Cost-Effectiveness of Supplemental Donor Milk Versus Formula for Very Low Birth Weight Infants

The Donor Milk for Improved Neurodevelopmental Outcomes (ISRCTN35317141) study was a double-blinded RCT in which 363 VLBW infants were enrolled from tertiary NICUs in southern Ontario between October 2010 and December 2012. The aim of the RCT was to determine if nutrient-enriched DHM compared with PTF as a supplement to mother’s milk reduced neonatal morbidity, supported growth, and improved neurodevelopment in VLBW infants. Infants were randomly assigned within 96 hours of birth to receive either supplemental DHM or PTF for 90 days or discharged from the hospital, whichever occurred first. Enteral feeds were initiated on the third postnatal day (interquartile range [IQR] 2–4) and advanced 10 to 25 mL/kg per day up to 160 mL/kg per day.¹³ Nutrient fortification of human milk commenced at ≥120 mL/kg per day using bovine-based multinutrient fortifiers. Once fortification of DHM commenced, a protein module was added to bring the estimated protein concentration to that of mature mother’s milk. The feeding intervention continued after transfer to community level II NICUs and follow-up of infants continued until 18 months’ CA. Baseline demographics and feeding metrics are provided in Table 1. The study protocol and neurodevelopmental, growth and clinical outcomes are published.^13,14 Research ethics boards at each participating hospital approved the study.

Cost-effectiveness analysis from a societal perspective was planned a priori and conducted alongside the RCT.^15,16 The RCT sample size was established to detect a 5-point difference in mean cognitive composite scores on the Bayley Scales of Infant and Toddler Development, Third Edition (Bayley-III) between infants randomly assigned to the DHM versus PTF group, the primary outcome.^13,14 As reported, we found no difference in Bayley-III cognitive composite scores between feeding groups (adjusted scores 92.9 in DHM versus 94.5 in PTF groups; fully adjusted mean difference, −2.0 [95% CI, −5.8 to 1.8]).¹⁴ We did find a significant reduction in NEC stage ≥ II among infants randomly assigned to the DHM (1.7%) versus PTF (6.6%) group (risk difference of −4.9% [95% CI, −9.0 to −0.9; P = .02]). The primary cost-effectiveness analysis in this manuscript, then, was conducted with a time horizon of birth to 18 months’ CA as intended a priori but focuses on the prevention of NEC rather than improvement in neurodevelopment. In secondary analyses, costs with a birth to hospital discharge time horizon were assessed in terms of total formal and informal health care and nonhealth care costs and cost-effectiveness in preventing NEC. Given the low incidence of NEC stage ≥ II, we used NEC stage ≥ I in these analyses, defined as demonstrating clinical symptoms according to Bells criteria followed by treatment involving suspension of enteral feeds and antibiotics for a minimum of 7 days.^14,17 A secondary analysis of costs postdischarge to 18 months’ CA was also included. Since the study’s completion, the Second Panel on Cost-Effectiveness in Health and Medicine recommended that in addition to presenting results from a societal perspective, they be presented from a health care sector perspective.¹⁸ Cost-effectiveness analysis from a health care sector perspective include formal health sector costs (medical) costs whether paid by a third-party or out-of-pocket by caregivers, whereas from a societal perspective includes all costs and health effects regardless of who incurred them or obtained the health effects. The cost components included in each reference case perspective are summarized in Supplemental Table 5 and costs from a health care sector perspective can be calculated from the tables summarizing component costs (Table 2, Supplemental Tables 6 and 7).

Formal health care sector medical costs incurred for each infant during their initial hospital stay were secured from the finance departments of all tertiary NICUs where >80% of initial hospitalization costs occurred and 12 of 17 community NICUs. Hospitals in the province of Ontario, Canada follow a standardized costing methodology developed by the Ontario Case Costing Initiative (OCCI).¹⁹ Formal health care medical costs captured in this system include expenditures associated with professional salaries (eg, nursing and other allied health but not physician), surgery, laboratory services, diagnostic imaging, and pharmacy. Indirect costs incurred by institutions, such as administration, human resources, and plant operations were allocated to each infant based on length of stay. Infant-specific health care costs were unavailable from 5 participating community NICUs because the OCCI standardized accounting system was not fully implemented. For these centers, a per diem cost was imputed by using mean costs calculated from the other 12 community NICUs, matching for the infant’s gestational age at birth to the nearest week.

Physician costs, not captured as part of OCCI, were calculated by using Ontario fee schedules.²⁰ Costs of enteral feeds were calculated by applying DHM, fortifier, and formula unit prices to volumes and recipes of feeds recorded daily. The unit cost of DHM used in our analysis was $4.95 Canadian dollars (CAD) per ounce; the cost of DHM purchased for the study adjusted to 2015 CAD. At the time of the study, most participating hospitals received fortifier and formula free of charge from vendors. To capture costs from a societal perspective, we used the price for formula and fortifier paid by 1 of our community NICUs participating in the Baby Friendly Initiative.²¹ The cost of bovine-based PTF, for example, was $0.13 CAD per ounce. The cost to fortify one ounce of human milk was $0.14 CAD per ounce.

A standardized Family Health Economic Questionnaire, previously employed by the investigators, was used to calculate caregiver informal health sector and nonhealth sector costs, including transportation to the hospital, meals away from home, and labor market earnings lost.^22,23 

This questionnaire was completed monthly by parents from the time of birth up to 18 months’ CA when either visiting their child in the NICU or by telephone with research staff.

Family Health Economic questionnaires were also used to capture the frequency of re-hospitalizations, medical or follow-up clinic visits, health care professional visits, prescription and over-the-counter medications, as well as reported loss of caregivers’ labor market earnings. Physician and other health care professional costs incurred postdischarge were calculated by using the province of Ontario fee schedules.²⁰ If an appropriate fee schedule was unavailable, costs were estimated by using average provincial wages posted on regulatory college and government Web sites. Costs of medications were assigned by using provincial reimbursement rates.²⁴ Costs of re-hospitalization were calculated based on the mean cost per episode, as defined by the principal reason for admission by using the OCCI database.¹⁹ 

If families were unavailable by telephone after discharge to complete the questionnaire, they were approached at their infant’s routine neonatal follow-up visits at 4, 8, 12, and 18 months’ CA. In situations in which Family Health Economic questionnaires were incomplete, a mean per diem cost for the infant was imputed based on the mean costs from infants with complete data matched by family’s income. At least some costing data were available on all infants in the study. Formal health care sector (medical) costs not requiring any imputation for missing days were available for 240 infants for their entire initial hospitalization and 361 for their tertiary NICU stay (Supplemental Fig 3). Complete cost data not requiring imputation were available for 223 infants from birth to 18 months’ CA.

Statistical analyses were conducted by using SAS Version 9.4 (SAS Institute, Inc, Cary, NC) using an intent-to-treat approach. All statistical tests were 2-tailed and P < .05 was considered statistically significant. Costs are reported in 2015 CAD. In 2015, 1.00 CAD was equivalent to 0.78 US dollars. Comparisons of clinical outcomes between groups were conducted by using Cochran–Mantel–Haenszel statistics with adjustment for site for categorical variables. For continuous outcomes, linear regression statistics adjusted for recruitment site and birth weight stratum (<1000 or ≥1000 g) or Wilcoxon rank tests for nonnormally distributed data were conducted.

Comparison of log-transformed mean total and cost components between groups were performed by using generalized estimating equation (GEE) adjusting for repeated measures taken at the same recruitment site and birth weight stratum. For comparing individual expense categories with a large number of 0 values (eg, hospital readmissions), repeated-measures nonparametric modeling (PROC MIXED, SAS) was used.

To assess sampling uncertainty in estimates of cost-effectiveness, we randomly sampled 1000 times with replacement all 363 subjects in the original data set to obtain 363 pairs of costs and effects (eg, NEC). For each of the 1000 samples, the mean costs and effects were plotted (eg, Fig 1) and the incremental cost-effectiveness ratio (iCER) was calculated. Specifically, iCER is the difference in the mean total cost between infants randomly assigned to the DHM and PTF groups divided by the difference in mean effectiveness (eg, NEC) between groups. We report sampling uncertainty using a cost-effectiveness acceptability curve (CEAC) (Fig 2),^25,26 which illustrates the probability that the intervention would be cost-effective given a willingness-to-pay threshold from $0 to $600 000 per case of NEC prevented.

Parameter uncertainty was assessed by using sensitivity analyses, in which the cost components were calculated first excluding infants who received exclusively mother’s milk during the intervention (Supplemental Table 6) and second excluding infants who had incomplete Family Health Economic questionnaires (Supplemental Table 7). Additional sensitivity analyses included calculation of the iCER using a range of plausible costs for DHM ($3.00–$7.60 per ounce), formal health care sector medical costs (initial and readmission), physician fees from birth to 18 months’ CA (70%–130%), and caregiver wages to reflect the Ontario minimum wage and national Canadian wage,^27,28 and NEC stage ≥ II instead of NEC stage ≥ I as the health outcome (Table 3).

Detailed description of progress of infants through the trial and clinical outcomes were previously published.¹⁴ Briefly, of 363 infants enrolled, the mean ± SD gestational age and weight at birth was 27.7 ± 2.6 weeks and 996 ± 272 g, respectively (Table 1). Infants born <1250 g comprised 75.8% (n = 275) of those enrolled and 12.4% (n = 45) were born small for gestational age. No statistically significant differences were found between infants randomly assigned to the DHM and PTF group with respect to the number of days they remained in the feeding intervention or mother’s milk use (Table 1). As previously reported, there were no statistically significant differences between infants randomly assigned to the DHM versus PTF group in adjusted composite cognitive (92.9 [95% CI, 89.8 to 95.9] and 94.5 [91.4 to 97.5]), language (87.3 [95% CI, 83.8 to 90.8] and 90.3 [95% CI, 86.7 to 93.9]), motor (91.8 [95% CI, 88.8 to 94.9] and 94.0 [95% CI, 91.0 to 97.0]) scores on the Bayley-III at 18 months’ CA. There were no differences in major clinical outcomes during initial hospitalization except for the incidence of NEC.

Considering all randomly assigned infants, mean total cost accrued from a societal perspective from birth to 18 months’ CA in the DHM group was $217 624 ($197 697–$237 551) and did not differ statistically from $217 245 ($196 494–$237 995) in the PTF group (P = .74) (Table 2). Likewise, infants randomly assigned to the DHM and PTF group had similar total initial hospitalization costs (P = .40). Examination of postdischarge to 18 months’ CA costs revealed lower costs for infants randomly assigned to the DHM versus PTF group (P = .04). Postdischarge, caregivers of infants randomly assigned to the DHM group had significantly lower productivity losses than infants randomly assigned to the PTF group (P = .03).

Using an 18-month time horizon, cost-effectiveness analysis revealed that DHM cost slightly more on average and had higher effectiveness (Table 4). Figure 1 illustrates the distribution of the differences in costs and effects between treatment groups using either a birth to 18-month time horizon (Fig 1A) or birth to discharge time horizon (Fig 1B) in which each point on the graphs represents a single bootstrap replication. As shown, a preponderance of replications in both figures are found on the right side indicating DHM was, on average, protective against NEC. In addition, Fig 1A reveals a 52.1% chance that DHM was more expensive using an 18-month time horizon. When the time horizon was restricted to initial hospitalization, there was a 71.7% chance that DHM was more costly and more effective in reducing NEC and a 28.0% chance that DHM would result in cost-savings and be more effective (ie, “dominant”).

Examination of the CEAC, however, reveals broad uncertainty in the estimation of the cost-effectiveness ratio (Fig 2). The CEAC shows the probability that an intervention is acceptable at various thresholds of willingness to pay for an outcome. There is a 67% probability that the therapy is acceptable to a decision-maker with a threshold of $100 000 per case of NEC prevented. Using a threshold that is analogous to the traditional upper limit of a CI, there is a 95% probability that DHM would be acceptable to a decision-maker whose threshold is $540 000 per case of NEC prevented.

Calculated total costs and iCER estimates were sensitive to hospitalization costs, physician fees, and the cost of DHM and caregiver wages (Table 3). If hospitalization and physician fees from birth to 18 months’ CA were reduced to 70% of the base case estimate, DHM would result in a mean cost-savings of $2119. Similarly, when the unit price of DHM was decreased to $3.00 per ounce, DHM would result in lower costs compared with PTF. In a separate sensitivity analysis, we re-ran the cost comparison analyses in the sample of infants with complete cost data to 18 months’ CA (n = 223, Supplemental Table 7). As illustrated, total costs from birth to 18 months’ CA remain similar between groups.

In this prospective cost-effectiveness analysis conducted for a RCT, the total costs of using DHM as a supplement to mother’s milk did not differ from PTF from a societal perspective to 18 months’ CA. Our data suggest that it would cost ∼$5328 (2015 CAD) per case of NEC averted to provide supplemental DHM; however, it should be emphasized that there was broad uncertainty around this estimate. In fact, formal analysis of this uncertainty suggests that there is only a 48% probability that DHM was simultaneously more effective and less costly than PTF in preventing NEC using a birth to 18-months’ CA time horizon (Fig 1A), and only a 67% probability that the intervention would be acceptable to decision-makers who were willing to spend $100 000 to prevent one case of NEC.

Interestingly, when comparison of total costs was restricted to postdischarge, there was a statistically significant cost advantage of using DHM as a supplement (Table 2). However, this difference was no longer statistically significant in sensitivity analyses in which exclusively mother’s milk-fed infants were removed (Supplemental Table 6). The difference in the base case analysis appeared to be mediated by lower productivity losses as a result of caring for their infant. Because there were no differences in the number of visits to health care professionals or hospital readmissions between treatments, we speculate that infants in the PTF group may have been slightly more fragile, requiring a parent to stay home from work but not so ill they required incrementally more medical attention.

Our findings differ from estimates of cost-savings associated with using DHM instead of PTF reported in the literature.^10,–12 Arnold^10,11 suggested potential cost-savings associated with DHM in both a case study of a formula-intolerant preterm infant with severe bronchopulmonary dysplasia whose mother was unable to provide breast milk and, in a second paper, using 3 modeling exercises based on assumptions of reduction in length of stay and of NEC. Similarly, Wight¹² estimated direct hospital cost-savings of supplemental DHM versus PTF for VLBW infants using published clinical findings from a single-site RCT and actual cost-estimates of different components of care from a different center.²⁹ DHM was assumed to be as effective as mother’s milk compared with PTF in preventing NEC and sepsis and shortening length of stay; DHM was not studied in this trial. In 2 other retrospective studies, authors also showed cost-saving associated with the use of human milk-based milk fortifiers to achieve an exclusive human milk diet.^30,31 Although the aforementioned cost-estimates spurred interest in DHM research and helped to engage administrators in a discussion of the merits of DHM, the risk of bias associated with these modeling and retrospective approaches is high and does not consider all costs associated with preterm birth.

There are a number of strengths of the current study. This is the first prospective cost-effectiveness analysis of DHM alongside a blinded RCT, and involved comprehensive analysis of infant-level costs. This experimental approach to health policy minimizes bias, optimizes internal validity, and permits simultaneous consideration of actual cost and effect distributions without the assumptions inherent in modeling studies. Moreover, unlike most formal economic evaluations in the neonatal trial literature, this analysis took a societal perspective, in which an attempt was made to capture all costs related to preterm birth to 18 months’ CA, including parental out-of-pocket expenses. In addition to being significant contributors to the validity of our cost estimates, the measurement of such costs highlights the high financial burden accruing to parents of preterm infants in general.

Despite the substantial protections against bias, certain limitations must be acknowledged. First, our study was undertaken in a single Canadian urban area, and the extent to which the results are generalizable to other settings in which mother’s milk feeding or costs might differ is uncertain. Of note, sensitivity analysis does show that the cost advantage postdischarge for DHM is not significant when exclusively mother’s milk-fed infants were removed from consideration, although there were minimal changes seen with other sensitivity analyses. A second limitation concerns the reliance on caregivers to recall out-of-pocket expenses and certain other health resource use; however, the monthly questionnaire administration is much more frequent than most other prospective economic evaluations in the neonatology literature, and would likely have minimized this problem.

In a high mother’s milk use setting, total costs from a societal perspective to 18 months’ CA of providing DHM as a supplement compared with PTF did not differ. Although the total costs were similar, supplemental DHM reduced the risk of NEC compared with supplemental PTF. Given the short- and long-term sequelae associated with NEC, and an additional cost of only $5328 per case of NEC averted, use of DHM over PTF as supplement is justified. Given the observation of cost-savings after hospital discharge associated with supplemental DHM, further exploration of the impact of early supplemental feeding on health outcomes and family costs after discharge is warranted.

Bayley-III

Bayley Scales of Infant and Toddler Development, Third Edition

CA

corrected age

CAD

Canadian dollars

CEAC

cost-effectiveness acceptability curve

CI

confidence interval

DHM

donor human milk

GEE

generalized linear modeling

iCER

incremental cost-effectiveness ratio

IQR

interquartile range

NEC

necrotizing enterocolitis

OCCI

Ontario Case Costing Initiative

PTF

preterm formula

RCT

randomized control trial

VLBW

very low birth weight

Additional members of the GTA DoMINO Feeding Group are as follows: Andrea Nash (Sunnybrook Health Sciences Centre, Toronto, Canada); Michael Jory, Joanne Rovet, and Christopher Tomlinson (The Hospital for Sick Children, Toronto, Canada); Kirsten Kotsopoulos and Karel O’Brien (Mount Sinai Hospital, Toronto, Canada); Anwar Asady, Ann Bayliss, and Sandra Gabriele (Trillium Health Partners, Mississauga, Canada); Shirley Sit and Sue Ekserci (Humber River Hospital, Toronto, Canada); Mahmud AlMadani (Lakeridge Health, Toronto, Canada); Munesh Singh (Markham Stouffville Hospital, Markham, Canada); Shaheen Doctor (North York General Hospital, North York, Canada); Debbie Stone (Rogers Hixon Ontario Human Milk Bank, Toronto, Canada); Karen Chang (Rouge Valley Health System, Toronto, Canada); Peter Azzopardi (The Scarborough Hospital, Scarborough, Canada); David Gryn (Mackenzie Health, Richmond Hill, Canada); Jelena Popovic (Michael Garron Hospital, East York, Canada); Debby Arts-Rodas (St Joseph’s Health Centre, Toronto, Canada); Carol Williams and Charmaine van Schaik (Southlake Regional Health Centre, Newmarket, Canada); Ilona Burkot and Judy Gibson-Stoliar (William Osler Health System, Brampton and Etobicoke, Canada). Members of the Data and Safety Monitoring Committee are as follows: Frank Greer (chair) MD, Professor Emeritus, Department of Pediatrics, University of Wisconsin School of Medicine and Public Health (Madison, WI); Sharon Groh-Wargo, PhD, RD, Professor, Nutrition and Pediatrics and Senior Nutritionist, Case Western Reserve University at MetroHealth Medical Center (Cleveland, OH); Ardythe Morrow, PhD, Director, Center for Interdisciplinary Research in Human Milk and Lactation Global Health Center, Professor, Department of Pediatrics, University of Cincinnati (Cincinnati, OH).

We thank the study families for their participation and ongoing support of this work. We acknowledge the Human Milk Banking Association of North America, specifically the Mother’s Milk Bank of Ohio and the NorthernStar Mothers’ Milk Bank in Calgary, Alberta for providing the DHM for this study.