Economic costs and health utility values associated with extremely preterm birth: Evidence from the EPICure2 cohort study

Abstract Background Preterm birth is associated with adverse health and developmental sequelae that impose a burden on finite resources and significant challenges for individuals, families and societies. Objectives To estimate economic outcomes at age 11 associated with extremely preterm birth using evidence from a whole population study (EPICure2 study). Methods The study population comprised a sample of children born at ≤26 completed weeks of gestation during 2006 in England (n = 200) and a comparison group of classmates born at term (n = 143). Societal costs were estimated using parent and teacher reports of service utilisation, and valuations of work losses and additional care costs to families. Utility scores for the Health Utilities Index Mark 2 (HUI2) and Mark 3 (HUI3) were generated using UK and Canadian value sets. Generalised linear regression was used to estimate the impact of extremely preterm birth on societal costs and utility scores. Results Unadjusted mean societal costs that excluded provision of special educational support in mainstream schools during the 11th year after birth were £6536 for the extremely preterm group and £3275 for their classmates, generating a difference of £3262 (95% confidence interval [CI] £1912, £5543). The mean adjusted cost difference was £2916 (95% CI £1609, £4224), including special educational needs provision in mainstream schools increased the adjusted cost difference to £4772 (95% CI £3166, £6378). Compared with birth at term, extremely preterm birth generated mean‐adjusted utility decrements ranging from 0.13 (95% CI 0.09, 0.18) based on the UK HUI2 statistical inference tariff to 0.28 (95% CI 0.18, 0.37) based on the Canadian HUI3 tariff. Conclusions The adverse economic impact of extremely preterm birth persists into late childhood. Further longitudinal studies conducted from multiple perspectives are needed to understand the magnitude, trajectory and underpinning mechanisms of economic outcomes following extremely preterm birth.


| BACKG ROU N D
The adverse health and developmental sequelae arising from preterm birth impose a burden on finite resources and significant challenges for health services, individuals, families and societies. 1,2 Compared with term-born babies, preterm babies are more likely to have neurodevelopmental impairments, such as cerebral palsy, visual disorders, cognitive deficits and learning difficulties, which impact on long-term physical health and development. 3 These challenges extend beyond childhood into adolescence and adulthood. 4 The rate of live preterm births in England and Wales has ranged between 7% and 8% since 2010. 5 Given the inverse association between gestational age at birth and developmental sequelae, this subgroup of infants is at greater risk of adverse outcomes in both early and middle childhood. 6,7 Evidence from both the 1995 and 2006 EPICure cohorts suggests that gestational age at birth was the single most important predictor of survival and neurodevelopmental outcomes for extremely preterm babies born in 2006. 4 Cerebral palsy was present in 14% of survivors at age three whilst neurodevelopmental impairment decreased with increasing gestational age from 45% for those born at 22-23 completed weeks to 20% at 26 completed weeks gestation. 4 Economic outcomes associated with extremely preterm birth include economic costs borne by the health services, other sectors of the economy and families, and preference-based health utility values that act as inputs into cost-utility analyses. [8][9][10][11][12][13][14] A systematic literature review suggests that initial hospitalisation costs alone range between $111,152 and $576,972 on average per infant born at 24 weeks' gestation compared with $930 to $7114 on average per infant born at term (2015 US dollar prices). 15 However, the evidence on the longer-term economic costs associated with extremely preterm birth was sparse.
In the UK, the prevalence of special educational needs (SEN) in children of school-going age remains stable from primary to secondary school in which 12.6% and 11.5% of pupils have SEN support, respectively. 16 These figures suggest SEN support at 11 years could be considered as representative of secondary school outcomes. 16 Using data from the 1995 EPICure birth cohort, we previously estimated that extremely preterm birth was associated with increased public sector costs of £2477 (2006-7 prices) and a decrement in health utility of 0.312 as measured by the Health Utilities Index Mark 3, on average, during the 11th year. In our most recent analysis of the EPICure2 cohort, we found little to no improvement in the rates of neurodevelopmental impairment or low academic attainment at 11 years compared with the 1995 EPICure birth cohort despite over a decade of improvements in neonatal care and survival. 17,18 Using data from the EPICure2 cohort, this study aimed to estimate the impact of extremely preterm birth on economic costs for the public sector, families and society, and on health utility values, for these births in 2006.

| Study population
The data used in this investigation are drawn from a whole popula-

Study question
What are the economic consequences of extremely preterm birth for individuals, families and society during the early to late childhood years?

What is already known
We previously showed that extremely preterm birth is associated with increased public sector costs by £2477

What this study adds
We have expanded our previous estimates of the economic effects of extremely preterm birth in early to late childhood to capture societal costs such as lost productivity and additional care costs to families attributable to the child's health.

K E Y W O R D S
costs, economic, extremely preterm, health utilities

| Estimation of costs
Parents and teachers completed questionnaires about the child's health and utilisation of services over the preceding year (see Appendix S1). Parents also provided details of time-off work and the additional costs borne by families over the preceding year related to the child's health status. Details of the type of school the child attended (mainstream school, special school or mainstream school with special unit attached) were obtained from the study assessment records. For children attending a mainstream school, teachers provided details of SEN provision, including whether the child had an Education, Health and Care Plan.

| Estimation of health utilities
The main parent (usually the mother) completed the 15-item proxyassessed usual health status assessment questionnaire for the Health Utilities Index (HUI), which encompasses both Mark 2 (HUI2) and Mark 3 (HUI3) health status classification systems. 31 The HUI2 was developed specifically for use with children and covers seven attributes: sensation, mobility, emotion, cognition, self-care, pain and fertility, each with three to five levels. 31,32 The HUI3 covers eight attributes: cognition, vision, hearing, speech, ambulation, dexterity, emotion and pain. Function within each attribute is graded on a 5-or 6-point scale ranging from normal function (level 1) to severe impairment (levels 3, 4, 5 or 6). The HUI2 has value sets for the UK and Canada but the HUI3 only has a Canadian value set. We applied UK algorithms 33 for generating HUI2 utilities and Canadian HUI 31,32 algorithms to generate values for HUI2 and HUI3 health utilities.

| Statistical analysis
Baseline characteristics for the extremely preterm children and their term-born classmates were summarised in tables as means and standard deviations for continuous variables and percentages for categorical variables. Comparisons of costs associated with each category of resource use and total public sector and total societal costs were made between the extremely preterm children and their classmates, and between prespecified groups of extremely preterm children of varying gestational age at birth and those born at term. Between-group differences in mean costs were estimated together with 95% confidence intervals (CIs) generated using nonparametric bootstrapping with replacement, based on 1000 replications. 34 For each of the seven attributes of the HUI2 (and eight attributes for HUI3), we compared the proportion of children with suboptimal levels of function (defined as below level 1) between the comparator groups using Fisher's exact test. Between-group differences in health utility values for each instrument were generated together with the associated 95% confidence intervals.
Multivariable generalised linear regressions were fitted assuming gamma distribution and a logarithmic link function for costs and utilities. Covariates in the regression equations included age, sex (male, female), marital status (married, single, cohabiting), race (White, Non-White), Index of Multiple Deprivation score (1st to 4th most deprived vs. 5th to 10th least-deprived decile) and number of smokers in the household (one or more vs. none) at child aged 11.
Covariates were selected based on clinical and epidemiological relevance and an assessment of whether the observed differences in the distribution of baseline characteristics between the two groups were meaningful. Analyses were performed using statistical package R version 4.0.1. 35

| Missing data
Multiple imputation using chained equations with predictive mean matching was used to predict values for missing costs and utilities, assuming data were missing at random. Fifty imputed datasets were generated at the level of the type of service or resource category stratified by gestational age at birth in line with current best practice recommendations. 36,37

| Ethics approval
Written informed consent was obtained from both children and parents prior to participation. Ethical approval was obtained from both the University College London and the University of Leicester Research Ethics Committees.

| Study population
Baseline characteristics of the study population are summarised in Table 1. A total of 343 children were assessed at age 11 of which 200 (58.3%) were born extremely preterm and 143 (41.7%) born at term. In the extremely preterm group, 15 (7.5%) were born at ≤23 completed weeks of gestation, 28 (14%) at 24 weeks, 69 (34.5%) at 25 weeks and 88 (44%) at 26 weeks. Children in the extremely preterm group were comparable to those born at term in age and sex distribution, parental smoking status and IMD score. Because of the adopted sampling strategy, the extremely preterm group was more likely to be of non-white ethnicity (32.8% vs. 15.3%) and to receive SEN support (12.5% vs. 0.0%), but less likely to speak English at home (55% vs. 71.3%).
Resource use questionnaire completion rates varied by the respondent, with greater completion rates for parent questionnaires (range between 73% and 94% of sample) than teacher questionnaires (range between 68% and 78%

| Service utilisation and costs
Utilisation rates were, on average, higher in the extremely preterm group than for their classmates (Table S1).

| Health utility values
The proportion reporting suboptimal levels of function was higher in the extremely preterm group compared with classmates across HUI2 and HUI3 attributes ( Table 3). When the analysis was stratified by gestational age, there were higher proportions of the suboptimal level of function for the extremely preterm children compared with classmates ( Table 3).
The extremely preterm group had lower mean scores on both HUI instruments and associated algorithms than their classmates born at term ( Note: Note that school costs and total societal costs include costs of mainstream education, special schools and special units attached to mainstream schools but excludes provision of special education services in mainstream schools.
a Bias-corrected bootstrap confidence interval; 95% CI refers to 95% confidence interval. b Not possible to estimate a confidence interval.

| Principal findings
We report here the impact of extremely preterm birth (≤26 com-

| Strengths of the study
Our analysis was based on a prospective population-based sample drawn from defined geographic areas of England; hence selection biases are unlikely to represent a major concern. We extended our cost estimates to include direct non-medical costs borne by families and indirect costs associated with lost productivity. The 2006 EPICure data thus provided a more complete picture of the cost of extremely preterm birth to society than the 1995 data, which were restricted to public sector costs only. Our estimates of the utility decrements for the HUI2 and HUI3 can be used to inform cost-utility analyses as they are based on values derived from Canadian and UK general populations, making them applicable for use across a wide range of evaluative studies and settings.

| Limitations of the data
The term-born controls in our study were sampled from the classmates of preterm-born children in mainstream schools only.
Recruiting a term-born classmate for every preterm child in special school would result in a substantially higher proportion of controls with complex special educational needs relative to the general population. We acknowledge that this sampling approach may have resulted in the recruitment of a term-born control group that is slightly healthier than the general population and therefore we may have slightly under-estimated the costs for term-born children. In addition, the response rate was lower for the teacher completed questionnaires. This meant that our data did not fully capture the costs TA B L E 3 Number (%) of children with suboptimal levels of function (below level 1) within attributes of HUI2 and HUI3

| Interpretation
The present analyses included direct non-medical costs borne by families and indirect costs associated with lost productivity and so are not directly comparable to our previous cost estimates based on the 1995 EPICure cohort, 38 which were limited to public sector costs. Furthermore, the extremely preterm group in the 1995 EPICure cohort was limited to children born at ≤25 weeks' gestation.
In an additional analysis aimed at comparing our results with those  Note: HUI2 Canada MAUF = HUI2 utility score generated via the Canadian multiattribute utility function value set for Canada.
HUI3 Canada MAUF = HUI3 utility score generated via the Canadian multiattribute utility function value set for Canada.
HUI2 UK MAUF = HUI2 utility score generated via the UK multiattribute utility function value set.
HUI2 UK MAUF = HUI2 utility score generated via the UK statistical inference value set.
in the perinatal care of mothers and newborn infants with substantial improvement in survival after preterm birth. 17  the EPICure1 cohort. 38 The discrepancy between our estimates of utility decrement based on the HUI2 and HUI3 is a consequence of differences in the populations surveyed to generate the respective HUI2 and HUI3 value sets and the valuation protocols applied.

| CON CLUS IONS
In conclusion, despite improvements in neonatal care in the decadelong period that separate the 1995 and 2006 EPICure cohorts, there is no evidence that the adverse economic impact of extremely preterm birth in the late childhood years has ameliorated. Further longitudinal studies conducted from multiple perspectives are needed to understand the magnitude, trajectory and underpinning mechanisms of economic outcomes following extremely preterm birth.

ACK N OWLED G EM ENTS
We thank all the EPICure participants and families, our Participant Advisory Group and members of the EPICure research team