Is There an Association between Dryland Salinity and RossRiver Virus Disease in Southwestern Australia?

Andrew Jardine,1 Peter Speldewinde,1,2 Michael D. A. Lindsay,3 Angus Cook,1 Cheryl A. Johansen,4

and Philip Weinstein1

1School of Population Health, M431, The University of Western Australia, 35 Stirling Highway, Crawley, Western Australia 6009, Australia2Centre of Excellence in Natural Resource Management, The University of Western Australia, Albany, Western Australia 6330, Australia3Mosquito-Borne Disease Control Branch, WA Department of Health, Shenton Park, Western Australia 6008, Australia4Arbovirus Surveillance and Research Laboratory, Discipline of Microbiology and Immunology, The University of Western Australia, Crawley, Western

Australia 6009, Australia

Abstract: Land use change has the potential to cause severe ecosystem degradation and drive changes in

disease transmission and emergence. Broadscale clearing of native vegetation for agriculture in southwestern

Australia has resulted in severe ecosystem degradation, which has been compounded by the subsequent

development of large areas of dryland salinity. The mosquito-borne disease, Ross River virus (RRV), has been

noted as a potential adverse human health outcome in these salinity affected regions. The association between

dryland salinity and RRV disease was therefore tested by undertaking a spatial analysis of disease notification

records using standard and Bayesian techniques. To overcome inherent limitations with notification data,

serological RRV antibody prevalence was also investigated. Neither method revealed a significant association

with dryland salinity, however, the spatial scale imposed limited the sensitivity of both studies. Thus, further

multidisciplinary studies are required to overcome these limitations and advance understanding of this eco-

system health issue, particularly using variables that can be investigated on a finer scale.

Keywords: Dryland salinity, Waterlogging, Ross River virus, Mosquito-borne disease, Western Australia

INTRODUCTION

Rapid and broadscale land clearing for agriculture in inland

southwest Western Australia (WA) has resulted in severe

ecosystem degradation, leaving only small highly frag-

mented patches of natural vegetation (Hobbs, 1993). Land

clearing alters the water balance because annual shallow-

rooted pasture crops use less water than native deep-rooted

perennial native vegetation, resulting in increased recharge

and runoff. This ultimately leads to rising of the water

table, bringing saline water into the root zone of plants

(Clarke et al., 2002). The combination of hypoxia due to

waterlogging and high salinity strongly adversely affects

growth (Barrett-Lennard, 2003), further compounding

ecosystem degradation and biodiversity loss (Cramer and

Hobbs, 2002). Currently, 1.047 million hectares of south-

west WA are affected by waterlogging and dryland salinity,

and this area may expand up to a further 1.7–3.4 million

hectares if trends continue (George et al., 2006).

Soil and water salinization resulting from agricultural

development has been noted to potentially influence

mosquito population composition and associated disease

Published online: February 20, 2008

Correspondence to: Andrew Jardine, e-mail: andrew.jardine@uwa.edu.au

EcoHealth 5, 58–68, 2008DOI: 10.1007/s10393-007-0151-z

Original Contribution

� 2008 International Association for Ecology and Health

risk (Mulla et al., 1987) but has received little research

attention to date. The limited available evidence supports

this observation, indicating that disease risk can both in-

crease or decrease depending on a range of factors specific

to the local setting (Klinkenberg et al., 2004; Temel, 2004).

These two case studies are detailed further in Table 1. These

studies on irrigation-induced salinity and malaria overseas

permit the tentative hypothesis that waterlogging and land

salinization have the potential to alter the species compo-

sition of native mosquito fauna, which may in turn alter the

health risk from mosquito-borne diseases to the local hu-

man population.

Recent research in southwestern Australia has investi-

gated this concept further and indicates that dryland

salinity and waterlogging may have altered the mosquito

ecology in the region by facilitating the inland spread of

Aedes camptorhynchus (Lindsay et al., 2007). This species

has been confirmed as the major vector of Ross River Virus

(RRV) in southern Australia (Ballard and Marshall, 1986;

Russell, 1995) and breeds in brackish to highly saline water,

frequently feeds on humans, and can disperse a long dis-

tance from breeding sites (Lee et al., 1984). Breeding is

primarily associated with coastal saltmarsh habitat, how-

ever, preliminary mosquito collections in inland regions of

southwestern Australia indicate that the species composi-

tion has changed over the past 50 years as the severity of

waterlogging and salinity has increased, and Ae. camp-

torhynchus is now widely established in salinity affected

areas (Lindsay et al., 2007).

Increased surface water salinity and soil waterlogging

may act synergistically to augment breeding of Ae. camp-

torhynchus (Jardine et al., 2007). Increasing salinity nega-

tively influences aquatic biodiversity (Pinder et al., 2005),

resulting in a reduction of salt-tolerant mosquito larval

predators and interspecific resource competition from

other salt-sensitive aquatic biota. Additionally, waterlog-

ging may also provide a greater area of potential mosquito

breeding habitat due to increased volume and duration of

surface water retention induced by waterlogging (Jardine

et al., 2007).

RRV disease is the most common arboviral infection in

Australia with 67,621 laboratory-diagnosed cases reported

in Australia to the end of 2006, since reporting began in

1991 (Communicable Diseases Network Australia, 2007).

RRV disease (also referred to as ‘‘epidemic polyarthritis’’) is

characterized by rheumatic joint manifestations, rash,

constitutional effects and other presentations including

splenomegaly, hematuria, glomerulonephritis, paresthesia, Tab

le1.

Exi

stin

gSt

ud

ies

Inve

stig

atin

ga

Lin

kb

etw

een

Soil

Sali

niz

atio

nan

dM

osq

uit

o-b

orn

eD

isea

se

Ref

eren

ceL

oca

tio

nan

dye

ar(s

)Sa

lin

ity

typ

eE

ffec

tso

nm

osq

uit

op

op

ula

tio

ns

Eff

ects

on

dis

ease

tran

smis

sio

n

Kli

nke

nb

erg

etal

.,20

04P

un

jab

,P

akis

tan

1970

–19

99Ir

riga

tio

n-i

nd

uce

dso

il

sali

niz

atio

nan

dw

ater

logg

ing

Ch

ange

inre

lati

ved

om

inan

ceo

f

the

pri

mar

ym

alar

iave

cto

r,

An

oph

eles

culi

cifa

cies

,to

the

mo

resa

lt-t

ole

ran

tb

ut

less

-effi

cien

tve

cto

r,

An

.st

eph

ensi

Mal

aria

tran

smis

sio

nn

egat

ivel

y

asso

ciat

edw

ith

sali

nit

y

dev

elo

pm

ent

ove

rti

me

Tem

el,

2004

Aze

rbij

an,

1999

Irri

gati

on

-in

du

ced

soil

sali

niz

atio

n

No

tin

vest

igat

ed.

Ho

wev

er,

the

pri

mar

ym

alar

iave

cto

rin

Aze

rbij

an,

An

.sa

char

ovi,

is

mo

der

atel

ysa

lt-t

ole

ran

t

Mal

aria

inci

den

cep

osi

tive

ly

corr

elat

edw

ith

incr

easi

ng

area

of

sali

ne

lan

d

Dryland Salinity and Ross River Virus Disease 59

headache, neck stiffness, and photophobia (Russell, 2002).

Symptomatic disease predominantly affects those aged be-

tween 20–60 years, peaking in the 30–40 years age group.

Infected individuals who develop disease completely re-

cover after a widely variable duration of a few days to over a

year, and thereafter possess long-lasting immunity (Russell,

2002).

In the southwest of WA, RRV activity is primarily

focused in the Swan Coastal Plain between Perth and

Dunsborough (Lindsay et al., 1992, 1996). Accordingly,

regular intensive mosquito and virus surveillance is carried

out in this region (Johansen et al., 2005a). However, small

clusters of cases have also occurred further inland (Lindsay,

1995; Lindsay et al., 2005) where Ae. camptorhynchus has

now become established (Lindsay et al., 2007).

The possibility of an association between dryland

salinity in inland southwest WA and RRV disease has been

suggested previously (Horwitz et al., 2001; Jardine et al.,

2007; Lindsay et al., 2007). The aim of this study was to

directly test this association using RRV disease notification

records and data from a serological survey.

Only two previous surveys of human RRV neutralizing

antibody prevalence have been conducted in WA, both in

known RRV endemic coastal regions of the southwest. A

survey of 271 serum samples from blood donors tested by

neutralization assay in the Peel region (Fig. 1) of WA in

1993 determined a seroprevalence of 8.5% (Johansen,

1998). A subsequent random serological survey was con-

ducted on 806 samples from coastal localities in the

southwest of WA, revealing an overall seroprevalence of

7.2% (Dodsley et al., 2001). Both studies showed a slightly

higher RRV antibody prevalence in males (�60%), and a

greatest prevalence among the 40–60 year-old-age group,

consistent with the studies from the eastern states, reviewed

in Harley et al. (2001).

Interestingly, both studies also showed evidence of a

relatively low antibody prevalence in the region where the

majority of virus notifications occur in the state (Lindsay

et al., 1992, 1996, 2005). Serological studies have not pre-

viously been undertaken on residents from inland regions

of southwest WA.

MATERIALS AND METHODS

Study Region

The study region is illustrated in Figure 1. The estimated

resident population at June 2005 in this region was 31,639

(ABS, 2006b). Future population estimates for this region

indicate that this figure will remain relatively static (in con-

trast to strong population growth forecast for coastal regions

of the southwest), with the estimated resident population in

2031 predicted to increase slightly to 32,480 (Western Aus-

tralian Planning Commission, 2005). The climate is tem-

perate with hot dry summers and cool wet winters. The

average annual rainfall ranges from 700 mm in the west,

decreasing to less than 400 mm in the east. The primary

industry in the region is agriculture, including broad acre

cropping, sheep grazing, and wool production. This region

has been almost completely cleared of all native vegetation

for agriculture and, as a result, a significant area of this region

is now affected by dryland salinity (George et al., 1995).

Data Sources

Data on all doctor-notified and laboratory-reported cases

of RRV from the study region between July 1988 and June

2006 were extracted from the enhanced surveillance data-

base by the Mosquito-Borne Disease Control branch

(MBDC), WA Department of Health (WADOH). In WA, it

is mandatory for doctors to report serologically confirmed

Figure 1. Map of southwestern Australia with the study region

indicated by shading. Numbered areas indicate subregions used in

the serological survey. A more detailed map of the study region is

shown in Figure 2.

60 Andrew Jardine et al.

cases of RRV to the state Department of Health and each

case is then recorded based on place of residence. Further

follow-up is then undertaken to determine the most likely

location of infection, which is not necessarily the place of

residence. The notification is sent to the local government

in which the patient resides and an Environmental Health

Officer in the region administers an enhanced surveillance

questionnaire for each case to determine timing of symp-

tom onset, exposure to biting mosquitoes, and travel his-

tory prior to becoming ill. The MBDC branch then use this

information to record the probable location of infection in

the enhanced surveillance database, rather than simply

recording it as the individual’s residential address as occurs

in other states (Lindsay et al., 2005). This case follow-up

process is completed for approximately 70% of notifica-

tions [M. Lindsay, 2006, unpublished data].

Soil-Landscape Systems data (metadata date: 03/02/

2005) developed by the Natural Resource Assessment

Group of the WA Department of Agriculture was used to

calculate the salinity and waterlogging variables in this

study. Soil-landscape mapping contains a nested hierarchy

of landscape and associated soil units, reflecting both

landscape and soil processes, and also incorporates geo-

logical and environmental variables. The dataset is de-

scribed in detail in Schoknecht et al. (2004).

Population data for each Local Government Area

(LGA) in the study region were obtained from the 1996

Census of Population and Housing (ABS, 1996). This

census dataset was used as it lies closest to the middle of the

RRV notification data range (i.e., July 1988–June 2006).

Calculation of Area of Salinity and Waterlogging

ArcGIS 9 (ESRI, Redlands, CA) was used to map the Soil-

Landscape Systems data and calculate the area of salinity

and waterlogging as a percentage of the total area in each

LGA. The area of salinity was calculated by multiplying the

area of each polygon in the LGA by the percentage of land

defined in the dataset as either at high risk of salinity or

presently saline. All polygons in each LGA were then added

together to give a total area of saline land in each LGA,

which was divided by the total area of the LGA to give a

percentage. Salinity was then coded into the following

tertile categories based on percentage of land affected; 1

(<7.89%), 2 (7.89%–14.32%), and 3 (>14.33%).

Percentage area of waterlogging for each LGA was

calculated in the same way as for salinity using the per-

centage of each polygon defined in the dataset as either at

high or very high risk of waterlogging, then coded into

tertile categories; 1 (<4.53%), 2 (4.53% – 8.54%), and 3

(>8.54%).

Negative Binomial Regression Modeling of

Cumulative Incidence Rate

Negative binomial regression was undertaken in STATA

v9.2 (StataCorp LP, College Station, TX) to test the

hypothesis that increasing area of salinity and/or water-

logging was associated with a high cumulative incidence

rate of RRV. This regression model was selected as good-

ness-of-fit tests using the Pearson statistic indicated that

standard Poisson analysis was inappropriate. The correla-

tion coefficient of 0.673 between the salinity and water-

logging variables indicated that these two variables were

strongly correlated; therefore, both variables were examined

separately to limit problems of colinearity. A combined

indicator of dryland salinity was also constructed by adding

the numerical value of the salinity and waterlogging cate-

gories together, and this indicator was modeled separately

to investigate possible interaction between the two vari-

ables.

Therefore, in the negative binomial regression model,

RRV cumulative incidence was the dependent variable, and

either salinity, waterlogging, or combined dryland salinity

indicator categories were the independent variables. The

1996 population based on census data was used as the offset

variable.

Bayesian Analysis

The mapping of standard incidence rates (SIR) can be

problematic. A small change in the expected value or in the

observed value can produce a large change in the SIR

(Lawson et al., 2003). This process tends to produce highly

unstable estimates with low frequency diseases, particularly

where the base populations are also relatively small (Mar-

shall, 1991). Additionally, SIRs of geographically proximate

areas tend to display positive spatial autocorrelation (Pas-

cutto et al., 2000). Bayesian methods overcome these

concerns regarding classic mapping of disease rates to a

certain extent.

Empirical Bayes models are the most robust as they can

model any random effects arising from the prior distribu-

tion (Lawson et al., 2000). In this context, a random effect

may be considered as an extra quantity of variation (or

variance component) which is estimable within a map and

Dryland Salinity and Ross River Virus Disease 61

which can be ascribed a defined probabilistic structure

(Lawson et al., 2003).

A Besag, York, and Mollie (BYM) model (Besag et al.,

1991; Lawson et al., 2003) was used as it is considered to be

one of the most robust and well-validated for spatial

analysis of disease (Lawson et al., 2000). This model

decomposes area-specific random effects into two compo-

nents. One component takes into account correlated het-

erogeneity (effects that vary in a spatially structured way

and represent the correlation of neighboring locations on a

map), and the other component represents uncorrelated

heterogeneity (effects that vary in an unstructured way and

represent unobserved covariates or confounding variables).

Mathematically, the BYM model is expressed as:

logðhiÞ ¼ aþ bxi þ bi þ hi

where hi corresponds to the relative risk, bxi the explana-

tory variable, bi the correlated heterogeneity and hi the

uncorrelated heterogeneity for the area i.

Models were run through WinBUGS� (MRC Biosta-

tistics Unit, Cambridge, UK) and compared using the

deviance information criterion (DIC) (Spiegelhalter et al.,

2002). Models with lower DIC value are usually selected as

providing the best representation of the data. A burn in of

10,000 iterations was used to ensure convergence before a

sample of 5000 iterations was used to calculate sample

means and DIC. For each WinBUGS simulation, alpha and

betas were monitored to ensure the simulation was

exploring all of the possible distribution space. At the end

of each simulation, the distribution of the alpha and beta

parameters was checked to ensure the simulation had re-

turned a normal distribution. The history of each simula-

tion was also checked to ensure convergence. In most cases,

convergence occurred within 200 iterations.

Serological Survey

The serological study methodology was approved by The

University of Western Australia Human Research Ethics

Committee (Project No RA/4/1/1300). A sampling method

similar to Dodsley et al. (2001) was used to enable com-

parison of results between the studies. Serum samples were

obtained from an existing serum bank maintained by

PathWest Laboratory Medicine WA, the primary public

pathology facility in WA. All samples sent for testing at

PathWest from across the state are stored in the serum

bank for 12 months before being discarded.

Four subregions were selected from within the study

region as detailed in Table 4 (see Fig. 1 for locations of

each subregion). These subregions were selected as they

varied significantly by area of salinity and waterlogging.

Deidentified samples were selected from the PathWest

database by residential postcode as required by the ethics

committee, with dates of sample collection between 10/11/

2004 and 8/29/2005. As mentioned previously, samples sent

for testing for suspected arbovirus infection were excluded

to avoid introducing bias. Every second sample (in chro-

nological order of collection) was excluded in subregion 2

to give approximately equal numbers of samples between

subregions. Duplicate blood samples were excluded so each

sample represented a unique individual.

Each serum sample was tested for the presence of

neutralizing antibodies to RRV in a serum neutralization

assay (Rosenbaum et al., 1972) modified as described by

Johansen (2005b), to determine the seroprevalence of

antibodies to this virus in each subregion. Observed dif-

ferences were tested for statistical significance by chi-

squared tests corrected for continuity using SPSS 14.0

(SPSS Incorporated, Chicago, IL).

RESULTS

RRV Disease Cases in the Study Region

The majority (70%) of the 142 RRV case notifications from

the study region from July 1988 to June 2006 occurred

between December and March (Fig. 2). Distinct variation

in case notifications between years is also evident with

elevated numbers apparent approximately every 3 years

(Fig. 3).

Negative Binomial Regression Model

Figure 4 illustrates the cumulative incidence rate for RRV

between July 1988 and June 2006 for each LGA overlaid

Figure 2. Monthly breakdown of notified RRV disease cases within

the study region between July 1988 and June 2006.

62 Andrew Jardine et al.

on the percentage of salinity in each polygon. Note that

the circles representing the cumulative incidence rate

within each LGA are placed at the centroid, represent the

rate for the whole LGA, and are a continuous variable,

with circle diameter directly proportionate to the inci-

dence rate.

Table 2 contains the cumulative incidence rate ratios

for each independent variable category derived from the

negative binomial regression analysis. The lowest category

of each independent variable (i.e., salinity, waterlogging

= 1 or combined indicator = 2) was set as the baseline

(incidence rate ratio = 1) to which each other strata was

compared. No significant associations were observed be-

tween the RRV cumulative incidence rate and the indi-

vidual nor combined waterlogging and salinity variables.

Bayesian Analysis

The addition of either salinity or waterlogging variables to

the Bayesian model did not significantly lower the DIC

value from the baseline model a + b + h (DIC = 114.069)

Figure 3. Annual (July–June) incidence of RRV disease notifications

from the study region.

Figure 4. Salinity (percentage of

area at high risk or presently

saline) overlaid by RRV cumula-

tive incidence rate for each LGA.

Note that the incidence rate is a

continuous variable and the circle

diameter is directly proportional to

the incidence rate for the whole

LGA, as indicated in the legend.

Table 2. Cumulative Incidence Rate Ratios for Each Indepen-

dent Variable Category Derived from the Negative Binomial

Regression Analysis

RRV cases

Cumulative incidence

rate ratio

95% CI

Salinity category

1 (baseline) 1.00

2 1.96 0.85–4.55

3 1.22 0.83–1.78

Waterlogging category

1 (baseline) 1.00

2 0.54 0.23–1.25

3 0.95 0.65–1.37

Combined index of environmental degradation

(waterlogging + salinity)

2 (baseline) 1.00

3 2.16 0.68–6.82

4 0.81 0.32–2.05

5 1.18 0.79–1.75

6 1.15 0.88–1.50

Dryland Salinity and Ross River Virus Disease 63

(Table 3), and therefore it is unlikely that salinity or

waterlogging explains the distribution of RRV cases using

this Bayesian model.

Serosurvey

The greatest prevalence of neutralizing antibodies to RRV

was observed in Kojonup (Table 4), one of the lower saline

and waterlogged subregions; however, none of the differ-

ences between the subregions were significant.

Across the four subregions, there was a large statisti-

cally significant (P value = <0.01) disparity in RRV anti-

body seroprevalence between the genders, with a male-to-

female prevalence rate ratio of 4.5:1.0 (Table 5). There was

a general trend of increasing seroprevalence with age group,

however, this was not statistically significant (P value =

0.103) (Table 5). After adjustment for age and sex based

on 2005 estimated resident population data (ABS, 2006a),

the seroprevalence across the four subregions was calcu-

lated to be 5.4%.

DISCUSSION

Notification Data

The negative binomial regression modeling using cumula-

tive incidence rates did not show any significant association

between RRV notification incidence between July 1988 and

June 2006 and salinity, waterlogging, or a combined indi-

cator of both.

It is important to note that although the best available

enhanced case notification and environmental exposure

data were used, the analysis was constrained by the fact that

a relatively rare disease was being investigated in a sparse

human population. It is thus subject both to possible

exposure misclassification and unstable disease rate esti-

mates (Marshall, 1991).

Table 3. Deviance Information Criterion (DIC) for Ross River

Virus at the Local Government Area Levela

Model DIC

a 132.177

a + b 113.531

a + b + h 114.069

a + sal + b + h 114.922

a + watl + b + h 113.394

asal, presence of salinity; watl, presence of waterlogging.

Table 5. RRV Antibody Prevalence from Inland Southwest WA

by Gender and Age Group

Gender RRV-antibody-positive

samples

Samples

tested

Percentage

with antibodies

Male 19 231 8.23

Female 6 320 1.88

Age group

£ 19 years 2 67 2.99

20–29 years 2 136 1.47

30–39 years 6 127 4.72

40–49 years 7 93 7.53

50–59 years 2 61 3.28

‡60 years 6 67 8.96

Total 25 551 4.54

Table 4. Subregions Used in the Serosurvey and the Percentage of Their Area Affected by Salinity and Waterlogginga

Subregion Salinityb (%) Water loggingb (%) No. of samplesc No. posd Preve (%) RRV cuml inc./1000f

1: Boddington, Williams 4.15 0.59 103 5 4.85 2.94

2: Narrogin Town and Shire 14.32 6.22 160 5 3.13 2.37

3: Wagin, Dumbleyung 16.06 5.43 142 6 4.26 2.44

4: Kojonup 7.89 1.86 146 9 6.12 5.82

Total 551 25 4.54

aSee Figure 1 for location within study region. Prevalence of neutralizing antibodies to RRV in residents of four subregions with differing severity of salinity

and waterlogging.bPercentage of subregion affected.cNumber of blood samples tested.dNumber of samples positive for neutralizing antibodies to RRV.ePrevalence of neutralizing antibodies to RRV.fCumulative incidence rate (7/88–6/06) per 1000 population.

64 Andrew Jardine et al.

These limitations in spatial analysis of RRV have been

reported in other studies. An assessment of RRV disease at

the LGA scale has previously been carried out in Queens-

land, the state with the highest RRV incidence in Australia

(Gatton et al., 2004). This study used cumulative notifi-

cation data over a 10-year period to estimate an average

seasonal incidence rate for each LGA. Despite the large

number of cases included in the analysis, the authors rec-

ognized that a spatial analysis at the broad LGA scale could

only provide a general indication of disease patterns across

the state.

Ryan et al. (2006) conducted a spatial analysis of RRV

cases in Redland Shire, a densely populated urban region in

coastal southeast Queensland. They conducted the analyses

over a range of scales and, with the dense population and

large numbers of RRV notifications in the region, were able

to utilize a maximum resolution of a 0.25 km2 grid. They

demonstrated clearly that ‘‘the use of arbitrary adminis-

trative boundaries for spatial analyses has the potential to

mask any small-scale heterogeneity in disease patterns’’

(Ryan et al., 2006). This is also equally true for the envi-

ronmental data. Salinity and waterlogging are, of course,

variable within each LGA, and by assignment of an overall

LGA percentage, this spatial heterogeneity was not taken

into account. However, given the sparse population and

resultant low case numbers in the present study region,

aggregation of data to the LGA level was the only option

available.

The Bayesian analysis overcame some of the limitations

relating to the small resident population size and resultant

low number of RRV notifications in the study region. Again

no evidence of a significant link between RRV cumulative

incidence and either salinity or waterlogging in inland

southwest WA was determined. However, RRV notification

data are thought to represent a significant underestimate of

the true incidence (Russell, 2002). Therefore, the seropre-

valence of neutralizing antibodies to RRV in residents of

four regions of inland southwest WA were also determined

with respect to dryland salinity and waterlogging.

RRV Antibody Seroprevalence in Inland Southwest

WA

The results from this cross-sectional study indicated (i) an

overall crude seroprevalence of 4.5%; and (ii) an age–sex

adjusted seroprevalence of 5.4% across the four subregions

studied. Direct comparisons to other studies can be prob-

lematic due to different sampling procedures and study

design. A similar sample selection procedure was used by

Dodsley et al. (2001), although this group used the hem-

agglutination inhibition method to initially screen all

samples. The crude prevalence of 4.3% across the four in-

land regions surveyed in the southwest of the state found in

our study was lower than the 7.8% reported by Dodsley

et al. (2001) from coastal regions. The only other previous

RRV antibody survey in southern WA reported a preva-

lence of 8.4% from a sample of blood donors in the coastal

Peel region, where RRV is known to be enzootic (Johansen,

1998). This difference in seroprevalance between the inland

and coastal regions is reflected by the enhanced surveillance

notification data. The annual average notification rate of

RRV per 100,000 residents between July 1995 and June

2007 across the four inland regions included in the current

serosurvey was 20.33, compared to 61.12 in the coastal Peel

region [M. Lindsay, unpublished data].

An elevated prevalence of RRV infection with age was

evident in the current study. This trend is commonly ob-

served in antibody prevalence studies in regions where RRV

is endemic (Harley et al., 2001), however, the increasing

trend was not statistically significant and cannot be used as

evidence of RRV being endemic in the study region. The

male-to-female ratio of 4.5:1.0 was statistically significant

and much larger than previously observed in WA (Dodsley

et al., 2001; Johansen, 1998) or other parts of Australia

(reviewed in Harley et al., 2001). Some of this disparity may

be explained by gender-based differences in the location

and nature of occupations which put individuals at risk of

exposure to RRV. Statistics show that across the four

subregions, 20% of males work in the typically outdoor

occupations of agriculture, forestry, or fishing, compared

with only 8% of females. Females were more likely to work

in typically indoor jobs, with 22% working in retail, edu-

cation, or health and community services, compared with

only 7% of males (ABS, 2001). Therefore, the greater ten-

dency for males to work in outdoor occupations may lead

to increased exposure to potentially infected mosquitoes

and explain the increased seroprevalence observed.

Seroprevalence and Dryland Salinity/Waterlogging

There was no significant association between the prevalence

of antibodies to RRV in humans and residence in areas

affected by salinity or waterlogging. Indeed, the subregion

with the greatest seroprevalence had a relatively low per-

centage area of both variables at the LGA level. The results

reflect the analysis of RRV notifications and do not provide

Dryland Salinity and Ross River Virus Disease 65

evidence to support the hypothesis that salinity and

waterlogging in inland agricultural regions of southwest

WA has significantly increased the risk of RRV infection in

the resident human population.

Limitations

The fact that a relatively uncommon disease was being

investigated in a sparse human population meant it was

necessary to use a large base unit of analysis both spatially

(LGA) and temporally (cumulative incidence over 18

years). This limited the capacity of the studies described in

this article to detect small changes in disease risk.

Another important factor to consider in this study is

population movement between regions, as individuals with

RRV may have acquired the infection elsewhere. Popula-

tion mobility data show that between the 1996 and 2001

censuses, there was a net population efflux of 9.5% and

0.7% from the Upper and Lower Great Southern statistical

divisions, respectively (which incorporate the study region)

(ABS, 2003). More detailed information regarding regional

population movements in WA were obtained in a series of

reports commissioned by the Regional Development

Council, Government of Western Australia using focus

groups (2–3 per region) and telephone interviews (150 per

region). The reports relevant to the study region indicated

relatively high population stability compared with other

regions in the state (Patterson Market Research, 1998ab).

While the relative stability of the population in the study

region indicates it is unlikely that population mobility will

exert significant bias on the results, the possibility that

some RRV-antibody-positive individuals identified in this

study may have acquired the infection elsewhere still exists.

Travel behavior of residents from the study region is

also important; unfortunately, no information is available

regarding specific travel behavior of residents in the study

region. However, there is no reason to assume that travel

behavior would differ between those living in saline versus

non-saline regions.

CONCLUSION

Using both disease notification and serological data, no

evidence could be found to support the hypothesis that

there is a significant association between dryland salinity

and RRV infection in inland southwest WA. In certain sit-

uations (such as the sparse human population spread over

a large area, as encountered in the current study), varia-

tions in disease rates are difficult to detect. Thus, the po-

tential for dryland salinity to increase RRV disease risk

cannot be entirely ruled out; however, if any association

does exist, it is likely to be small.

The studies described in this article represent an

important first step and raise a number of avenues for

further research. A similar study in dryland salinity affected

regions of eastern Australia where human population

densities are greater may overcome some of the limitations

faced in the current study, and further develop a better

understanding of the potential for dryland salinity to

influence RRV disease risk. Epidemiological studies also

need to be supplemented by further ecological or envi-

ronmental data to more clearly delineate health hazard.

Field-based studies of variables that can be investigated on

a smaller scale would be most useful, such as mosquito

population dynamics, and mosquito and vertebrate host

infection rates. Until such further studies are undertaken,

there is no evidence to suggest that changes are necessary to

currently used mosquito-borne disease control practices in

saline regions.

It is also worth noting that the area affected by dryland

salinity in southwest WA is predicted to expand by up to

three times to 3.4 million hectares over coming decades

(George et al., 2006). While increasing efforts to ameliorate

dryland salinity on the grounds of reducing RRV risk is

currently not justified based on the available evidence, this

situation may change as dryland salinity spreads and new

evidence comes to light. Ultimately, however, amelioration

of dryland salinity is clearly warranted on environmental

grounds alone; therefore, any potential reduction in human

health risk is simply an added benefit from an ecosystem

health perspective, and is consistent with the current rec-

ommendations for sustainable environmental management

(Millennium Ecosystem Assessment, 2005).

ACKNOWLEDGMENTS

Parts of this work were supported by the Cooperative

Research Centre for the Plant Based Management of

Dryland Salinity. Sarah Power provided technical support

in the laboratory, and Scott Carver assisted with useful

discussions regarding some of the issues presented. The

Department of Agriculture and Food Western Australia is

acknowledged for providing the Soil-Landscape Systems

Data.

66 Andrew Jardine et al.

REFERENCES

ABS (1996) Census of Population and Housing, 1996. Canberra,Australia: Australian Bureau of Statistics

ABS (2001) Census of Population and Housing, 2001. Canberra,Australia: Australian Bureau of Statistics

ABS (2003) Census of Population and Housing: Population Growthand Distribution, Australia, 2001. ABS Catalogue No 2035.0,Canberra, Australia: Australian Bureau of Statistics

ABS (2006a) Population by age and sex, Western Australia, June2005. ABS Catalogue No 3235.5.55.001, Canberra, Australia:Australian Bureau of Statistics

ABS (2006b) Regional Population Growth Australia, 2004–2005.ABS Catalogue No 3218.0, Canberra, Australia: Australian Bu-reau of Statistics

Ballard JWO, Marshall ID (1986) An investigation of the potentialof Aedes camptorhynchus as a vector of Ross River virus. Aus-tralian Journal of Experimental Biology and Medical Science64:197–200

Barrett-Lennard EG (2003) The interaction between waterloggingand salinity in higher plants: causes, consequences and impli-cations. Plant and Soil 253:35–54

Besag J, York J, Mollie A (1991) Bayesian image restoration withtwo applications in spatial statistics. Annals of the Institute ofStatistical Mathematics 43:1–59

Clarke CJ, George RJ, Bell RW, Hatton TJ (2002) Dryland salinityin south-western Australia: its origins, remedies, and futureresearch directions. Australian Journal of Soil Research 40:93–113

Communicable Diseases Network Australia (2007) NationalNotifiable Diseases Surveillance System. Communicable DiseasesNetwork Australia, Department of Health and Ageing. Avail-able: http://www9.health.gov.au/cda/source/Rpt_4_sel.cfm [ac-cessed January15, 2007]

Cramer VA, Hobbs RJ (2002) Ecological consequences of alteredhydrological regimes in fragmented ecosystems in southernAustralia: impacts and possible management responses. AustralEcology 27:546–564

Dodsley NA, Broom AK, Smith DW, Plant AJ, Lindsay MD (2001)Ross River virus: determining the prevalence in the south westof Western Australia. Arbovirus Research in Australia 8:122–125

Gatton ML, Kelly-Hope LA, Kay BH, Ryan PA (2004) Spatial-temporal analysis of Ross River virus disease patterns inQueensland, Australia. American Journal of Tropical Medicineand Hygiene 71:629–635

George R, Clarke JA, English P (2006) Modern and palaeogeo-graphic trends in the salinisation of the Western Australianwheatbelt. Proceedings of the Australian Earth Sciences Conven-tion 2006, Melbourne. Available: http://www.earth2006.org.au/papers/extendedpdf/George%20Richard%20-%20Modern%20and%20palaeogeographic-extended.pdf [accessed September22, 2006]

George RJ, McFarlane DJ, Speed RJ (1995) The consequences of achanging hydrologic environment for native vegetation insouthwestern Australia. In: Saunders DA, Craig JL, Mattiske EM(editors), Nature Conservation 4: the Roles of Networks. Syd-ney, Australia: Surrey Beatty & Sons, pp 9–22

Harley D, Sleigh A, Ritchie S (2001) Ross River virus transmission,infection, and disease: a cross-disciplinary review. ClinicalMicrobiology Reviews 14:909–932

Hobbs RJ (1993) Effects of landscape fragmentation on ecosystemprocesses in the Western Australian wheatbelt. Biological Con-servation 64:193–201

Horwitz P, Lindsay M, O’Connor M (2001) Biodiversity, ende-mism, sense of place, and public health: inter-relationships forAustralian inland aquatic systems. Ecosystem Health 7:253–265

Jardine A, Speldewinde P, Carver S, Weinstein P (2007) Drylandsalinity and Ecosystem Distress Syndrome: human healthimplications. EcoHealth 4:10–17

Johansen CA (1998) The Ecology and Epidemiology of CertainArboviruses in the South of Western Australia. MSc Thesis,Department of Microbiology, University of Western Australia,Perth, Australia

Johansen CA, Broom AK, Lindsay MDA, Maley FM, Power SL,Gordon CJ, et al. (2005) Surveillance of arboviruses in mos-quitoes from the southwest of Western Australia between 2000and 2004. Arbovirus Research in Australia 9:159–163

Johansen CA, Mackenzie JS, Smith DW, Lindsay MD (2005)Prevalence of neutralising antibodies to Barmah Forest, Sindbisand Trubanaman viruses in animals and humans in the south-west of Western Australia. Australian Journal of Zoology 53:51–58

Klinkenberg E, Konradsen F, Herrel N, Mukhtar M, van der HoekW, Amerasinghe FP (2004) Malaria vectors in the changingenvironment of the southern Punjab, Pakistan. Transactions ofthe Royal Society of Tropical Medicine and Hygiene 98:442–449

Lawson AB, Biggeri AB, Boehning D, Lesaffre E, Viel J-F, Clark A,et al. (2000) Disease mapping models: an empirical evaluation.Statistics in Medicine 19:2217–2241

Lawson AB, Browne WJ, Vidal Rodeiro CL (2003) Disease Map-ping with WinBUGS and MlwinN. Senn S (editor), Chichester,UK: Wiley

Lee DJ, Hicks MM, Griffiths M, Russell RC, Marks EN (1984) TheCulicidae of the Australasian Region, Vol 3. EntomologyMonograph No 2, Canberra, Australia: Australian GovernmentPublishing Service

Lindsay M, Oliveira N, Jasinska E, Johansen C, Harrington S,Wright AE, et al. (1996) An outbreak of Ross River virus diseasein southwestern Australia. Emerging Infectious Diseases 2:117–120

Lindsay MD (1995) Ecology and Epidemiology of Ross River Virusin Western Australia. PhD Thesis, Department of Microbiology,University of Western Australia, Perth, Australia

Lindsay MD, Condon RJ, Mackenzie JS, Johansen CA, D’ErcoleM, Smith DW (1992) A major outbreak of Ross River virusinfection in the south-west of Western Australia and the Perthmetropolitan area. Communicable Diseases Intelligence 16:290–294

Lindsay MDA, Breeze AL, Harrington SA, Johansen CA, BroomAK, Gordon CJ, et al. (2005) Ross River and Barmah Forestviruses in Western Australia, 2000/01-2003/04: contrastingpatterns of disease activity. Arbovirus Research in Australia9:194–201

Lindsay MDA, Jardine A, Johansen CA, Wright AE, HarringtonSA, Weinstein P (2007) Mosquito (Diptera: Culicidae) faunaand the potential for mosquito-borne disease in inland areas ofsouth west Western Australia. Australian Journal of Entomology46:60–64

Marshall M (1991) A review of methods for the statistical analysisof spatial patterns of disease. Journal of the Royal StatisticalSociety–Series A 154:421–441

Dryland Salinity and Ross River Virus Disease 67

Millennium Ecosystem Assessment (2005) Ecosystems and HumanWell-being: Health Synthesis. Washington, DC: Island Press

Mulla MS, Mian LS, Gratz NG (1987) Agricultural managementpractices—their impacts on production of vector and pestmosquitoes. Journal of Agricultural Entomology 4:97–131

Pascutto C, Wakefield J, Best NG, Richardson S, Bernardinelli L,Staines A, et al. (2000) Statistical issues in the analysis of diseasemapping data. Statistics in Medicine 19:2493–2519

Patterson Market Research (1998a) Living in the regions: theWestern Australians: the Great Southern report. Patterson MarketResearch in conjunction with Focused Management and Ha-mesSharley: Perth. Available: http://www.dlgrd.wa.gov.au/Publications/StatInfo/LivingInRegions [accessed July 20, 2006]

Patterson Market Research (1998b) Living in the regions: theWestern Australians: the Wheatbelt report. Patterson MarketResearch in conjunction with Focused Management and HamesSharley: Perth. Available: http://www.dlgrd.wa.gov.au/Publications/StatInfo/LivingInRegions [accessed July 20, 2006]

Pinder AM, Halse SA, McRae JM, Shiel RJ (2005) Occurrence ofaquatic invertebrates of the wheatbelt region of Western Aus-tralia in relation to salinity. Hydrobiologia 543:1–24

Rosenbaum MJ, Sullivan EJ, Edwards EA (1972) Techniques forcell cultivation in plastic microtitration plates and their appli-cation in biological assays. In: Wasley GD (editor), AnimalTissue Culture. Advances in Techniques. London: Butterworth,pp 49–81

Russell RC (1995) Arboviruses and their vectors in Australia: anupdate on the ecology and epidemiology of some mosquito-borne arboviruses. Review of Medical and Veterinary Entomology83:141–158

Russell RC (2002) Ross River virus: ecology and distribution.Annual Review of Entomology 47:1–31

Ryan PA, Alsemgeest D, Gatton ML, Kay BH (2006) Ross Rivervirus disease clusters and spatial relationship with mosquitobiting exposure in Redland Shire, Southern Queensland, Aus-tralia. Journal of Medical Entomology 43:1042–1059

Schoknecht N, Tille P, Purdie B (2004) Soil-landscape mapping insouth-western Australia: overview of methodology and inputs.Department of Agriculture, Government of Western Australia.Resource management technical report 280. Available:http://www.agric.wa.gov.au/pls/portal30/docs/FOLDER/IKMP/LWE/LAND/TR280.PDF [accessed August 20, 2004]

Spiegelhalter DJ, Best NG, Carlin BP, van der Linde A (2002)Bayesian measures of model complexity and fit. Journal of theRoyal Statistical Society—Series B 64:583–639

Temel T (2004) Malaria from the gap: need for cross-sector co-operation in Azerbaijan. Acta Tropica 89:249–259

Western Australian Planning Commission (2005) Western Aus-tralia Tomorrow. Western Australian Planning Commission.Population report No 6. Available: http://www.wapc.wa.gov.au/Publications/723.aspx [accessed August 21, 2006]

68 Andrew Jardine et al.