Secondary Logo

Journal Logo

Original Article

Early Warning of Ross River Virus Epidemics: Combining Surveillance Data on Climate and Mosquitoes

Woodruff, Rosalie E.*; Guest, Charles S.*; Garner, Michael G.; Becker, Niels*; Lindsay, Michael

Author Information
doi: 10.1097/01.ede.0000229467.92742.7b
  • Free


Ross River virus is an arbovirus transmitted by mosquitoes. It causes a painful polyarthritic disease and, in most people, the symptoms persist for 3–6 months.1,2 The arbovirus is established throughout Australia, and outbreaks of the disease have been reported in Papua New Guinea3 and several countries in the Pacific.4,5 The average number of new cases in Australia each year is approximately 5000. Although all ages are at risk for infection, people ages 20–55 years are most at risk for contracting the disease. A partial estimate of the annual costs of symptom management and productivity losses is AUD$ 3.1–6 million,6 although this sum does not account for public health responses or full diagnostic and medical costs.7 The lack of a treatment or a vaccine for this disease makes prevention of infection the main public health strategy.

Ross River virus disease is one of the few infectious diseases that can be predicted by climate-based early warning systems.8 Changes in climate strongly influence the replication of the virus,9 the breeding, abundance and survival of the mosquito vector,10–12 and the breeding patterns of alternative hosts.13 Furthermore, the incidence of the disease varies with climatic conditions.14,15 The epidemiology of the disease differs across the Australian continent,16 with an endemic pattern of notifications in the northern tropical region. In the temperate and arid parts of the country there is wide interannual variation in disease occurrence, filling one criterion for early warning of outbreaks.7,8

The projected level of global warming within the next 80 years (1.4–5.8°C by 2080)17 is anticipated to cause changes in the distribution and pattern of mosquito-borne diseases. Whether these changes result in increased numbers of cases of disease will depend on many interacting environmental, behavioral, and adaptive factors, as well as the dynamics of regional transmission of Ross River virus. To plan for future climate change, it is important to document the sensitivity of the relation between climate and Ross River virus transmission.

In previous research, we demonstrated that data on climate at relatively coarse temporal and spatial resolution can be used to predict Ross River virus disease epidemics in south-eastern Australia, with enough accuracy and lead-time to be of use for planning public health actions.14 In this report, we describe a study conducted in the southwest of Western Australia, where large outbreaks of Ross River virus disease have been reported since the late 1980s16,18,19; the largest outbreak, with 1507 cases, occurred in 2003–2004. A number of environmental and climatic factors have been observed to coincide with disease outbreaks in this region; these include anomalously high mean sea levels, tide heights, and rainfall. No model of this relationship has been developed. The aim of this study was two-fold: to ascertain whether the methods we used to develop climate-based early warning models in our previous research could be applied to a different region and to test whether adding mosquito surveillance data to routinely collected data on climate could increase the accuracy of prediction models.


Ross River virus inhabits multiple ecological niches, each with different vector and vertebrate species and temporal transmission patterns. Previously, we described a method for defining the boundaries of bioclimatic regions for the purpose of modeling Ross River virus transmission.14 In brief, we identified 43 different regions on the basis of long-term average climate and environmental factors that influence the distribution and seasonal breeding patterns of mosquitoes and of kangaroos, which are the putative main vertebrate host of Ross River virus in nonurban settings.6 From these bioclimatic regions, we chose one in south-west Western Australia (the “Southwest”), which comprises 14 Statistical Local Areas, the base spatial unit for analysis (Fig. 1).

Statistical local areas and location of mosquito trapping areas within the southwest study region of Australia.

Table 1 shows the main characteristics of the region. The coastal region from Mandurah to Busselton (the Swan coastal plain) comprises major estuaries, tidal wetlands and freshwater lakes. High tides in midspring through summer inundate broad areas of salt marsh and provide ideal breeding conditions for huge numbers of mosquitoes.20 Previous studies16,21,22 indicate that the western gray kangaroo (Macropus fuliginosis) is the primary vertebrate host of the virus in this region. Kangaroos are thought to have a short period of viraemia (6 days),9 after which they are immune from reinfection. In this region, Ochlerotatus camptorhynchus is the main mosquito vector (approximately 75% of total mosquitoes trapped during an 8-year period) and Oc. vigilax is a minor vector (8% of total).

Profile of the Southwest Study Region of Australia, July 1991 to June 1999

The region has a temperate climate with strongly seasonal rainfall. Typically, there is a pronounced summer drought23 and, on average, 88% of rain falls between April and October (the austral winter). Figure 2 shows the long-term average rainfall and the maximum and minimum temperature values for the Southwest. Rainfall over the southern tip of Western Australia has been weakly related to El Niño events, with an estimated average 10% less rain falling at those times.24 Total rainfall has been more closely related to La Niña events, increasing by more than 30% at those times.

Long-term average rainfall and maximum and minimum temperature profiles for the southwest study region. Average maximum temperature (°C) ▴; Average minimum temperature (°C) ▪; Rainfall (mm) •.


Ross River virus disease is a notifiable condition in Australia and reporting quality is considered to be high. We obtained notifications of the disease by Statistical Local Area from the Commonwealth Department of Health and Ageing for the period July 1991 to June 1999. The ratio of subclinical to clinical infections is likely to lie between 3:1 and 1.2:1.6 The protocol and mechanisms for reporting, the public health system, and the general practitioner and laboratory networks are the same for all Statistical Local Areas across this region. In addition, Western Australia is one of only 2 States in Australia that conducts routine case follow-up for Ross River virus disease to determine, among other things, the likely place of infection. For these several reasons the risk of spatial misclassification of notifications is low. Public Health Units and Local Governments liaise with general practitioner networks throughout the risk season (in both epidemic and nonepidemic years) to reinforce the importance of testing suspected cases. We think that year-to-year differences in reporting practice (ie, potential misclassification of epidemic years as nonepidemic ones) is minimal in this region.

Monthly climate data for the period and long-term averages were obtained from the Queensland Department of Natural Resources and the Bureau of Meteorology25 and averaged over each of the Statistical Local Areas. The interpolation method, density of stations in the region, and topography of the area make the possibility of exposure misclassification low. Data on tide height at Bunbury were obtained from the Coastal Management Branch of the Maritime Division of Transport, Western Australia. These data and their quality have been described previously.14Table 2 summarizes the climate data, which include tidal data.

Explanatory Variables Used in Multiple Logistic Regression Models, Units of Measurement, and Methods for Calculating Monthly Values

The University of Western Australia provided data on mosquito populations for the spring and summer months of the study period. Adult mosquito populations and Ross River virus activity have been monitored regularly since 1987 at up to 40 sites between Rockingham and Busselton (Fig. 1).22 The data for this study come from a “core set” of these traps that have not changed their location across this period. Mosquitoes were collected each fortnight between late spring and early autumn and once a month for the remainder of the year. Traps had been placed in regions where Ross River virus was endemic, where high attack rates had been reported and where viral activity appears to commence and then spread to other areas.20 Mosquito trap numbers per month were categorized as: 1 (0–99 mosquitoes), 2 (100–199 mosquitoes), 3 (200–299 mosquitoes), or 4 (>300 mosquitoes). Because mosquito data were available for 9 of the 14 Statistical Local Areas in the Southwest (Fig. 1), the contribution of mosquito counts to the predictive model was tested for these areas only. Mosquito numbers for the epidemic and pre-epidemic year (“lagged mosquito numbers”) were explanatory variables. We assumed that mosquito numbers in the pre-epidemic year were a proxy for vertebrate host immunity levels as, when mosquito numbers are high, more kangaroos are infected with the virus and the population immunity levels are likely to be high. Conversely, when there are few mosquitoes we assumed that a low level of virus would be circulating in the kangaroo population, and that a greater proportion of juvenile vertebrates would be susceptible. For this reason climate variables that influence mosquito abundance were also lagged by 1 year.

Model Structure and Validation

To summarize the methods,14 data were entered into multivariate logistic regression models, with the statistical package Stata (StataCorp, College Station, TX). The outcome variable was an epidemic in a Statistical Local Area (defined as more cases than the mean plus 1 standard deviation of all cases during the period for that area). Thus, for the 14 Statistical Local Areas, there was a possible range of 0–14 epidemics each year of the 8 years of the study. The unit for the explanatory variables was 1 month. We developed “early” and “later” warning models. The early warning model included variables for July to November (austral winter to late spring). The earliest cases commenced in October, and mosquito control in December was assumed to be effective in reducing mosquito breeding. In the later warning model, we added variables for December to January (austral summer). In most years, the majority of cases did not appear until February–March. Public alerts at that time could still make it possible to reduce the magnitude of an epidemic if the results of the modeling were positive. For both the early and later warning periods, we developed models including climate variables alone (“climate models”), to which we later added mosquito variables (“climate + mosquito models”). We used the definition of best fit as in the previous report,14 which is that the model should have more than 90% accuracy in predicting outcomes and be significant on a Hosmer-Lemeshow goodness-of-fit test.26

We validated the models within the dataset. In a rotating fashion, 1 year was systematically dropped and the remaining 7 years were used to derive the parameter estimates for the model and the predicted probability of an epidemic in the remaining one-eighth of the data. From these probabilities, we calculated the accuracy, sensitivity, specificity, positive predictive value and negative predictive value of the model for predicting epidemics for each year and Statistical Local Area, during the study and for the period overall. Validation of the climate +mosquito models was affected by a lack of mosquito data for the year 1996–97 and hence of a lagged mosquito variable for 1997–1998. Consequently, the validation tests for these models show the range of test results that would be expected if data for the additional 1 (or 2) years had been available.


There were 1507 notifications for the Southwest, 75% of which were for persons age 25–59 years. The 8-year average annual notification rate for the region was 85 per 100 000 (Table 1), which is substantially higher than the national average of 25 per 100 000 for the same period. Because it is likely that most of the notifications were of cases with typical and severe clinical symptoms, rather than mild symptoms, the number probably represent an underestimate of the true incidence of disease.

Early Warning Model


The sea surface temperature in October–November (the later months of spring) was the climate variable most strongly associated with epidemics (Table 3). Late spring rainfall and high tides were also associated. The number of days of rain in November was a better predictor than total rainfall in that month. The best predictor of tidal influence was the absolute tide height in October–November. The mean for this variable over the period was 110 cm, and 86% of epidemics occurred in years when tide heights were greater than this level; 54% of epidemics in the period occurred during the year of the highest tide height. Lagged days of rain in August–December (ie, in the pre-epidemic year) was negatively associated with epidemics. We retained this variable in the model as a proxy for vertebrate host immunity levels, an essential component of the transmission cycle. The model with the best fit had the form

Logistic Regression Models for the Early and Later Warning Periods in the Southwest

Climate + Mosquito

The logistic regression model with the best fit included the variables from the climate model and lagged November mosquito numbers, which was negatively associated with epidemics in the following year. “Oc. camptorhynchus count” contributed more to all models than either “Oc. vigilax count” or “total mosquito count.” The model with the best fit had the form

Later Warning Model


None of the climate variables in December or January improved the prediction of epidemics over the level already achieved with the early warning climate model.

Climate + Mosquito

Other than lagged days of rain in August–December, the climate variables from the early warning model dropped out of the model when November–December mosquito numbers (in the epidemic year) was added. This variable had a significant, positive association with epidemics (Table 3). The best model was determined by


Table 4 shows the results of the validation test for all models. The climate early warning model had a sensitivity of 64% and a specificity of 96%. The accuracy for the period was 89%. The addition of mosquito data to this model dramatically improved the sensitivity to 90% (range, 56–94%), with only a small drop in specificity. The sensitivity for the climate + mosquito later warning model was 5% lower; however, the confidence intervals for this sensitivity were narrower (69–88%), and the specificity and accuracy were higher. This model was estimated from data on mosquito numbers over 7 years and is consequently likely to be more precise than the early warning model (estimated from 6 years’ data). All the models predicted nonepidemics better than epidemics.

Sensitivity Parameters of the Early and Later Warning Models for the Southwest


Epidemics in the Southwest region of Australia can be predicted with high sensitivity and specificity by a model that combines data from mosquito trapping and climate surveillance. Climate data alone predicted Ross River virus disease epidemics with moderate sensitivity (64%) by November. Adding mosquito surveillance data to the model increased prediction to 85–90%. In the early warning period, this would provide time for mosquito control. At the later warning stage, public education campaigns that promote the effectiveness of using mosquito repellents27 could reduce the number of cases. These findings indicate that the methods designed to predict epidemics of Ross River virus disease in south-eastern Australia14 can be applied successfully to other epidemic regions of Australia.

Our understanding of the biology of Ross River virus transmission suggests that the major determinants of Ross River virus infection are (1) infected mosquitoes and (2) susceptible nonhuman vertebrate hosts. The motivation for using data on climate is that these are surrogates for aforementioned (1) and (2). Indeed, data on climate do provide predictions of value. However, once data on mosquito counts are included, climate data do not improve predictions—with the exception of lagged climate variables which act as a surrogate for (2).

The following factors should be considered in evaluating the results. First, with respect to the exposure, continuous data on tides were available for only one fixed point along the coast, at Bunbury (at the center of the Swan Coastal Plain). We assumed these to be broadly representative of trends for the affected regions. Tide heights might, however, vary to a greater extent to the north and south of this point, although there is no information about the direction or magnitude of such possible differences. Second, data on mosquito numbers were not available for one year of the study, thus increasing the confidence interval around the validation test results. Third, mosquitoes were not trapped in all Statistical Local Areas in the Southwest, and we assumed that the numbers in Statistical Local Areas adjacent to collection sites would be the same. As trap levels within a single Local Area can vary,20 however, further work is needed to establish whether this assumption is reasonable. We would expect more detailed information on tides and mosquitoes to improve prediction.

In any ecological study, specific factors may modify or confound the association between exposure and disease. Changes in public health management of Ross River virus disease over time, such as mosquito control and campaigns aimed at increasing public practice of personal protection activities, may reduce human population vulnerability. Low-level adulticiding and application of larvicides was conducted in some northern and central regions of the study area. Given the large geographic areas over which spraying was conducted during this period, we consider that the affect of vector control on the outcome (epidemic/nonepidemic) was small. Public health protection messages were issued on a fairly consistent basis throughout the study period. Regarding the outcome, it is possible that people with asymptomatic infection—as well as those with disease—could have contributed to the amplification cycle, in addition to mosquitoes. However, because there is no evidence to suggest that people with asymptomatic infection would favor any particular location we think this is an unlikely source of bias.

The data on climate for the later warning period (December and January) did not increase the sensitivity of the model above that already provided by the data for previous months, suggesting that the preconditions for an epidemic have been established by November and that climate conditions after this do not determine the occurrence of epidemics in this region, although they can be expected to have substantial bearing on the scale of an epidemic.

The relation between the main climatic drivers (sea surface temperature, tide height, and rainfall) and an abundance of infectious mosquitoes is mediated by host population dynamics. Surface cooling in the Pacific (La Niña) corresponds to higher mean sea levels and tides around Western Australia.28,29 Between 1991 and mid-1995, there were 5 consecutive years of El Niño events,30 and the absolute tide heights in late spring were low (96–110 cm). The highest tide in the period (124 cm) was recorded in late 1995, concurrent with a “weak ” La Niña.30 In that year, 12 of the 14 Statistical Local Areas recorded epidemics. In the following year, cool sea surface temperatures persisted, and the second highest tide was recorded (120 cm). Only 4 epidemics occurred that year, 2 in Statistical Local Areas that had not recorded epidemics the previous year. This pattern suggests that consecutive years of suitable climatic conditions and high mosquito numbers would not allow sufficient time for an increase in the proportion of susceptible kangaroos. Improving our knowledge of the ecological conditions supporting maintenance of Ross River virus, especially the interaction between kangaroo breeding and mosquito infectivity, is the next step to improving prediction of this disease.

The positive association between late spring sea surface temperatures and epidemics is noteworthy. During a La Niña phase, monthly rainfall and cloud cover in south-west Western Australia increase and average monthly temperatures fall.24 The converse also occurs: local cooling of the oceans increases the likelihood of warmer temperatures in this region. This could increase transmission by lengthening the life span of mosquitoes and by reducing the time for the virus to amplify to infectious levels within the mosquito. The role of El Niño in changing temperature has been noted previously in relation to dengue epidemics.31

The projected average increase in temperature in this region caused by global warming is likely to be generally favorable for mosquito development and viral amplification, resulting in potentially greater numbers of infectious mosquitoes. However, mosquito survival under these conditions will depend on whether the extreme temperatures are accompanied by sufficient humidity. A marked, step-like, decline in rainfall has been observed in south-west Western Australia since the 1970s, which has been attributed to a combination of climate change and natural variability.32 Further major changes have been predicted to the Southwest's rainfall pattern for the next 50 years, including a drying trend during winter and spring, a reduction in the number of heavy rainfall days and intense rainfall, and lower total rainfall (years of very low rainfall may increase by a factor of 5).29 The longer spacing between high rainfall years could increase the interval between Ross River virus disease epidemic years (and thus reduce the frequency of epidemics). Such a pattern might also result in an accumulated proportion of susceptible kangaroo hosts, and, in years of good rainfall, the number of epidemic events might increase. Given the importance of tide height for the initiation of Ross River virus disease epidemics, the projected increase of 9–88 cm in global mean sea level by 2100 is relevant.17 Tidal inundations may occur more frequently than at present, still allowing mosquito breeding along the Swan coastal plain in what are expected to be drier springs and summers.

Climate data are relatively inexpensive and easy to collect. By improving our ability to predict Ross River virus disease epidemics, they could enhance the effectiveness of response actions such as the timely and efficacious spraying of mosquito breeding sites,33 and public education about personal protection (application of effective mosquito repellents, screening of houses). El Niño events can be predicted, on average, from winter onwards in Australia.34 This, coupled with knowledge of the preceding year's rainfall, provides several early indicators of the probability of an impending epidemic season. Education campaigns could thus be used only in predicted “high risk” years and areas, to avoid overload for people exposed to the same message every year. Active surveillance of mosquitoes is a more expensive early warning indicator but has additional predictive value. Recent studies (C. Gordon, unpublished data) in this region suggest that monitoring kangaroo seroprevalence before the start of the Ross River virus disease season may also have value in epidemic prediction. The decision to survey mosquito numbers or sentinel kangaroos as part of a Ross River virus disease early warning system would include a cost–benefit assessment comparing the additional predictive value and the number of cases averted with the opportunity for expenditure on other priority diseases.


The authors acknowledge the Communicable Diseases Network of Australia and New Zealand for the data on Ross River virus disease, and the Microbiology Department of Western Australia for providing mosquito trapping data. Staff at SILO (Queensland Department of Natural Resources and Mines) assisted with the climate data. The authors thank staff at the Animal Health Science Unit, Department of Agriculture Fisheries and Forestry Australia, for use of computing resources.


1. Harley D, Bossingham D, Purdie DM, et al. Ross River virus disease in tropical Queensland: evolution of rheumatic manifestations in an inception cohort followed for six months. Med J Aust. 2002;177:352–355.
2. Mylonas AD, Brown AM, Carthew TL, et al. Natural history of Ross River virus-induced epidemic polyarthritis. Med J Aust. 2002;177:356–360.
3. Scrimgeour EM, Aaskoz JG, Matz LR. Ross River virus arthritis in Papua New Guinea. Trans R Soc Trop Med Hyg. 1987;81:833–834.
4. Rosen L, Gubler DG, Bennett PH. Epidemic polyarthritis (Ross River) virus infection in the Cook Islands. Am J Trop Med Hyg. 1981;30:1294–1302.
5. Aaskov JG, Mataika JU, Lawrence GW, et al. An epidemic of Ross River virus infection in Fiji. Am J Trop Med Hyg. 1981;30:1053.
6. Harley D, Sleigh A, Ritchie S. Ross River virus transmission, infection, and disease: a cross-disciplinary review. Clin Microbiol Rev. 2001;Oct:909–932.
7. Woodruff RE. Epidemic early warning systems: Ross River virus disease in Australia. In: Ebi KL, eds. Integration of Public Health with Adaptation to Climate Change: Lessons Learned and New Directions. London: Taylor & Francis Group; 2005:91–113.
8. WHO. Using Climate to Predict Infectious Disease Outbreaks: A Review. Geneva: World Health Organization; 2004;55.
9. Kay BH, Aaskov JG. Ross River virus (epidemic polyarthritis). In: Monath TP, ed. The Arboviruses: Epidemiology and Ecology. Vol. 4. Boca Raton, FL: CRC Press; 1989:93–112.
10. Sellers RF. Weather, host and vector—their interplay in the spread of insect-borne animal viral diseases. J Hyg (Lond). 1980;85:65–102.
11. Reeves WC, Hardy JL, Reisen W, et al. Potential effect of global warming on mosquito-borne arboviruses. J Med Entomol. 1994;31:323–332.
12. Lee DJ, Hicks MM, Debenham ML, et al. The Culicidae of the Australasian Region. Monograph Series, Entomology Monograph No 2. Vol. 7. Canberra: Australian Government Publishing Service; 1989.
13. Caughley G, Shepherd N, Short J. Kangaroos: Their Ecology and Management in the Sheep Rangelands of Australia. Cambridge: Cambridge University Press; 1987.
14. Woodruff RE, Guest CS, Garner MG, et al. Predicting Ross River virus epidemics from regional weather data. Epidemiology. 2002;13:384–393.
15. Tong S, Hu W, McMichael AJ. Climate variability and Ross River virus transmission in Townsville Region, Australia, 1985–1996. Trop Med Int Hlth. 2004;9:298–304.
16. Russell RC. Ross River virus: disease trends and vector ecology in Australia. Bull Soc Vector Ecol. 1994;19:73–81.
17. IPCC, ed. Climate change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge: Cambridge University Press; 2001.
18. Lindsay MD, Latchford JA, Wright AE. Studies on the ecology of Ross River virus in the south-west of Western Australia. In: Uren MF, Blok J, Manderson LH, eds. Arbovirus Research in Australia. Vol. 5. Brisbane: CSIRO Division of Tropical Animal Science, and Queensland Institute of Medical Research; 1989:28–34.
19. Lindsay MD, Condon R, Mackenzie JS, et al. A major outbreak of Ross River virus infection in the south-west of Western Australia and the Perth metropolitan area. Commun Dis Intell. 1992;16:290–294.
20. Lindsay M, Broom A, Oliveira N, et al. Western Australian arbovirus and research program. Annual report: 1997–1998. Perth: Department of Microbiology, University of Western Australia and the Health Department of Western Australia; 1998;172.
21. Stanley NF, Choo SB. Studies of arboviruses in Western Australia. Bull World Health Organ. 1966;30:221–226.
22. Lindsay MD, Oliveira N, Jasinska E, et al. An outbreak of Ross River virus infection in the south-west of Western Australia. Commun Dis Intell. 1996;20:136–139.
23. Gioia P, Pigott JP. Biodiversity assessments: a case study in predicting richness from the potential distributions of plant species in the forests of south-western Australia. J Biogeography. 2000;27:1049–1064.
24. Partridge IJ, ed. Will it Rain? The Effects of the Southern Oscillation and El Niño on Australia. Information Series Q194015. 2nd ed. Brisbane, Queensland: Department of Primary Industries; 1994.
25. Queensland Department of Natural Resources. The Data Drill. Brisbane: Queensland Centre for Climate Applications, Department of Natural Resources; 2000.
26. Hosmer DW, Lemeshow S. Goodness-of-fit tests for the multiple logistic regression model. Communications Statistics. 1980;A9:1043–1069.
27. Weinstein P, Cameron AS. A houseboat outbreak of epidemic polyarthritis. Med J Aust. 1991;155:721–722.
28. Allan R, Lindesay J, Parker D. El Niño Southern Oscillation and Climatic Variability. Melbourne: CSIRO Publishing; 1996.
29. Timbal B. Southwest Australia and future rainfall trends. Clim Res. 2004;26:233–249.
30. Climate Prediction Center NOAA. Cold and Warm Episodes by Season. Camp Springs, MD: Climate Prediction Center NOAA; 2001.
31. Gagnon AS, Bush ABG, Smoyer-Tomic KE. Dengue epidemics and the El Niño Southern Oscillation. Clim Research. 2001;19:35–43.
32. IOCI. Climate variability and change in south west Western Australia. Perth: Indian Ocean Climate Initiative, 2002;34.
33. Whelan P, Merianos A, Hayes G. Ross River virus transmission in Darwin, Northern Territory, Australia. In: Kay BH, Brown MD, Aaskov JG, eds. Arbovirus Research in Australia. Vol. 7. Brisbane: Queensland Institute of Medical Research; 1996.
34. Stone RC, Hammer GL, Marcussen T. Prediction of global rainfall probabilities using phases of the Southern Oscillation Index. Nature. 1996;384:252–255.
© 2006 Lippincott Williams & Wilkins, Inc.