Is the cost‐effectiveness of an early‐childhood sleep intervention to prevent obesity affected by socioeconomic position?

Abstract

Objective

This study aimed to determine whether the cost‐effectiveness of an infant sleep intervention from the Prevention of Overweight in Infancy (POI) trial was influenced by socioeconomic position (SEP).

Methods

An SEP‐specific economic evaluation of the sleep intervention was conducted. SEP‐specific intervention costs and effects at age 5 years, derived from the trial data, were applied to a representative cohort of 4,898 4‐ to 5‐year‐old Australian children. Quality‐adjusted life years and health care costs were simulated until age 17 years using a purpose‐built SEP‐specific model. Incremental cost‐effectiveness ratios and acceptability curves were derived for each SEP group.

Results

The incremental cost‐effectiveness ratios, in Australian dollars per quality‐adjusted life year gained, were smaller in the low‐ ($23,010) and mid‐SEP ($18,206) groups compared with the high‐SEP group ($31,981). The probability that the intervention was cost‐effective was very high in the low‐ and mid‐SEP groups (92%‐100%) and moderately high in the high‐SEP group (79%).

Conclusions

An infant sleep intervention is more cost‐effective in low‐ and mid‐SEP groups compared with high‐SEP groups. Targeting this intervention to low‐SEP groups would not require trade‐offs between efficiency and equity.

Study Importance.

What is already known?

• The Prevention of Overweight in Infancy (POI) trial showed that a semitailored sleep‐guidance intervention was cost‐effective at preventing childhood obesity at an early age.

What does this study add?

• The medium‐term cost‐effectiveness and cost‐utility of a sleep intervention, examined in the POI trial, were affected by socioeconomic position.

• The sleep intervention was highly cost‐effective in low‐ and middle‐socioeconomic groups and moderately cost‐effective in a high‐socioeconomic group.

How might these results change the direction of research or the focus of clinical practice?

• Offering semitailored infant sleep guidance to parents from low socioeconomic backgrounds may be both a cost‐effective and equitable strategy to prevent childhood obesity.

INTRODUCTION

Socioeconomic inequality in the prevalence of childhood overweight and obesity is a stubborn problem in many countries. Recent national health surveys have shown that obesity prevalence among children in low socioeconomic areas is twice that in high socioeconomic areas in Australia [1] and New Zealand [2]. Furthermore, there is concern that these inequalities are increasing over time [3].

Considering that this inequality can likely be attributed to differences in social circumstances and opportunities, not biological inevitability, it should be regarded as an inequity. Improving such health inequities is frequently considered a responsibility of government health policy and funding [4]; however, obesity policies in the UK [5] and Australia [6] have been criticized for failing to address inequities. The economic evidence to support improvements in health inequities is limited [7] but imperative; in particular, analyses of the balance between resources expended and outcomes achieved in different socioeconomic groups are crucial to making decisions on how best to target resources [8].

Existing evidence regarding the effectiveness and cost‐effectiveness of interventions to prevent childhood overweight and obesity in different socioeconomic groups is mixed. In the past decade, several reviews have suggested that the mode, target, and context of interventions impact their differential effectiveness by socioeconomic group [9, 10, 11]. These reviews emphasize that the reporting of effectiveness by socioeconomic group is infrequent and call for study designs to routinely allow such analyses. Examination of differential cost‐effectiveness by socioeconomic groups is even rarer. In a recent systematic review of equity‐informative cost‐effectiveness analyses [12], only three of fifty‐four eligible papers focused on obesity, and each of these evaluated hypothetical interventions and were not informed by randomized trial results [13, 14, 15].

In a randomized trial, the Prevention of Overweight in Infancy (POI) trial, a novel, evidence‐based [16] educational sleep intervention delivered to the caregivers of infants in their first 2 years of life to reduce the risk of obesity was examined [17]. The intervention was found to be both effective [17] and cost‐effective [18] at reducing body mass index (BMI). The cost‐effectiveness study modeled future costs and BMI outcomes into adolescence but it did not investigate cost‐effectiveness across different socioeconomic groups [18]. Therefore, this sleep intervention holds considerable promise from an efficiency perspective, but its potential from an equity perspective is, as yet, unknown. The objective of the current study was to incorporate an equity perspective into the economic evaluation of the POI sleep intervention by examining its cost‐effectiveness in different socioeconomic groups.

METHODS

The POI trial

The POI trial aimed to determine the effectiveness of different behavioral interventions to prevent early‐childhood obesity. Participants were pregnant women recruited between 2009 and 2010 from Dunedin, New Zealand. Interventions started antenatally and continued until the child was up to 2 years of age. The trial protocol and primary results have been detailed elsewhere [17, 19, 20]. Families were randomized to a group receiving a traditional food and physical activity intervention, a novel sleep intervention group, a group combining both interventions (combination intervention), or usual care. Randomization was stratified by parity and the New Zealand Deprivation Index 2013 (NZDep13), an area‐level measure of socioeconomic position (SEP) categorized in deciles. The sleep intervention was effective at both age 2 and 5 years [17] and cost‐effective over a 15‐year time horizon [18], whereas the food and physical activity intervention and combination intervention were not. Therefore, the present study investigated the sleep intervention in comparison with usual care and whether its cost‐effectiveness was affected by SEP.

The sleep intervention and usual care

The sleep intervention involved an antenatal group session in which advice on how to promote healthy infant sleep patterns was provided and a home visit delivered by trained staff when the child was 3 weeks of age. When the child was 6 months of age, parents were asked whether their child had a sleep problem. Those who said yes were offered additional support of their choice: 1) simple advice (e.g., changes to sleeping arrangements or feeding patterns); 2) advice around settling techniques; 3) a partial sleep intervention (e.g., identifying a single behavior or limited‐duration behavior change); or 4) a full sleep intervention (usually a modified extinction technique such as parental presence, camping out, or controlled comforting) [21, 22]. Parents were contacted again when the child was 12 and 18 months of age to be offered this extra support but they could also request extra support at any time up to 24 months. Each parent could request any support options and could request the same option multiple times.

The usual care group received a government‐funded “Well Child” service in which nurses provide free home and clinic visits and offer guidance on a variety of early‐childhood health issues [23]. Intervention groups also received this standard service.

Economic evaluation

We conducted a modeled economic evaluation of the sleep intervention, stratified by three SEP groups. We estimated cost‐effectiveness, which determines the incremental cost per BMI unit avoided, and cost‐utility, which determines the cost per quality‐adjusted life year (QALY), of the intervention compared with control. Although the trial was conducted in New Zealand, we used an Australian health funder perspective because we were interested in whether the intervention could be cost‐effective in this context. A health funder perspective was taken because such an intervention would most likely be implemented by a government agency. We used a 17‐year time horizon because this was long enough to capture important health and cost effects of the intervention but short enough to be relevant for most time frames of policy targets. This is 2 years longer than the time horizon of the earlier unstratified economic evaluation of the sleep intervention [18], as an extra wave of data was available for the study which informed the modeling [24]. All costs were valued in 2018 Australian dollars. Costs, BMI, and QALYs were discounted at 5% per year as recommended by Australian health funding guidelines [25]. The reporting of our analysis adheres to the Consolidated Health Economic Evaluation Reporting Standards (CHEERS; online Supporting Information) [26].

Intervention costs

The costs of delivering the sleep intervention were calculated for each participant using the same microcosting approach as the earlier unstratified economic evaluation [18]. As costs were calculated at an individual level, this approach allowed for potential differences in use of the optional additional support across SEP groups. Although there would likely be differences in “usual care” between Australia and New Zealand, the costs of these services would be equivalent between intervention and control arms. Therefore, the usual care costs were not calculated specifically.

The EQuity‐informative Early Prevention of Obesity in Childhood model

We adapted the model used for the economic evaluation of POI sleep (the EPOCH model) to allow child BMI trajectories to be predicted by SEP. The EQuity‐informative Early Prevention of Obesity in Childhood (EQ‐EPOCH) model is a discrete‐time, deterministic microsimulation model in which BMI, associated health care costs, and quality of life are computed over annual cycles. The BMI growth part of the model was developed using 12 years of data from the kindergarten (K) cohort of the Longitudinal Study of Australian Children (LSAC) release 7.1 [24]. Anthropometric and SEP measures were collected biennially from age 4 to 5 years until age 16 to 17 years. An SEP z score was derived accounting for parent education, parent occupation, and family income [27]. We used these data to derive equations predicting annual BMI growth for high‐ and low‐SEP groups (above and below median SEP) and based on child age, sex, and current BMI.

Direct health care costs and utilities (a preference‐based measure of quality of life) were predicted from modeled weight status categorized as healthy weight (including underweight) and overweight (including obesity) based on World Health Organization (WHO) growth standards [28]. Underweight was included in the healthy weight category because, while the association between underweight and health care costs and utilities is uncertain [29, 30, 31], underweight represented less than 1% of the K cohort in LSAC; therefore, any potential effects were unlikely to influence our results. Direct health care costs by weight status were determined using a top‐down approach using national Australian data (online Supporting Information). Utilities and utility decrements, which are measures of quality of life, were derived from a meta‐analysis for the calculation of QALYs (online Supporting Information) [32].

To validate the EQ‐EPOCH model for this study, we inputted the BMI measurements, age, sex, and SEP group from the first wave of the LSAC K cohort, when participants were aged 4 to 5 years, and simulated BMI every year until children were aged 16 to 17 years. We then compared predicted SEP‐specific BMI trajectories with the actual trajectories from the K cohort at biennial waves of data collection.

Further details about the EQ‐EPOCH model have been described elsewhere [33] and in online Supporting Information.

Simulating intervention effects and costs

The POI trial results for the sleep and usual care groups were stratified into three SEP groups based on the NZDep13 as per standard categorization [34] and those used in the original randomized controlled trial [19, 20]: NZDep13 1 to 3 was classified as high SEP; NZDep13 4 to 7 as mid SEP; and NZDep13 8 to 10 as low SEP. As this resulted in small group numbers (Table 1), a non‐parametric approach was used to apply SEP‐specific intervention effects and costs derived from the POI study to the larger LSAC K cohort. The first wave of the LSAC K cohort (mean age: 5 years) was similarly categorized into three SEP groups based on SEP z scores: low SEP = deciles 1 to 3; mid SEP = deciles 4 to 7; and high SEP = deciles 8 to 10. We then took random samples from the intervention and control arms for each SEP group in the POI trial, using bootstrapping with replacement. An intervention effect was calculated for each pair of intervention and control samples by taking the mean difference in BMI at age 5 years. These effects were randomly applied to individual participants in the corresponding SEP group of the LSAC K cohort. Similarly, random samples of intervention costs were taken from each SEP group in the sleep arm of the trial and applied to the corresponding SEP group of the LSAC cohort.

TABLE 1.

Demographic and anthropometric characteristics of participants in the POI RCT ^a and the simulation input cohorts at age 5 years

	Low SEP		Mid SEP		High SEP
	Control	Intervention	Control	Intervention	Control	Intervention
POI RCT
Sample size	39	43	93	84	74	65
Age (y), mean (SD)	4.99 (0.06)	5.00 (0.07)	4.98 (0.05)	4.99 (0.06)	4.99 (0.04)	4.99 (0.07)
Female, n (%)	21 (54)	19 (44)	51 (55)	36 (43)	37 (50)	27 (42)
BMI (kg/m2), mean (SD) b	15.95 (1.04)	15.74 (1.19)	16.07 (1.37)	15.81 (1.39)	15.94 (1.35)	15.76 (1.22)
Difference in BMI (95% CI)	0.22 (−0.27 to 0.71)		0.27 (−0.15 to 0.68)		0.18 (−0.25 to 0.62)
Overweight b, , c , n (%)	7 (18)	4 (9)	26 (28)	20 (24)	16 (22)	13 (20)
Simulation input cohorts
Sample size	1470	1470	1959	1959	1469	1469
Age (y), mean (SD)	4.79 (0.22)	4.79 (0.22)	4.78 (0.22)	4.78 (0.22)	4.79 (0.22)	4.79 (0.22)
Female, n (%)	747 (51)	747 (51)	933 (48)	933 (48)	732 (50)	732 (50)
BMI (kg/m2), mean (SD)	16.45 (1.73)	16.23 (1.74)	16.29 (1.62)	16.02 (1.63)	16.17 (1.48)	15.98 (1.50)
Overweight c , n (%)	550 (37)	477 (32)	641 (33)	521 (27)	423 (29)	370 (25)

Abbreviations: POI, Prevention of Overweight in Infancy; RCT, randomized controlled trial; SEP, socioeconomic position.

^a Three participants in the control group had missing data on the New Zealand Deprivation Index 2013 and, therefore, could not be categorized as high or low SEP. These participants have been excluded from analysis.

^b Including 159 imputed BMI values.

^c Including obesity. Determined from age, sex, and BMI using World Health Organization growth standards.

Running the model

We then ran the EQ‐EPOCH model for a “control” cohort, in which no intervention effects and costs had been applied, and for an “intervention” cohort, with intervention effects applied, to predict BMI for each child every year until age 17 years. The model equations for “low SEP” were used for children with an SEP z score under 0, and those for “high SEP” were used for children with an SEP z score higher than or equal to 0.

Base case analysis

Incremental cost‐effectiveness ratios (ICERs) were calculated for low‐, mid‐, and high‐SEP groups. For the cost‐effectiveness analysis, the incremental cost per BMI unit avoided at age 17 years, as predicted by the EQ‐EPOCH model, was calculated. For the cost‐utility analysis, an incremental cost per QALY saved was calculated by summing the predicted utilities over the 12 years simulated. We assumed that there were no differences in QALYs among weight‐status groups before the age of 5 years based on evidence indicating little effect on QALYs from excess weight in young children [35]. We also assumed that utilities for overweight were not SEP‐dependent, as suggested by our recent study [36].

Characterizing uncertainty and sensitivity analyses

To estimate uncertainty in the costs and outcomes, we took 1000 samples of the modeled data sets using bootstrapping with replacement. We estimated 95% confidence bounds of costs, BMI, QALYs, and ICERs by taking the 2.5th and 97.5th percentiles of these bootstrapped values. The bootstrapped samples were also used to estimate the probability that the intervention, in each SEP group, was cost‐effective under a range of ICER thresholds.

Sensitivity analyses were conducted to test the potential impact of assumptions made about weight‐specific utilities, health care costs, BMI gain, and discount rates in the base case analysis (online Supporting Information).

RESULTS

The distribution of most maternal and family characteristics across the POI trial arms was similar among the three SEP groups, indicating that the randomization remained intact after stratification (Table 1, Supporting Information Table A2a). At age 5 years, the difference in mean BMI among arms (i.e., the effect size) was smallest in the high‐SEP group (0.18 kg/m²) and highest in the mid‐SEP group (0.27 kg/m²), but the 95% confidence intervals (CI) were overlapped (Table 1). The characteristics of the simulation input control and intervention cohorts are presented in Table 1 and Supporting Information Table A2b. At age 5 years, these characteristics across SEP groups were generally similar to the POI trial participants, although mean BMI and prevalence of overweight were greater in the input cohorts for all SEP groups. In the input control cohort, the prevalence of overweight reduced with increasing SEP (Table 1).

In internal validation of the EQ‐EPOCH model, the simulated BMI trajectories predicted the actual BMI trajectories very closely in all SEP groups (Figure 1). Simulated mean BMI values lay within the 95% CI of most of the actual mean BMI values. Our model reflects known trends that children at lower SEP have steeper trajectories than those at higher SEP [37], even if they start with the same BMI.

FIGURE 1.

Internal validation of the EQuity‐informative Early Prevention of Obesity in Childhood (EQ‐EPOCH) model. Modeled and actual BMI trajectories from Longitudinal Study of Australian Children, by SEP, are presented with their 95% CI. SEP, socioeconomic position

Mean intervention costs in the POI trial were highest in the mid‐SEP group ($198) and similar across the high‐ ($171) and low‐SEP ($179) groups (Table 2). These differences were owing to greater uptake in the personalized intervention components and the more intensive options in the mid‐SEP group compared with the other groups. The greatest offsets in health care costs were in the mid‐SEP group ($55), which meant that the overall incremental costs were similar across all SEP groups (Table 3). At age 17 years, incremental BMI and incremental QALYs for the sleep intervention compared with usual care were greatest in mid‐SEP group and lowest in the high‐SEP group (Table 3).

TABLE 2.

Sleep intervention costs per child by SEP group in 2018 Australian dollars

Item		Low SEP (n = 43)		Mid SEP (n = 84)		High SEP (n = 65)
	Cost of item per use ($) a	Number of times used a , b	Total cost for item (% of total cost for group)	Number of times used a , b	Total cost for item (% of total cost for group)	Number of times used a , b	Total cost for item (% of total cost for group)
Standard intervention
Staff training costs (establishment costs)	NA	NA	642 (5)	NA	1255 (5)	NA	971 (5)
Equipment and overhead cost, including printed educational materials, telephone calls, and venue hire (ongoing costs)	NA	NA	376 (3)	NA	734 (3)	NA	568 (3)
Standard group intervention cost (ongoing costs)	83.94	43	3609 (29)	84	7051 (27)	65	5456 (30)
Subtotal			4628 (37)	84	9040 (35)	65	6995 (38)
Personalized intervention (ongoing costs)
Level 1	137.86	13	1792 (14)	16	2206 (9)	12	1654 (9)
Level 2	107.44	3	322 (3)	10	1074 (4)	1	107 (1)
Level 3	182.36	1	182 (1)	5	912 (4)	0	0 (0)
Level 4	408.08	2	816 (7)	9	3673 (14)	6	2448 (13)
Mean personalized intervention cost per child		72.4		93.63		64.77
Total intervention cost (undiscounted)		7741		16,905		11,206
Total intervention cost (discounted at 5%)		7678		16,635		11,084
Mean intervention cost per child (discounted at 5%)		178.55		198.04		170.52

Abbreviations: NA, not applicable; SEP, socioeconomic position.

^a Cost of item per use and number of times used are provided where relevant. As staff training costs and equipment and overhead costs were not incurred at the level of use by an individual trial participant, only the total cost attributed to the SEP group is provided.

^b The standard intervention was provided to all participants allocated to the sleep intervention group. The personalized components of the intervention were provided optionally after the standard intervention was complete, and participants could access the personalized intervention levels multiple times until the child was 2 years old.

TABLE 3.

Incremental costs, outcomes, and cost‐effectiveness ratios by SEP at age 17 years

	Low SEP (n = 1470)	Mid SEP (n = 1959)	High SEP (n = 1469)
Incremental health care costs, mean (BS 95% CI)	−44 (−73 to −13)	−55 (−82 to −32)	−31 (−58 to −5)
Incremental costs, mean (BS 95% CI)	140 (110 to 173)	141 (114 to 166)	143 (115 to 172)
Incremental BMI (discounted; BS 95% CI)	−0.151 (−0.232 to −0.064)	−0.175 (−0.245 to −0.111)	−0.110 (−0.169 to −0.049)
Incremental QALYs (BS 95% CI)	0.006 (0.002 to 0.010)	0.008 (0.005 to 0.011)	0.004 (0.001 to 0.008)
ICER (A$ per unit BMI avoided; BS 95% CI)	927 (493 to 2687)	806 (485 to 1451)	1308 (688 to 3374)
ICER (A$ per QALY gained; BS 95% CI)	23,010 (10,689 to 88,877)	18,206 (10,150 to 35,973)	31,981 (13,577 to 188,433)
Probability cost‐effective at $50,000/QALY threshold	91.7%	99.6%	78.5%

Note: All incremental costs and outcomes were calculated as intervention minus control.

Abbreviations: A$, Australian dollar; BS, bootstrapped; ICER, incremental cost‐effectiveness ratio; QALY, quality‐adjusted life year; SEP, socioeconomic position.

The cost‐effectiveness and cost‐utility ICERs indicated that the sleep intervention was most cost‐effective in the mid‐SEP group ($806 per BMI unit avoided/$18,206 per QALY gained) and least cost‐effective in the high‐SEP group ($1308 per BMI unit avoided/$31,981 per QALY gained; Table 3). When accounting for joint uncertainty in outcomes (BMI and QALYs) and costs, almost all bootstrapped samples in all SEP groups lay within the top‐right quadrant of the cost‐effectiveness planes (Figures 2A and 3A), indicating that the intervention was more costly and more effective than the usual care group. The cost‐effectiveness acceptability curves (Figures 2B and 3B) indicate that the sleep intervention has the highest probability of cost‐effectiveness in the mid‐SEP group at all cost‐effectiveness thresholds. At the $50,000 per QALY threshold, commonly used in Australia [38], the intervention had a 99.6% probability of being cost‐effective in the mid‐SEP group, 91.7% probability in the low‐SEP group, and 78.5% probability in the high‐SEP group. The probability of cost‐effectiveness remained moderate to high for all SEP groups under a range of cost‐effectiveness thresholds, including that identified from a more recent estimation of the marginal productivity of the Australian health system (~$28,000 per QALY gained; Figure 3B) [39].

FIGURE 2.

BMI outcomes at age 17 years. (A) Cost‐effectiveness plane presenting incremental costs and BMI for each bootstrapped sample, as well as the point estimate from the primary sample, for each SEP group. (B) Cost‐effectiveness acceptability curves for each SEP group presenting the probability of cost‐effectiveness under a range of willingness to pay thresholds. AUD, Australian dollar; SEP, socioeconomic position

FIGURE 3.

QALY outcomes at age 17 years. (A) Cost‐effectiveness plane presenting incremental costs and QALYs for each bootstrapped sample, as well as the point estimate from the primary sample, for each SEP group. (B) Cost‐effectiveness acceptability curves for each SEP group presenting the probability of cost‐effectiveness under a range of willingness to pay thresholds. AUD, Australian dollar; QALY, quality‐adjusted life year; SEP, socioeconomic position

Sensitivity analyses are presented in Figure 4 and online Supporting Information. The ICERs were most sensitive to changes in the utilities assigned to each weight status and surpassed the $50,000 per QALY gained threshold at the lower bound of the 95% CI for the disutility of overweight for all three SEP groups. ICERs were minimally sensitive to changes in BMI growth, assumptions regarding health care costs, and discount rates. In all sensitivity analyses conducted, the cost‐effectiveness and cost‐utility were consistently highest in the mid‐SEP group and poorest in the high‐SEP group.

FIGURE 4.

Sensitivity analyses. Tornado plots presenting the range of cost‐effectiveness and cost‐utility ICERs using alternative analysis parameters. Further description is provided in online Supporting Information. AUD, Australian dollar; HW, healthy weight; ICER, incremental cost‐effectiveness ratio; OW, overweight; QALY, quality‐adjusted life year; SEP, socioeconomic position

DISCUSSION

This economic evaluation of a sleep intervention to prevent childhood overweight investigated a policy‐relevant but rarely examined issue: whether the cost‐effectiveness of the intervention is affected by SEP. We developed a fit‐for‐purpose health economic model that accurately predicts BMI trajectories, with children at a lower SEP gaining BMI faster than those at a higher SEP. Simulations identified differences in cost‐effectiveness across socioeconomic groups; although the intervention had a moderate‐to‐high cost‐effectiveness in all groups, it had considerably better cost‐effectiveness in the mid‐ and low‐SEP groups compared with the high‐SEP group. When accounting for uncertainty, the intervention had a very high probability of cost‐effectiveness in the low‐ and mid‐SEP groups and a moderate‐to‐high probability in the high‐SEP group.

This is one of very few SEP‐specific economic evaluations of childhood obesity interventions and the only one, to our knowledge, targeted to early childhood specifically. One study of a policy targeted at Australian children of all ages (not just early childhood) considered the impact of a hypothetical ban on television advertising of unhealthy food and beverages and found that the ban was likely to be more cost‐effective in the lower‐SEP groups [14]. In another study, the potential impact of a sugar‐sweetened beverage tax on the whole Australian population (children and adults) was examined and, again, was found to be more cost‐effective for the low‐SEP group [13]. Another study of a hypothetical mass media campaign to promote weight‐loss mobile applications in New Zealand predicted greater cost‐effectiveness in the indigenous (Māori) versus nonindigenous groups [15]. Each of these studies estimated the intervention effect in population subgroups based on distributions of health needs and behaviors rather than measuring these effect sizes from intervention trials. One other study of two school‐based behavioral interventions in the Netherlands derived intervention effects from a nonrandomized trial and also identified higher cost‐effectiveness in the lower‐SEP groups [40]. By contrast, our evaluation is data informed and based on stronger evidence than the earlier studies, which more heavily relied on assumptions. These studies also modeled outcomes over a lifetime horizon, with costs and benefits only incurred in adulthood, whereas our study used a medium‐term time horizon and counted costs and benefits accrued during childhood and adolescence.

The differential cost‐effectiveness across the SEP groups was likely driven by a combination of the differences in the starting BMI distributions, BMI trajectories, and BMI effect size at age 5 years. In the simulation control cohort, the prevalence of overweight in 5‐year‐old children was higher with a lower SEP pre‐intervention (Table 1). The model simulates faster BMI growth at a lower SEP, reflecting actual trends [37], and the intervention effect size was larger in the mid‐ and low‐SEP groups compared with the high‐SEP group. Together, these differences across SEP groups resulted in higher modeled effectiveness by age 17 years in mid‐ and low‐SEP groups and, therefore, better cost‐effectiveness.

The sensitivity analyses showed that the results were most sensitive to uncertainty in the weight status‐specific utilities used in our model. This is unsurprising in the context of the known difficulties in measuring health‐related quality of life (HRQoL) in children [41]. Being an active area of research, progress in this work may reduce uncertainty in weight status‐specific utilities.

The findings of this study bear important implications for decisions on how best to target the sleep intervention. As the intervention is highly cost‐effective in the low‐ and mid‐SEP groups, and the needs of the low‐SEP group are the highest (having the highest rates of overweight and obesity), an approach that targets the intervention to low‐SEP groups would satisfy both efficiency and equity objectives. The decision‐maker would not need to make any trade‐offs between these goals.

By showing that a targeted implementation approach could be both equitable and efficient, this study is one of several ways in which an economic evaluation can provide equity‐relevant information [42]. Another technique, distributional cost‐effectiveness analysis [43], could be applied to the sleep intervention in future work to assess its impact on the distribution of costs and effects across a population.

A major strength of our analysis is the development, validation, and use of an SEP‐specific BMI trajectory model: the EQ‐EPOCH model. This is one of the first health economic models to predict BMI trajectories by SEP. Although a handful of models have incorporated SEP, they have either been cohort‐based [13] or have predicted weight status rather than BMI [44]. Predicting BMI allows for nuanced analyses in which cost per unit BMI avoided can be calculated and compared with various other evaluations using the same outcome. The model was developed using a nationally representative data set and was validated before use. The economic components of the model (utilities and health care costs) were informed by the best available evidence: utility decrements by weight status were derived from a systematic review and meta‐analysis, and health care costs were estimated from national level data on health care expenditure.

Furthermore, a microsimulation model is well suited to exploring effect variation across SEP groups. This feature captures heterogeneity in BMI within and across each SEP group as well as individual‐level uncertainty in modeled outcomes. Our evaluation is also unique in using a midlength time horizon (17 years), which is particularly valuable for funding decisions that account for important health and economic benefits of obesity prevention within a relevant time frame for the policy maker. Finally, we conducted sensitivity analyses on uncertain parameters in our model and evaluation methods, including utilities, health care costs, annual BMI gain, and discounting rates, providing confidence in the main finding that the intervention's cost‐effectiveness was affected by SEP.

However, there are some limitations to our analysis, some of which are in common with the limitations of the unstratified economic evaluation of the sleep intervention [18]. First, the effectiveness results come from a relatively small trial conducted 12 years ago, in one country context. Moreover, input parameters of the model were derived from data collected in different years. Therefore, further work is needed to determine the generalizability of our findings to the current context and to other contexts with different demographic profiles and health care systems. Furthermore, the trial was not powered to detect effects within SEP groups, and, consequently, the full extent of differences in effect sizes across SEP groups may not have been captured. Another limitation is that we have applied clinical trial results collected in New Zealand to an Australian population and model with the assumption that similar intervention effects could be achieved. Furthermore, as our analysis models HRQoL impacts from BMI effects, we have not been able to account for the impact of the intervention itself on the HRQoL of the child or parents. As improvements in sleep are likely to increase HRQoL in both the child and parents, our analysis has likely underestimated HRQoL impacts and overestimated the cost‐utility ICER. We have also assumed that the classification of SEP groups based on the NZDep13 corresponds to the classification using the SEP z score available in LSAC. Finally, there is the inherent assumption of the EQ‐EPOCH model that starting BMI, sex, age, and SEP are sufficient predictors of annual BMI gain when there are other known risk factors for child overweight and obesity [45]. The purpose of this model, however, is for sample‐ or population‐level economic evaluations, and it has been validated for this use.

CONCLUSION

Our study provides economic evidence to support policy making that addresses health inequities. By demonstrating that the cost‐effectiveness of an early‐childhood sleep intervention is dependent on SEP, we have shown that a trade‐off between efficiency and equity would not need to be made if choosing to target low‐SEP groups. To promote policy approaches that address inequities, analyses such as these should be conducted routinely for candidate interventions. Further work is needed in the reporting of intervention effect sizes by SEP, health care usage by SEP, and the collection of intervention resource use and SEP measures at an individual level to allow for routine stratified economic evaluations of interventions. This, in turn, should facilitate the choice of appropriate interventions for the right contexts and progress attempts at tackling inequities in childhood obesity.

AUTHOR CONTRIBUTIONS

Anagha Killedar conducted the analyses and wrote the initial draft of the manuscript. Anagha Killedar, Alison Hayes, Thomas Lung, and Rachael W. Taylor contributed to the study design, and all authors interpreted analyses and made revisions to the manuscript drafts.

FUNDING INFORMATION

This study was funded by the National Health and Medical Research Council (NHMRC) Center of Research Excellence in Early Prevention of Obesity in Childhood (EPOCH CRE) (APP1101675). Anagha Killedar was supported by the NHMRC Postgraduate Scholarship (APP1169039). Thomas Lung is supported by an NHMRC Early Career Fellowship (APP1141392) and a Heart Foundation Postdoctoral Fellowship (award ID 101956). Rachael W. Taylor is supported by the Karitane Fellowship in Early Childhood Obesity. None of the funding bodies had any input into the design and conduct of the study.

CONFLICT OF INTEREST

The authors declared no conflict of interest.

CLINICAL TRIAL REGISTRATION

ClinicalTrials.gov identifier NCT00892983.

Supporting information

APPENDIX S1 Supporting Information

Killedar A, Lung T, Taylor RW, Taylor BJ, Hayes A. Is the cost‐effectiveness of an early‐childhood sleep intervention to prevent obesity affected by socioeconomic position? Obesity (Silver Spring). 2023;31(1):192‐202. doi: 10.1002/oby.23592