| Policy and rural accessibility | Measuring potential accessibility | Travel time | |
| Modelling potential accessibility | Simulation models | Algorithm approaches | |
| Cover range models | Conclusion | ||
Introduction
It has long been recognized that geography affects the cost and organization of service delivery. In general, the cost of supplying services is higher per capita in rural than urban areas (Ellis Williams, 1987). This reflects a number of factors including the difficulties of achieving economies of scale in sparse and/or peripheral areas, the duplication of facilities, additional travel costs, the high level of unproductive time spent travelling and the need for additional time for management and networking (Woollett, 1990). With few exceptions (e.g. the sparsity adjustment to the SSA), however, resource allocation systems for public services in England do not incorporate rural dimensions to need or take into account the impact of rurality on the costs of supplying health services.
With the exception of the adjustment made for emergency ambulance services, the English resource allocation system has not adjusted for the additional costs of providing health services in rural areas (Asthana et al, 2002). Concerns have consequently been expressed about the implications of urban-rural differences in supply for accessibility to and utilization of primary and secondary health care. For example, Smith and Ramana's (1998) study of deprivation in south Cambridgeshire showed that although some rural areas experienced equivalent levels of deprivation to urban areas (even using urban biased indicators) they nevertheless encountered problems with regard to the provision of levels of mental health services and accessibility to these. Health care services where Shucksmith et al. (1996b) found access to be particularly problematic were chemists, opticians, family planning and hospitals.
Difficulties of access, isolation and lower levels of service provision in rural areas can actively contribute to health and social care problems for the elderly. However, the provision of social care services also tends to be based on policy guidelines developed for more urban areas (Craig and Manthorpe, 2000). According to a recent study of home care services for the elderly, 1 in 16 people over the age of 65 received home care services in their urban areas. In sparse areas this reduced to 1 in 23 and in super-sparse areas to 1 in 30, almost half the rate in urban areas (Cumbria County Council, 1997).
There are alternative solutions to the problems of service provision in remote areas. These include the provision of multiple services from one fixed site, the deployment of mobile services, multi-skilling of health and social care professionals 21 and the use of ICT to provide remote access to shopping and banking services. However, the coverage of such schemes is, as yet, patchy (Hogg, 2000).
Turning to rural/urban differences in service utilization, several studies have noted a distance decay effect in rates of GP consultations (Haynes and Bentham, 1982; Bentham and Hayes, 1985; Jones et al, 1998). However, the age, gender and socio-economic circumstances of individuals may again have a more profound influence on the uptake of services than distance; ‘elderly poor women living at a distance from services are the most disadvantaged’ (Wiltshire Health Authority, 1997).
Access to acute hospital services (in terms of the numbers of hospital beds per head of population and the distances which potential patients have to travel to hospital) is lower in the most rural Health Authorities (Hale et al, 1996). Hospital referral and admission rates have also been shown to be significantly lower with increasing distance from hospitals (Haynes and Bentham, 1979).
Such studies suggest the need for rural service delivery to be given greater attention on the policy agenda. However, one of the obstacles to developing policies to improve service access in rural areas has been the difficulty of quantifying the problem. The investigation of patterns of utilization is complex, due to the difficulties of disentangling the impact of need, demand and supply factors on use. For example, Doogan et al. (1997) note the difficulty of interpreting the meaning of low GP consultation rates in rural Wales. On the one hand, low attendance could reflect relatively high levels of health. On the other, it could suggest problems of accessibility.
The development of direct morbidity estimates does provide a methodological advance in this respect, as it provides a measure of need that can be distinguished from demand and supply factors that are associated with variations in use. For example, focusing on the use of cardiology services, Gibson et al. (2002) find that the practice populations of rural surgeries make significantly lower use of inpatient services according to need than urban practice populations. In their analysis, fairly simple measures of service accessibility, namely distance from acute hospitals to GP surgeries and population sparsity, are used. There are, however, alternative measures that can be used to this end. In this section, we review existing methods of measuring distance and access, focusing on work conducted in both the academic and commercial sectors to measure and to model service accessibility.
The ultimate aim of work that has been conducted to isolate accessibility effects on patterns of utilization is, of course, to support rural agencies in their attempts to achieve equitable service provision to rural communities. We therefore begin this section by considering how problems of accessibility have been approached within the policy arena.
Policy and rural accessibility
When the market is left to work in rural areas, there is a dearth of provision. Retail outlets are sparsely distributed and there are declining numbers of village shops. Rural transport, never abundant, has retreated even further since the deregulation of the industry and there is a far more restricted choice of jobs than in an urban setting.
The distribution of privately provided services reflects the difficulties of fulfilling basic threshold requirements in rural areas. However, although public service providers face the same difficulties, provision of public goods such as health care has not been primarily influenced by market considerations in Britain. Even if rural areas have lower levels of demand than urban settings, public services still have to be provided, without the economies of scale that can be achieved for urban provision. Thus, for public authorities charged with making services available to all, higher unit costs for lower catchment populations will be unavoidable.
The additional costs associated with rural service provision have been gradually gaining more attention on the policy agenda. For example, groups such as the fire service started their own sparsity group to lobby about the increased costs of maintaining cover in difficult rural areas. There has been formal Home Office acknowledgement of work conducted to demonstrate the sparsity effect on costs of providing rural police services. The current Welsh formula includes a weighting for the additional costs of providing services in sparsely populated areas in relation to community and ambulance services and cash limited General Medical Services. The major review of the distribution of funds for health care in Scotland conducted by the Arbuthnott Working Party also makes explicit the need to adjust for the excess cost of delivering services in remote and rural areas (SHED, 1999a; 1999b).
In England, the problems faced by rural health service providers have not been strongly acknowledged (White, 2001; Asthana et al., 2002). The English Formula for Hospital and Community Services does apply a weighting for sparsity to expenditure on the emergency ambulance service, but this only represents less than 2% of the HCHS budget. Rurality does not emerge as a focus for inquiry in the Acheson Report on Inequalities in Health. Similarly, when the authors of a recent NHS scoping exercise 22 on access to health care were asked about their views on problems of access to health care in rural areas, they admitted that they not identified this as a potential theme for research and development funding.
Rural access is identified as an issue in the Rural White Paper 23 that was published in November 2000. This highlights the publication of a 'Rural Service Standard' setting out what 'rural users' can expect. Although it acknowledges that 'we don't have firm access standards', it promises an annual independent audit so that problems can be identified and targets set. National standards for library services and less frequent attendance at job centres for unemployed people in rural areas are envisaged. With regards to response time targets, the only new target is the requirement for ambulances to respond to 75% of life threatening calls within 8 minutes by March 2001 and other 999 calls within 19 minutes in 95% of cases. Reviewing the White Paper, Hale (2001) concludes that the proposals are unlikely to do much to narrow the gap between levels of service provision in urban and rural areas.
As we have argued above, this partly reflects the difficulties that have been encountered in empirically demonstrating the impact of accessibility on service use. The use of direct needs estimates overcomes one methodological problem, namely the difficulty of disentangling the effects of need from socio-economic status, accessibility, etc., on patterns of utilization. However, a series of other methodological problems affect the investigation of rural access. For example, variable quality of provision in rural areas has received scant attention (Watt et al., 1994). The impact of a relative mix of primary, secondary and community care is another area where further research could be usefully undertaken. It is possible that lower than expected rates of hospital care are countered by higher levels of primary and community management in rural areas. Finally, there is the issue of how to measure accessibility itself.
Measuring potential accessibility
In the literature on health service access and utilisation, the distinction is sometimes drawn between potential and effective accessibility. The former refers to the physical availability of services. What makes people actually use services is the combined impact of a number of factors - physical service availability is just one of these. Effective accessibility therefore refers to the actual utilisation of services.
The investigation of utilisation is of key interest to service providers who are seeking to monitor equity of access. Nevertheless, it is generally accepted that, because utilisation is influenced by factors that lie outside the control of service providers, potential accessibility is a more realistic and legitimate target for policy intervention. We therefore examine literature that has focused on the development and application of methods to measure and model potential accessibility.
One of the simplest ways of capturing potential accessibility is to examine the location of rural services in relation to settlement type. For example, the Rural Development Commission (now called the Countryside Agency) has undertaken a survey of English parishes every four years. Until recently, the approach taken was to report on the percentage of parishes comprising particular services. In 1994, for example, only 1% of rural parishes housed hospitals with Accident and Emergency and outpatient facilities, 7% a public nursery, 8% homes for the mentally or physically disabled, 17% a GP practice based in the parish, 21% a pharmacy of any kind, 24% residential and nursing homes for the elderly, 25% a daily bus service, 30% a general store, 50% a school and 57% a post office (RDC, 1995).
The RDC surveys clearly demonstrate the lack of public and private services in rural parishes. However, the problem with this approach is that it takes no account of services that may be accessible to parishioners but that are located beyond the physical boundaries of a parish. The Countryside Agency is therefore now exploring the use of postcoded sources of information in order to examine aggregate accessibility to services by households in rural parishes. To this end, it is using data from the Royal Mail's Address Manager database and postcoded addresses in the Yellow Pages.
Rather than counting services within the arbitrary areal unit of a parish, a grid approach is being used to analyse service distribution. Residents and services are allocated to hectare grid cells and a surrounding buffer zone ensures that there are no distorting edge effects. The key to the allocation process is the unit postcode of each service outlet and the number of delivery points within each postcode contained within the Postcode Address File.
Once dwellings and services are allocated to a fine grid, distances are calculated between dwellings and services, and the number of service outlets available to each dwelling in each cell within a selected distance is calculated. The results can be converted into more conventionally used reporting units by measuring the numbers of residents in each area, such as a parish, or the proportions of residents in that area who have specified levels of potential access to a given number of service outlets.
This approach has been used extensively in the 2001 Parish Survey and provides a very different picture to the previous Rural Survey method of potential service accessibility. For example, only 46% of parishes in Lincolnshire have a post office whereas 89% of households within the county are within 2 km of a post office. Although the use of distance offers clear advantages over the 'container' view of space, straight line distance does not equate to travel time distance to rural services. This will vary according to the class of road. Straight line distance will also misrepresent access in areas where geographical features such as rivers and estuaries act as barriers to direct access.
Travel time has been incorporated as a measure of access in a number of studies, notably in part of the NHS Resource Allocation Review commissioned by the Welsh Assembly. This used ED level census data to estimate the numbers of people resident in areas more than thirty minutes travel time from a hospital and fifteen minutes travel time from the general practice. In Wales, 57,944 people (2.05% of the total population) live outside these access times, whilst comparable analysis in Scotland found 1,205,518 people (2.45% of the total population) living in 'under-served areas'.
An interesting focus of a study based in Norfolk (Jones et al., 1998) study has been to examine patterns of service access by both public and private transport. Estimated car travel time to primary care services was calculated using a digitised road network (based on the 1:25,000 Bartholomew map). Road speed estimates were adjusted for six different road types (from a dual carriageway A road to a minor road) and for urban/rural journeys. Thus, average road speeds on an A road dual carriageway in a rural area are given as 54 mph compared to 28 mph in an urban area. Using these measures, travel times from patients' addresses to each facility (in the case of general practices, both the nearest surgery and the practice at which a patient was registered) were calculated. The results suggest that a similar proportion of the population are 'under-served' in terms of access to GP surgeries as in Scotland and Wales. Only 2.5% of the population were outside fifteen minutes car travel time; 8% lived more than ten minutes drive from a surgery; 67% were within five minutes drive of a main or branch surgery and a further 23% between five and ten minutes drive away. Good car access to one facility was associated with good access to other services. For example, of the people who lived more than ten minutes car journey from the nearest GP surgery, 87% were more than ten minutes distant from both a pharmacist and NHS dentist.
The Norfolk project also examined access using public transport. Details of bus services in East Anglia were obtained by examining public timetables and route maps and a buffer of 800m was placed around patients' addresses and facilities to represent an acceptable walking distance for most of the population. Bus services were classified into good, moderate and not useable according to the numbers of return daytime journeys available in weekdays. Community transport schemes were also recorded (e.g. community care schemes, dial-a-ride services) and a GIS package, Arc Info, used to calculate travel times using public transport to the nearest GP practice, pharmacy or NHS dentist.
For the majority of Norfolk's residents (82%), there was a reasonably frequent daytime bus service to a GP surgery (four or more daytime return journeys every weekday). Five percent of the population could reach a GP surgery using a less frequent bus service (1–3 return journeys per weekday), whilst 13% lived in areas with no return daytime services to a surgery. The latter group was also unable to access dental and pharmacy services. Some of these, of course, will live within a reasonable car travel time (though not all will have access to a car). However, the study estimates that 5% of the population live more than ten minutes car journey and have no useable bus service to a surgery.
The Norfolk study provides just one example of a growing number of attempts to use travel time rather than conventional straight line distance to measure potential service accessibility. Most rely on existing quantitative databases, although questionnaire surveys have also been used to assess patient travel times to GP services (e.g. Saunders, 1998). Operational Research in Health (ORH) based in Reading has produced software that calculates node to node times (snapping postcodes to the nearest node) and can be used to create travel time networks for particular areas. A travel time matrix has also been developed by researchers at the Northern Transport Laboratory at Lancaster University as part of a review of local authority resource allocation. This links varying levels of service delivery to population centres of varying size. For example, a town comprising 20,000 residents would be expected to have a community hospital and a library, whilst a town with a population of 100,000 would have a District General Hospital and a local government function. One of the difficulties encountered in this research has been the fact that thresholds such as these do not apply to all parts of the country. In peripheral areas, for example, more facilities tend to be concentrated in smaller settlements.
Modelling potential accessibility
The modelling of potential accessibility involves the quantification of the relationship between demand, provision and 'standards' achieved. A good access model for a rural area should also work in an urban area, not least so that comparisons can be made between areas that are more or less sparse.
Access models also need to be appropriate for different types of services. Broad distinctions can be drawn between services that are delivered to the client and services that the client travels to. Services also differ according to the extent to which daily activity is predictable or unpredictable. For example, the emergency services provide an example of services that are delivered to clients which have unpredictable daily activity. The numbers of patients accessing open GP surgeries are also unpredictable but in this case the client travels to the service. Examples of predictable daily activity would be home care services and refuse collection (where the service is delivered to the client) and day care (where the client travels to a facility).
In the following sub-sections, three approaches to the modelling of service accessibility (predictable and unpredictable, service to client and client to service) are described. These include simulation models, algorithm approaches and cover range models.
Unpredictable service delivery to clients is the most challenging service type to model. For this type, simulation models have been a preferred option. A simulation approach could, of course, be used to model access for all of the service types described above. However, it requires significant developmental effort. The data requirements can be onerous and the model outputs need careful analysis and interpretation. Once the basic simulation model has been developed for a service, it can be used to examine a range of issues and has great flexibility. Service use can be simulated on a travel time network in response to incidents, generated to match a projected or actual demand profile, or according to defined parameters about access standards.
Operational Research in Health, based in Reading, has been undertaking simulation modelling for the emergency services for a number of years. One example of their work was a research study undertaken for the Home Office which examined the effect of population sparsity on the costs of providing police services (ORH, 1998). This centred on the development and use of a simulation model that allowed the sparsity effect to be isolated and quantified in a way that could inform the Police Funding Formula.
The key resources modeled in the ORH study were single- and double-manned police vehicles, although a facility was also designed for assigning 'delayed response' calls to foot patrol officers. The response of manned vehicles to demand levels was analysed, both overall and by distinguishing between immediate and delayed calls. Various measures of demand were used, including formula-projected levels for 1998/99, demand levels recorded in HMIC returns for 1997/98, by shift according to the average pattern found in fourteen sample police forces, and by area in proportion to the resident population by ward. As a first stage, a ward model for each force was used, distributing incidents in relation to ward population size and constructing travel times on the basis of neighbouring ward centroid distances. Analysing variations in resource/demand ratios, sparsity was found to have a significant effect on resource use.
Having isolated the sparsity effect, a simulation model was run to examine the resources required by different areas to achieve a common set of response targets (90% of immediate response calls within ten minutes across urban wards and twenty minutes across rural wards, and 90% of delayed response calls within forty minutes of the scheduled time). Due to the size of the geographical area, the more dispersed clustering pattern of calls, the lower potential for inter-station back-up, the longer 'turnaround' times per call, and the greater variability in call rates around the mean (because of the typically low number of calls), rural areas would have to maintain relatively high availability in order to respond to a high proportion of calls within the target time. There was also less flexibility in the more rural forces in determining the optimum balance between single-manned and double-manned deployed vehicles. Therefore the sparsity effect manifested itself principally as additional availability per call. This effect was far more significant than simply increased travel times per call in sparse forces.
Although the research demonstrated a clear sparsity effect, translating these results into a formula change process had to take account of the assumptions behind the simulation approach and the structure of the formula itself. The sparsity effect is caused by two factors (longer travel time and higher level of availability required), the second of which was by far the most dominant. However, given the formula structure and the specified requirement to focus the study on the call response function, the only component amenable to re-allocation was the 'travel to scene' part of call management. A range of options for applying the simulation results to this component were tested with the dual aim of a) using a sparsity indicator that explained a high level of cost variation from the model outputs (R-squared values in excess of 80 per cent) and b) retaining a close link between model outputs and 'travel to scene'. The best sparsity indicator, based on the sparsity research and the simulation modelling results, was found to be the same 'ED sparsity' measure as used at present.
As indicated above, simulation modelling involves considerable resources to set up and run. Thus, whilst it is a useful approach for modelling access to services that are characterised by unpredictable daily activity, other methods may be more efficient for the modelling of predictable daily activity. Tony Hindle and his team at Lancaster University have developed an algorithm approach to this end, termed Simplified Modelling of Spatial Systems or SMOSS. Adapted from earlier work by Christofides and Eilon (1969) and Fernandez et al. (1974), the essence of SMOSS is to devise simple mathematical functions that can yield satisfactory approximations of the travel distances (and times) in a routing situation that would be obtained from the repeated application of more exacting routing algorithms. The aim has been to make feasible (in time and effort) the evaluation of large-scale strategic distributional options, such as facility location, resource requirements and resource allocation.
SMOSS has been applied to a number of service accessibility studies, including the cost of home care and day care service provision (especially for the 'elderly') in England across the urban/rural continuum; health and social service provision in urban/rural areas of Northern Ireland (particularly the cost of providing health visitors, district nurses, social workers, day centres and passenger service vehicles); emergency ambulance provision in Northern Ireland; and facility location in Northern Ireland (particularly of Accident and Emergency services). Thus, the approach can be applied to services that are delivered to clients and to services to which clients travel.
The SMOSS package can cope with geographical data, census data and need indicators from any selected administrative units and models can be built into a simple spreadsheet system. Typically, service co-ordinates and clients are linked to the nearest road nodes associated with enumeration district centroids. Thus, clients are assumed to be randomly located within the area of the relevant ED. Travel times within the SMOSS system are initially calculated using default values for expected speeds on different categories of road. However, a range of options are also available within the package for modifying these assumed speeds on the basis of local knowledge or observed data. A key feature of this package is its flexibility. For example, estimates of service capacity have been adapted to take account of such factors as length of consultation, repeat visiting, service area overlap, hours available for carrying out particular tours of duty and so on.
Using simulation studies (that have typically involved the comparison of SMOSS approximations of travel distances with those obtained from more exact routing applications) as well as empirical work (where travel distances and times have been collected in the field), the results obtained using simplified modelling have been validated. The approach provides results of comparable validity and accuracy to those obtained from more expensive data-intensive techniques such as detailed simulation and scheduling/routing models. SMOSS is also designed to be user-friendly, and public service managers have found the spreadsheet tools to be both useful and easy to interpret.
An example of the application of this approach can be found in a project funded by the County Councils Network of England which investigated the impact of population sparsity on the provision of domiciliary care for the elderly (ORH, 2000). This proposed a simplified model of distance per visit as a function of demand and sparsity (measured as the average area in hectares per population aged 65+). Weightings were obtained by running SMOSS for sample districts that spanned the urban/rural continuum. The validity of the model was tested by comparing the results to those obtained from survey data during the project. Using only the two variables of demand rates and sparsity, the model generated expected distances per visit that compared well with the survey outcomes.
This is one of the simplest approaches to modelling accessibility and, as such, is perhaps most attractive to users untrained in modelling techniques. In essence, incidents are distributed on a travel time network and the range of cover provided by a particular service location is expressed as the percentage of incidents that occur within a specified time.
An adaptation of the cover range approach has been used by ORH to model Accident and Emergency cover. In this, a travel time model was developed to predict isochrones and catchment areas for hospitals and ambulance stations. This incorporated blue light and standard travel times between any two points on a travel time network (based on distance and road type). Link times were verified with actual times achieved under blue light and emergency response conditions.
Cover range models are easily used in-house on spreadsheets, and can be adapted for initial analysis of a number of access problems. With a little knowledge they can be used as a first step to defining a problem for the model types already discussed, or to calculate approximate results for situations where time and resources are limited.
This review of the measurement and modelling of accessibility has uncovered the dominance of a few universities and commercial consultants in work on rural access to health services. The use of GIS within Health Authorities appears to be largely restricted to presentational maps where factors such as deprivation are spatially depicted. There is also a division between the academic and commercial sectors in the focus of work conducted. Academic studies tend to have concentrated on examining who does or does not get a certain level of service, whilst work on service provision, particularly its complex modelling, has been the preserve of a few specialist consultancy companies.
With the introduction of less complex systems such as SMOSS, there is certainly scope for service modelling to be carried out in-house. Routing problems such as the routing of refuse collection are already undertaken by county councils where the application of GIS in diverse areas such as housing, planning and transport has allowed sufficient expertise to develop. A range of commercial packages is available and, as efforts are being made to ensure that such software can be easily integrated with other packages, organisations within the health sector could quite feasibly do more work on issues of access themselves.
Regarding access measures, the use of travel time is generally superseding the use of straight line distance. In order to make studies comparable, however, travel speeds for different road types could be usefully standardised. For example, conventions could be agreed for rush hour and non-rush hour speeds in urban and rural areas. Some agreement on what constitutes a reasonable travel time to a GP surgery or hospital would also be useful. A number of studies have used fifteen minutes travel time to a general practice as a proxy for reasonable acceptability. There are grounds for adapting measures of accessibility with regard to the use of public transport, however. For example, a fifteen minute drive time from home to a surgery car park would produce a very different catchment population to a fifteen minute travel time using public transport where the walk to the bus stop, the wait at the bus stop and the walk from the bus stop to the surgery are included.
The results of recent studies also suggest that common understandings about poor access in rural areas may need to be revisited. The most recent analysis of rural parishes suggests that more people in rural areas are within crow-fly distances of basic services than has been thought in the past. Similarly, access to primary care is not a problem for the vast majority of rural inhabitants. This is not to say that there no problem of access exist. Rurality does appear to be associated with lower levels of hospital use and there will be certain groups of people living within rural areas (often referred to as the transport-poor) for whom access is a real problem. One of the key issues for policy makers is how such populations can be more precisely and effectively targeted.