Indigenous patient migration patterns after hospitalisation and the potential impacts on mortality estimates.
Zhao, Yuejen ; Condon, John R. ; Li, Shu Qin 等
This study analysed interregional migration for Indigenous patients
in the Northern Territory, Australia. Individual-level linked
hospitalisation data between July 1998 and June 2011 were used to
describe the migration patterns and associated factors.
Micro-simulations were conducted to assess the impacts on mortality
estimates. Indigenous patients were 35% more likely to migrate from
remote to urban areas after hospitalisation than in the reverse
direction (risk ratio 1.35, 95% confidence interval 1.30-1.41). The
likelihood was positively associated with hospitalisations, age and the
Central Australia region. Indigenous patients with diabetes, renal
disease or chronic obstructive pulmonary disease had higher risks of
urban migration. Non-Indigenous patients were included for comparison.
The micro-simulations indicated the patient migration may result in a 6%
under-estimation of Indigenous mortality in remote and very remote areas
and 3% over-estimation of mortality in urban areas. The results are
pertinent to a sound understanding of health outcomes across remoteness
categories.
KEY WORDS: Hospital services; Health and inequality;
Micro-simulation; Migration; Mortality.
1. INTRODUCTION
Migration and health are interrelated (International Organization
for Migration, 2005). Health can influence migration decisions and
migration may impact on an individual's health. Health status
influences migration in complex ways. Good health at younger ages tends
to promote migration from remote to regional/urban areas for education,
employment and business opportunities. People in poor health at older
ages tend to migrate to urban areas seeking better access to health care
(Bentham, 1988; Norman et al., 2005). Place of birth versus residence
has attracted much attention, and many studies have shown that place of
birth is a more important factor than place of residence (Vigotti et
al., 1988; Fascioli et al., 1995). The cause-effect relationships
between migration and health are also interchangeable. Migration causes
ill-health, whereas health status also influences the decisions on
migration (Saarela and Finnas, 2008). Migrants may also differ in
mortality and morbidity from non-migrants. Various studies have
investigated rural to urban movements and focused on the potential
negative impacts of migration on the population. The negative impacts
may be due to the stress of the movement as well as the adoption of a
less healthy lifestyle (Verheij et al., 1998). Atherly and colleagues
(2003) used health-related migration patterns to assess primary care
workforce shortages in remote and rural areas and to inform the
development of primary care services. Health data have also been used to
estimate general migration statistics (Chappell et al., 2000). Patient
migration has been regarded as an important factor in local health
services planning and economic development (Wismar et al., 2011) and may
be associated with a lack of local services, lack of capacity to address
regional health care needs, patients' perception of service
quality, and medically-induced demand (Hodgkin, 1996). In this study,
the term patient migration is used to describe people travelling from
one residence to another to access hospital care (Kirch, 2004).
Patient migration may bias comparisons of health status measures
(such as mortality rates) between urban, regional and remote areas, if
people with serious chronic conditions migrate from regional and remote
to urban areas shortly before their death. In Australia, most studies
have found that mortality rates are higher in remote and very remote
areas than in regional and metropolitan areas (see, for example,
Wilkinson et al., 2000; Cass et al., 2001; Wakerman, 2008; Begg et al.,
2008), while two studies have reported that mortality rates for
Indigenous people (Aboriginal and/or Torres Strait Islander) in the
Northern Territory (NT) were lower in very remote areas than in other
areas (Rowley et al., 2008; Andreasyan and Hoy, 2010). However, patient
migration may have affected these results, because people with serious
chronic illnesses may have migrated from very remote areas to be closer
to hospital and other health and support services in less remote service
centres.
Little is known about the impact of patient migration on
population-based mortality rates. To the best of our knowledge, there
have been no comprehensive studies that have measured Indigenous patient
migration patterns and the impact on mortality estimation. This study
aims to understand the patterns, underlying factors and consequences of
Indigenous patient migration within the NT. We investigate whether
Indigenous patients in the NT have migrated from very remote areas to
access health services, especially during the last few years prior to
death, and if so, the extent of this patient migration and its impact on
mortality rates by remoteness category.
One of the obstacles in evaluating patient migration patterns has
been a lack of suitable data. In the NT, the five public hospitals have
used a single client information system (Caresys, the NT public
hospitals' patient administration and clinical information system),
with a common unique personal identifier, namely the hospital
registration number (HRN), since 1992. A summary of each inpatient
episode is recorded in Caresys and stored in the NT hospital morbidity
dataset (HMD), in which all inpatient episodes for each person can be
linked. Place of residence at the time of each inpatient episode is also
recorded. Indigenous identification is known to be very accurate (Foley
et al., 2012). Thirty percent of the NT population are Indigenous, and
56% of the NT Indigenous population live in very remote areas
(Australian Bureau of Statistics, 2007). Only a very small proportion of
the NT Indigenous population use the one private NT hospital.
Hospitalisation is frequent for NT Indigenous people; in 2008-2009 their
average annual public hospital admission rate was 1.7 inpatient episodes
per person (Australian Institute of Health and Welfare, 2010). The NT
HMD is therefore a suitable data source to examine patient migration
trends among the NT Indigenous people.
2. METHOD
NT public hospital inpatient episodes for patients discharged
between 1 July 1998 and 30 June 2011 were used for the analysis. The HMD
included admission date, discharge date, discharge mode (ie. whether
died in hospital), usual resident locality, age, sex, Indigenous status,
HRN and diagnosis.
Inclusion Criteria
All NT Indigenous and non-Indigenous residents with at least two
hospitalisations in the study period were included. Patient mobility was
assessed using changes in the Statistical Local Area (SLA) of residence
between episodes, considering only intra-Territory regional migration,
assuming the Indigenous population was closed to interstate migration.
Interstate residents, NT residents who moved interstate, and those
transferred to an interstate hospital were excluded.
Patient migration estimates were derived by counting moves, based
on changes in residential address between inpatient episodes. Usual
residence is recorded at the time of each hospital admission as the
address where the patient lives or intends to live for at least three
months.
Residential addresses were translated in Caresys into locality
codes (suburbs within cities and towns, remote communities, cattle
stations, mine sites, etc). The individual episode data previously coded
to respective year SLAs of the Australian Standard Geographic
Classifications were rebased to 2006 SLAs, which contain 96 SLAs for the
NT. The HMD contains the locality codes, which were mapped to SLAs for
this project using a concordance file developed by the NT Department of
Health (DoH). The Accessibility/Remoteness Index of Australia (ARIA)
score was used to determine remoteness. ARIA describes remoteness based
on road distance between each locality and service centres of various
sizes (Commonwealth Department of Health and Aged Care, 1999). The road
distances are converted to a continuous measure, ranging from 0 for
least remote to 12 to most remote. For each SLA, the average ARIA score
was determined using ARIA values of all localities within the SLA. The
average ARIA scores for the SLA were then assigned to each inpatient
episode. Discharge mode was used to identify in-hospital deaths.
Measuring Migration
Migration was measured using a net transition model as the
difference in ARIA scores of the residential locality between the first
inpatient episode and the last episode. Suppose that a patient migrates
m times during the study period. The remoteness is changed in ARIA
scores: [A.sub.0] [right arrow] [A.sub.1] [right arrow] ... [right
arrow] [A.sub.m], where [A.sub.j-1] and [A.sub.j] represent the starting
and finishing ARIA scores for the jth hospitalisation (j = 1, ... m).
The net transition is measured by
([A.sub.0] - [A.sub.1]) + ([A.sub.1]/[A.sub.2]) + ... +([A.sub.m-1]
- [A.sub.m]) = [A.sub.0] - [A.sub.m]. (1)
The ratio is measured by
([A.sub.0]/[A.sub.1]) x ([A.sub.1]/[A.sub.2]) x ... x
([A.sub.m-1]/[A.sub.m]) = [A.sub.0]/[A.sub.m]. (2)
It is clearly seen in equations (1) and (2) that all the ARIA
scores are algebraically eliminated except for first ([A.sub.0]) and
last ([A.sub.m]) The transition was therefore counted just once for each
patient using the first and last inpatient episode.
Based on the ARIA, there were three levels of remoteness categories
relevant to the NT, which were Outer Regional, Remote and Very Remote.
Numerical ARIA scores were used for measuring patient migration (rather
than the three remoteness categories). Changes were compared to estimate
the likelihood of moving to "less remote" (inflows) with that
of moving to "more remote" (outflows). As illustrated in Table
1, the number of inflows is patients moving away from more remote areas
towards less remote areas, ie. [N.sub.21] + [N.sub.31] + [N.sub.32]; the
number of outflows represents patients moving from less to more remote
areas, which is [N.sub.12] + [N.sub.13] + [N.sub.23]; the number of
patients who did not change remoteness category is [N.sub.11] +
[N.sub.22] + [N.sub.33]. The risk ratio (RR) between inflows and
outflows is obtained dividing [N.sub.21] + [N.sub.31] + [N.sub.32] by
[N.sub.12] + [N.sub.13] + [N.sub.23]. The percentage of net migration is
[([N.sub.21] + [N.sub.31] + [N.sub.32]) - ([N.sub.12] + [N.sub.13]
+ [N.sub.23])/[3.summation over (i=1)][3.summation over (j=1)]
[N.sub.ij]] (3)
Statistical Analysis
Demographic variables included sex, age group and region at first
episode. Due to the difference in geographic location and population
characteristics, the analysis was also stratified into two regions: the
Top End and Central Australia. Health conditions were identified by the
Australian refined diagnosis related groups (Australian Department of
Health and Aged Care, 1998). Due to repeated hospital admissions, a
person was allowed to have more than one condition.
A linear regression model was used for multivariate analysis with
the last admission ARIA score as the dependent variable, and the first
admission ARIA score and the time (in years) between the first and last
admission as independent variables to estimate average ARIA score
changes per year.
Multivariate analysis using log-linear regression was performed to
assess the association of migration from more to less remote areas with
age group, sex, region and severity measures (such as number of
hospitalisations). The log-linear model is a regression model with
dependent variable being the frequency of a contingency table and all
explanatory factors as independent variables, including age, sex,
region, hospitalisation and patient migration in this study. It is
commonly used to assess association between two independent variables by
setting interaction terms. The RR estimates can be derived from the
log-linear model by examining the interaction terms between patient
migration and associated variables. Refer to Christensen (1997) for
details of log-linear modelling. Forward selection was undertaken for
modelling. There was no statistically significant difference between
males and females, so a term for sex was not included in the final
model. A separate multivariate analysis (using the log-linear model)
restricted to only patients who died in hospital was performed to assess
the patient migration patterns in the last few years before death.
Straight line distance was derived from MapInfo (Professional 10.5)
on the basis of the farthest distance of SLAs using distance calculator
(Pitney Bowes Software, 2010). Straight lines were measured to describe
distance of patient migration.
Spider Diagrams
Spider diagrams were used to illustrate inflows and outflows of
migration separately. In the inflow diagram, a straight line represents
migrations from a locality with a higher ARIA score to one with a lower
ARIA score, whereas in the outflow diagram, migrations in the opposite
direction are illustrated. To simplify the spider diagrams, one line in
the spider diagram represents ten patients. Migration lines with less
than ten patients were not included. To avoid overlapping, the migration
lines are reassigned to different combinations of Collection Districts
of the SLAs. Patients with positive [A.sub.0] - [A.sub.m] were regarded
as inflows (see panel a in Figure 1) and those with negative [A.sub.0] -
[A.sub.m] as outflows (panel b), in comparison with the NT map (panel
c).
Impact of Migration on Mortality Rates
The patient migration patterns were then incorporated into a
microsimulation model to evaluate the impacts of patient migration on
mortality estimates (Schmertmann and Sawyer, 1996). The microsimulation
model is devised to imitate migration patterns and key demographic
indicators, and to compare different scenarios with and without patient
migration. In the micro-simulations, three ARIA categories of very
remote, remote and outer regional were considered with distinct
mortality patterns, taken from the 2006 NT Indigenous life tables by
remoteness (there are no inner regional or major city categories in the
NT). Over lifetimes, individuals may switch back and forth between
regions via migration, in which the level, age and sex patterns of
mobility were determined by the earlier patient migration results. The
model assumptions broadly matched the mortality and migration patterns
of the NT Indigenous population. The modelling results mirror the
implication of interregional patient migration on mortality. The
migration bias was assessed by comparing mortality estimates for
different scenarios. This was of course a simple representation of
reality, but this lead to some interesting insights of the impacts of
patient migration on mortality outcomes. The micro-simulation involves
the following steps:
(a) Estimate probabilities of death based on current mortality
data, conditional on age, sex and region.
(b) Generate individual level population micro-data and applied
Bernoulli random numbers to simulate death. The Bernoulli random numbers
were generated in MS Excel using RAND() function. A macro and
user-defined functions were written in Visual Basic to generate binomial
and Poisson random numbers. The attributes of individuals included age,
sex, Indigenous status, region, death and migration. The bottom-up
approach was applied to micro-simulation.
(c) Validate the micro-simulated data in comparison with current
mortality and population data. Mortality rates were calculated for
comparison. Run the model multiple times to ensure the micro-simulated
data were not biased. Absolute differences in mortality rate between the
actual and synthetic data were used for validation.
(d) Use the micro-simulated data as baseline scenario, which is
subject to patient migration. The no patient migration scenario was
constructed by assuming that the RRs were applicable to the deaths (the
numerator of mortality rate) and the population (denominator). The RRs
were derived from equation (3).
(e) Compare the no patient migration scenario with the baseline to
estimate bias introduced by patient migration.
For a more detailed discussion of micro-simulation method, see
O'Donoghue et al. (O'Donoghue et al., 2013).
The study was approved by the Human Research Ethics Committee of
the NT DoH and Menzies School of Health Research (HREC-20101401).
3. RESULTS
Between July 1998 and June 2011, 46,309 Indigenous NT residents had
at least two inpatient episodes (Table 2). This was almost
three-quarters (72%) of the total NT Indigenous population
(approximately 64,000 in 2006). Approximately 10% of patients moved over
a straight line distance of longer than 100 kilometres at least once,
among whom 20% made multiple movements ranging from 2 to 36 times. Ten
percent of the patients moved to a less remote area after
hospitalisation, 7% to a more remote area, and 82% remained in the same
remoteness category. Among those who migrated, migration was 35% more
likely to be to a less remote area than to a more remote area, with 2.7%
net migration.
Indigenous patient migration from more remote to less remote areas
increased with increasing age and number of inpatient episodes, and was
higher in Central Australia than in the Top End, but was similar for
males and females (Table 2). Indigenous patients with diabetes, renal
disease, chronic obstructive pulmonary disease (COPD) and injury have a
higher RR than those with ischaemic heart disease (IHD), hypertension,
or pregnancy. This patient migration pattern was not unique to
Indigenous residents, and non-Indigenous residents had a similar but
stronger pattern of patient migration with a greater risk differential
in males, youth and Central Australia region (Table 3). Pregnancy played
a more important role in non-Indigenous patient migration.
The spider diagrams in Figure 1 compare the inflows and outflows of
Indigenous patient migrations in the NT maps. By comparing the density
of panels (a) and (b), it is apparent that the inflows were generally
more frequent than outflows. The inflows were mainly from South East Top
End, Luritja-Pintupi, South East Arnhem, Maningrida and Anmatjere, to
Darwin, Alice Springs, and Katherine. The outflows were mainly from
Darwin, Alice Springs and Katherine. Areas to receive larger outflows
were Katherine East, North East Arnhem, Eastern Arrernte-Alyawa,
Pitjantjatjara and Port Keats.
The patient migration pattern was further investigated using
proximity to death (in years). The movements were more pronounced for
Indigenous people close to the end of their lives. The closer the
proximity to death, the more likely they migrated to less remote areas
(Table 2). Figure 2 shows Indigenous patient migration flows by the
proximity to death. It is evident that the transits moving from the
upper-left to the lower-right corner were more frequent than those in
the opposite direction, suggesting that patient inflows are associated
with proximity to death.
The Indigenous patient migration pattern data were then analysed by
multivariate linear and log-linear models to quantify the magnitude of
the effect. The linear models suggest that on average, the ARIA score of
patient decreased by 0.16 of one ARIA score annually prior to death
(Table 4). The log-linear models show that the risk of moving to less
remote areas after hospitalisation was mainly driven by the number of
hospitalisations, age and Central Australia region (P<0.01) (Table
5). Sex was not selected in the final model.
Micro-simulation results suggest the Indigenous patient migration
prior to death leads to over-estimation of mortality rate in urban areas
by 3%, and under-estimation of mortality in remote and very remote areas
by 6%.
Interstate hospital transfer is relatively minor. Of all
hospitalisations, 0.8% and 1.6% involved interstate hospital transfer
for Indigenous and non-Indigenous NT residents, respectively.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
4. DISCUSSION
Migration is a continuous, dynamic and complex social phenomenon,
and an important factor for health planning and service delivery.
Selective migration may affect mortality and morbidity estimates
(Brimblecombe et al., 2000; Boyle, 2004; Norman et al., 2005), and be
the result of the movement of a selective group of either healthy or
unhealthy migrants (Kington et al., 1998). A longitudinal case-control
study on Indigenous alcohol treatment in the NT indicated that alcohol
users who were treated in the program and resumed drinking at the same
level were less likely to return to remote communities with alcohol
restrictions (Dingwall, 2011). People in very remote areas often
relocate to access health, education and other goods and services in
urban areas, because there is a lack of these services in their remote
communities. In this study, we documented the interregional Indigenous
patient migration patterns using the HMD. The hospital Indigenous
patients who migrated were 35% more likely to move to less remote areas
than to more remote areas. For those who died in hospital, ARIA score
decreased by an average of 0.16 of one ARIA score annually prior to
death. This health-related migration is broadly in line with one element
of the selective migration hypothesis (Bentham, 1988; Norman et al.,
2005), which is that the unhealthier old people tend to move to urban
areas. Unfortunately it is difficult to investigate the other element
(that young and healthier population also tend to move to urban areas to
access employment and social opportunities) using the HMD because it
lacks relevant data.
Patient migration appears to be a component of late-life migration
in the Indigenous population. It is linked to morbidity in later-life,
rather than retirement, because of the low labour force participation in
the Indigenous population (Australian Bureau of Statistics, 2004).
However, the need for health care is not the only reason. Other life
course events such as housing, employment, education, and better access
to other services may all play an important part in population mobility
(Cohn et al., 1994). It is also evident that 82% of hospital Indigenous
patients and 93% of hospital non-Indigenous patients did not change
remoteness category between their first and last inpatient episode. The
net effect in general is not overwhelming (2.7% and 1.2% for Indigenous
and non-Indigenous patients respectively), but it is notably elevated in
Central Australia, and increased with number of hospitalisations. Based
on this information, the existing patient migration pattern is unlikely
to play a major role in changing service planning. Indigenous patient
migration increased with age (Table 2) and non-Indigenous patient
migration peaked at ages 10-19 years (Table 3). Overall, the RRs were
driven mainly by disease severity (hospitalisations, disease and death)
for Indigenous patients and by demographic and geographic variables
(age, sex and region) for non-Indigenous patients.
Patient migration may deflate the number of deaths among the
longterm population of very remote areas. The masking effect for remote
area mortality may lead to confusions in differences of mortality and
health outcome between remote and urban Indigenous populations. By data
simulations, we can demonstrate how the Indigenous patient migration
affects mortality estimates. The effect is to raise mortality rates in
less remote areas (by 3%) and lower them in more remote areas (6%).
Patient migration can distort the true picture of regional variation in
health risk.
This result is consistent with international study (Rogerson and
Han, 2002), and may partially explain why the Indigenous mortality rate
was underestimated in very remote areas (Zhao et al, 2009). Based on the
micro-simulation and empirical data from this study, illness-related
migration may lead to under-estimation of morbidity and mortality in
very remote areas and over-estimation in better serviced regional and
urban areas.
The analyses by condition support the general argument that those
with more protracted and progressive conditions have a greater risk of
urban migration. Among the listed health conditions, the extremes were
pregnancy, which is a short term event, in which a woman is likely to
return to her community, versus diabetes, a progressive condition
requiring ongoing management of increasing complexity. Injury and
alcohol use possibly demonstrate reverse causality: the migration may
not have been associated with ill health and health care need, but
rather indicated migration to urban areas to access and consume alcohol,
which increased the risk of alcohol-related disease and injury that in
turn lead to more inpatient care.
Several limitations require consideration in interpreting these
findings.
Firstly, this study assumed a stable NT Indigenous population
closed to net interstate migration. In 2006, the net interstate
migration of NT Indigenous people was estimated to be close to zero
(Australian Bureau of Statistics, 2009). The non-Indigenous patient
migration results may be less reliable than the Indigenous results, due
to higher levels of interstate migration among non-Indigenous
population. Admissions to private hospital and lower hospitalisation
rate further limit the utility of non-Indigenous results. Secondly, this
study is dependent on the reliability of the HMD. The residential
locality data in the HMD have been validated in the national hospital
data quality surveys and the data accuracy was assessed to be good (88%)
(Foley et al., 2012). Thirdly, the net transition model utilised only
the first and last ARIA scores as a proxy of change in remoteness. The
ARIA scores in between have been ignored. This may be potentially
biased. As part of the sensitivity tests, all moves were also analysed
to compare with the net transition model. The results were very close,
and therefore omitted for brevity. Another alternative method we tested
was to calculate regression coefficients at the individual level using
all moves and a Stata module "bcoeff" (Wang and Cox, 2000).
The results were also broadly consistent with those presented in Section
3.
Finally, the causes of patient migration are complex and vary
geographically. This analysis shows that severity of disease may offer
partial explanation of Indigenous patient migration in the NT. More
research is needed for further investigation of this issue. Indigenous
people are highly mobile, regardless of whether or not health and
hospital services are provided. The extent to which our findings can be
generalised to other parts of Australia is unknown, particularly in
relation to inner regional and metropolitan areas which could not be
included in this study. A future study designed to investigate this
issue nationally seems warranted. Although general patterns apply to
similar environments, individual situations do need careful
consideration.
Unpacking the key elements of patient migration is an important
tool to understanding population health, and will benefit health care
planning and service delivery.
There appears to be a patient movement from more remote to less
remote areas after hospital admission in remote Indigenous and
non-Indigenous populations. The patient migration pattern is an
important factor to consider in interpreting health outcome measures and
health planning. More research is needed to further explore why
Indigenous patients migrate, and how health and other services can be
better managed for health outcome improvement.
REFERENCES
Andreasyan, K. and Hoy, W. E. (2010). Patterns of mortality in
Indigenous adults in the Northern Territory, 1998-2003: are people
living in more remote areas worse off? Medical Journal of Australia,
190, pp. 307-311.
Atherly, A., Kralewski, J., Johnson, C. and Brasure, M. (2003).
Developing rural primary health care service areas: an analysis of
patient migration patterns. Health Care Management Review, 28, pp.
189-197.
Australian Bureau of Statistics (2004). Australian Social Trends.
ABS: Canberra.
Australian Bureau of Statistics (2007). Population Distribution,
Aboriginal and Torres Strait Islander Australians. ABS: Canberra.
Australian Bureau of Statistics (2009). Experimental Estimates and
Projections, Aboriginal and Torres Strait Islander Australians. ABS:
Canberra.
Australian Department of Health and Aged Care (1998). Development
of the Australian Refined Diagnosis Related Group (AR-DRG)
Classification: Development of AR-DRG classification version 4.1
(includes appendices on computer disk). Department of Health and Aged
Care: Canberra.
Australian Institute of Health and Welfare (2010). Australian
Hospital Statistics 2008-09. AIHW: Canberra.
Begg, S. J., Vos, T., Barker, B., Stanley, L. and Lopez, A. D.
(2008). Burden of disease and injury in Australia in the new millennium:
measuring health loss from diseases, injuries and risk factors. Medical
Journal of Australia, 188, pp. 36-40.
Bentham, G. (1988). Migration and morbidity: implications for
geographical studies of disease. Social Science & Medicine, 26, pp.
49-54.
Boyle, P. (2004). Population geography: migration and inequalities
in mortality and morbidity. Progress in Human Geography, 28, pp.
767-776.
Brimblecombe, N., Dorling, D. and Shaw, M. (2000). Migration and
geographical inequalities in health in Britain. Social Science &
Medicine, 50, pp. 861-878.
Cass, A., Cunningham, J., Wang, Z. and Hoy, W. (2001). Regional
variation in the incidence of end-stage renal disease in Indigenous
Australians. Medical Journal of Australia, 175, pp. 24-28.
Chappell, R., Vickers, L. and Evans, H. (2000). The use of patient
registers to estimate migration. Population Trends, pp. 19-24.
Christensen, R. (1997). Log-Linear Models and Logistic Regression.
Springer: New York.
Cohn, S. E., Klein, J. D., Mohr, J. E., Van Der Horst, C. M. and
Weber, D. J. (1994). The geography of AIDS: patterns of urban and rural
migration. Southern Medical Journal, 87, pp. 599-606.
Commonwealth Department of Health and Aged Care (1999). Measuring
Remoteness. Accessibility/Remoteness Index of Australia (ARIA). DOHA:
Canberra.
Dingwall, K. M., Maruff, P. and Cairney, S. (2011). Similar profile
of cognitive impairment and recovery for Aboriginal Australians in
treatment for episodic or chronic alcohol use. Addiction, 106, pp.
1419-1426.
Fascioli, S., Capocaccia, R. and Mariotti, S. (1995). Cancer
mortality in migrant populations within Italy. International Journal of
Epidemiology, 24, pp. 8-18.
Foley, M., Zhao, Y. and Condon, J. (2012). Demographic Data Quality
Assessment for Northern Territory Public Hospitals 2011. Department of
Health: Darwin.
Hodgkin, D. (1996). Specialized service offerings and
patients' choice of hospital: The case of cardiac catheterization.
Journal of Health Economics, 15, pp. 305-332.
International Organization for Migration (2005). Health and
Migration: Bridging the Gap. Migration Policy and Research Programme of
the International Organization: Geneva.
Kington, R., Carlisle, D., McCaffrey, D., Myers, H. and Allen, W.
(1998). Racial differences in functional status among elderly US
migrants from the south. Social Science & Medicine, 47, pp. 831-840.
Kirch, W. (2004). Public Health in Europe: 10 Years European Public
Health Association. Springer: New York.
Norman, P., Boyle, P. and Rees, P. (2005). Selective migration,
health and deprivation: a longitudinal analysis. Social Science &
Medicine, 60, pp. 2755-2771.
O'Donoghue, C., Ballas, D., Clarke, G. P., Hynes, S. and
Morrissey, K. (2013). Spatial Microsimulation for Rural Policy Analysis.
Springer: New York.
Pitney Bowes Software (2010). MapInfo Professional 10.5 User Guide.
Pitney Bowes Business Insight: New York.
Rogerson, P. A. and Han, D. (2002). The effects of migration on the
detection of geographic differences in disease risk. Social Science
& Medicine, 55, pp. 1817-1828.
Rowley, K. G., O'Dea, K., Anderson, I., McDermott, R.,
Saraswati, K., Tilmouth, R., Roberts, I., Fitz, J., Wang, Z. and
Jenkins, A. (2008).
Lower than expected morbidity and mortality for an Australian
Aboriginal population: 10-year follow-up in a decentralised community.
Medical Journal of Australia, 188, pp. 283-286.
Saarela, J. and Finnas, F. (2008). Internal Migration and
Mortality: The Case of Finland. Environmental Health Insights, 2, pp. 1.
Schmertmann, C. P. and Sawyer, D. O. (1996). Migration bias in
indirect estimates of regional childhood mortality levels. Mathematical
Population Studies, 6, pp. 69-93.
Verheij, R. A., Van De Mheen, H. D., De Bakker, D. H., Groenewegen,
P. P. and Mackenbach, J. P. (1998). Urban-rural variations in health in
The Netherlands: does selective migration play a part? Journal of
Epidemiology and Community Health, 52, pp. 487-493.
Vigotti, M. A., Cislaghi, C., Balzi, D., Giorgi, D., La Vecchia,
C., Marchi, M., Decarli, A. and Zanetti, R. (1988). Cancer mortality in
migrant populations within Italy. Tumori, 74, pp. 107.
Wakerman, J. (2008). Rural and remote public health in Australia:
building on our strengths. Australian Journal of Rural Health, 16, pp.
52-55.
Wang, Z. and Cox, N. J. (2000). BCOEFF: Stata Module to Save
Regression Coefficients to New Variable. Chestnut Hill IDEAS. Online
version accessed 26 September 2012,
http://ideas.repec.org/c7boc/bocode/s412201.html
Wilkinson, D., Hiller, J., Moss, J., Ryan, P. and Worsley, T.
(2000). Mortality variation across Australia: descriptive data for
States and Territories, and statistical divisions. Australian and New
Zealand Journal of Public Health, 24, pp. 226-233.
Wismar, M., Palm, W., Figueras, J., Ernst, K. and Ginneken, E.
(2011). Cross-border Health Care in the European Union: Mapping and
Analysing Practices and Policies. European Observatory on Health Systems
and Polcies: Brussels.
Zhao, Y., Guthridge, S., Li, S. and Connors, C. (2009). Patterns of
mortality in Indigenous adults in the Northern Territory, 1998-2003.
Medical Journal of Australia, 191, pp. 581-582.
Yuejen Zhao
Principal Health Economist, Health Gains Planning Branch,
Department of Health, Northern Territory, PO Box 40596, Casuarina NT,
0811, Australia.
Centre for Remote Health, Flinders University and Charles Darwin
University, Alice Springs NT, 0871, Australia. Email:
[email protected]
John R Condon
Department of Health, Northern Territory, PO Box 40596, Casuarina
NT, 0811, Australia.
Menzies School of Health Research, Casuarina NT, 0810, Australia.
Shu Qin Li
Department of Health, Northern Territory, PO Box 40596, Casuarina
NT, 0811, Australia.
Steven Guthridge
Department of Health, Northern Territory, PO Box 40596, Casuarina
NT, 0811, Australia.
Centre for Remote Health, Flinders University and Charles Darwin
University, Alice Springs NT, 0871, Australia.
Ramakrishna Chondur
Department of Health, Northern Territory, PO Box 40596, Casuarina
NT, 0811, Australia.
Table 1. Notational Illustration of Patient Migration in
ARIA Categories and Changes between the First and Last
Hospitalisations.
ARIA category ARIA category at
at first last hospitalisation
hospitalisation
Outer Remote
regional
(Urban)
Outer regional [N.sub.11] [N.sub.12]
(Urban) (Urban stayers) (Urban to
remote
movers)
Remote [N.sub.21] (Remote [N.sub.22]
to urban movers) (Remote
stayers)
Very remote [N.sub.31] (Very [N.sub.32] (Very
remote to urban remote to
movers) remote movers)
ARIA category ARIA category at
at first last hospitalisation
hospitalisation
Very remote
Outer regional [N.sub.13] (Urban
(Urban) to very remote
movers)
Remote [N.sub.23] (Remote
to very remote
movers)
Very remote [N.sub.33 (Very
remote stayers)
Note: ARIA = Accessibility/Remoteness Index of Australia;
[N.sub.ij] = number of people (i = 1, 2, 3; j = 1, 2, 3).
Source: the Authors.
Table 2. Remoteness Changes and Risks of Inflow by
Demographic and Geographic Variables and Health Category,
Indigenous Population, Northern Territory, 1998/9-2010/1.
Persons Migration (%) RR
Less More
remote remote
All patients 46,309 10.1 7.5 1.35 **
Age
<10 16,338 7.4 6.7 1.10 *
10-19 7,117 10.3 8.1 1.28 **
20-39 14,242 12.3 8.6 1.44 **
>=40 8,612 11.6 6.7 1.72 **
Sex
Males 20,201 10.1 7.5 1.35 **
Females 26,108 10.2 7.5 1.36 **
Region
Top End 28,938 9.3 8.5 1.09 **
Central 17,371 11.6 5.8 1.99 **
Australia
Hospitalisations
2 10,931 4.6 4.5 1.01
3-4 13,137 7.8 7.2 1.08
5-9 12,726 12.1 9.0 1.34 **
10-19 6,382 15.0 9.4 1.60 **
>19 3,133 21.8 9.0 2.43 **
Disease category
Pregnancy 9,535 10.7 8.6 1.24 **
Hypertension 196 16.8 12.8 1.32
IHD 2,699 13.1 8.9 1.48 **
COPD 3,080 16.1 7.8 2.07 **
Renal disease 2,325 17.8 8.3 2.14 **
Diabetes 1,170 20.2 8.9 2.27 **
Alcohol use 1,525 18.6 10.4 1.79 **
Injury 317 18.6 8.8 2.11 **
Cancer 319 11.6 6.3 1.85 *
Deaths 1,817 16.5 7.2 2.29 **
Proximity to death
0-1 574 9.8 3.8 2.55 **
2-4) 501 16.4 7.2 2.28 **
>4 742 21.8 9.8 2.22 **
(RR 95% CI) % net
migration
All patients (1.30, 1.41) 2.7%
Age
<10 (1.02, 1.20) 0.7%
10-19 (1.15, 1.42) 2.3%
20-39 (1.34, 1.55) 3.8%
>=40 (1.56, 1.91) 4.9%
Sex
Males (1.26, 1.44) 2.6%
Females (1.28, 1.44) 2.7%
Region
Top End (1.04, 1.15) 0.8%
Central (1.85, 2.14) 5.8%
Australia
Hospitalisations
2 (0.89, 1.14) 0.0%
3-4 (0.99, 1.18) 0.6%
5-9 (1.24, 1.44) 3.1%
10-19 (1.45, 1.77) 5.6%
>19 (2.13, 2.78) 12.7%
Disease category
Pregnancy (1.14, 1.36) 2.1%
Hypertension (0.80, 2.17) 3.5%
IHD (1.26, 1.73) 4.2%
COPD (1.78, 2.40) 8.2%
Renal disease (1.82, 2.52) 9.4%
Diabetes (1.82, 2.83) 11.0%
Alcohol use (1.49, 2.16) 8.0%
Injury (1.37, 3.25) 9.0%
Cancer (1.09, 3.15) 5.0%
Deaths (1.88, 2.79) 9.2%
Proximity to death
0-1 (1.56, 4.15) 5.8%
2-4) (1.56, 3.33) 8.8%
>4 (1.71, 2.89) 11.5%
Note: * 0.01<P<0.05; ** P<0.01; CI=confidence interval; COPD
= chronic obstructive pulmonary disease; IHD = ischaemic
heart disease; RR=risk ratio. Source: the Authors.
Table 3. Remoteness Changes and Risks of Inflow by
Demographic and Geographic Variables and Health Category,
Non/Indigenous Population, Northern Territory, 1998/9-2010/1.
Persons Remoteness RR
Less More
remote remote
All patients 59,010 3.9 2.7 1.44 **
Age
<10 10,705 3.2 2.1 1.50 **
10-19 4,846 4.8 2.3 2.10 **
20-39 22,148 4.7 3.3 1.43 **
>=40 21,311 3.3 2.5 1.31 **
Sex
Males 26,997 4.2 2.8 1.52 **
Females 32,013 3.7 2.7 1.38 **
Region
Top End 44,229 2.9 3.2 0.93
Central 14,781 6.9 1.4 4.91 **
Australia
Hospitalisations
2 24,758 2.6 1.8 1.41 **
3-4 18,948 4.3 2.9 1.48 **
5-9 10,959 5.5 3.9 1.41 **
10-19 3,191 6.0 4.1 1.48 **
>19 1,154 5.5 3.3 1.66 *
Disease category
Pregnancy 14,426 4.5 3.1 1.45 **
Hypertension 95 3.2 5.3 0.60
IHD 2,486 4.5 3.3 1.38 *
COPD 4,822 4.5 3.6 1.25 *
Renal disease 1,465 3.7 2.5 1.50
Diabetes 692 4.2 2.5 1.71
Alcohol use 763 9.2 5.6 1.63 *
Injury 296 5.1 4.4 1.15
Cancer 1,136 4.3 2.3 1.88 *
Deaths 2,256 3.5 2.0 1.74 **
Proximity to death (years)
0-1 817 2.7 1.2 2.20 *
2-4 619 3.1 1.1 2.71 *
>4 820 4.8 3.5 1.34
(RR 95% CI) % net
migration
All patients (1.36, 1.54) 1.2%
Age
<10 (1.27, 1.78) 1.1%
10-19 (1.67, 2.63) 2.5%
20-39 (1.3, 1.57) 1.4%
>=40 (1.17, 1.46) 0.8%
Sex
Males (1.39, 1.67) 1.4%
Females (1.26, 1.5) 1.0%
Region
Top End (0.86, 1) -0.2%
Central (4.22, 5.71) 5.5%
Australia
Hospitalisations
2 (1.25, 1.6) 0.8%
3-4 (1.32, 1.65) 1.4%
5-9 (1.24, 1.59) 1.6%
10-19 (1.18, 1.84) 1.9%
>19 (1.11, 2.48) 2.2%
Disease category
Pregnancy (1.29, 1.64) 1.4%
Hypertension (0.14, 2.52) -1.9%
IHD (1.04, 1.84) 1.2%
COPD (1.02, 1.53) 0.9%
Renal disease (0.98, 2.29) 1.2%
Diabetes (0.93, 3.11) 1.7%
Alcohol use (1.12, 2.37) 3.5%
Injury (0.55, 2.42) 0.7%
Cancer (1.17, 3.04) 2.0%
Deaths (1.21, 2.51) 1.5%
Proximity to death (years)
0-1 (1.03, 4.69) 1.5%
2-4 (1.13, 6.52) 1.9%
>4 (0.83, 2.17) 1.2%
Note: * 0.01<P<0.05; ** P<0.01; CI=confidence interval; COPD
= chronic obstructive pulmonary disease; IHD = ischaemic
heart disease; RR=risk ratio. Source: the Authors.
Table 4. Parameters Estimated by Linear Regression Model for
Indigenous Patient Migration after Hospitalisation Prior to Death
Coefficient Standard t-value * 95% interval
error confidence
ARIA score of first admission
4 4.65 0.88 5.30 2.93 6.37
5 3.69 0.81 4.54 2.10 5.28
6 6.22 0.17 36.90 5.89 6.55
7 7.40 1.35 5.47 4.75 10.05
8 7.60 0.45 16.95 6.72 8.47
9 7.82 0.23 33.79 7.36 8.27
10 7.80 0.28 28.35 7.26 8.34
11 9.22 0.15 60.10 8.92 9.52
12 9.69 0.28 34.72 9.14 10.23
Proximity to death
Years -0.16 0.02 -8.59 -0.19 -0.12
* All t-values are statistically significant (P<0.01). Source:
the Authors.
Table 5. Risk Ratios Estimated by Log-Linear
Regression Model for Indigenous Patient Migration
RR * 95% CI
Hospitalisations 3-9 1.33 1.16 1.52
10-19 1.72 1.46 2.02
20+ 2.14 1.78 2.58
Age (years) 10-39 1.31 1.18 1.45
40+ 1.60 1.40 1.82
Region Central 1.82 1.66 2.00
Australia
Note: CI=confidence interval; RR=risk ratio;
* All RRs are statistically significant (P<0.01).
Source: the Authors.
