Workstream 3: comparing the costs and effects of care

In NICUs and LNUs for very preterm babies in England: a retrospective analysis of a national birth cohort.

Introduction

Our OPTI-PREM programme of work employed a mixed-methods design to identify the optimal setting in which to provide care at birth for very preterm babies born at 27–31 weeks gestation. In Chapter 4 we used the NNRD, to evaluate whether being born into maternity services colocated with NICU compared with LNU and admitted to their neonatal unit for care affected mortality (while in neonatal care and in infancy) and key secondary neonatal morbidity outcomes. In Chapter 6 we estimated total costs of care for very preterm babies born at 27–31 weeks gestation and admitted to all but one neonatal unit in England. We identified that total costs were ~0.25 billion GBP per year, that costs were driven by daily costs of care and that costs of care were inversely proportional to gestational age at birth. Using the same data sources, we then conducted an allied assessment of the cost-effectiveness of care in each type of unit from an NHS perspective and using additional number of lives saved as measure of benefit in the economic evaluation. We report our findings in this section.

Aim and objective

To estimate the cost-effectiveness of care of very preterm babies born at 27–31 weeks gestation and admitted to NICU compared with LNU, in England, up to discharge from neonatal care.

Methods

We utilised individual patient-level data on admissions to all but one of the neonatal units in England, for babies born at 27–31 weeks gestation and who were discharged or died between 1 January 2014 and 31 December 2018. In addition, the cohort also included babies who were inadvertently not born in a hospital with a LNU or NICU, but who were later transferred to one of these for continued care. Admissions were identifiable through the NNRD, as described in Chapter 5. Data for the analysis were extracted from the NNRD on 5 October 2020.

Resource use and costs

We conducted the analysis from the perspective of the NHS in England. The time horizon was up to the end of neonatal unit care. To calculate the costs associated with the routine daily neonatal care received by each baby, we multiplied the number of days spent receiving each level of care by level-specific national average bed-day costs sourced from the 2018–9 National Schedule of NHS Costs.28 We used HRG codes, and their unit costs as described in Chapter 6. We included five high-cost non-routine procedures as described in Chapter 6, that is nitric oxide, surfactant replacement, total parental nutrition (TPN) (over 14 days of use), palivizumab use and surgical care. Similarly, we identified types of neonatal surgery from their corresponding OPCS and ICD-10 codes for costing purposes, and costs for other activities were obtained from various sources as detailed in Appendix 5, Table 34. Again, the NNRD data for these five non-routine procedures were assumed to be complete, with no entry in the corresponding data set fields taken as an indication that the procedure did not take place.

Survival

Given the retrospective nature of the analysis, we used survival status at the end of neonatal care as the measure of outcome for the cost-effectiveness analysis, with incremental effectiveness expressed as the number of lives saved.

Statistical analysis

As per the statistical analysis methods used for the OPTI-PREM clinical analyses and described in Chapter 4, comparisons of costs and effects between unit types were made after adjusting for measured confounders, and also unmeasured confounders using an IV approach. We utilised the same measured confounders (see Chapter 4) and the same IV – excess travel time to a NICU. Our IV model was estimated using two-stage linear least squares for the cost (continuous) variables and seemingly unrelated bivariate probit regression for the mortality (binary) variable. The choice of the bivariate probit instead of the standard probit model was the binary nature of ‘unit type’ (NICU or LNU). We also conducted a sensitivity analysis and ran costs and mortality models without any measured confounders to understand the impact of this adjustment on the overall results.

For information and the benefit of the reader, we present unadjusted crude means and mean differences between ‘unit types’, and an analysis adjusted for measured confounders only.

Results

Figure 7 describes the flow of study participants. Records were initially extracted for 29,842 babies born at 27–31 weeks gestation and discharged from neonatal unit care between 1 January 2014 and 31 December 2018. Following exclusion of ineligible cases (n = 3557) and those with missing data (n = 7673), the final analysis was performed using data from 18,612 babies.

FIGURE 7 Flow of NNRD babies included in the health economic analysis. a, Missing neonatal and maternal characteristics were not mutually exclusive and it was possible to have missing information in more than one characteristic. (Figure not reproduced here; see the original article.)

Table 23 presents maternal and baby baseline characteristics of the final complete-case sample and those excluded due to missing data. In general, we observed similar distribution between both samples for the majority of characteristics. However, the complete-case analysis sample has slightly more multiple births and caesarean sections, and comparatively a smaller proportion of white women. Table 24 shows baseline characteristics by neonatal unit type. These data show babies managed within NICU were born at earlier gestations and had a lower birthweight than those born within LNU. Mothers of babies cared for in NICU were also less likely to be of white ethnicity and to have higher levels of deprivation.

Figure 8 details the distribution of excess travel time to NICU of the IV used within the analysis to adjust for unmeasured confounders. The median excess travel time was estimated to be 3.9 minutes suggesting on average women lived slightly closer to a LNU than to a NICU in our sample. Table 25 shows the baseline characteristics of the cohort when recategorised according to the median IV < 3.9 or IV > 3.9. The groups were well-balanced across most characteristics with an absolute standardised difference < 0.15, with the exception of the IMD, which showed a higher proportion of the most deprived mothers living in closer proximity to a NICU than to a LNU.

FIGURE 8 Cohort distribution of the IV used in the analysis – excess travel time to a NICU as compared with a LNU. (Figure not reproduced here; see the original article.)

Costs

Appendix 7, Table 36 presents an unadjusted analysis of the mean number of days for which babies cared for in NICU and LNU received different intensities of care. As expected, significantly longer durations of intensive care, high-dependency care and special care were observed with NICU. However, after adjusting for measured confounders (see Appendix 8, Table 37), differences in intensive and high-dependency care were no longer present. Table 26 presents the final analysis in which resource use and costs were adjusted for both measured confounders and unmeasured confounders using an IV approach. For each care setting, mean (SE) resource use and costs are reported along with mean cost differences (99% CI). On average, and compared to babies cared for in LNU, those managed in NICU spent fewer days receiving intensive care and high-dependency care, and more days receiving special care. Costing these daily levels of care resulted in a mean per baby cost of NICU-based care of £44,177 and LNU-based care of £46,179 [mean cost difference −£2003 (99% CI −£3432 to −£573)]. This difference in the levels of care provided accounted for 80% (£2003/£2534) of the lower overall total neonatal cost with NICU as compared to LNU, shown in the final row of Table 26. Lower costs of hospital transfers and other major non-routine procedures in the NICU setting accounted for the remaining 20%.

Table 27 shows the final analysis adjusted mean total cost of neonatal care for each setting by week of gestation at birth. Costs were significantly lower in the NICU setting for babies born at 30 and 31 weeks gestation. Appendix 9, Table 38 shows the associated mean number of days provided at each level of intensity and week of gestation at birth and reveals that the significant differences in overall costs are driven by differences in intensive care days at 30 weeks and intensive care, high-dependency and special care days at 31 weeks.

Effects

Table 28 shows the proportions of deaths at neonatal discharge in each care setting, for the overall cohort of 18,612 and by weeks of gestation at birth. Across all babies, there was no difference in death proportions between care settings. Similarly, no differences were observed when analysing mortality data by gestational age at birth.

Cost-effectiveness

Figure 9 plots, for the overall cohort, the difference in mean costs and difference in mean effects between NICU and LNU care settings on the cost-effectiveness plane. Difference in mean effects were expressed as additional number of babies surviving to facilitate interpretation. The point plotted is in the lower half of the plane, reflecting the cost saving in the NICU setting, but it is also close to the origin of the x-axis, illustrating the absence of any difference in infant mortality. Also shown are the 95% CIs around the differences and a 95% confidence ellipse illustrating the likely uncertainty of the 95% of the joint distribution of cost and effect differences. The latter suggested that uncertainty on the joint distribution of costs and effects lies in the south-east and south-west of the cost-effectiveness plane. Almost 100% of the joint density of costs and effects wase below willingness to pay for health gain between £50,000 and £100,000.

FIGURE 9 Mean cost and effect differences and associated uncertainty between NICU and LNU care settings plotted on the cost-effectiveness plane. (Figure not reproduced here; see the original article.)

Sensitivity analysis

Appendix 10, Table 39 shows the results of the overall and gestation-specific mortality and total costs using an IV model without any adjustment to measured confounders. The results suggested that very little or negligible adjustment was needed using measured confounders in the IV model.

Discussion

Utilising high-quality data from a national database and an IV modelling approach, this study has compared the costs and mortality to discharge of babies born 27–31 weeks gestation in NICU and LNU settings. To the best of our knowledge, no previous studies have assessed the cost-effectiveness of these care settings for preterm babies born at this or indeed any other gestational age.

Findings in context

The results generated by this work require careful interpretation and placing within the context of the organisation of neonatal unit services in England as well as the nature of preterm birth per se. If solely following the conventions of cost-effectiveness analysis, these data appear to advocate for NICU units, suggesting that the setting of post-delivery neonatal care does not impact infant survival but that when care is provided in a NICU, and particularly so for babies born at 30 and 31 weeks gestation, significant cost savings can be realised. In practice however, it would be challenging to try to redirect the post-delivery care of many of these babies from LNU to NICU. The geographical dispersion of neonatal care units within England provides one reason for this. NICU tend to be housed within large, tertiary, city-based hospitals and a great many pregnant women do not live in close proximity to these hospitals. Another and related reason is that unlike a planned elective procedure for which a date and location can be arranged in advance, a preterm birth is an unexpected and frightening event for a woman, who will therefore most likely present to her local hospital regardless of the type of associated neonatal unit. Finally, shifting the care of large numbers of ‘less sick’ preterm babies to cots within NICU units solely for the purposes of saving costs would be considered unethical and detrimental to those preterm babies born at earlier gestations, for whom care within a NICU is essential for their survival.

Recognising both practical and ethical constraints to the reorganisation of neonatal care for babies born at 27–31 weeks gestation, it would perhaps be prudent to use the findings from this work, which have been able to detail where such cost differences arise, to hypothesise why they might arise. Table 26 showed that 80% of the mean total cost difference observed between NICU and LNU was attributable to variations in the intensity of care provided to babies daily. Babies cared for in NICU spent just over 1 day longer receiving special care, while those in LNU received around 1 day more of intensive care and high-dependency care. A range of factors with the potential to vary by type of neonatal unit could explain these findings. For example, and as noted above, NICUs are usually located within large tertiary hospitals that often have specialist obstetric and feto-maternal medicine services, and where staff are experienced in managing preterm babies with a range of complex conditions. Highly skilled and specialised clinicians at these sites may be adept at recognising and making timely amendments to the intensity of care required by less sick babies in the unit. Further, in these larger hospitals with greater capacity, staff may more easily be able to facilitate such changes. In smaller LNU with fewer beds, a timely step down of care may prove challenging, with resulting delays, if bed occupancy in the unit is high or at capacity.

The gestation-specific cost estimates shown in Table 27 would support these hypotheses. While costs for babies born at the earliest gestations (27, 28 and 29 weeks), and thus requiring the highest levels of care, were not significantly different between NICU and LNU settings, costs for babies born at 30 and 31 weeks gestation, and who may be considered ‘less sick’, were significantly lower with NICU care. Additionally, within LNU, the duration of usage of TPN was longer, and the total amount of surfactant replacement was higher than in NICU. Further mixed-methods research should seek to fully understand the reasons for the observed differences in care practices between NICU and LNU along with whether any of these factors are modifiable such that the management of babies in both types of units could be more closely aligned and cost savings safely realised.

The discussion generated thus far is based upon the observation that there would appear to be no difference in baby mortality between each setting of care. Alongside mortality, however, it is also important to consider the morbidity of these babies. Key morbidities were assessed as secondary outcomes for the OPTI-PREM clinical effectiveness analysis and included ROP at Stage 3+, oxygen dependency, NEC, SBI and breast milk at discharge (see Chapter 4). Small yet statistical differences in the proportions of babies with oxygen dependency (1.8% more in NICU) and with SBI (1.1% more in LNU) were observed, with these differences driven by the 11% of babies in the cohort born at 27 weeks gestation. When analysing data only for the 89% of babies born at later gestations, no differences between care settings for any of the morbidities studied were detected.

Placing this research within the context of existing work in the field is challenging, because as noted previously, we are not aware of any other studies that have formally compared the costs and effects of NICU and LNU care for preterm babies. It is possible however to compare certain aspects of this work, and in particular the costing components, with other studies in this field.

Limitations

This study has several limitations, most of which stem from the retrospective cohort approach used for the analysis. The obvious drawback is the lack of a randomised comparison and the risk of bias this exposes the analyses to. We attempted to minimise these risks as much as possible by using appropriate statistical analysis techniques to account for measured as well as unmeasured confounders. Using IV techniques, we were able to generate study groups, which aside from the setting of neonatal care, were well-balanced in terms of mother and baby characteristics (see Table 1).

The retrospective approach taken also limited the study outcomes to those available within the NNRD. This analysis used survival status at discharge from the neonatal unit as its measure of effectiveness, which led to further challenges around how to interpret and explore the uncertainty seen in the results. Cost-effectiveness analyses typically measure the effect of an intervention in terms of the number of life-years or quality-adjusted life-years (QALYs) gained. Established threshold or ceiling values reflecting the maximum amount that society is willing to pay for one of these units of outcome (e.g. one QALY) are frequently used by analysts to interpret their results and to construct cost-effectiveness acceptability curves for the purposes of quantifying the level of confidence in their findings.119,120 As the unit of effect in this case is lives saved, which are not commensurate with life-years saved, such analyses could not be conducted here.

The time horizon used for the study was also dictated by the availability of data and may be considered too short given that the health implications of preterm birth are known to extend through early and late childhood and into adulthood.121,122 If differences exist in the levels of longer-term morbidity experienced by babies treated in each setting, then the implications of such differences ought to be captured and included in the analysis. Although it was not possible to observe these data, the OPTI-PREM secondary analysis of key morbidities showed no real differences between settings and thus provides some reassurance that this is unlikely to be the case.

Around 38% (n = 11,230) of the original 29,842 babies available for the analysis were excluded because they were not eligible or because of missing data due to neonatal or maternal characteristics, missing postcode information to estimate excess travel or missing cost information. Most of the missing data were due to missing mode of delivery (4%), maternal ethnicity (13%) and IMD (8%). Similar rates of missingness have been observed in previous studies using NNRD data for the period of time utilised in OPTI-PREM.123,124 Some babies in our study had one or more of these characteristics missing that were needed in the modelling to adjust for measured confounders resulting in a significant drop in the final sample size included in the analysis. We explored whether the baseline characteristics of the complete-case sample were somewhat different from the sample with missing data. Our descriptive analysis reported in Table 1 suggested that only minor differences were observed in the distribution of multiple births, caesarean sections and maternal ethnicity. This was further explored in our sensitivity analysis where we removed measured confounders from the IV model and derived a sample excluding only missing data due to missing postcodes or cost data. This resulted in a data set with a sample of 25,352 babies for the overall and age-specific mortality and a sample of 25,046 babies for the total costs. The results of this analysis were virtually the same as the complete-case analysis suggesting the little or negligible impact of adding measured confounders to the model. To some extent, this was expected as we showed in Table 25 that groups were well-balanced in terms of neonatal and maternal characteristics when using our IV. Our conclusion to this evaluation is that the sample excluded due to missing data was not significantly different from the complete-case sample and that missing data impacted very little our overall results.

Conclusions

To the best of our knowledge, this is the first study to compare the costs and outcomes of babies born at 27–31 weeks gestation in either NICU or LNU and using high-quality data from a national cohort. Results from OPTI-PREM suggested no difference between the settings in terms of baby outcomes at neonatal unit discharge but identified that those babies born at later gestations (30 and 31 weeks) in NICU had shorter durations of intensive and high-dependency care than their counterparts born in LNU. Factors specifically related to the organisation of neonatal care including the location, size and staff present in these units may explain these findings.

Practical and ethical concerns surround the redirecting of neonatal care for very preterm babies from LNU to NICU purely on the grounds of cost savings. Further mixed-methods research should seek to understand the reasons for the observed differences in care practices between NICU and LNU along with whether any of these factors are modifiable such that the management of babies in both types of units could be more closely aligned and cost savings safely realised.

Summary points from OPTI-PREM costs and effects of care study, workstream 3

What was already known prior to the study

• We are not aware of any previous studies comparing the costs and effects of care provided to very preterm babies born in NICUs and LNUs.
• Such work is required to inform policy-makers on the most appropriate place of care for very preterm babies born at 27–31 weeks gestation.

What this study adds

• The costs of care for babies born at 27–29 weeks gestation were not significantly different between NICU and LNU.
• Costs were significantly lower for babies born into NICU settings at 30 and 31 weeks. These were driven by differences in the durations of different levels of care provided (intensive care and special care days).

How this study may affect policy

• Research is needed to understand the reasons for the differences in the durations of intensive care between settings and how care processes can be more aligned.