Impact of cuts to local government spending on Sure Start children’s centres on childhood obesity in England: a longitudinal ecological study

Background Childhood obesity is rising in disadvantaged areas in England. Sure Start children’s centres provide community-based services for children <5 years and their parents, including many services that can support healthy weight, directly or indirectly. Since 2010, austerity-driven cuts to local authority (LA) budgets have led to substantially reduced public expenditure on Sure Start services. We assessed whether childhood obesity prevalence has increased more since 2010 in those areas in England that experienced greater cuts to spending on Sure Start. Methods This longitudinal ecological study covers the period 2010/2011–2017/2018. Our exposure was LA expenditure on Sure Start, using Department for Education data. Our main outcome was LA obesity prevalence at age 4–5 years, using National Child Measurement Programme data. We used fixed-effects panel regression to quantify the association between change in spending and change in the prevalence of childhood obesity. Results Spending on Sure Start children’s centres decreased on average 53% over the study period, with deeper cuts in more deprived LAs. Each 10% spending cut was associated with a 0.34% relative increase in obesity prevalence the following year (95% CI 0.15% to 0.53%). We estimated there were an additional 4575 children with obesity (95% CI 1751 to 7399) and 9174 overweight or obese (95% CI 2689 to 15 660) compared with expected numbers had funding levels been maintained. Conclusions Cuts to spending on Sure Start children’s centres were associated with increased childhood obesity. With deprived areas experiencing bigger spending cuts, reinvesting in these services may, alongside wider benefits for child development, contribute to reducing inequalities in childhood obesity.


SUPPLEMENTARY MATERIAL
Impact of cuts to local government spending on Sure Start Children's Centres on childhood obesity in England: a longitudinal ecological study KE Mason, A Alexiou, D Bennett, C Summerbell, B Barr, D Taylor-Robinson

Directed acyclic graph and description of potential confounders
Our use of fixed-effects regression to estimate the within-area change in obesity associated with changes in spending removes the need to account for time-invariant confounders. We identified three potential time-varying confounders and adjusted our models for these, reflecting the directed acyclic graph in Figure S1. First, local economic conditions will reflect need and also influence the level of revenue from business rates and central government, thus determining how much is available to be spent on children's services, while also influencing child health through the socioeconomic conditions of parents. We use Gross Disposable Household Income (GDHI) as an annual measure of this. [1] Second, the level of child poverty in an area is both an indicator of need for services and a determinant of child health. We use HMRC data on the proportion of children living in low-income families in each year to capture this. Third, spending on other services for children may also confound the association, especially since LA budgets are finite and constrained. For instance, more spending on statutory child protection services leaves less money available for discretionary services such as Sure Start, and spending on other services may influence child obesity directly or indirectly. Figure S1. Directed acyclic graph of the relationship between spending on Sure Start and obesity at age 4-5 years.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)

Linear model rather than Poisson
To test the robustness of our findings to alternative model specifications we repeated the primary analyses using linear rather than Poisson regression models. When we used a linear model, a 10% decrease in spending was associated with an absolute change in obesity prevalence of 0.033 percentage points (95% CI: 0.015, 0.052; Table S1). In the context of the baseline (2010)  The multiplicative interaction detected with prior trend in obesity prevalence persisted on the difference scale.

Alternative lags between exposure and outcome
We also examined the effect of not lagging the outcome, and of lagging the outcome by two years rather than one. In both cases we observed an attenuation of the association between spending and obesity, compared with our primary analysis with a one-year lag. For a 10% cut in spending, we estimated a 0.19% increase in obesity prevalence with no lag in the outcome (95%CI: 0.07%, 0.32%) and a 0.10% increase in obesity prevalence with a two-year lag (95%CI: -0.17%, 0.38%).

Alternative adjustment sets and alternative exposure definition
We estimated unadjusted and partially adjusted models to examine the influence of confounder adjustment on the main results. Excluding any of the potential confounders had only a small impact -all combinations from none to all covariates gave a point estimate somewhere between 0.33% and 0.41%, so our primary analysis errs on the conservative side. We also examined the effect of defining the exposure as total spend rather than per capita spend. The primary result was robust to operationalising spend without adjusting for the number of 0-4 year olds in the local authority. (Table S3) Overweight at reception (including obesity) as outcome In a set of secondary analyses we examined overweight and obesity as the outcome (defined as BMI greater than or equal to the 85 th centile of the growth reference), rather than obesity only. Results were very largely unchanged when expanding the outcome definition to include overweight. Each 10% decrease in spend was associated with a 0.28% relative increase in prevalence of overweight or obesity (Table S4) Figure S2). The negative control again showed the expected null association, and again the trend was stronger in areas where overweight and obesity had been declining prior to 2010, while no effect modification by deprivation was observed.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)