An economic analysis of human milk supplementation for very low birth weight babies in the USA

An economic model was undertaken to explore the costs and benefits of using an EHMD in VLBW babies, compared to usual practice of care in which cow’s milk based products are used. The model was based on a variant of economic evaluation called cost-consequence analysis, in which estimates of cost-effectiveness are supplemented by further information on wider costs and benefits so that decision-makers may form their own judgements on which feeding strategy offers greatest value in terms of costs and benefits. The main analysis, or ‘base case’, considers the clinical effects of an EHMD, as well as the costs to the health care payer of the diet (including the costs of the initial hospital stay and follow up from treatment of retinopathy of prematurity and cerebral palsy), and the subsequent clinical effects. Sensitivity analyses explore the additional costs to society of cerebral palsy and retinopathy of prematurity. Future costs and benefits are discounted to their present values in 2016 using an annual discount rate of 3%, as recommended by the US Second Panel on Cost-Effectiveness in Health and Medicine [11].

The model uses a hypothetical population of 1000 VLBW babies, all of whom are assumed to be admitted to a NICU. They are assigned to either an EHMD or to usual practice of care. The babies may then develop NEC, late onset sepsis, both of these, or neither, with the probability of each event dependent on the diet that they have received. Babies who develop NEC can be treated medically or surgically.

All babies also have a probability of developing the following sequelae: BPD, a chronic lung disease; retinopathy of prematurity, which causes abnormal blood vessels to grow in the retina and can lead to blindness; and cerebral palsy, a neurodevelopmental problem. Babies who receive surgical NEC treatment can also develop short bowel syndrome, a condition which can lead to long term or lifelong parenteral feeding. The probability of developing these adverse outcomes is assumed to differ according to whether the baby received usual practice of care or an EHMD.

Figure 1 shows the structure of the model. The model takes the form of a decision tree and uses a cohort approach. It was developed in Microsoft Excel©.

Babies in the model receive one of the following diets:

All other aspects of care are assumed to be the same between groups.

The model is populated using the results of a literature search undertaken to identify parameters for the model. See Table 1 for full details.

The literature review was based on bidirectional citation searching [18] starting from two key papers in this area [5, 12] that were known to the authors, and supplemented by searches in Google Scholar. We looked for RCTs or observational studies that compared clinical outcomes following an EHMD compared to standard of care. We also considered studies that provided evidence of clinical outcomes resulting from NEC or sepsis, as these are key clinical outcomes for EHMD.

The two key studies were:

A third RCT [13] has also included the retinopathy of prematurity outcome, which was not included in these two RCTs. No additional RCTs were found via the literature search.

Best practice in economic modelling is to use clinical practice data to estimate clinical outcomes in the comparator group - in this case, those receiving usual practice of care - and to apply relative treatment effects from RCTs to estimate outcomes in the intervention group – in this case, those receiving an EHMD. The retrospective cohort study [12] provides the best clinical practice data available to us to represent infants receiving usual practice of care as defined above. Whilst other data is available [14, 15] on the incidence of NEC in this population, it does not differentiate between babies receiving bovine-based fortifier and an EHMD. The preventative effect of an EHMD on NEC, mortality and RoP incidence was stronger in the RCTs than in the retrospective study, whereas the relative effect of EHMD on BPD and late onset sepsis was stronger in the retrospective study, although note that the RCTs report sepsis rather than late onset sepsis. We use the RCT data on relative effects in the base case, and utilise alternative values in a sensitivity analysis. There is one exception to this, where for mortality we use estimates from the retrospective cohort study [12] to estimate mortality in the usual practice of care and EHMD groups in the base case. This is because, for this outcome, combining the data sources resulted in what appeared to be an unrealistically high number of lives saved by the EHMD. These results are presented separately as a sensitivity analysis.

The three key papers did not provide data on the probability of developing cerebral palsy or short bowel syndrome, or the impact on survivors’ subsequent IQ. The probability of developing these outcomes were thus based on the best available evidence identified through the literature search.

There is currently insufficient evidence about the direct impact of an EHMD on the probability of developing cerebral palsy. In our model, this outcome therefore depends only on whether the baby developed NEC, late onset sepsis or both during the initial hospital stay. The data are obtained from a meta-analysis of 4377 VLBW babies with and without NEC and sepsis across five cohort and case-control studies [15].

To calculate the incidence of short bowel syndrome following surgical NEC, we used data from Cole et al. (2008) [19], who conducted an analysis of 12,316 VLBW babies. They found that 0.7% of the cohort had short bowel syndrome, 96% of which was caused by NEC. We assume that all these NEC-related cases of short bowel syndrome are due to surgical NEC, and calculated the incidence rate of short bowel syndrome amongst babies with surgical NEC accordingly. Some of these patients would require lifelong Total Parenteral Nutrition or an intestinal transplant, but we were unable to include these long term effects in the model because of a lack of robust data on their incidence.

The reduction in IQ points resulting from NEC and late onset sepsis was taken from Roze et al. [16] and Van der Ree et al. [17] respectively. Babies who had both NEC and late onset sepsis were assumed to experience the loss of IQ associated with NEC only, as this was the higher of the two estimates. We do not assume any additive effect.

In each case, the assumptions are conservative about the impact of an EHMD, in that they are likely to underestimate its beneficial effects, although the number of patients likely to be affected will be small.

The model includes the costs of the diets, plus the costs of treating any complications that arise during the initial hospital stay. We also include costs for follow up from treatment of retinopathy of prematurity, and lifetime costs to the health care payer of cerebral palsy. Using this approach, the long-term health care costs following the initial hospital stay and resolution of NEC, late onset sepsis and sequelae for all patients in the model who do not have cerebral palsy or retinopathy or prematurity is assumed to be the same for all babies. Table 2 shows the key cost parameters used in the model. As before the costs were identified via bidirectional citation searching and Google scholar, starting from two key papers that were known to the authors [6, 7], with the aim of identifying the most recent and applicable cost estimates. The cost of the EHMD was provided by Prolacta Bioscience. All costs were expressed in 2016 USD prices, updated using the medical care component of the Consumer Price Index [24].

Different strengths of human milk fortifier are available. The exact product used is likely to differ according to local protocols, and in many cases will change throughout the infant’s stay. We made the simplifying assumption that the Prolact+ 6® is used; this product is mid-range in terms of strength and cost and is the most commonly used product [Prolacta, personal communication]. The + 6 is 30 mL and is mixed with 70 mL of donor milk. The cost of 30 mL Prolact+ 6 product in the US is $187.50, and the cost of donor milk was assumed to be $133 per litre in 2008$ (updated here to $183 in 2016$), based on a retrospective evaluation of the use of donor milk for feeding very preterm infants in the NICU [20]. The estimated total cost of the EHMD is therefore $7731.

The cost of cow’s milk based products as reported in Ganapathy et al. [21] ($195 in 2011$, updated here to $226 in 2016$) was subtracted from the cost of the EHMD to give an estimate of the incremental cost of the EHMD.

The incremental costs of NEC, late onset sepsis, BPD and retinopathy of prematurity were taken from economic analyses of treatments of these conditions in the US [6, 21, 22]. The cost of short bowel syndrome was included in the cost of surgically treated NEC. These costs were added to the cost of an NICU stay for a VLBW baby with no complications, provided by a retrospective analysis of the costs of comorbidities amongst 425 VLBW babies [6]. Costs were obtained after controlling for birth weight, gestational age, sociodemographic characteristics, and various clinical outcomes.

The discounted lifetime costs of cerebral palsy were calculated from data provided by the Centres for Disease Control and Prevention [23] to the health system.

Note that all costs are incremental to the cost of a baby that does not develop NEC, late onset sepsis or sequelae. The costs are considered to be additive in all cases: data from Johnson et al. [6] suggests that this is a reasonable simplifying assumption as treatment costs significantly increase with the number of morbidities.

Sensitivity analyses are undertaken to explore the impact on the results if alternative input values are used in the model, for example different baseline estimates of NEC prevalence. We conducted four types of sensitivity analysis:

Tables 3 and 4 show the results for the base case analysis. An EHMD reduces occurrence of most adverse clinical outcomes during the initial hospital stay, including preventing 36 deaths in the hypothetical 1000 infant cohort, but not BPD. The increased incidence of BPD is because more babies survive with a EHMD. This reduction in mortality outweighs the reduction in risk of getting BPD.

In addition to clinical improvements, the EHMD generates overall lower health care costs, because the lower costs due to the reduction in adverse clinical outcomes more than offset the increase in dietary costs. This means that the EHMD is dominant in cost-effectiveness terms in a US setting.

Holding other factors constant, an EHMD would still reduce costs if the baseline incidence of NEC in the usual practice of care group was as low as 7%. This is shown diagrammatically in Fig. 2. Below a baseline incidence of 7% the EHMD leads to small increases in cost per patient ($2010 at 5%; $6707 at 2%) but still gives significant clinical benefits, for example at 5% 34 cases of NEC would be avoided, at 2%, 14 cases. Ignoring all other clinical benefits, this means $59 per case prevented at 5% incidence and $479 at 2%. The EHMD would still be cost saving if the baseline incidence of late onset sepsis was zero. Any increases in the incidence of NEC or late onset sepsis increase the cost savings associated with EHMD.

When wider societal costs are included the cost savings from an EHMD are even greater at $117,239 per infant. In the case lower cost scenario, the cost savings per infant increase to $22,226, and in the higher cost scenario they reduce to $10,416 (Fig. 3). Finally, when the retrospective cohort data and trial data are combined to estimate mortality for the EHDM group, an additional 131 lives are saved per 1000 babies in the EHMD group, leading to a decrease in cost savings ($12,164 per infant).