Data source and sample construction
Data were drawn from the Social Health Atlas (SHA) of Australia by SLA, which was released by the Public Health Information Development Unit (PHIDU).[1] The SHA brings together a range of data on population health, health service use and the social determinants of health. SLAs are the principal regional building blocks defined by the Australian Bureau of Statistics (ABS) and, in aggregate, cover the whole of Australia without gaps or overlaps. In total, there are 1,397 SLAs [26],[2] although only 1,094 SLAs are included in the SHA data.[3] Although SHA data were released in 2010, 2011, 2012, 2013, and 2014 respectively, some variables have not been updated over years. For example, for the number of GP visits, the variable of main interest, only one year’s observation (2009-10) has been reported. The final sample size is 756 after dropping observations with missing values.[4] Comparing with the original sample, the SLAs in our analysis sample seem to be more socio-economically advantaged and with a higher proportion in urban areas (Additional file 1). SLAs dropped from the original sample are mainly those located in remote areas; given the geography and population distribution of Australia, the provision of health care faces challenges that are substantially different to those in larger population centres and are addressed through different policies. Therefore, our results provide estimates on the effect of demand and supply factors on regional variations in GP use primarily for non-remote Australia.
Dependent variable
The use of GPs was measured by the number of GP visits per capita in 2009-10 by SLAs. This was calculated by dividing the total number of GP services by SLAs, including those within the MBS and DVA, by the population size in each SLA.
Independent variables
In general, variation in regional health care utilisation is related to differences in populations’ needs for health care and in supply factors that include accessibility of services, practice patterns of health care providers, and health care system characteristics.
Need-related factors include the wide-ranging determinants of population health, burden of disease, demographics, and socioeconomic status. These factors reflect justified causes of variation in healthcare utilisation. Demographics were captured by the proportion of each age subgroup (age 0-9, 10-29, 30-44, 45-64, and 65 and above) in 2009, the percentage of males within the population in SLAs in 2009, and the percentage of the Aboriginal and Torres Strait Islander population in 2006.
We measured the health status of each local population using four indicators. First is the proportion of people who reported fair or poor health in each SLA. Second is the proportion of people with profound or severe disability living in the community. Third is the share of people aged 18 years and over with high or very high level of psychological distress. The last group of variables describes chronic diseases and conditions, which were measured by the proportions of people with Type 2 diabetes, circulatory system disease, and respiratory system disease, respectively. In this study, we used health status variables from previous years (2007-08) rather than those reported in 2009-10, to minimise the risk of bias due to reverse causality, because health status variables from the same year as GP usage (2009-10) may measure population health status after receiving care or treatment by GPs.
Health-related behaviours or health indicator variables were taken into account by the four variables: (1) percentage of current smokers among those aged 18 years and over, (2) percentage of people consuming alcohol at levels considered to be a high risk to death among those aged 18 years and over, (3) percentage of people who are physically inactive among those aged 15 years and over, and (4) percentage of obese people among those aged 18 years and over. The four variables were calculated by dividing the number of people who had these health-related behaviours in 2007-08 by the population size of each SLA. Concession cards provide access to cheaper medicines and concessions on health services in Australia, therefore the concession card status of the population was also controlled for.
The Socio-Economic Indexes for Areas (SEIFA) - Index of Relative Socioeconomic Disadvantage (IRSD) reported in 2006 was utilised to measure the socio-economic characteristics of each SLA. The IRSD identifies and ranks areas in terms of their relative socio-economic disadvantage. A low index score on the IRSD indicates relatively greater disadvantage in general, while a high score on it corresponds to a relative lack of disadvantage.[5] To account for the non-linear effect of the IRSD, we introduced a four-category variable, where the lowest quartile consists of areas with the lowest IRSD scores (most disadvantaged).
Rurality of people’s location of residence has been shown to play a role in population health status and their accessibility to health care services. The measure of remoteness was obtained from matching the SLAs with the remoteness areas defined by the Australian Standard Geographical Classification (ASGC) remoteness index in 2011 [27]. The ASGC remoteness index provided by the ABS comprises major cities, inner regional, outer regional, remote, and very remote areas [28]. Due to the small number of SLAs in the last two categories of the ASGC, we combined the last three groups and adopted a three-level measure: (1) major cities, (2) inner regional areas, (3) rural and remote areas, including outer regional, remote, and very remote areas.
A series of measures of people’s access to health care in SHA data was also taken into consideration. They were: The proportion of people aged 18 and over who delayed purchasing prescribed medication because they could not afford it and the proportion of people who often has a difficulty with transport or cannot get to places needed. Also included is a general measure of service availability where services include banking, legal, employment and other government services as well as health care; this is the proportion of people who reported difficulty in accessing services in 2007-08.
Supply-related factors, generally relating to unjustified variation, were also included in the analysis. In this study, we used the density of GPs and specialists to capture the capacity of the health care system and the accessibility of health care. GP and specialist densities by local government areas (LGAs) were constructed from the Health Workforce Data provided by Australian Institute of Health and Welfare (AIHW). The two variables were measured by the number of GPs and specialists per 1,000 population at LGA level in 2010 separately, based on the correspondence between LGA and SLA [29, 30]. Each LGA is formed by one or more SLAs and there was a total of 667 LGAs in Australia in 2011. Additionally, to account for the substitutability between ED treatment and GP usage, the number of EDs in each SLA was constructed and included in the analysis.[6]
The variable names and definitions used in this paper and the mean and standard deviations of these variables are summarised in Table 1. The number of GP attendances per person by SLAs ranged from 2.35 to 9.27, with an overall average of 5.58, indicating substantial regional variation in GP use in Australia.
Table 1: Definitions of variables and descriptive statistics
Variable name
|
Definition
|
Mean
|
SD
|
Dependent variables
|
|
|
|
Number of GP visits per capita
|
=Total GP services (MBS and DVA) by SLAs/population in each SLA in 2009-10
|
5.58
|
1.08
|
Explanatory variables: demand-side factors
|
|
|
|
Age distribution
|
|
|
|
Age 0-9
|
Proportion of population aged 0-9
|
0.12
|
0.02
|
Age 10-29
|
Proportion of population aged 10-29
|
0.26
|
0.05
|
Age 30-44 (base group)
|
Proportion of population aged 30-44
|
0.20
|
0.03
|
Age 45-64
|
Proportion of population aged 45-64
|
0.27
|
0.04
|
Age 65 and above
|
Proportion of population aged 65 and over
|
0.14
|
0.05
|
Share of male
|
Proportion of male population
|
50.33
|
1.86
|
ASGC remoteness index
|
|
|
|
Major city (base group)
|
=1 if in major city
|
0.46
|
0.50
|
Inner regional area
|
=1 if in inner regional areas
|
0.26
|
0.44
|
Remote and very remote areas
|
=1 if in outer regional, remote, and very remote areas
|
0.28
|
0.45
|
SEIFA-IRSD index
|
|
|
|
25th percentile and below (the most disadvantaged)
|
=1 if below 25th percentile of SEIFA
|
0.20
|
0.40
|
25th-50th percentile
|
=1 if 25th-50th percentile and below of SEIFA
|
0.27
|
0.44
|
50th-75th percentile
|
=1 if 50th-75th percentile and below of SEIFA
|
0.25
|
0.44
|
Above 75th percentile (base group - the most advantaged)
|
=1 if above 75th percentile of SEIFA
|
0.28
|
0.45
|
Proportion of Aboriginal population
|
Proportion taken up by Aboriginal population
|
3.53
|
1.08
|
Proportion of concession card holders
|
Proportion of population holding concession cards
|
10.98
|
1.92
|
Share of fair or poor self-assessed health population
|
Share of people who report fair or poor self-reported health in each SLA
|
14.67
|
3.75
|
Chronic disease and conditions (%)
|
|
|
|
Type 2 diabetes
|
Proportion of population having Type 2 diabetes
|
3.53
|
0.83
|
Circulatory system disease
|
Proportion of population having circulatory system disease
|
22.71
|
5.34
|
Respiratory system disease
|
Proportion of population having respiratory system disease
|
26.10
|
2.96
|
Proportion of people with profound or severe disability living in the community
|
Proportion of people who have profound or severe disability living in the community
|
3.53
|
1.08
|
Proportion of people with high/very high level of psychological distress
|
Proportion of people who have high/very high level of psychological distress
|
10.98
|
1.92
|
Health-related factors (%)
|
|
|
|
Current smokers
|
Proportion of current smokers (aged 18 and above)
|
20.35
|
3.75
|
Alcohol consumption at levels of high risk to health
|
Proportion of people consuming alcohol at levels of a high risk to health (aged 18 and above)
|
5.83
|
2.26
|
Physical inactivity
|
Proportion of persons who are physically inactive (aged 18 and above)
|
35.99
|
6.05
|
Obese persons
|
Proportion of persons who are obese (aged 18 and above)
|
18.03
|
3.23
|
Access to services (%)
|
|
|
|
Delayed purchasing prescribed medication
|
Proportion of people aged 18 years and over who delayed purchasing prescribed medication because they could not afford it
|
8.67
|
2.75
|
Have difficulty in accessing service
|
Proportion of people aged 18 years and over who had difficulty in accessing services
|
25.03
|
4.54
|
Have difficulty in transportation
|
Proportion of people aged 18 years and over who often has a difficulty with transport or cannot get to places needed
|
3.03
|
0.87
|
Explanatory variables: supply-side factors
|
|
|
|
Physician density
|
|
|
|
Number of specialists per 1,000 population
|
Number of GPs per 1,000 population in each LGA
|
0.85
|
2.59
|
Number of GPs per 1,000 population
|
Number of specialists per 1,000 population in each LGA
|
1.09
|
0.50
|
Number of EDs by SLAs
|
|
|
|
No ED (base group)
|
=1 if there is no EDs in a SLA
|
0.38
|
0.49
|
1-2 EDs
|
=1 if there are 1-2 EDs in a SLA
|
0.44
|
0.50
|
3 or more EDs
|
=1 if there are 3 or more EDs in a SLA
|
0.18
|
0.38
|
Empirical strategy
The analysis was undertaken in two stages. The first aim of this study is to examine the factors that influence regional variation in GP use and to ascertain the relative impact of various control variables on the magnitude of the differences in GP usage. To begin, the ordinary least squares (OLS) regression model was estimated with the number of GP visits per capita by SLAs as the outcome variable. Following this, to further explore the variability in effects of the explanatory variables across the distribution of GP use, quantile regression (QR) models were also utilised. Since there are noticeable differences in the provision of primary health care between rural and remote areas and major cities in terms of GPs’ services hours, travelling distances for GPs, and models of medical care [31], we also performed a subsample analysis by rurality to allow the effect of factors that influence the use of GP services to vary between urban and rural areas.
In the second stage of the analysis, we targeted unexplained differences in the use of GP consultations between two extreme groups — areas in the top and bottom quintiles of the distribution of the GP usage. It is the comparison between these two groups that is more challenging and makes us think about whether there is under or over use. We estimated a series of multiple linear regression models that initially include only categorical indicators representing areas’ quintiles rankings of GP usage. The coefficients of the quintile dummy variables measure the difference in GP visits per capita between quintile 1 (with the lowest number of visits) and each of the other four quintiles. With no other control variables in the model, these initial coefficients are precisely the differences in the number of GP visits per capita across the quintiles. We then expanded the number of explanatory variables in the model with demand-side and supply-side variables entering into the model sequentially. The coefficients of the quintile dummy variables change as each set of additional measures were included; therefore, the changes in these coefficients represent the amount of the initial regional difference that can be “explained” by the additional measures. The coefficients of the quintile dummy variables from the final regression model that includes all the observable independent variables represent the amount of the difference that is due to unidentified factors.
[1] These data are part of the Public Health Information Development Unit’s Social Health Atlas series.
[2] The delimitation of SLAs is based on the boundaries of incorporated bodies of local government. These bodies are the Local Government Councils and the geographical areas which they administer.
[3] The statistics for Australian Capital Territory are missing and there is no data for the areas that are unincorporated in the corresponding state or with unknown ABS cell adjustment.
[5] The score for Australia is 1,000 as a benchmark.
[6] The information on EDs, such as name, hospital type (public or private), postcode, and address are obtained from MyHospitals, accessed at <http://www.myhospitals.gov.au/>.