Introduction
As described elsewhere in this report, the BEEP trial sought to determine if advising parents to apply an all-over-body application of emollient to their child’s skin throughout the first year of life in addition to standard infant skin-care advice might prevent the onset of eczema in children at high risk of allergic disease. The trial took a pragmatic, randomised, controlled, parallel group, multicentre assessor-blind design in which parents were advised to follow the skin-care advice for their child at home with minimal clinical contact. One thousand three hundred and ninety-four newborns at high risk of developing eczema were randomly allocated to the intervention or control arm. Standard practice infant skin-care advice was provided to all parents and those randomised to the intervention arm were also given advice to apply emollient daily to the child’s entire body surface area for the first year of the child’s life. Families were given a choice of two emollients (Doublebase Gel and Diprobase Cream). Diagnosis of eczema between 12 and 24 months of age (defined as meeting the UKWP Diagnostic criteria) was the primary outcome. The methods and results have been published1 showing no evidence that daily emollient during the first year of life prevents eczema in high-risk children and some evidence to suggest there may be an increased risk of skin infections.
Data on health resource use and quality of life were captured alongside the trial in order to undertake a within trial economic evaluation. Though concerns have been raised in the literature about so-called paradoxical (not clinically effective but yet found cost effective) conclusions between clinical and economic analyses of the same studies,68 we report the economic results despite the main clinical findings for several reasons. Firstly, the intervention was preventative in nature and when designing the study, we considered it may be possible for emollients not to be found statistically clinically effective yet estimated to be cost-effective. This is because the intervention is relatively cheap and thus even a small insignificant improvement across a lot of people could potentially be deemed cost-effective. Cost-effectiveness, via changes in resource use and costs, may also have been driven by improved secondary outcomes, such as change in severity of eczema, were such differences found. Secondly, Xu et al.69 published a decision tree model estimating the cost-effectiveness of seven moisturisers used in the first 6 months of life to prevent eczema in high-risk individuals. The study concluded that daily moisturisation is a cost-effective preventative strategy that can reduce the burden of eczema. It is therefore important to report the economic evaluation for this preventative strategy based on individual-level data collected alongside the definitive trial in order to add to the evidence base around whether this preventative strategy is cost effective or not. Thirdly, since the data were collected, it is important to analyse and report it so as to help inform any future or related research that might find this analysis relevant. The current level of economic evidence available for interventions aimed at preventing and treating eczema is limited.70,71 Understanding resource use, costs and quality of life in early life in high-risk children is of value in its own right and is something this study adds to the literature.
The primary objective of this within-trial economic evaluation was to estimate the cost-effectiveness of advice to provide daily all-over-body application of emollient during the first year of life for preventing AE in high-risk children over a 2-year time horizon. A sensitivity analysis is undertaken to estimate the medium-term cost-effectiveness at 5 years using a within-trial analysis.
Methods
Overview of economic analysis
The primary analysis forms the base (or reference) case within-trial economic analysis using individual participant-level data collected over 2 years from the BEEP trial to compare the cost-effectiveness of providing parents with advice to apply an all-over-body application of emollient daily in addition to standard skin-care advice compared to standard skin-care advice alone. This comparison was chosen to enable us to estimate the additional costs and benefits over usual care. The time horizon was chosen to be consistent with the time points used for the clinical study. The base-case analysis undertakes a cost-effectiveness analysis (CEA) from an NHS perspective in terms of the difference in proportion of cases without eczema. A secondary analysis was undertaken using a cost–utility approach; this was chosen as the secondary analysis due to uncertainties about how best to capture child health utilities especially in the very young.72–74
The evaluation was undertaken in line with published guidelines for the economic evaluation of health-care interventions as appropriate.75–79 After analysis had started NICE published updated guidance,80 which changed the preferred mapping function to be used in reference case analyses from that published by van Hout et al.81 to that published by Hernandez Alava et al.82,83 Since the analysis started prior to the change in guidelines we have adhered to the earlier guidelines, which for the most part is unchanged, and given the study results is unlikely to be significant in the conclusions reached.78
A health economic analysis plan was written and reviewed prior to the trial database being locked, the health economics analysis plan (HEAP) is publicly available.47
Identifying and measuring resource use
In keeping with the trial being conducted in the UK, where the NHS provides publicly funded health care, the analysis takes a health service perspective which is also in line with the NICE reference case.78 Disease-specific (eczema, asthma, and rhinitis) resource use was collected via online or postal paper questionnaires completed by participants at 3, 6, 12, 18 and 24 months. Personal social service (PSS) resource use was not captured explicitly, as it was anticipated that these types of services would not be accessed for the diseases of interest. Since the trial was light touch, in that after recruitment participants only had one face-to-face contact at 2 years, the resource use data collected were limited to NHS resource use relevant to the diseases of interest and did not collect any costs incurred by the family or wider society. This enabled us to keep the respondent burden low while still meeting the requirement of the base case to capture the intervention costs to the NHS and the participant’s wider disease-specific resource use of the NHS. Resource use related to food allergies was not collected as this aspect was added to the trial rather than included from the outset. Patient and public involvement members of the research team were involved in the design of the questionnaires.
Valuing costs
The cost of the intervention was estimated using data collected by the NCTU and costed using published unit costs for Doublebase Gel and Diprobase Cream in the prescription cost analysis (PCA).84 In the trial, the emollients were posted to participants homes but in practice if these were rolled out, people would collect these via repeat prescription from their GP surgery/pharmacy. As a result, these postage costs were not included in the economic evaluation.
Resource use relevant to the NHS perspective was valued using UK unit costs (Great British pounds) for the most current price year available at the start of the analysis (2019–20). Unit costs were identified from published sources and are clearly reported in the Results section.
A mean cost (SD) per participant per arm was estimated for 2 years.
Identifying, measuring and valuing outcomes
The primary economic outcome measure was incremental cost per percentage decrease in risk of eczema, a CEA. Secondary analysis reports a cost–utility analysis (CUA) where QALYs are estimated using utility scores obtained from the parental proxy CHU-9D instrument at 24 months. The CHU-9D is a generic preference-based measure of health-related quality-of-life instrument that asks how a child is today on nine questions (worries, sad, pain, tired, annoyed, schoolwork/homework, sleep, daily routine, activities) each with five response levels (ranging from no difficulty through to a lot or cannot do). We used the additional guidance given to us by the developer of the CHU-9D. This guidance provided extra wording to help parents of younger children understand how to answer questions 6 (schoolwork/homework), 8 (daily routine) and 9 (join in activities) given their child was of pre-school age. Utility ranges from 0.33 (worst health-related quality of life) through to 1 (best health-related quality of life).50 We also elicited utility scores for the main parent/carer using the EQ-5D-5L at baseline, and 24 months. The EQ-5D-5L is a generic preference-based measure of health-related quality-of-life instrument with five dimensions (mobility, pain/discomfort, usual activities, self-care, anxiety/depression) with five levels ranging from no problems through to unable to or extreme problems. Responses were converted to utility scores using the EQ-5D-5L Crosswalk UK preference weights as this was in line with recommendations at the point analysis started. Utility ranges from −0.594 to 1.81
In the CUA, the responses received on the quality-of-life instruments were converted to utility scores using the Stevens, 2012 valuation set for the CHU-9D and UK preference weights published by van Hout et al.81 for parental EQ-5D-5L. Utility values were used to estimate QALYs over 24 and 60 months, using both linear interpolation and area under the curve analysis with and without baseline adjustment.85 Child utility was assumed to be 1 (perfect health) at baseline, where baseline represented birth, for all participants. This was assumed because such instruments are not appropriate to ask at such a young age and babies were not eligible for inclusion in the study if they had a serious health issue or severe widespread skin condition at birth. The health-related quality of life for the main carer was elicited using the EQ-5D-5L in order to explore whether, if significantly fewer infants developed eczema, there might be any health spillover effects. Parental QALYs were estimated in the same way using baseline, and 24-month utility derived from the EQ-5D-5L. Parental QALYs were not incorporated into the CUA because the appropriate methods to do this are unclear86 and changes to main parent/carers health resource use and productivity were not also collected.
Economic analysis at primary outcome end point (2 years)
A CEA was planned regardless of what the full clinical results showed in terms of change in eczema cases for the reasons highlighted in the Introduction section of Chapter 4.
The economic base-case analysis included all randomised participants with complete cost and outcome data available. As the time horizon was 24 months, costs and benefits in months 13–24 were discounted using recommended rates, 3.5% for both.78
The main base-case analysis was a CEA, to estimate the incremental cost of preventing an additional case of eczema. Since it is unknown how much decision-makers would be willing to pay to prevent an additional case of eczema, a secondary analysis using the proxy reported CHU-9D to estimate QALYs in a CUA was undertaken and reports incremental cost per QALY as the outcome measure. A cost-effectiveness threshold (ʎ) of £20,000 (£30,000) per QALY is used in line with NICE guidance.78
Mean (SD) resource use and mean (SD) cost per participant are estimated per randomised arm. Mean difference (95% CI) in cost per participant between arms is estimated unadjusted and adjusted [for centre and number of immediate family members with atopic disease (one, two or more than two)]. Mean (SD) utility and mean (SD) QALYs per participant per randomised arm are presented along with mean difference (95% CI) in QALYs between arms unadjusted and adjusted. QALYs for the main carer were also estimated using responses to the EQ-5D-5L and are presented separately.
The unadjusted CEA was analysed using the heabs command87 in STATA (for which explanatory variables cannot be added to the regression command), while the adjusted analysis used a generalised linear model (GLM) for continuous and binary outcomes assuming costs and outcomes are uncorrelated. The Gaussian distribution was used for the cost GLM model and the binomial for the outcome GLM model. The identity option is used for the link function for both GLM models. The CUA uses a regression-based approach (seemingly unrelated regression equations).88
To determine the level of sampling uncertainty surrounding the mean incremental cost-effectiveness ratios (ICERs) non-parametric bootstrapping was undertaken generating 10,000 estimates of incremental costs and benefits. Cost-effectiveness acceptability curves (CEACs) were also produced, which show the probability that the intervention is cost effective at different values of willingness to pay. STATA MP version 17 was used to conduct the analysis. No subgroup analysis was undertaken as FLG mutation was not shown to be important in terms of the clinical effect.1
Sensitivity analysis
In the 2-year analysis the cost of emollients was varied in a threshold analysis to explore what cost of emollients would switch the results from cost effective to cost ineffective or vice versa.
The main sensitivity analysis was focused on taking a longer time horizon. It focused on the 5-year economic evaluation, where the primary 2-year CEA and secondary 2-year CUA [see Economic analysis at primary outcome end point (2 years)] were repeated using the 5-year data. Disease-specific (eczema, wheezing and rhinitis) resource use was collected via online or postal paper questionnaires completed by participants at 36, 48 and 60 months. The costs are not broken down and reported by individual year (e.g. year 3, 4 and 5) because secondary inpatient stays were collected in a format that did not enable us to see which year they pertained to. As the time horizon was 60 months, costs and benefits in months 13–60 were discounted using recommended rates, 3.5% for both.89 As part of these analyses, we re-ran the 5-year analysis after removing inpatient hospital costs incurred for wheezing, as it seems unlikely these would have been incurred as a result of having the intervention or not. We did this in order to explore how influential these costs were on the results of the 5-year analysis. In addition, we report the mean utility at 36, 48 and 60 months alongside the mean QALYs at 5 years per participant using the CHU-9D and the EQ-5D-5L for the main carer.
In the base-case economic evaluation, we did not impute missing data; instead, a complete-case analysis was undertaken in line with the clinical analysis. To evaluate the impact of missing data on the cost-effectiveness estimates, multiple imputation was employed for the cost–utility analyses, assuming that the data were MAR and using chained equations to handle the missing cost and outcome data to assess the impact on the conclusions reached.90 We did not undertake multiple imputation for the cost-effectiveness analyses as we had no baseline costs or effects, and all children were the same age.
There were plans to model the longer-term (i.e. beyond 5 years) costs and benefits of preventing or reducing the severity of eczema if the intervention had been found effective as detailed in the HEAP but this was not undertaken due to the clinical results.
We did not plan to look at the distributional effects, describing how costs and outcomes were distributed across different individuals, as this was not part of the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) criteria at the time of writing the HEAP, but given the findings of the clinical study such analysis is unlikely to add much to this study.
This economic evaluation is reported in line with the CHEERS guidance.77
Results
The full trial paper1 provides a detailed description of the final sample size and characteristics at 24 months. Of the 1394 babies randomly assigned to the emollient or control arm at the start of the study, 184 infants and their families either dropped out or were lost to follow-up by the 24-month period. This resulted in a sample of 1210 infants at 24 months: 598 allocated the emollient intervention alongside standard skin-care advice and 612 allocated standard skin-care advice only. Of these, 598 had complete cost and outcome data in the emollient intervention arm and 610 had this in the control arm (less than % missing) and it is this sample (n = 1208) that is analysed in the base-case CEA at 2 years.
Resource use and costs
Unit costs, together with their source, are presented in Table 21.
| Resource item | Unit cost (£) | Source (notes) |
|---|---|---|
| Intervention | ||
| Doublebase Gel | 6.63059 | PCA 202084 |
| Diprobase Cream | 6.67394 | PCA 202084 |
| NHS care | ||
| GP | 39.23 | PSSRU 202091 (9.22 minutes) |
| Practice nurse | 21 | PSSRU 202091 (0.33 of an hour per patient) |
| Hospital doctor | 85 | PSSRU 202091 (per half hour) |
| Hospital nurse | 19.8 | PSSRU 202091 (per 20 minutes) |
| Health visitor | 15.84 | PSSRU 202091(0.33 of an hour per patient) |
| Dietician | 92 | PSSRU 202091 (1 hour) |
| Physiotherapy | 82 | PSSRU 202091 (1 hour) |
| Pharmacist | 12 | PSSRU 202091 (1 hour) |
| Paediatric endocrinologist | 244 | NHS unit costs 2019/202089 (1 hour) |
| Paediatric dermatologist | 170 | NHS unit costs 2019/202089 (1 hour) |
| Paediatric respiratory medicine | 229 | NHS unit costs 2019/202089 (1 hour) |
| Paediatric ear nose and throat | 124 | NHS unit costs 2019/202089 (1 hour) |
| Paediatric clinical immunology and allergy service | 247 | NHS unit costs 2019/202089 (1 hour) |
| Paediatric ophthalmologist | 103 | NHS unit costs 2019/202089 (1 hour) |
| Paediatric dentistry | 152 | NHS unit costs 2019/202089 (1 hour) |
| Hospital admission/overnight stay | 1889 | NHS unit costs 2019/202089 |
| Medication | Range from 0.74 to 583 | PCA 202084 |
Intervention resource use and costs
In the 12-month intervention period, the mean number of pots issued per participant in the emollient arm was 4.21 (SD 2.20) (Table 22). The associated mean cost of emollients was £28.00 (SD £14.65) per participant in the base case (Table 23).
| Emollient arm | Control arm | Mean difference (95% CI) | |
|---|---|---|---|
| Mean ± SD (n) | Mean ± SD (n) | ||
| Intervention | 4.21 ± 2.20 (693) | 0.00 ± 0.00 (637) | 4.21 (4.04 to 4.38) |
| Doublebase Gel | 2.52 ± 2.05 (693) | 0.00 ± 0.00 (637) | 2.52 (2.36 to 2.68) |
| Diprobase Cream | 1.70 ± 1.54 (693) | 0.00 ± 0.00 (637) | 1.70 (1.58 to 1.81) |
| Wider NHS resource use | |||
| GP | 1.73 ± 3.31 (693) | 1.73 ± 2.81 (637) | 0.0003 (−0.33 to 0.33) |
| Practice nurse | 0.12 ± 0.53 (693) | 0.17 ± 0.80 (637) | −0.06 (−0.13 to 0.02) |
| Hospital doctor | 0.35 ± 1.27 (693) | 0.38 ± 1.48 (637) | −0.03 (−0.18 to 0.12) |
| Hospital nurse | 0.02 ± 0.18 (693) | 0.03 ± 0.27 (637) | −0.01 (−0.03 to 0.02) |
| Other health professional | 0.14 ± 0.68 (693) | 0.18 ± 0.75 (637) | −0.04 (−0.12 to 0.03) |
| Hospital episode | 0.17 ± 1.50 (693) | 0.19 ± 1.81 (637) | −0.02 (−0.19 to 0.16) |
| Medication | 3.74 ± 9.84 (693) | 3.50 ± 7.71 (637) | 0.24 (−0.72 to 1.20) |
| Emollient arm | Control arm | Mean difference (95% CI) | |
|---|---|---|---|
| Mean ± SD (n) | Mean ± SD (n) | ||
| Total intervention | 28.00 ± 14.65 (693) | 0.00 ± 0.00 (637) | 28.00 (26.86 to 29.14) |
| Doublebase Gel | 16.69 ± 13.59 (693) | 0.00 ± 0.00 (637) | 16.69 (15.63 to 17.74) |
| Diprobase Cream | 11.32 ± 10.25 (693) | 0.00 ± 0.00 (637) | 11.32 (10.52 to 12.11) |
| Wider NHS cost | |||
| GP | 67.33 ± 129.01 (693) | 67.30 ± 109.54 (637) | 0.03 (−12.90 to 12.96) |
| Practice nurse | 2.40 ± 10.89 (693) | 3.59 ± 16.77 (637) | −1.19 (−2.70 to 0.32) |
| Hospital doctor | 50.14 ± 207.69 (693) | 51.17 ± 226.93 (637) | −1.03 (−24.40 to 22.35) |
| Hospital nurse | 0.40 ± 3.31 (693) | 0.56 ± 5.18 (637) | −0.16 (−0.63 to 0.30) |
| Other health professional (eczema) | 6.67 ± 32.99 (693) | 10.46 ± 47.42 (637) | −3.79 (−8.15 to 0.58) |
| Hospital episode | 170.32 ± 1106.85 (693) | 145.29 ± 950.98 (637) | 25.03 (−86.43 to 136.50) |
| Medication | 24.06 ± 83.95 (693) | 23.58 ± 79.39 (637) | 0.48 (−8.33 to 9.28) |
| Mean total cost (Int+NHS) | 349.32 ± 1314.29 (693) | 301.94 ± 1083.61 (637) | 47.37 (−82.84 to 177.59) |
Other health resource use and costs
Resource use between the emollient and control arms was not significantly different (see Table 22). Table 23 reports the disaggregated mean discounted costs per infant for both arms using available case data. In the complete-case analysis, mean cost was £398.23 (SD 1408.39) per child in the emollient arm (n = 598) and £312.16 (SD 1105.04) in the control arm (n = 610). When intervention use was combined with other health resource use, the unadjusted mean incremental cost per infant was £86.07 (95% CI £ −57.77 to 229.90; Table 23). The difference in total costs between groups largely reflects the cost of the intervention and greater hospital inpatient stays; other NHS costs were not significantly different between groups. The largest component of cost was overnight hospital stays, especially for those infants reporting wheezing.
Outcome measures
Table 24 presents the outcomes for both arms unadjusted on the available-case sample at 2 years. In the complete-case analysis, the proportion of cases without eczema according to the UKWP-AD definition for the emollient arm was 0.7676 (SD 0.4227) (n = 598) and 0.7557 (SD 0.4300) for the control arm (n = 610) over the 24-month period. In the complete-case analysis, the adjusted incremental difference in proportion of cases without eczema at 2 years was 0.0164 (95% −0.0329 to 0.0656) adjusted or 0.0118 (−0.0359 to 0.0595) unadjusted. Complete responses to the CHU-9D were received from 75.6% (77.3%) of all participants in the intervention (control) arm at 2 years (or 88% of those completing and returning the study questionnaire booklet). Complete responses to the EQ-5D-5L were received from 71.6% to 82.7% (72.5–84.2%) of participants in the intervention (control) arm at baseline and 2 years. The outcomes for the CHU-9D and the EQ-5D-5L are not too dissimilar between the two arms.
| Intervention (n = 693) | Control (n = 701) | Mean difference (95% CI) | |||
|---|---|---|---|---|---|
| Mean ± SD (n) | Missing | Mean ± SD (n) | Missing | ||
| Child participants | |||||
| Proportion without eczema at 24 months (based on UKWP-AD) | 0.7676 ± 0.4227 (598) | 99 | 0.7549 ± 0.4305 (612) | 89 | 0.0127 (−0.0355 to 0.0608) |
| CHU-9D 24 months | 0.9349 ± 0.0690 (524) | 169 | 0.9338 ± 0.0685 (541) | 160 | 0.0010 (−0.0071 to 0.0091) |
| QALYs at 24 months | 1.9030 ± 0.0672 (524) | 169 | 1.9020 ± 0.0642 (541) | 160 | 0.0010 (−0.0069 to 0.0089) |
| Main carer | |||||
| EQ-5D-5L at baseline | 0.8560 ± 0.1513 (496) | 198 | 0.8520 ± 0.1580 (508) | 193 | 0.0040 (−0.0152 to 0.0232) |
| EQ-5D-5L at 24 months | 0.9212 ± 0.1417 (573) | 120 | 0.9187 ± 0.1303 (592) | 109 | 0.0024 (−0.0132 to 0.0181) |
| QALYs – EQ-5D-5L (parent) (24 months) | 1.7441 ± 0.2279 (457) | 236 | 1.7453 ± 0. 2314 (467) | 234 | −0.0012 (−0.0308 to 0.0285) |
Base-case cost-effectiveness analysis at 2 years
Table 25 presents the unadjusted and adjusted results of the CEA in terms of the number of eczema cases diagnosed together with an estimate of the ICER and separately the CEAC for the adjusted analysis (Figure 10). The incremental difference in cost for the emollient arm (n = 598) compared to the control arm (n = 610) was £87.45 (95% CI −54.31 to 229.27) (unadjusted this was £86.07, 95% CI £ −57.77 to 229.90). The adjusted incremental difference in proportion without eczema for the emollient arm compared with the control arm was 0.0164 (95% −0.0329 to 0.0656) (unadjusted was 0.012, 95% CI −0.0359 to 0.0600), meaning that the emollient arm had less cases (by a very small margin) of eczema at 24 months. The ICER was £5337 (£7281 unadjusted) per percentage decrease in risk of eczema. The amount decision-makers would be willing to pay per percentage decrease in risk of eczema is unknown, but it can be seen from the CEAC that there is a lot of uncertainty around the probability of the intervention being cost effective, at willingness-to -pay values over £10,000 per percentage decrease in risk of eczema the probability of cost-effectiveness is always < 60 to 72%.
| Analysis (N e, N c) | Incremental cost (UK£) (95% CI) | Incremental effect (95% CI) | ICER | % Cost effective at £20k (£30K) |
|---|---|---|---|---|
| CEA base case (CCA, unadjusted)(598, 610) | 86.07 (−57.77 to 229.90) | 0.012 (−0.0359 to 0.0600) | £7281 per percentage decrease in risk of eczema | Willingness-to-pay threshold per percentage decrease in risk of eczema unknown. |
| CEA base case (CCA, adjusted)(598, 610) | 87.45 (−54.31 to 229.27) | 0.0164 (−0.0329, 0.0656) | £5337 per percentage decrease in risk of eczema | |
| CUA (CCA, CHU-9D, unadjusted)(524, 542) | 81.47 (−80.21 to 243.14) | 0.0010 (−0.0069 to 0.0089) | £82,250 per QALY | 30% (36%) |
| CUA (CCA, CHU-9D, adjusted)(524, 542) | 84.28 (−78.36 to 246.93) | 0.0010 (−0.0068 to 0.0089) | £82,580 per QALY | 29% (36%) |
FIGURE 10 Cost-effectiveness acceptability curve for the complete-case adjusted CEA analysis (UKWP-AD) at 2 years. (Figure not reproduced here; see the original article.)
Secondary cost–utility analysis at 2 years
Proxy-reported Child Health Utility instrument-9 domains for the infant
The adjusted incremental difference in cost per infant for the emollient arm (n = 524) compared to the control arm (n = 542) was £82.28 (95% CI −78.36 to 246.93). The adjusted mean QALYs for infants was very slightly more in the emollient arm, incremental mean difference of 0.0010 (95% CI of −0.0068 to 0.0089), see Table 25. This means that the emollient arm was more expensive and slightly more effective than the control arm. The adjusted ICER was £82,580 which, when using a threshold value of £20,000 (£30,000), would not be considered cost-effective.78 The probability of the intervention being cost-effective at 2 years was 29% (36%) for a threshold value of £20,000 (£30,000) (Figure 11).
FIGURE 11 Cost-effectiveness acceptability curve for the complete-case adjusted CUA (CHU-9D) at 2 years. (Figure not reproduced here; see the original article.)
EuroQol-5 Dimensions, five-level version for the main carer
The adjusted mean difference in the QALYs was close to zero, −0.0012 (95% CI −0.0308 to 0.0285) (see Table 24). We did not combine these results into a CUA as we only collected NHS resource use for the infants. However, on average main carers in the emollient arm had slightly lower QALYs than main carers in the control arm.
Sensitivity analyses
Cost of emollients
Given we do not know what a decision-maker’s willingness to pay per percentage decrease in risk of eczema is, the cost of the emollient was explored in the 2-year adjusted CUA by undertaking a threshold analysis (i.e. the cost of emollients that would switch the intervention from cost-ineffective to cost effective at a willingness to pay of £30,000 per QALY). Doing this, we observed that even at a cost of £0 per emollient item it would not be enough to make this intervention cost-effective unless reduced healthcare use/NHS cost savings could also be found elsewhere.
Cost-effectiveness analysis over 5 years
The complete-case analysis for the CEA at 5 years consisted of 383 (55.3%) children in the intervention arm and 411 (58.6%) children in the control arm and it is this sample that is analysed in the 5-year CEA analysis. In the CUA analysis, this was 354 (36.7%) and 263 (37.5), respectively. Questionnaire response rates declined significantly in years 3–5.
Table 26 reports the mean outcomes per child for both arms for years 3–5 for available case data. The proportion without eczema is slightly higher at 5 years in the control arm, 0.8939 (SD 0.3082) in the intervention arm versus 0.9315 (SD 0.2529) in the control arm [incremental effect −0.0375 (95% CI −0.0732 to 0.0019)], though it should be noted this is based on parental report of a clinical diagnosis of eczema in the previous year unlike at 2 years. The adjusted incremental difference in QALYs (based on proxy report for the CHU-9D) at 5 years for the emollient arm compared with the control was 0.0166 (95% −0.0135 to 0.0467) (unadjusted was 0.0171, 95% CI −0.0138 to 0.0480). Mean QALYs for parental health-related quality of life using the EQ-5D-5L at 5 years was 4.2063 (SD 0.4707) in the intervention arm and 4.1684 (SD 0.5135) in the control arm, with an incremental difference of 0.0379 (95% CI −0.0319 to 0.1077) QALYs (available case). These incremental effects are small, just as they were at 2 years.
| Intervention (n = 693) | Control (n = 701) | Mean difference (95% CI) | |||
|---|---|---|---|---|---|
| Mean ± SD (n) | Missing | Mean ± SD (n) | Missing | ||
| Child participants | |||||
| Proportion without eczema at 60 months (Parent report of a clinical diagnosis of eczema in the previous year at 60 months) | 0.8939 ± 0.3082 (462) | 231 | 0.9315 ± 0.2529 (496) | 205 | −0.0375 (−0.0732 to 0.0019) |
| CHU-9D 36 months | 0.9283 ± 0.0713 (335) | 358 | 0.9235 ± 0.0671 (350) | 351 | 0.0048 (−0.0056 to 0.0152) |
| CHU-9D 48 months | 0.9337 ± 0.0718 (445) | 248 | 0.9296 ± 0.0754 (480) | 221 | 0.0048 (−0.0056 to 0.0152) |
| CHU-9D 60 months | 0.9398 ± 0.0660 (451) | 242 | 0.9349 ± 0.0696 (472) | 229 | 0.0049 (−0.0039 to 0.0137) |
| QALYs (CHU-9D) at 60 months | 4.424 ± 0.1820 (255) | 438 | 4.4053 ± 0.1740 (263) | 438 | 0.0181 (−0.0126 to 0.0488) |
| Main carer | |||||
| EQ-5D-5L at 36 months | 0.8957 ± 0.1436 (506) | 187 | 0.8931 ± 0.1404 (539) | 162 | 0.0026 (−0.0148 to 0.0200) |
| EQ-5D-5L at 48 months | 0.8813 ± 0.1605 (452) | 241 | 0.8764 ± 0.1666 (491) | 210 | 0.0049 (−0.0160 to 0.0259) |
| EQ-5D-5L at 60 months | 0.8953 ± 0.1504 (452) | 241 | 0.8771 ± 0.1635 (477) | 224 | 0.0181 (−0.0021 to 0.0384) |
| QALYs (EQ-5D-5L) at 60 months | 4.2063 ± 0.4707 (379) | 314 | 4.1684 ± 0.5135 (390) | 311 | 0.0379 (−0.0319 to 0.1077) |
The results of the 5-year analysis are shown in Table 27. The adjusted incremental cost at 5 years for the emollient arm (n = 383) compared to control (n = 411) was £−106.89 (95% CI −354.66 to 140.88) [unadjusted this was £−123.74 (95% CI £−382.55 to 127.08)]. The adjusted incremental difference in proportion without eczema at 5 years for the emollient arm compared with the control was –0.0329 (95% −0.0659 to 0.0002) (unadjusted –0.0386, 95% CI –0.0776 to 0.0004), meaning that the intervention was cost saving but less effective (i.e. associated with slightly more cases of eczema) at 5 years. The adjusted ICER was £3201 (£3312 unadjusted) per percentage decrease in risk of eczema. The amount decision-makers would be willing to pay per percentage decrease in risk of eczema is unknown. It should be noted, however, that the driver for the intervention to be cost saving was the number and cost of inpatient hospital stays incurred as a result of wheezing which were higher in the control group. The control arm had a greater number of participants reporting such costs at 5 years than the intervention arm (22 compared to 7 in the intervention arm) and the control arm had a higher range on the cost incurred than the intervention arm, thus leading the incremental cost to look cost saving for the intervention. When these costs were removed from the analysis, incremental cost became positive (i.e. those in the intervention arm cost more on average) such that since the proportion without eczema was lower in the intervention arm, we would conclude that the intervention is dominated by the control (i.e. we would not recommend the intervention). Such a finding is possible because the costs observed are small and as such a single large cost item can sway the result. It seems quite implausible that the intervention could impact asthma inpatient stays given it was not significantly effective at preventing eczema which would be the first change required in the mechanism of action for it to impact resource use for other atopic diseases.
| Analysis (N e, N c) | Incremental cost (UK£) (95% CI) | Incremental effect (95% CI) | ICER |
|---|---|---|---|
| 5-year CEA (CCA, unadjusted)(383, 411) | −123.74 (−382.55 to 127.08) | −0.0386 (−0.0776 to 0.0004) | £3312 per percentage decrease in risk of eczema |
| 5-year CEA (CCA, adjusted)a(383, 411) | −106.89 (−354.66 to 140.88) | −0.0329 (−0.0659 to 0.0002) | £3201 per percentage decrease in risk of eczema |
| 5-year CEA (CCA, adjusted)a without inpatient costs due to wheezing(383, 411) | 100.34 (−30.09 to 230.83) | −0.0329 (−0.0658 to 0.0001) | Dominated |
| 5-year CUA (CCA, CHU-9D, unadjusted)(254, 263) | −336.02 (−670.22 to −1.82) | 0.0171 (−0.0138 to 0.0480) | Dominant |
| 5-year CUA (CCA, CHU-9D, adjusted)(254, 263) | −312.68 (−645.17 to 19.82) | 0.0166 (−0.0135 to 0.0467) | Dominant |
| 5-year CUA (MI, CHU-9D, adjusted)(693, 701) | −49.80 (−306.83 to 207.22) | 0.0174 (−0.0045 to 0.0394) | Dominant |
| 5-year CUA (MI, CHU-9D, adjusted) without inpatient costs due to wheezing (693, 701) | 90.92 (−17.06 to 198.90) | 0.0173 (−0.0051 to 0.0396) | £5268 per QALY |
Table 27 also shows the cost–utility results at 5 years, despite more cases having eczema in the emollient group the intervention was dominant (cheaper and slightly more effective). Dealing with the missing data using multiple imputation did not change this finding. However, incremental costs were higher for the emollient group when high-cost wheezing inpatient stays were excluded, although the ICER was lower than NICE’s implicit threshold. It seems likely the CEA is more plausible, although all results reflect the very small values involved.
Discussion
Main findings
In our economic analysis of the main multicentre, pragmatic RCT of high-risk infants, we find no evidence that regular emollient use for the first year of life is cost-effective at 2 years of age. Incremental costs and effects were very small. These results support the findings of the original clinical study.1 We find that the intervention is more expensive, with potential to prevent only marginally more cases of eczema and generates very slightly more QALYs as measured using the proxy version of the CHU-9D. When the results were adjusted for covariates, the same conclusion held for the primary and secondary outcome measures.
At 5 years almost half the sample had missing cost data. The complete-case CEA found the intervention to be cost saving at 5 years but to have a slightly lower proportion without eczema. However, when inpatient hospital costs for wheezing were removed from the analysis (because < 4% of the sample incurred them and they were associated with high cost), it was found the cost saving of the intervention at 5 years disappeared, and that the CEA was dominated by control (as the intervention was both more costly and less effective).
In the 5-year CUA (as complete case or using multiple imputation) the intervention was found to dominate; the incremental cost was negative indicating the intervention was cost saving and the incremental effect was higher (better quality of life) in the intervention arm. Given the intervention was not clinically effective, the plausibility of these results needs to be questioned. In part, the cost savings found in the 5-year analyses were driven by differences in number and cost of inpatient hospital stays for wheezing between study arms in years 3–5; when these were removed as part of sensitivity analysis, the incremental cost was very small in the CUA. Given there is no evidence the intervention was clinically effective at preventing eczema, it seems highly unlikely it will have had any impact on wheezing resource use such that the 5-year CUA results are likely to be spurious. That we find this seemingly ‘paradoxical’68 finding at 5 years (particularly in the CUA) is indicative of there being non-significant small incremental costs and effects.
Our results, alongside those of the full trial, have important implications for the existing evidence on whether to use emollients prophylactically in the early years of a high-risk infant’s life. This is important to report as a previous economic analysis of emollient use in infants prophylactically69 reached a different conclusion, that daily emollient use was a cost-effective strategy. Their study was a secondary analysis based on pilot evidence about the effectiveness of emollients, primarily relying on a very large RR estimate of preventing 50% of new eczema cases (n = 108 in complete-case analysis) reported in Simpson et al (2014).92 Using this estimate along with assumptions over the amount of emollient that would be used and equivalent effectiveness across the seven products compared, the decision tree was reportedly analysed using a CUA approach for a 6-month time horizon. Using data from a larger, definitive study, our results suggest that emollient use as a preventative measure early in life for high-risk infants is not effective or cost-effective taking a 2-year time horizon and this conclusion holds at 5 years given the caveats discussed.
Strengths and limitations
Issues about how to best capture utility for infants and young children for use in economic evaluations are well known.72–74 In this study, we used the CHU-9D by parental proxy for infants at ages 2, 3, 4 and 5 years old. The CHU-9D is currently validated for children 7 years and upwards50 and for children aged 5–7 years by parental proxy.93 We have been unable to find any published studies using the CHU-9D in children under 5 years of age, although the MAGIC trial94 plans to use the CHU-9D with children as young as 3 years old (Sheffield Clinical Trials Research Unit, 2020). While it is a strength to explore a new approach to add to the growing academic discussion on appropriate HR-QoL measures for young children and infants, it is important to interpret the results tentatively until the use of this instrument in this context is less experimental. Other potential issues such as using parental proxies to value child health also need to be acknowledged.95
Response rates to the 36-month questionnaire was comparatively low due to issues distributing this to those first randomised into the study (i.e. the CHU-9D and EQ-5D was inadvertently left out of the questionnaire initially). Completion rates for the questionnaires declined in years 3–5. Together, these may affect the quality and conclusions that can be taken away from the 5-year sensitivity analysis.
The study collected disease-specific resource use but did not capture resource use related to food allergy as this component was not included at the outset of the study. Given the small number of patients who experience clinically diagnosed food allergy and the outcome of the study, this is unlikely to be important. However, it may be appropriate to include resource use related to food allergy in future studies seeking to explore interventions to prevent eczema and other allergic disease.
If a favourable difference in risk had been found at 24 or 60 months and costs and outcomes had not converged at 60 months, a longer-term CEA from birth to 16 years using a model-based analysis would have been undertaken. Given the results, a decision was taken not to develop a longer-term model for this intervention, but future preventative strategies might well benefit from such an approach and from considering a longer time horizon.
Further research
This study is the first we know to have published results using the proxy version of the CHU-9D in children aged as young as 2 years. We did this using the additional guidance developed by the developer of the CHU-9D. We plan to analyse these data further to help inform the design of future studies of very young children, but in terms of practicality it can be seen in the response rate reported in this chapter that most parents did not have difficulties completing this instrument at 2 years (84% provided complete responses which was not too dissimilar to the completion rate for the main clinical outcome, see Outcome measures for further details). There is a need for further research to confirm the validity of the CHU-9D for children and infants aged under 5 years.
Given the results of the clinical and economic study, there is a need for further research looking for potentially preventative strategies that are both effective and cost-effective in order to help reduce the burden of AE.
Conclusions
The daily use of all-over-body application of emollient during the first year of life as described in this study is unlikely to be a cost-effective intervention in preventing AE in high-risk children under 2 years of age if a 2-year time horizon is taken. At 5 years the intervention appears to be cost-effective, but this seemed to reflect the high cost of inpatient hospital stays due to wheezing amongst a small proportion (< 4%) of the sample in a study that shows no clinical benefit for wheezing prevention at 5 years. The 5-year results in this study illustrate what Raftery et al.68 call ‘doubly null’ results, that is no significant difference in both primary outcome and cost per patient. We seek to avoid ‘paradoxical’ conclusions at 5 years by exploring and acknowledging the issue (small incremental costs and effects) and factors (such as the differentially high wheezing inpatient costs between arms over years 3–5) that might explain the findings68 regarding not clinically effective but potentially cost-effective results. Resource allocation decisions should be based on the totality of evidence established by systematic reviews and meta-analysis. Our results add an important contribution to the evidence available to inform any future decisions in this clinical area. Our results also indicate that in terms of practicality it is possible to use the proxy CHU-9D with children as young as 2 years old and further research to explore the validity of this would be useful.