Understanding the impact funding cuts on Environmental and regulatory services and gastrointestinal infections: a longitudinal ecological study

Patient and Public Involvement

This project will incorporate public and patient involvement and engagement (PPIE) throughout. To date, a PPIE panel has been held, in which preliminary plans were presented and variables were discussed. Input was used in deliberation of variables of analysis. The PPIE group will be involved in the development of the protocol in addition to steps following this throughout the remainder of the study.

Study aim and objectives

The aim of the study will be to understand the impact of changes to ER service expenditure on ER services indicators, and how this may impact GI infection.

The main objectives of the study are to:

Study setting and location

The location of the study is England. The data collected will only pertain to LAs in England and the populations of these LAs. There are a total of 314 lower tier LAs in England (April 2020 geography). Local authorities are responsible for food safety at the district and single tier level¹⁴. The area of the LAs involved range from 5020 km² to 12 km² and populations range from 1,142,494 to 41,381 persons.

Study overview and design

The study will be an observational study, it will use a longitudinal ecological study design, looking at data from 2010/11 to 2018/19. The data used will be secondary data and will be aggregated to LA level. The use of an ecological study design is pragmatic and will allow us to analyse the data at aggregated local authority level, over time each year. The study will be quantitative and use spatial and statistical analysis throughout. Furthermore, the study will be split over several results sections, each section focusing on different the objectives as stated. We have constructed a Directed Acyclic Graph (DAG) to clarify the study aims and objectives (Figure 1). The DAG allows us conceptualise theories, assumptions and bias in causal pathways between exposures and outcomes, which relate to the study objectives. Additionally, we specify a-priori the minimal set of covariates required for study in the DAG and in the next section of the protocol.

Figure 1. Directed Acyclic Graph (DAG), showing causal relationships between the exposure Environmental and regulatory service expenditure, and Environmental and regulatory service outcomes and gastrointestinal infection outcomes, also showing interactions with possible observed and unobserved confounders.

Study data

ER expenditure

Indicator(s):

Local authority expenditure data is published by the Local Government Finance - Data Collection Analysis and Accountancy division of the Department for Communities and Local Government¹⁵. This data describes the spending for Environmental and regulatory services, this is also known as the revenue outturn data (RO5). The data we will use throughout the project are data that have been prepared by the Place Based Longitudinal Data Resource (PLDR) to provide consistent measures at the LA 2020 geography over time.

ER services

Indicator(s):

The Consumer Data Research Centre (CDRC) provides information on the food hygiene rating result given to customer-facing food establishments participating on the Food Standard Agency’s (FSA) food hygiene rating scheme (FHRS). The data are held on behalf of local authorities. This data will be used to provide the food hygiene rating score which indicated the compliance level on a food establishment.

The FSA collate, analyse and publish data collected by LA food hygiene enforcement activities annually as local authority enforcement monitoring system (LAEMS) returns. This data will be used to derive ER indicators; the number of FTE positions, percentage of interventions achieved, the percentage of interventions achieved at unrated establishments, and proportion of enforcements.

GI infection indicators

Indicator(s):

Hospital admissions due to GI infections are reported in hospital episode statistics (HES) which provide inpatient admissions for National Health Service (NHS) hospitals in England. The study will measure hospitalisations in England and will include cases defined using ICD-10 codes (A00–A09, K52.9) for all-cause acute gastroenteritis episodes (AGE).

The NHS telephone help line NHS 111, superseded by NHS Direct, allows people to access medical advice over the phone instead of in person consultations. The study will use data on the number of calls to NHS 111 reporting diarrhoea and vomiting symptoms in England.

The study will use laboratory pathogen data from Second Generation Surveillance System (SGSS) provided by the United Kingdom Health Security Agency (UKHSA), formerly Public Health England (PHE) which provide routinely collected national and regional surveillance data. The SGSS is the primary collection process by which data on positive cases of clinical significance and antimicrobial resistance (AMR) in England are recorded. Data requested from UKHSA SGSS will include positive laboratory detections: Salmonella non-typhi, Campylobacter, E. coli, Clostridium perfringens, and Listeria. These pathogens have been selected through an expert group of health professionals affiliated to the NIHR Health Protection Research Unit in Gastrointestinal Infections.

The study will also request data from HPZone at UKHSA, a national system that monitors confirmed and suspected outbreaks of food poisoning, gastroenteritis, haemolytic uraemic syndrome and infectious bloody diarrhoea¹⁶.

In addition, we will access data from EPINorth3, to provide detailed data on notifiable infections reported in the North East of England. The data will be linked to detailed information regarding exposures prior to infection by surveillance questionnaires and cover pathogens such as Salmonella, Shiga toxin producing E. coli.

Community demographics

The measure of deprivation used will be the English Indices of Multiple Deprivation (IMD), which is produced by the UK Department for Communities and Local Government using local administrative data and data from the census¹⁷. The IMD measure the relative levels of deprivation in England based on seven different bases of deprivation, these include: income deprivation, employment deprivation, education, skills and training deprivation, health deprivation and disability, crime, barriers to housing and services, and living environment deprivation¹⁸. Population density data will be calculated using midyear population estimates and the area of LA’s in square kilometres provided by Office for National Statistics (ONS). As a proxy for accessibility to healthcare we explore the utility of data on distance from GP surgery.

We will use unemployment rate data, represented as the percentage of the economically active population aged 16 and above. In addition we will use gross domestic household income (GDHI), this measure details the amount of money that the population in the household sector have available to spend or save after income distribution measures. These data will be accessed from ONS and used to control for overall economic growth in each LA.

The data will cover years 2010/11 to 2018/19, and all data will be aggregated to the LA level. All outcome measures will be linked to a geocode which is then matched to the corresponding LA. Summary information on data, the source of the data and the corresponding measures can be found in Table 1.

Software

This project will use R studio version 4.3.0 for all data analysis.

Ethical considerations

Ethical approval has been sought for this research; the University of Liverpool Ethics committee confirmed ethical approval will not be required. All data will be aggregated and anonymised therefore only data sharing agreements will be required.

Data analysis

Once collected the data will be cleaned and aggregated to LA level, and variables will be derived. This will produce annual (repeated) local authority level measures. Exploratory and descriptive analysis will be carried out on the variables. Spatial mapping approaches will be used to visualise and explore geographically referenced data, to determine preliminary evidence supporting or contradicting a priori hypotheses. Missing data will be examined and dealt with where appropriate, using techniques such as imputation or elimination depending on the nature, extent, and reasons for missing data. Re sampling techniques may be used by incorporating a Bootstrap approach for multiple imputation purposes. Robustness of the variables will be assessed, as variables will be checked for collinearity and interdependency. Once the indicators to be carried forward for final analysis are determined, we will then carry out more formal statistical analysis as outlined below.

Changes in spending (objective 1)

Funding data will be analysed spatially and over time, to describe any changes. Annual expenditure per capita will be analysed between 2010/11 and 2018/19. Following a graphical assessment of trends in spending, a multivariable regression model incorporating random effects for LA’s and/or time as necessary will be fitted to quantify how changes in expenditure over time for LA’s association with explanatory variables such as deprivation, using IMD, and population density and LA type. This will allow us to begin exploration of how expenditure per capita varies in line with socioeconomic and demographic characteristics.

Changes in ER indicators (objective 2)

The ER indicators will be described spatially and overtime. Descriptive analysis will be carried out between the indicators and factors such as deprivation, and population density and food establishment density. We will also aim explore the level of tourism for each LA, as this may create additional pressure to local services of some areas.

Association between changes in spending and changes in ER indicators (objective 3)

To analyse the association between changes ER expenditure and trends in ER indicators we will use a fixed effects, methods which have been previously used in panel models for interrogating LA finance data^19–22.

For the main analysis, we will use fixed effects panel regression, with fixed effects for LAs and individual years, to estimate the mean within-LA relationship between ER spend and ER indicators. We will account for secular trends across all LAs and observed and unobserved time invariant confounders (see Figure 1), including GDHI and unemployment. The fixed-effects approach removes unobserved confounders that vary between local authorities but are constant over time. We will log-transformed the exposure variable to account for expected diminishing returns on investment. Model results are, therefore, interpretable as change in ER indicators relative to the percentage change in spend. Initially we will use linear regression fixed effects models with exposure and outcome expressed as a rate per 100 population. In robustness tests we will use a fixed effects Poisson and linear regression model, including the log of the population as an offset.

Changes in GI infection indicators (objective 4)

The GI indicators will be described over time and spatially in the UK. This will allow the identification of times and areas where prevalence looks potentially high for further statistical investigation. Spatially referenced explanatory variables of interest include measures of deprivation and population density.

Association between changes in ER services and changes in GI infection indicators (objective 5)

We will develop spatiotemporal models to estimate the association between exposures and GI outcomes accounting for socioeconomic, demographic, and environmental factors. We will explore Poisson mixed model approach and explore the spatial clustering of residuals, such as using Moan’s I tests to examine if we need to incorporate a spatial model. As the outcomes are count data, we will utilise Poisson/negative binomial multivariable and multivariate regression models, depending on over-dispersion and zero inflation in the data. We will also evaluate the use of fixed effects for this objective, as this approach has been used to analyse LA spending association with health indicators^19–22. The models will adjust for spatial autocorrelation if we detect its presence in the residuals by using ‘BYM’ models that extent the count regression modelling frameworks.

Methods will be developed, and a full data analysis plan will be established with further exploration and knowledge of the data.

Power calculation

Power estimations are used in reference to sample size estimations, where they determine the number of subjects required to form the sample of the population, in aim to avoid type 2 error²³. The current study will not use sample data, rather it will use population level data. For this reason, a power calculation will not be required. However, sensitivity analysis and tests for robustness of our population data will be carried out to our assumptions.

Timeline

The study will take place over 2 years, beginning in September 2022. Before the study begins, the relevant procedures regarding data sharing agreements and permissions will be actioned.

Project governance

A stakeholder group will be established, consisting of ER workers from different LA’s who work with food hygiene. Initial meetings will be set up to introduce the project and gain relevant information. Throughout the project they will be able to contribute to the project. They will be updated periodically and be able to discuss and offer insights. or long-term secure storage of research data in a format.

The HES data to be used was made available to PLDR by NHS Digital under the data sharing agreement DARS-NIC-16656-D9B5T-v5.2 and has been risk-assessed by PLDR team and complies with HES small number analysis guidance. The data, as with all other data will be anonymised, non-identifiable and aggregated to the LA level. Safeguarded data will only be accessed, analysed, and stored on the university secure network, in line with management and practice guidelines.

Dissemination and research findings

We will produce a report and policy brief of the key implications of our work to be shared with the stakeholder group and more widely to disseminate to UKHSA, public health professionals and policy makers. Findings should inform government and public health policy and strategy. The findings will be presented at professional and scientific conferences. The results will also be published in peer review open access publications.