Reorganisation of health-care services requires an evaluation of the clinical benefit and costs to the health service and to the wider society. The aims of this chapter are to provide the building blocks for this evaluation by exploring the impact that service reconfiguration has on clinical outcomes (mortality), costs (neonatal bed-days, LOS and parent costs) and to undertake qualitative research on the factors that families and policy-makers would like to see taken into consideration in determining service reconfiguration.
Clinical outcome
Literature review
During the last 20 years, many models have been developed to estimate infant and neonatal mortality. The majority of these models explored the impact of infant characteristics on mortality and estimated mortality for either the VLBW infants, weighing 800–1500 g, or very preterm infants, with a gestational age of 22–32 weeks. Most studies used a logistic regression approach with covariates that have been found to be strong predictors for neonatal mortality (i.e. of gestational age, small for gestational age, sex, birthweight and birthweight z-score).70–73 A few studies considered sociodemographic and socioeconomic variables including race and ethnicity,18,20,74–76 education, insurance status, percentage of inhabitants living below the poverty line18,20,77 and the lowest decile of the IMD score.19
A smaller number of studies have investigated the impact of service configuration in terms of (1) working patterns and staffing and (2) organisational level. One of the first studies that looked at the impact of workload and staffing was by the Tucker and UK Neonatal Staffing Study Group,78 which explored the impact of the availability of consultants (high availability defined as ≥ 2 consultants) and nurses relative to the BAPM nurse-to-cot ratio guidelines (high availability is defined as a nurse-to-infant ratio of ≥ 0.84). The UK Neonatal Staffing Study Group78 used three workload measures, including occupancy, which measured the maximum number of infants present in a unit over their study period. The authors reported that for every 10% increase in percentage of maximum occupancy at admission, the odds of mortality increased by 1.09 (95% CI 1.01 to 1.18). Rautava et al.79 explored the impact of working patterns on mortality and found that the risk of mortality for very preterm infants increased when the infant was born in non-office hours, and that mortality rates could be consequently improved by an increase of resources. Finally, Watson et al.14 estimated the effect of the 1 : 1 nurse-to-patient ratio for intensive care neonates and found that a 1 : 1 nurse-to-patient ratio reduces infant mortality. A further set of studies explored the degree to which mortality varied depending on the care levels by which services were organised (e.g. NICUs, LNUs and SCUs in the UK). For example, Cifuentes et al.20 estimated the mortality of low-birthweight infants for different levels of care in the California area and found that the level of care in NICUs can influence the probability of survival. The 2010 systematic review by Lasswell et al.80 explored the association between the designation level of hospital and VLBW infant mortality based on studies that evaluated the regionalisation of perinatal services for very preterm or VLBW infants. Most studies that looked at the impact of organisation in terms of staffing or hospital level used logistic regression, with the exception of Watson et al.,19 who used an instrumental variable approach. Care is needed in interpreting estimation of the logistic model when there is a small number of events, as is the case in neonatal mortality, and so these estimates may be affected by small-sample bias.
Other studies have aimed to look at the impact of organisational factors, such as the number or volume of infants treated in neonatal units, based on the idea that staff can become more experienced and skilful as they treat a larger number of different and complex infants. A high-volume unit has been defined as one that treats at least a fixed number of VLBW or very preterm infants per year76,78 or as one that is in the top quartile of all neonatal units.19 Rogowski et al.76 explored the impact of hospital-level determinants of mortality among VLBW infants and showed that the number of VLBW infants admitted to a neonatal unit reduced infant mortality. More recently, studies have aimed to look at the causal relationship between infants born in high-volume units and infant mortality using instumental-variable (IV) approach methods.14,19,77 The effect of designation and volume of neonatal care for preterm birth reported in Watson et al.19 had a significant effect, especially on neonatal mortality and in-hospital morbidities (bronchopulmonary dysplasia, necrotising enterocolitis and retinopathy of prematurity). For infants with < 33 weeks of gestational age, the risk of death was reduced by 2.6 percentage points if admitted to a high-volume unit, and this effect was higher if infants had a gestational age of < 27 weeks. Finally, some studies explored the impact of hospital volume within different levels of care. For instance, Phibbs et al.18 examined the impact of hospital volume among different levels of neonatal care units and found that volume and level had a significant impact on risk of infant mortality, showing that the delivery of VLBW infants in NICUs with a high volume can reduce neonatal mortality.
summarises the neonatal mortality models that explored the impact of volume on mortality.
Brief description of effect of volume on neonatal outcomes reported in the literature
Thus, the few studies that have attempted to estimate causal effects suggest that there is a positive effect of birth in higher-volume hospitals, whereas there is no evidence that birth in a hospital with a NICU has any effect on mortality. However, previous studies have only compared NICUs against all other hospital designation categories combined without distinguishing between SCU and LNU hospitals. Furthermore, there is no study in England that has evaluated the causal impact on LOS and reimbursement costs. Our aim was to estimate the impact of volume and designation level on mortality and costs between NICUs, SNUs and LNUs, and separately between high-volume units and other hospitals of birth.
Data
Data relative to neonatal care were collected in 2014/15 from units in England as part of the BadgerNet data set. The distance in time and miles was evaluated using LSOAs of the mother and the postcode of the hospital. Mortality was defined as mortality during the in-hospital period from the admission to the discharge.
Mortality was registered between 2014 and 2015 for a total of 2010 infants, out of a total number of 165,450 admissions to neonatal units. Out of all registered deaths, 52% were for infants born with a gestational age of < 28 weeks. By adding infants born between 29 and 32 weeks of gestational age, 65% of deaths are covered; including infants born between 33 and 36 weeks, up to 83% of deaths are accounted for. illustrates the rate of death and survival per gestational age of infants admitted to neonatal units in our study.
Mortality and survival rates of infants by gestational age.
Method
High-risk infants tend to be treated in high-volume units, so in order to estimate the effect of volume on mortality, it is important to measure this effect in a representative sample of infants, including both high- and low-risk infants. More recent studies have aimed to estimate the causal effect of volume using an IV approach to control for confounding. The proximity to care can be used as an instrument that determines the chances of receiving care (e.g. birth in a hospital with a NICU) and should be independent of infant mortality. Proximity to care can be used to control confounding by estimating the difference in mortality between those who live close to NICUs and those who live far from NICUs, and both groups should be made up of a comparable mixture of high- and low-risk infants. The strength of an instrument can be tested by looking at the relationship between the instrument (i.e. travel time) and attendance at high-volume NICUs. In our case, a strong instrument would imply that those who live nearest to high-volume units are more likely to attend at those hospitals (i.e. the t-statistic on travel time, or the F-statistic if there is more than one instrument, is > 10).
Watson et al.19 used an IV approach based on travel distance alongside eight other instruments [i.e. surgical facilities (1 = yes, 0 = no), high volume (1 = yes, 0 = no), Level 3 (1 = yes, 0 = no), Level 2 (1 = yes, 0 = no), distance × surgical facility, distance × high volume, distance × Level 3 and distance × Level 2]. Level 3 represents NICUs and Level 2 represents LNUs.
The main analyses will use only one instrument (travel time or travel distance) to facilitate interpretation and to eliminate the need to identify and remove weak instruments when multiple instruments are used. The secondary analysis will use multiple instruments. We control for the following covariates in the model: age and age squared at birth, sex, deprivation of residence (quintiles of multiple deprivation), mode of delivery (emergency caesarean without labour, emergency caesarean with labour, vaginal non-spontaneous, elective section, unknown or vaginal spontaneous) and fetus number. A similar approach is used to estimate the impact of high volume on total LOS (sometimes referred as the super stay) and the associated reimbursement costs by level of care (BAPM) actually received, for infants referred to high-volume units, where length of hospital stay is defined as the number of days from admission to hospital discharge or death, whichever took place first. The LOS results are shown in the evaluation section of this report (see Chapter 9).
Two sets of neonatal mortality models are estimated: semiparametric and parametric models. The semiparametric model is a structural mean model (SMM)82 and serves to estimate the treatment effect on the treated, thus allowing for the possibility of different treatment effects between the treated (NICU-born) and untreated (non-NICU-born) infants. The parametric model instead adopts a bivariate (probit) distribution for mortality outcomes and the exposure status (birth in a hospital with a NICU vs. birth in a hospital without a NICU) and allows us to estimate the average treatment effect (i.e. the effect in the whole infant population, at the cost of imposing the assumption of homogeneity of treatment effect and normal distribution of unobservable confounding factors). We present both sets of results, but in the main discussion we focus on the semiparametric results, given that these results are based on less restrictive assumptions and so are more robust estimates of the effect. We have instrumented both the semiparametric and parametric models using an IV approach and, for comparison, run a linear probability model (LPM) that has no instrument [i.e. ordinary least squares (OLS)].
Two different units were considered: high-volume and tertiary units (i.e. NICUs). High-volume units are characterised by a minimum number of 100 admissions per year of infants with a birthweight of < 1500 g. Tertiary units are represented by NICUs that are the highest level of neonatal care, providing a service dedicated to babies needing respiratory support (ventilation) weighing < 1000 g, born at < 28 weeks’ gestation or needing significant continuous positive airway pressure support.
Like Watson et al.,19 we estimated infant mortality for infants born at a gestational age of < 32 weeks in a high-volume unit or hospital with a NICU. In addition, we conducted analysis to explore the effect of neonatal transfers to a NICU hospital from a lower-level hospital in infants born before 32 weeks’ gestational age. We conducted a sensitivity analysis on the results for those born between 26+0 and 31+6 weeks of gestational age, to account for the possibility of bias due to the effect of imbalance in the distribution of extremely premature babies across treatment (hospital of birth) groups.
Secondary analyses of the relative effects of birth in a hospital with a NICU versus a SCU and a LNU were conducted using three available instruments of travel time to these three types of hospitals. These IV models were formulated as seemingly unrelated equations and estimated by simulated maximum likelihood.83 In these analyses, neonatal mortality was analysed using multivariate probit distributions. Because we did not have complete information on the closest LNU and SCU units to some infants in the data set, we conducted these analyses excluding infants with incomplete data.
Results
We first checked if travel time was correlated with the exposure variable (i.e. birth in a hospital with a high-volume neonatal unit, and birth in a hospital with a NICU), thus supporting its use as an IV. Second, we looked at whether or not travel time is correlated with any other sample characteristics, such as birthweight and sex, to check if any possible effect of travel time on mortality operating through the exposure variable is confounded by other variables.
The descriptive statistics in summarise sample characteristics by tertiles of travel time for high-volume units and NICUs and shows that in most cases there are no systematic differences of sample characteristics across the travel time tertiles. In addition to the exposure variables (delivery at hospital with a NICU and delivery at hospital with a high-volume unit), systematic differences arise only for deprivation of residence and unknown delivery mode; this suggests the need to control for possible confounding by these variables in our analyses.
Sample characteristics by travel time to high-volume units and NICUs
shows the mortality model using LPM, semiparametric IVs and parametric IV bivariate probit model (marginal effect) for the high-volume units and hospitals with NICU. The controlled covariates used in these models are gestational age and age squared at birth, birthweight, sex, deprivation of residence, mode of delivery and fetus number.
Effect of neonatal care at high-volume units and NICUs (n = 12,687)
The instrument strength reported for the IV model estimates, both in the linear SMM and bivariate probit models, shows that the instrument is strong (e.g. the t-score test statistic on travel time is 32 for high-volume units; rule of thumb is that a robust instrument has an F-test statistic of > 10). In addition, travel time affects the treatment variable (the probability of delivery at a hospital with a high-volume unit in one analysis and delivery at a hospital with a NICU in another) in the expected direction and is negative, implying that the longer travel times are associated with a lower likelihood of birth in hospitals with high-volume units and of birth in hospitals with NICUs.
The estimated causal effects of the IV models indicate that delivery in high-volume units reduces neonatal mortality by 1.2 percentage points in accordance with the bivariate probit model and by 5.0 percentage points with the linear SMM. In contrast, the LPM reports high-volume units having an increased risk of death, but this result is likely to be affected by confounding issues (i.e. high-volume units treat high-risk infants and so are likely to have higher mortality), as indicated by the Hausman test statistic. Sensitivity analyses excluding infants born at < 26 weeks’ gestational age were also conducted and they confirmed these findings (see Appendix 2).
Birth in a hospital with a NICU does not appear to result in any difference in terms of mortality relative to other hospitals, as summarised in Table 13. These results are based only on the use of one instrument: travel time to closest NICU hospital. We ran additional analysis to explore the effectiveness of NICUs for those infants born in a hospital with a NICU or transferred within the first 48-hour period. Of all neonatal transfers with a recorded transfer time (n = 1519), 65% took place within 48 hours of birth. Infants who were born in a hospital with a NICU or who were transferred to a hospital with a NICU had no detectable effect on mortality, using the IV approach based on the single instrument of travel time to closest NICU (or distance to closest NICU). The same results were obtained when the sample was limited to those infants born between 26+0 and 31+6 weeks’ gestational age (see results in Appendix 2).
Secondary analysis of the relative effects of birth in a hospital with a NICU compared with a hospital with a SCU or a LNU were conducted using three available instruments of travel time to these three types of hospitals. In these additional analyses, we find that the NICU does appear to reduce mortality, compared with the other levels of care, by 2 percentage points, and so suggests that NICUs in themselves have some beneficial impact on mortality compared with other levels of care (see Appendix 2).
Discussion
We estimate infant mortality for infants born at a gestational age of < 32 weeks as a function of exposure to high-volume unit or hospital with a NICU at birth. We find that exposure to a high-volume unit at birth reduces mortality relative to other neonatal units. A very preterm infant born in a high-volume unit (≥ 100 babies weighing < 1500 g per year) has a 5-percentage-point lower risk of death in this unit than in other neonatal units when travel time is used as the instrument and mortality is estimated using a semiparametric method (linear SMM). This estimate drops to 1.2 percentage points when the same analysis is run using a parametric method (bivariate probit). There is a debate to be had about which result should be given greater credence. The semiparametric approach is based on less-restrictive assumptions and, therefore, is potentially more robust to violations of assumptions underpinning the analysis, so we have chosen to emphasis this result here; to allow comparisons with existing literature in this area, it is necessary to further discuss the parametric result. The parametric result is also the one that we use in the evaluation section of this report to calculate the incremental cost-effectiveness of high-volume units, to avoid potential problems with predicted values outside the 0–1 probability range.
Watson et al.19 found that a preterm infant born in a high-volume unit (defined as those in the top quartile of all neonatal units) has a 2.6-percentage-point lower risk of death than in other neonatal units when travel distance is used as the instrument and mortality is estimated using a parametric method. This percentage point risk reduction is not immediately apparent from the paper, but can be calculated from the reported OR of 0.68 for in-hospital mortality reported in the paper (which approximates the risk ratio in cases like this when the deaths are rare), and the in-hospital mortality for high-volume units reported as 5.5 percentage points in the descriptive statistics [giving an estimated percentage point reduction of approximately = (5.5/0.68) – 5.5 = 2.6 percentage points]. The 2014 estimate of Watson et al.19 is higher than the 1.2 percentage point reduction found by our parametric approach. An obvious explanation for the differences is the definitions used for high-volume units, but similar results were found when we defined high volume as those in the top quartile of all neonatal units. Another reason for the differences is the instruments used, given that Watson et al.19 used travel distance and we explored the use of both travel distance and travel time. We report here only the results based on travel time as an instrument because there is strong support for travel time accurately representing access to health-care services. Sensitivity analysis excluding infants born at < 26 weeks of gestational age halves the mortality effect of birth in high-volume units compared with other units (2 vs. 5 percentage points in all the infants aged < 32 weeks’ gestational age), but the estimates are imprecise.
A baby being born at < 32 weeks’ gestational age in a hospital with a NICU does not appear to result in any difference in terms of the risk of death compared with other units. Similar results are also found by Watson et al.19 We ran additional analysis to explore the effectiveness of NICUs for those infants born in a hospital with a NICU or transferred within the first 48-hour period. Birth at a hospital with a NICU or being transferred to a NICU within a 48-hour period had no detectable effect on mortality when using the IV approach based on the single instrument of travel time to closest NICU (or distance to closest NICU).
In our interviews, policy-makers raised queries about the robustness of the finding that NICUs did not affect the risk of death compared with other hospitals. To check the robustness of the result, we present additional analyses comparing the effectiveness of NICUs with other levels of care (i.e. NICU vs. LNU and NICU vs. SCU). All other results for the NICU were based on only one instrument: travel time to closest NICU. The advantage of considering other levels of care is that it opened up the approach to using three instruments: travel time to the closet NICU, LNU and SCU. A slight disadvantage of the approach is that the data set was less complete, because we did not have complete information on the closest LNU and SCU for some infants in the data set; therefore, we were restricted to performing these additional analyses on a smaller data set that excluded infants with incomplete data. In these additional analyses, we find that NICU does appear to reduce mortality by 2 percentage points, compared with the other levels of care.
A limitation of the data and analysis is that it is currently not possible to estimate the impact of mortality for those infants who are transferred into NICUs. The numbers of transfers are not insignificant; for example, for all infants with a gestational age of < 33 weeks at birth and who have ≥ 7 days of BAPM Level 1 (intensive) care, 61.2% are born in a hospital with a NICU, 18.9% are born in a hospital without a NICU and are transferred to a NICU and 19.8% are born in a hospital without a NICU and are not transferred to a NICU. We are therefore unable to separate out the benefits of antenatal care taking place prior to birth. It is clear that birth in a high-volume unit leads to improvements in mortality, and closing down low-volume units might lead to more infants being born in high-volume units, but the impact of changes in transfers is unclear.
Costs
In this part of the study we had planned to look at NHS neonatal costs in more detail by first gaining a better understanding of the national reference costs and how they inform HRGs. We had also planned to assess the components that make up the national reference costs by collecting data from the four main types of neonatal units that currently exist in the UK, and explore how these data are seen to vary by the number of infants. Finally, we had planned to estimate the costs of neonatal care for families, based on a survey of family costs by BLISS.
When looking at the national reference costs84 and how they inform HRGs, it became clear that units were still paid in accordance with the HRG 2001 data set,85 which did not accurately reflect resource usage by BAPM guidelines2 (i.e. nurse-to-infant ratios). The reference cost submissions in July 201786 were the first to ask for units to submit data in accordance with the revised HRG reference cost guidance that took BAPM 2011 guidelines2 into account. During the course of our study and interviews with unit staff, a lot of units were still trying to work out how best to apply the new guidance and so were reluctant to release cost data, making it hard to apportion costs to the different activities. To assess the impact on NHS costs, we decided to shift the focus from assessing the components that make up the reference costs to exploring the impact of high volume on the LOS of infants, to allow us to begin to explore the impacts of reorganisation. Further work on the cost components would be possible using the submissions under the new guidance that became available in January 2018, but which is beyond the scope of this current report.
Literature review
There are a number of papers that look at the impact of gestational age on hospital neonatal costs and families. Rogoswki et al.87 explored the impact of gestational age on neonatal and perinatal cost in the American Vermont Oxford hospital network for NICUs. Costs were classified into accommodation costs and ancillary costs; ancillary costs were divided into five subcategories: respiratory therapy, laboratory, radiology, pharmacy and other ancillary. The authors show how costs vary within gestational age, birthweight, location of birth and discharge status. The category of infants who have the highest costs are those born between 24 and 26 weeks, with a birthweight of < 1000 g and born outside the hospital. The study also shows an inverse relationship between costs and gestational age; this result is also confirmed for neonatal and childhood costs for extreme preterm88 and preterm births.89 Further work by Petrou et al.88 estimated costs for extremely preterm birth for families using evidence from a population study. Results show that extremely preterm births are associated with higher public sector costs and there is an inverse relationship between costs and gestational weeks. Several sociodemographic covariates were included in the model, but only long-term unemployment is associated with an increase in costs.
For service reorganisations, it is important to consider the impact on costs as services change and the effects of economies of scale and scope. One approach used to address this question is to look at the elements that make up the costs and analyse how they vary by case mix. The UK Neonatal Staffing Study Steering Group developed a cost function to evaluate the nature and the degree of economies of scale in the provision of care for NICUs.90 The economic analysis shows that volume and case mix interact to determine the degree of economies of scale, even if the determinants of costs and efficiency in neonatal costs have a high complexity. The treatment of the sickest infants centralised at a regional level can take advantage of economies of scale. Another study by O’Neill et al.91 investigated the relationship between activity (total days of care provided and total days of intensive care provided) and costs (clinical staffing, support services and overheads) using a multivariate regression model. They found an inverse relationship between average cost per day and scale of services provided, confirming the benefits of centralisation of intensive care in larger units. The authors91 also show that the adoption of a different form of estimation (i.e. the log–log or double-log function) provided the best fit to the data.
Another approach is to simply look at the costs of high-volume units compared with low-volume units. Watson et al.92 costed NICU services using the tariffs paid to hospitals to cost high-volume NICUs compared with low-volume NICUs, and compared their effectiveness in terms of reductions in the risk of mortality to estimate the cost-effectiveness of moving £100 to high-volume NICUs. The study estimated an incremental cost per life saved of £420,000 per life-year saved.92
Reference costs and how they inform Healthcare Resource Groups
Historically, HRGs for infant intensive care tended to be too low and HRGs for infant special care tended to be too high. Intensive care for infants should use similar costs to intensive care for adults and so should be much higher. For example, the ratio of HRGs in 2014/15 were intensive care = 2.8 × special care, high dependency = 2 × special care, special care = transitional care. This anomaly has arisen because of the way units have submitted reference costs; there is a tendency to average the nursing over all infants rather than to apportion nurses’ costs to the care needs of infants based on BAPM guidelines.2 Differences between units’ costs have also arisen as a result of the way that units apportion:
costs between neonates and paediatrics
costs between the different neonatal unbundled HRGs
diagnostic costs
layout and organisation costs, etc.
Reference cost submissions inform HRGs, but there is a lag whereby HRGs change more slowly than the reference cost submissions. Payments continue to be based on the 2001 HRGs despite the new BAPM classification in 2011. An update to BAPM 2011 was agreed in 2015, and this went through the appropriate systems to flow into the NHS data collection systems from December 2016. There are now two sets of data being collected: HRG 2001 for price and payment, and HRG 2016 for reference costs. shows the information flows relative to the two BAPM classifications.
Example of information flows for 2014/15 data.
Length of stay and costs
Methods
In this section, we explore the impact of service configuration on LOS and cost the LOS using a microcosting approach based on the HRG per diem reimbursement.
Length of stay (LOS) is defined as the number of days from admission to hospital discharge or death, whichever takes place first. In our analysis, we assumed that the infant spell was censored if the last episode for an infant was a transfer to another hospital (detailed results available from the authors). The costs of reimbursement for neonatal services for the whole inpatient spell were derived by applying HRG per diem reimbursement tariffs for 2015 based on the 2001 reference costs and payment system, which were the reimbursement opportunity cost to hospitals at the time of this study. In 2015, tariff costs were as follows: intensive care-days were reimbursed at £1176.47, high-dependency-days at £847.15, special care-days at £532.95, normal care-days at £424.35 and transitional care-days at £464.23.
We consider the impact on LOS and reimbursement of two service configurations: (1) high volume and (2) birth in a NICU compared with birth in a LNU or SCU. As with the mortality modelling, for the analysis of high-volume units we use the whole data set, whereas, for the comparison of NICUs with other levels of care (LNU and SCU), we use a slightly smaller data set that excluded infants with an incomplete number of data on the closest LNU and SCU.
The analysis used an IV approach similar to the one employed to analyse mortality. Following convention, LOS and costs were assumed to follow a log-normal distribution,93–95 whereas the two additional equations, for the SCU and LNU binary treatment indicators, were modelled as before using a probit equation in each case. The same instruments and covariates as for the analysis of mortality were used for obtaining estimates of these models, and included covariates for gestational age, gestational age squared, infant sex, last decile of IMD score, mode of delivery and number of fetuses. We present the results of naive OLS regressions of the LOS and cost equations for comparison.
In order to avoid problems in convergence of model estimation, the model that was developed included 172 cases that presented incomplete hospital spells without adjustment for censoring (1.5% of overall data).
Results
shows that the total LOS following birth in a high-volume unit is 9 days longer and costs £5715 more to commissioning bodies than birth in another neonatal unit.
Causal effect on LOS and costs of birth in a high-volume unit (n = 12,687)
shows that the mean total LOS following birth in a LNU is shorter by 1–2 days, whereas birth in a SCU results in 3–4 fewer total hospital days, relative to a NICU. Although the IV estimates are not significant, the diagnostic test results are consistent with the idea that the variables of interest, birth at LNU and birth at NICU, are not endogenous and, therefore, a simple OLS model may provide valid approximation to the true effects on LOS. The same result applies to IV estimates of effect on reimbursement costs, which were both statistically insignificant. A simpler OLS model suggests that a birth in a SCU is £1870 less costly to commissioning bodies than a birth in a NICU, although the effects are imprecisely estimated. In contrast, reimbursement costs for a birth in a LNU is £643 less costly to commissioning bodies than a birth in a NICU, but the result is not significant. These log-linear model estimates are back-transformed to original units, adjusting for the non-linear effect of the error variance on treatment effect estimates.
Causal effect on LOS and costs of birth in lower-level hospitals (LNU and SCU) relative to NICU (n = 11,037)
Discussion
Length of stay following birth in a high-volume unit has a mean duration of 9 days longer and a mean cost of £5715 more to commissioning bodies than LOS following birth in another neonatal unit. LOS following birth in a LNU is shorter by 1–2 days, whereas birth in a SCU results in 3–4 fewer total hospital days, relative to a NICU. For reimbursement costs, a birth in a SCU is £1770 less costly to commissioning bodies than a birth in a NICU, although the effects are imprecisely estimated (with overlapping CIs). The reimbursement costs to commissioning bodies for births in a LNU are no different from those for births in a NICU. This appears paradoxical given the longer total LOS for birth in a LNU; however, a possible explanation for these results is the different production functions between NICUs and LNUs in terms of their relative use of number of days at different levels of care.
We will return to the estimates of LOS and reimbursement again in Chapter 9, Evaluation of high-volume neonatal intensive care units compared with other hospitals, when we calculate the effectiveness of high-volume units and NICUs.
Parent costs
Data
Data were collected by BLISS on 1347 parents for neonatal events between 2010 and 2014 in the UK for infants with a gestational age of between 25 and 34 weeks, most of which took place in England (89%). The questionnaire collected clinical information about the pregnancy, infants and place of birth (gestational weeks, hospital and unit type, infant additional hospitalisation and LOS). Information was also collected on the financial status of individuals during the period when the infant was born and the expenses paid by the parents during the visit to the neonatal unit (overnight stays and the relative cost, costs of parking, food, public transport tickets, travel and childcare, how neonatal hospitalisation affected the family budget and access to new loans). Finally, information was collected on sociodemographic status (such as income, sex, age, relationship status and ethnic group).
Methods
An OLS cost model was developed in order to capture the factors that define and influence the costs borne by families during the event of a birth in a neonatal unit. These costs are considered ‘out of pocket’, meaning that they are not supported by the NHS, but they can have a significant impact on family budget, especially considering the long LOS of preterm infants in neonatal units.
The model evaluated the feasibility of a regression model in relation to several variables or characteristics of infants and families using a linear regression model. The dependent variable (the total costs for families) was defined by the following covariates: cost for food and travel (in GBP), the use of childcare and baby care when parents were away, overnights spent, the purchase of breast pumps, the use of parking, the use of unpaid leave, a dummy variable that represents parents both having had unpaid leave, if the employer of the partner was supportive in the maternity period, the average household income of parents, the days of visit per week in the neonatal unit, the presence of children at home and the relative number, the distance in miles from the birth hospital, the age of the parent, the use of public transport, the use of private and public transport and if the parent is in a couple.
Results
At an early stage of the analysis, concerns were raised over the missing data. Some answers were mostly complete; for example, infant LOS was only missing for 6% of cases. Other missing answers are attributable to families finding it difficult to recall such information; for example, the distance in miles (29% of cases were missing) and time (81% of cases were missing). The questionnaire contained 100 questions, with 22 questions relating to financial information, and some open-text questions. The final model is based on 614 complete observations, so caution is needed regarding the generalisability of these results.
summarises the main results and shows that the cost of food and travel, the use of baby care, the use of car parking, the unpaid leave and the presence of unpaid leave for both parents, the average income and the support of the employer of the partner during the maternity period are all significant. Factors that reduce costs are the partner’s employer (a higher availability of the partner can help to reduce parents’ expenses), the LOS of the infant, and if the mother is in a couple (even if all the covariates are not significant). The model shows a good fit to the data with an adjusted R2 of 0.5847.
Family costs regression model
Discussion
The model shows that unpaid leave, food, travel, baby care and parking all have a significant impact on costs. The support from a partner’s employer can reduce costs, as does the availability of the partner to help (e.g. with preparation of meals to take to the hospital and other facilities).
The questionnaire was long, which may have increased the likelihood of missing data. To improve completeness, we recommend fewer questions that did not use free text.
What is important to families?
The objective of this part of the study is to undertake qualitative research on the factors that families and policy-makers would like to see taken into consideration when determining service reconfiguration. The aim is then to assess the feasibility of including these aspects in a DCE, which is typically used in health economic evaluations.
The interest in DCEs in health-care decision-making has increased in recent years. DCE is a method that allows a number of characteristics to be traded off against each another.96,97 Janus et al.97 report a systematic review of preference elicitation studies, defining categories for studies that informed clinical decision-making, supported reimbursement decisions (as in health technology assessments) or elicited the perceived benefits and risks of health innovation for the market authorisation of drugs; in all of these types of decision, DCEs were adopted.
There are review articles summarising how qualitative research can be used to inform health-care research.98–100 There are also applied examples of how qualitative research has been used to inform outcomes important to families and policy-makers in other areas (e.g. for children and young people with neurodisability).101 In addition, we are aware of two methodological papers that give detailed advice about how attributes might be developed for a DCE,102,103 but very few DCEs report the attribute development stage in any detail.104 In this research, we follow the steps suggested by Coast et al.102 and, like Klojgaard et al.,103 we recommend cognitive interviews to help identify if any attribute might prove problematic in the DCE. Data collection may take the form of interviews or focus groups in the absence of a well-constructed meta-ethnography. Coast et al.102 suggest that the choice between these methods may be a result of practicalities: interviews are recommended for sensitive topics (attitudes towards end of life) and focus groups are recommended if discussion among those affected may reveal additional issues.
The preferences present in DCE studies are related to health and non-health outcomes, processes and service characteristics.104 A crucial element in the DCE process is the selection of appropriate attributes for outcomes and process characteristics. It is also likely that including qualitatively different attributes can also increase the complexity of the decision and may make the choices harder to complete. If we consider service reorganisation, such as centralising health services, the impacts can be present in both health, non-health and process characteristics.105 DCEs have been proposed to examine patients’ and decisions-makers’ preferences towards such reorganisations of care, such as those currently under review for centralisation of neonatal care.33 However, such applications face the additional challenge that the outcomes may take place at different points along the care pathway (e.g. risk of hospital mortality for the mother or child vs. risk of longer-term childhood disability) that increase decision complexity. Although no formal DCE study been undertaken here, the qualitative research will be used to inform the outcomes of interest that will be used in a later DCE and to assess the feasibility of using DCEs in future service reorganisations of neonatal care.
Systematic review
The current guidance for attribute selection in DCEs emphasises the need to achieve a balance between the competing objectives of the participants and the decision-maker, the relevance of the research question(s) and if attributes are related to one another.106–108 However, the challenges faced in selecting appropriate attributes for service provision, outcome or both have received less attention. The systematic review examines the extent to which researchers investigating antenatal and neonatal care through DCEs consider and justify in their design whether the attributes are for service provision, outcome or both (see Report Supplementary Material 1).
Methods
Systematic electronic searches of prespecified terms were performed in EconLit, EMBASE HMIC (Healthcare Management Information Consortium), MEDLINE, NHS EED (Economic Evaluation Database), PsycINFO and Web of Science databases, from 2000 to June 2016. DCE studies investigating topics on premature infants, neonates, newborns, or mothers/fathers/parents were included. Studies were excluded if the participants were children or adolescents aged < 16 years.
Results
After removing duplicates, 9701 unique results were identified but only 299 DCEs for antenatal and neonatal care were initially considered, of which 13 met the inclusion criteria. The selection of studies is presented in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram109 (see Appendix 3, ).
Most of the studies were conducted in industrialised countries (UK, n = 4;110–113 Australia, n = 3;114–116 the Netherlands, n = 2;117,118 and Canada, n = 1119), and three were conducted in developing countries.120–122
The objective of the elicitation exercise varied in all studies from types of obstetric services,117 screening tests for Down syndrome,113 induction of labour114 and delivery,118 perinatal experiences119 and supplementary diet,121 among others. Most of the studies involved women participants, and health-care providers were included in only two studies.115,122 Almost 80% (10/13) of the studies focused on antenatal care, including one that considered both antenatal and postnatal care. The number of individuals participating in the DCEs varied from 56 to 1464 (mean 284, median 130). The mode of administration was also diverse, with web-based questionnaires;110,115,118 postal questionnaires,111,117 and pen-and-paper questionnaires at clinics and teaching hospitals,113,114,119,121 with the rest reporting using a questionnaire but without specifying the mode of administration.112,116,120,121 The reduction in studies selected (from 299 to 27) due to eligibility is because the vast majority related to paediatric rather than neonatal care and/or did not relate directly to DCE. Overall, the studies showed a heterogeneous picture of selection of attributes for service provision, outcomes or a mix of both, but there is a lack of consideration or justification for these choices.
Summary of approaches used to develop attributes within these discrete choice experiments
Attributes were selected by reviewing existing studies,111 as part of the characteristics of the intervention being evaluated in a randomised controlled trial,113,118 qualitative interviews with patients and/or stakeholders,112,113,117,118,122–124 literature review and qualitative interviews,110,115,116,120,125–127 and international guidelines.121 However, in some studies it was unclear how the authors reduced the number of attributes obtained in qualitative research to a manageable set of attributes.116
Focus group
At the beginning of the study, we started with a set of attributes suggested by the pilot work, which included one process outcome (access to neonatal services) and three health outcomes: maternal mental health, infant death and risk of childhood health problems. We presented these attributes to the first focus group before conducting the qualitative interviews, giving examples of how they might be shown to parents:
maternal mental health – reduce anxiety and depression
risk of infant death for those born at 24 weeks of gestational age – reduce from 5 in 1000 to 2 in 1000 infants
risk of childhood health problems – eyesight problems increase from 1–2 in 100 to 2–5 in 100 infants
access to neonatal services – increases travel time for some families from 60 to 120 minutes.
Although some of the outcomes, like access and travel times, are more straightforward, the issues of risk and uncertainties over morbidity are clearly complex, and so the research set out to explore the feasibility of DCEs, or other approaches, to collect data to inform the weights that decision-makers can apply in this context. Examples of framing considered for uncertainty in focus groups were as follows:
Low confidence – our confidence in this effect estimate is limited.
Medium confidence – we are moderately confident in this effect estimate.
High confidence – we are very confident that the true effect lies close to our estimate.
Or:
Most parents felt that the presentation of these options was difficult and not easy to understand and compare, and two parents preferred to use the confidence levels rather than the values reported for risks. We also presented parents with a table to illustrate risk with a pictogram with 100 faces, showing 2% chance as two faces of different colours. Generally, it was felt to be a clear presentation of risk. We also showed parents the protocols we were aiming to use in the qualitative study, which aimed to discuss the parents’ experiences focusing on a set of open questions. Little additional feedback was given on these protocols in the focus group.
Patient preference interview
The qualitative study built on earlier public and patient involvement work,33 which aimed to identify outcomes that were important to families of children requiring neonatal care. The qualitative research explored these outcomes further and aimed to refine the language that is used to describe these outcomes, to explore in what ways these outcomes are perceived to vary by parents and to identify if any attribute might prove problematic in the DCE.
Methods
Recruitment of participants
We purposively sought to interview parents who had a child discharged from the neonatal service within the previous 6 months to 5 years, who were interested and willing to share their experience of using the neonatal service. We aimed to recruit a varied range of parents to provide further insight into the issues and outcomes that are important and how they describe those outcomes. Carrying out the interviews at participants’ houses broadened the scope of who we could interview.
Neonatal support groups, such as BLISS and SNUG (Supporting Neonatal Users and Graduates), hold a list of parents who have consented to be contacted for future research. We advertised the study through this list of parents and through a conference in south-west England on neonatal services to parents held in October 2016 in Torquay, as well as through families known to the neonatal unit at Royal Devon and Exeter NHS Trust through the lead link, Sue Prosser, who is Matron of the neonatal unit at Royal Devon and Exeter NHS Trust. The advertisement is provided in Report Supplementary Material 2.
At the recruitment stage, prospective participants were provided with written information about the purpose of the research, what the interviews would involve, how long the interviews would take and how their data will be used (see Report Supplementary Material 2). Prior to the beginning of the interview, participants were asked to sign a consent form (see Report Supplementary Material 2).
Flexible topic guide
A flexible topic guide for individual interviews with parents (see Report Supplementary Material 2) was developed. The semistructured probing questions were based on a series of attributes that were raised from patient and public involvement (PPI) in a previous study of neonatal care;33 a review of the literature; outcomes suggested by the National Perinatal Epidemiology Unit (www.npeu.ox.ac.uk/research; accessed January 2018); and PPI feedback from a NeoNet PPI group held in Exeter in July 2016. In 2017, a set of outcomes for neonatal care was under development for the COMET initiative (www.comet-initiative.org/; accessed January 2018),128 but the study was at too early a stage to inform our work. The topic guide was piloted with two parents and then amended in the light of feedback prior to commencing data collection.
Interview process
The 10 interviews were carried out by Katie Kelsey, five in the University of Exeter St Luke’s Campus, four in family homes and one in a children’s centre following a SNUG coffee morning. The lengths of the interviews were tailored to suit the participants, and varied from 35 minutes to 2 hours. Both parents were interviewed in two cases, and the mother was interviewed in all other cases.
Data management and analysis
The interviews were recorded digitally and transcribed verbatim. During transcription, all names and places were anonymised: any names of parents reported here are pseudonyms. Thematic analysis supported through a framework approach was used.99,129
Two researchers (KK and PL) initially read through the transcripts and developed a thematic coding framework that captured the factors that parents would like to see taken into consideration in determining the configuration of neonatal services and how they describe them. The coding framework was then used to code the transcripts to generate relevant themes and subthemes using NVivo version 11 software (QSR International, Warrington, UK). The framework (see Report Supplementary Material 2) was developed further as they (researchers KK and PL) became more familiar with the content.
Katie Kelsey and Paolo Landa separately coded two transcripts to check for comprehensiveness and consistency of coding. Differences that arose in interpretation between the researchers were discussed. Katie Kelsey and Paolo Landa then coded all remaining materials from the interviews.
Thematic charts were developed:
Thematic matrices were created in NVivo version 11 to help with the analysis.
The coded data for each transcript were summarised in the matrices so that for every theme and subtheme we had a summary of each person’s related dialogue.
Each of the themes were then synthesised further so that we had a condensed version of each theme.
The summaries for each of the main themes were presented to a final PPI group meeting on 5 June 2017, to which all of the interviewed parents were invited. This group validated the results and discussions took place to further refine the outcomes.
Ethics approval was granted by the University of Exeter Medical School Research Ethics Committee (reference Dec16/B/096Δ2).
Results
Ten families were interviewed using the flexible topic guide (see Report Supplementary Material 2), which explored the following characteristics:
Hospital environment.
Travel time and means of transport.
Impact on the family (emotional and financial).
What is important for the family?
Language used, sensitivity, medical language, how parents explain the events, services and treatment.
Understanding of risks – infant survival and long-term disabilities.
Mitigating aspects – family support, SNUG and BLISS support and hospital staff support.
Background – gestation weeks, LOS and other children in family.
The families that were interviewed all had very different stories. The gestational time ranged from 24 to 34 weeks and the LOS ranged from 2 to 17 weeks. For half of the families, this was their first child. Some of the births were in a NICU and some were in a LNU; some of the babies born in a LNU were transferred to a NICU.
A summary of the synthesis under each theme as presented to the PPI group is presented in the following sections. We identified the themes in the coding process (see Report Supplementary Material 2).
Hospital environment
The families that were interviewed reported having very profound experiences in the neonatal units. They reported feeling deep and contrasting emotions of trauma, anxiety, frustration and excitement as they lived through the first weeks of their baby’s life, often not knowing if their baby would survive. One mother spoke of the moment of excitement when the hospital staff began referring to the time ‘when’ she would take her baby home, implying the ‘if’ prior to that. They all had very vivid and detailed memories of their time in the neonatal units, even years after their baby was born. The parents became attached to the care teams/nurses and felt that they were ‘like family’. Communication was very important and parents felt supported by the staff and that they could always ask questions. They would frequently telephone for updates, and felt that their baby was getting the best care.
There were also some occasions when parents noticed a problem with their baby and they had felt that it took a long time for the hospital staff to take notice.
Parents found it very helpful to have a tour of the neonatal unit before the birth, as they were then mentally prepared for the sight of their baby in an incubator and attached to tubes and monitors. An issue highlighted by some parents was the difficulty they found with initial bonding. This was for various reasons: partly because of all the equipment, partly that there were so many other people involved in the care of their baby and also the anxiety around the baby’s vulnerability. It made a big difference when the hospital staff encouraged them to be involved.
The understandable attachment parents felt towards their care teams meant that transferring between hospitals could be a source of anxiety, even when moving nearer home to a lower level of care. On the other hand, those who had other children and were travelling every day to the neonatal unit and managing other childcare found that moving nearer to home was of great benefit. Some parents found the atmosphere in the NICU to be intense and preferred the more hands-on nature of the LNU.
Some families/mothers made good long-term friends while in the neonatal unit, as these were people with whom they had shared an extreme experience.
It was very important to families that they be involved in the care of their baby. If parents were travelling to the unit every day, and/or juggling other childcare, it made a big difference when the hospital staff took into account their timings so that they did not miss their baby being fed or washed.
It has been reported in recent studies130–132 that it is important to involve parents in the process of care when their infant is in a neonatal unit. This process usually consists of feeding the infant with breastmilk, cleaning and bathing, taking care of the infant and letting the infant feel the mother’s presence.
Some parents felt that it was crucial for them to have been able to stay at the neonatal unit. Those who did not stay spent every day there and found that the facilities made a big difference. It was important that they could bring their other children to the unit, and the nurses made this very possible:
[. . .] although it was a stressful, worrying situation you’re in, it was nice at the same time, it was really bizarre, it was . . . It’s just like another family, you know [. . .]
Family 3 d
[. . .] I think it’s just, trying to be there for your baby, because it’s such a strange sensation, because when you have a normal birth and you take your baby home straight away, there’s obviously that bonding that happens immediately. Whereas, with a premature birth, the doctors and nurses have all that first contact, so particularly when you first come on to the unit and you’re around your baby, you’re not really sure what you’re allowed to do or what you should be doing and, you know, when we first came on to the unit, I just didn’t think I’d be able to touch him or have any contact with him at all [. . .]
Family 6 m
Impact of travel
It should be noted that parents are willing to travel anywhere and make whatever sacrifices necessary to help make sure that their infant has the best care possible. Sometimes this is at the expense of the rest of the family. Some parents were travelling by car for ≥ 1 hour each way every day to the neonatal unit. Those parents without cars relied on public transport and family members, and in some cases spent ≥ 3 hours per day travelling. The impact of travel becomes much more pronounced when there are other children in the family, with the parents trying to fit their time at the neonatal unit around school times.
Parents can feel torn between their children at home and the need to look after their baby as much as possible, and they can feel guilty as soon as they leave the hospital. They had been advised by health-care professionals that the older children will remember their lack of attention whereas the baby will not, so they tried to make sure that they included the other children as much as they could:
[. . .] And all of this has a cost involved in it, which is irrelevant, really, because – well, it is relevant, but you’re gonna do everything you can for your child, for your baby, you know [. . .]
Family 2 d
[. . .] I would spend, like, a Sunday night, so I could get all her school uniform ready at home and a Wednesday night at home just so it broke the week up for her [. . .]
Family 7 m
[. . .] that was the toughest time. That was the hardest time, being away from everyone, not having any support [. . .]
Family 4 m
Family disruption
Families are disrupted in different ways and some of the effects are felt for years after the birth. When the mother and baby are in hospital, the separation is felt by the rest of the family. Parents report struggling with difficult behaviour because other children feel neglected and/or are worried about their sibling.
Families spoke about living in bubble away from family, in which parents, especially the mother, is focused only on the baby’s health and development. For parents, it is a stressful and traumatic time affecting the whole family. Parents feel guilty about leaving their baby behind and also feel guilty about neglecting their other children.
Strains are felt between parents who feel that they are on an emotional ‘rollercoaster’, with each parent feeling the strain differently and trying to stay strong for the other. When fathers had to go home, leaving mothers in hospital, this created other tensions and anxieties.
The wider family was also affected as people do not know what to say or how to act. Parents felt that:
[. . .] every day we ended up saying sorry [to each other] and starting each day [. . .]
Family 2 m
[. . .] They came in and saw [baby] all hooked up to god knows what else, you’re trying to explain to them that everything’s OK, but they’re like, ‘well, come on mum, he’s got, like, a tube down his throat, he’s got one in his tummy . . . he’s got a cannula in and he’s . . .’, you need someone to explain to your children, as well, . . . this is fine, . . . this is helping . . . it affects everybody, definitely, my mum and my dad [. . .]
Family 1 m
Information/language used
Parents would try to be in the unit for the doctors’ rounds and would ask nurses afterwards if they did not understand what the doctor had said. They felt that the doctors were good at explaining the information in lay terms. Parents seemed to learn the medical terminology very quickly so that they could follow what was going on.
Mostly, people would appreciate the practical information, in stages, so that they could be involved and know what to do, only having the information they needed for that day.
It took a bit of getting used to the language:
You hear the worst things – . . . ‘brain bleeds’, but then they tell you it’s nothing to be worried about, it’s ‘like a bruise’.
Family 9 m
Parents liked to be given information directly/bluntly, without health-care staff ‘beating round the bush’; they reported always wanting to know the situation, and not to be told that everything is fine. They felt that the neonatal unit was better than the maternity ward at providing information:
[. . .] I think they were pussyfooting around it a little bit too much, maybe . . . they were sort of indirectly telling us what was gonna happen or what would happen if we didn’t do this or didn’t do that, and I think if they’d just said ‘Look, if she stops growing there’ll be serious problems, so we’ve got to get her out, basically [. . .]
Family 3 d
[. . .] Doctor [name removed] was very good and talked to my husband . . . until his questions finished and he had resolutions or some kind of answers . . . the fact that he had been heard was really, really important. You know, you can’t always give an answer or solve the problem but at least he’d been heard [. . .]
Family 5 m
Understanding risks
Parents found discussions about risk difficult to process, particularly because they were immersed in such an emotionally charged experience, and they liked to focus on what they could do.
The possibility of their baby not surviving was always on their minds. For most parents, there had been some sort of discussion between them and the medical staff about the risks. Some people wanted more information than others. Mostly, parents wanted to know what might happen, but not necessarily the percentages (or likelihood).
In some cases, there were discussions about the potential risks of certain treatments (e.g. loss of use of limb when arterial line was put in, but it was to give their baby a better chance so there was no choice). Risks were also discussed in other cases in which the infant was transported by plane or ambulance. When there was a treatment or procedure with more than one possibility, parents tended to defer to the medical expertise.
Parents felt that they were given hope, but that it was realistic. Some felt that discussion was too broad and not specific enough to their case.
They used expressions like ‘if things went the other way’ and ‘still touch and go’ when referring to the condition of their baby:
[. . .] one of the neonatal doctors came to see me when I was in the labour ward . . . told me that he might not survive. They’ve got to tell you the pros and cons haven’t they? I’m just telling you [name removed], he might not survive, he’s very, very early [. . .]
Family 10 m
[. . .] we knew the risks and we know we have a long battle . . . But it was always done in a nice way that there was always hope, which was nice [. . .]
Family 2 m
[. . .] her chances of survival, we didn’t want to know that [. . .]
Family 8 m
[. . .] if something was gonna happen now which would affect in those few weeks, then just tell us straight . . . say – if we don’t do this now, in 3 or 4 weeks this could happen. Brilliant, that’s . . . direct [. . .]
Family 3 d
Mitigating aspects
We looked at the factors that parents felt had helped them. These are summarised in the following list:
Confidence in the hospital staff. Parents felt well supported by the hospital staff. In some cases, they had a care team and always knew that they could talk to any member of the team.
Being able to stay at the neonatal unit.
Being hands-on with their baby. They were able to and were encouraged to be involved in all aspects of their baby’s care. When parents were travelling to the neonatal unit each day, the unit staff would mostly try to hold back the feeding/washing times and doctors’ rounds until they got there.
Thoughtful practices, such as when a mother had to be transferred to the NICU after her baby and staff had prepared a room and taken photos of her baby, so that she arrived to a lovely welcome.
Playrooms for other children, so that they could come to the unit too and not be left out of the picture.
Families rallying together, helping with childcare and providing emotional support.
Supportive employers. Some fathers’ employers had given them extra leave, and all had been understanding.
SNUG and BLISS – parents were aware of them during their stay and had accessed information leaflets; however, most parents had contacted them once they had gone home and were hit (‘hit me like a tonne of bricks’) by both the trauma of what had just happened and the vulnerability of being at home without all the monitors and medical staff. Parents reported continuing their link with SNUG for years afterwards.
Financial support from charities and children’s centres.
Friends made in the neonatal unit.
[. . .] The nurses are like your counsellors there, you know, they’re the people that are just always there and listening, aren’t they? [. . .]
Family 8 m
[. . .] I sent a message to the SNUG people, I said ‘Yeah, I actually need someone and [SNUG representative] called me immediately, it was really nice and she told me to just talk through what happened – your journey and I was telling her the whole journey [. . .]
Family 9 m
[. . .] that was my precious bit, that’s what I could do for my baby, is get her dressed and try and feed her [. . .]
Family 7 m
Summary
What was most important to the parents is that their baby had the best health outcome possible, and they were willing to do whatever they could do to help make that happen. The sacrifices made by parents can be disruptive to the family both emotionally and financially.
There was variation in the parents’ preferences regarding staying in the neonatal unit or travelling to the unit each day. This did not depend on whether or not the baby was their first child. Some who had other children chose to stay in the unit if they had family support at home, others travelled to the unit each day to fit around childcare. Some parents for whom this was their first child chose to travel to the unit each day.
Patients also differed in their preferences towards feeding their babies. In one case, the parents did not want to feed their baby through a tube but instead wanted to wait until they could feed their baby with a bottle or breastfeed; other parents preferred to be involved in all aspects of the care.
In order to help understand how the results of the qualitative research can inform the feasibility of and attributes for a DCE, we have broken down the overall care picture into three components:
best care for the baby – this refers to/includes the medical care team, medical facilities and the health of the mother, including emotional support from family and friends
communication – including parents knowing what is happening to their baby, understanding what might happen (risks), what parents can do ‘now’, how they can prepare for the future (short and long term)
family involvement – including the care of the baby (washing, dressing and feeding), facilities for parents to stay, facilities for other children and preparing to take their baby home.
We suggest that parents would be unlikely to consider any attributes that compromise the first component; however, parents’ preferences and circumstances vary enough in components 2 and 3 that we could perhaps develop attributes around communication and family involvement.
Interviews with policy-makers
We conducted a series of 10 interviews with policy-makers, clinicians and staff to assess their views on outcomes of neonatal care, how they might prioritise outcomes and which types of economic measures might be useful for planning. This involved interviewing members of the Neonatal Critical Care Clinical Reference Group as well as specialist commissioning groups (NHS England) and representatives of the Maternity and Children’s Services Strategic Networks. In addition, the interviews were used to check the approach to costing neonatal services, and included questions on the factors that made up the reference cost submissions, how centralisation affects total costs and the facilities that should be provided for parents to mitigate some of the negative consequences of centralisation.
A recurring theme in the policy interviews was the importance of the well-being and health of the baby. Secondary issues were the availability of staff to cover the rotas and the difficulty of recruiting and retaining staff; for example, it became clear that in some areas, even though it may be optimal for parents to relocate to a neonatal unit, it was felt to be infeasible to staff such new centres.
Ten policy interviews were carried out by Paolo Landa and Anne Spencer. For those working within the neonatal units, we used the interviews to explore service-level issues that we may need to take into consideration when using financial and BAPM returns, such as staff composition and duties and exploring how infants were prioritised within the system when there were staff shortages. For those who were responsible for those co-ordinating and managing network services, we explored their role and the challenges arising from service reorganisation. For those working to support families with neonatal infants, we explored how their work helped to shape the neonatal service and the factors that they felt needed to be taken into consideration in service reorganisations. The second part of all of the interviews explored what policy-makers would like to take into consideration when determining service reorganisations.
A topic guide was developed and adapted as the interviews progressed (see Report Supplementary Material 2). All interviews were recorded and notes were then taken. The interviews involved representatives of a charity that works in neonatal care, two neonatologists that work in the organisation of a local neonatal network, two consultants of NICUs, four matrons, and a neonatologist who works both for a local neonatal network and as a consultant lead for a NICU.
Neonatal charity
The neonatal charity represented the families in the neonatal care process and the parents’ needs and preferences in terms of resources and organisation. The main objective of the neonatal charity was to improve health outcomes for neonates through their research, and they engaged with decision-making and the NHS to influence and improve neonatal care. The charity also provided a support service to families by providing leaflets and other sources of information (e.g. videos and multimedia) and offering helplines for psychological support. Parents needed to talk to someone who was able to listen and provide reassurance and to work as a guide in the complex neonatal care setting. In addition, the charity offered training to neonatal staff to encourage good practice.
If neonatal services were centralised, the charity noted the importance of travel distance and the need to provide facilities to reduce the costs and improve the accessibility of services for families, such as by providing accommodation, travel reimbursement, meals and free car parking. They also felt that several resources were still missing for families and were not provided by the NHS, and that these gaps increased when home care, after the infant is discharged from the neonatal unit, is taken into consideration.
Neonatal network neonatologists
The interviews with neonatal network staff were focused on the network organisation and the impact of different policies. The first problem that was reported regarding the neonatal service reconfiguration was the challenge of changing the skill mix of staff. Converting a SCU to a LNU or a LNU to a NICU can cause several difficulties in terms of attracting new staff and developing the supervisory structures, and these could take > 10 years to overcome in terms of training, education and organisation. The neonatologists also acknowledged that it was hard to model the staff composition in the unit, as this depended on several factors including local availability of a trained workforce, the size of the unit and the number of cots.
We asked the neonatologists what type of service organisation they envisaged (e.g. a NICU for each local neonatal network, which seemed to be the main working model). The network neonatologists acknowledged that centralisation would improve mortality for the VLBW infants, but that infants should be located near to home if appropriate services were available. They also acknowledged the regional differences in access to services (e.g. in South West England: in Cornwall and Devon, there is only a NICU but accessibility in terms of travel time is very different from the London area network). As a result, they were aware of the need to support those families that travelled long distances by providing overnight accommodation while their infant was in a neonatal unit. Some of these costs are covered by the NHS and charities, but it was felt that the provision was often not enough.
Neonatal unit neonatologists and matrons
Four hospitals were interviewed in order to understand the organisation in the unit, the availability of resources and the different unit configurations in terms of service delivery. In our interviews, we considered three English NICUs and one English LNU.
We started by asking questions about the organisation of the neonatal services in each area. Most of the units were inside a hospital with a large service availability in terms of diagnostic examinations, laboratory tests and consultation of specialists. One of the units represented in interviews was a women’s hospital in which the availability of some resources, such as psychological support to mothers and tests, were provided by another site, and the unit contracted separate service agreements for these ancillary services. The NICU interviews raised the issue that it was sometimes difficult to discharge back to a local hospital (LNU or SCU), creating some bottlenecks and delays of neonate transfer.
We then asked questions regarding staff and staff composition and how staff members were organised when there were staff shortages. Each unit had a different composition; for example, in one unit there was a large number of advanced neonatal nurse practitioners (ANNPs), whereas in the other units there was a high availability of junior doctors. This variability in the staff configuration can create large differences in terms of staff costs. We asked if some types of infant (e.g. infants in intensive care and infants in high-dependency care) were prioritised when the workload exceeded the BAPM standard and the staff availability could not cover the number of infants in accordance with the BAPM guidelines.2 Staff from each unit advised that there was no prioritisation of infants and that each infant is treated with the same level of priority, with staff being allocated in accordance with the BAPM guidelines2 where possible.
When we discussed the resources needed if the services were to be centralised further, and if NICUs became larger, all interviewees raised the need to increase staffing. In some areas, this may be a challenge because of the limited availability of training staff. Most units were already not working in accordance with BAPM standards for ≥ 80% of the time, and so there were concerns about staff shortage and the ability to attract staff to new sites if the unit had to be relocated to another part of the network. In terms of facilities offered to families, unit representatives noted the need to expand the provision of accommodation for families and services if services were further centralised. They also talked about the need to make efficient use of resources.
Discussion of qualitative study
The results of the qualitative interviews show that interviewees talked more about the infant as a whole, rather than separating the risks of death and childhood health problems. This notion of a combined attribute is similar to the idea of an aggregate measure of length of quality of life used by NICE,133 which may be more meaningful if extrapolated forward to consider the longer-term impacts on infants, and not just short-term prognoses. In addition, the families made a connection between the baby’s and the mother’s health, and the importance of the mother’s health was considered to be equal to that of the child by the focus group.
The qualitative interviews also raised other process outcomes including communication with the families and family support.
Furthermore, the qualitative study raised questions about the ability and willingness of parents to trade off health attributes for the process attributes. Although no trade-off questions were asked in the patient interviews, it became clear that mothers were unlikely to want to sacrifice these ‘core’ aspects of their baby’s health for improvements in process outcomes. Parents stated in interviews that once they knew that their baby was receiving the best care, they could then perhaps consider other factors. This raised questions about including a DCE with combined health and non-health outcomes, because parents would always choose the configuration that favoured the best health outcomes for the infant (lexicographic preferences). However, in configurations that maintain a high level of care but affect process outcomes, trade-offs and DCEs become more feasible.
