ more NPs, on average. The finding around advanced practice nurses in the organisation corroborates other research that has found improvement in birth outcomes in hospitals that employ midwives even if the midwives are not caring for the specific patients in question.49

Our findings bear similarities to recent qualitative research by Howell et al30 and to those of Betran et al.52 Even though Howell et al were looking at hospitals that were high performing with regard to severe maternal morbidity, they found that strong nurse-physician communication and teamwork were important for high-performing hospitals.30 We, too, found that the collegial nurse-physician relationships were statistically significantly better rated in high-performing hospitals. Betran et al52 also noted dysfunctional teamwork, the strength of interdisciplinary team culture and professional power relationships as being associated with the frequency of caesarean birth.52 Howell et al30 reported adequate physician and nurse staffing and supervision as hallmarks of high-performing hospitals. The variation in average nurse staffing between high- and low-performing hospitals in our sample was not statistically significant. On average, however, nurses in high-performing hospitals in our sample reported fewer patients per nurse, which aligns with the broader literature that staffing is associated with maternity outcomes, including higher rates of SVB.22 45 Nurse staffing is linked with patient morbidity and mortality, readmission and racial disparities therein, in other patient populations.24 39 53 Issues around teamwork and communication are critical for safety culture and have been independently linked with preventable maternal mortality, an outcome marked by racial disparity.54 55

Quality and safety did not vary in a statistically significant manner between high- and low-performing hospital categories. Our sensitivity analysis, however, suggested that high-performing hospitals that met the SVB target had better safety cultures, with nurses strongly disagreeing that mistakes were held against them (2.75 vs 2.57, p=0.046) or that they were not able to question those in authority (2.86 vs 3.01, p=0.01). In a prior study, researchers found that fewer than half (43.7%) of maternity nurses felt free to question the decisions or actions of those in authority, and only 26.7% disagreed that their mistakes were held against them.35 This suggests the presence of a relationship between safety culture and birth outcomes, especially with regard to the role of nurses, that has not been well delineated in the research previously.

Limitations

While the four states in the sample represent about a quarter of the births in the USA in 2016 and three different regions of the country, generalisations should be made with caution. For instance, few rural hospitals were represented in the sample. The data are old, but we think it is unlikely that the associations reported here have changed meaningfully, since the SVB rate has remained fairly stable over time and hospital organisational factors have not improved dramatically. Race and ethnicity data in administrative discharge abstracts have been critiqued for not necessarily representing how an individual identifies.53 Research has found that discharge abstracts have a high sensitivity and positive predictive value for the accurate identification of Black and White race (ie, race in the discharge abstract accords with birth certificate data or self-report), while validity for identifying Asian-Pacific Islander and American Indian and Alaskan Native is lower.53–57 Additionally, the Healthy People 2030 goal is specifically for women giving birth for the first time, but we included both nulliparous and multiparous women in the sample. Thus, the SVB goal is more conservative than it might be. As a result, we believe the associations we reported are accurate, but future work should stratify by parity. Finally, we used the term ‘equivalent’ to describe hospitals where there was no statistically significant difference in the SVB rate between Black and White women who were at low risk, but this does not mean that the care and outcomes are necessarily equitable. We believe it is important to note hospitals where a disparity (a difference adversely affecting disadvantaged populations) is not apparent, but improving the quality of care and outcomes to achieve equity is an ongoing effort.

Conclusion

We found that the nursing work environment and especially collegial nurse-physician relationships were statistically significantly better in high-performing hospitals. High-performing hospitals also had a higher average number of NPs employed in the organisation. Both the nursing work environment and the number of NPs employed by the organisation are hospital organisational factors amenable to intervention to improve patient outcomes. We also found evidence to suggest better safety culture—especially around questioning those in authority and not fearing that mistakes would be held against one—in high-performing hospitals, which is also a feasible organisational factor for hospitals to target to improve care and outcomes.

Data availability statement

Data are available upon reasonable request. Data may be obtained from a third party and are not publicly available. Multiple data sets were used for this study. Depending on the set, data may be made available upon reasonable request or be obtained from a third party and are not publicly available.

Ethics statementsPatient consent for publication

Not applicable.

Ethics approval

This study does not involve human participants as determined by the University of Pennsylvania Institutional Review Board.

Acknowledgments

The authors wish to acknowledge Jesse Chittams, Senior Biostatistician, and Octavio Alanis’ helpful contributions.