Abstract

Infection with arthropod-borne (arbo)viruses presents a significant and growing public health threat to the resident population of Queensland (QLD), the north-eastern state of Australia. Clinical infection with Ross River virus (RRV) is the most commonly detected, and arguably most debilitating, of Australia’s 75 known indigenous arbovirus species. Development of prediction models to forecast arbovirus epidemics aims to provide accurate and reliable tools that may facilitate planned interventions by local and state authorities to curb disease transmission. Acute immunoglobulin (Ig)M-positive enzyme-linked immunosorbent assay results are often misleading, with interpretation cautioned. As such, this serological testing was recently excluded as a means to confirm cases of arbovirus infection in Australia. The purpose of this study was to investigate the seroepidemiological value of acute IgM-positive results across QLD by correlating with RRV case reports and to develop a mathematical model to predict RRV outbreaks. Blood samples from patients throughout QLD suspected of arboviral infection were tested for RRV, with numbers for various serology results grouped by geographical region. The serology data were compared with case reports for each respective region by multiple regression in order to determine any relationships. RRV IgM-positive results correlated significantly to the number of case reports per region (P < 0.05). An estimated multiple regression equation was used to predict RRV case reports from a subset of data extracted for the period December 2015 and January 2016. Predicted cases based on IgM-positive/IgG-negative serology showed no significant correlation to the respective case reports for each region (P > 0.05). Hence, these findings failed to validate the potential use of IgM-positive seroprevalence to predict RRV infection with sufficient accuracy for diagnostic purposes. A possible indirect value may exist, however, in analysing pooled seroprevalence data, which may better inform concurrent surveillance measures and thereby enhance the accuracy of RRV outbreak forecasts.

1. Introduction

Vector-borne infectious diseases, principally those transmitted by arthropod insects such as mosquitoes and ticks, are a growing public health threat. Many are emerging or re-emerging diseases, but their patterns of distribution are extremely volatile due to the high variability among pathogens, hosts, vectors, regional environments and landscapes ^{1, 2, 3}. The geographical range of causative agents is restricted to the areas inhabited by their vectors and, if zoonotic, their reservoir hosts, so the escalating impact of climate change and dramatic weather events on habitat suitability is of major concern ⁴. Shifts in societal, demographic and epidemiological characteristics of human populations, including urbanisation, international travel and migrant integration, also influence how environmental risk affects the frequency of disease cases, patient outcomes and ultimately the burden of these vector-borne diseases. Mathematical modelling is a technology that is pivotal to integrating infectious disease data retrieval and analysis with vector-borne disease research in order to enhance prevention and control measures and to inform public health management policy.

The development of surveillance-based prediction models for arthropod-borne (arbo)virus epidemics offers a potentially invaluable tool to help combat the far-reaching emerging health impact that is caused globally by mosquito transmission of flaviviruses (notably dengue, yellow fever, Japanese encephalitis, West Nile and Zika) and alphaviruses (notably chikungunya and Ross River virus, RRV) ⁵. These statistical models could be utilised to predict outbreak risk and thereby to reduce epidemics by such early interventional actions as drainage of still water, spraying residual insecticides and/or warning communities via news channels, mobile device apps and social media networks ^{6, 7, 8}.

The mosquito-transmitted viral disease Ross River fever is the cause of profound public health problems in Australia, especially Queensland (QLD), the Northern Territory and the Kimberley region of Western Australia ^{9, 10, 11}. For almost a decade after the identification of RRV ¹², very few patients were identified as having a clinical infection with this pathogenic agent because diagnostic screening was restricted to a research setting using an in-house developed test method. Following the 1985 commercial release of an enzyme-linked immunosorbent assay (ELISA) to detect anti-RRV immunoglobulin (Ig)M antibody, epidemic polyarthritis (EPA; clinical RRV infection) became a nationally notifiable disease in 1990 ¹³. Nationwide annual RRV diagnoses rose from 20-50 pre-commercial testing to a current figure of several thousand ¹⁴.

The Australian National Notifiable Diseases Surveillance Scheme received notification of 47,256 cases of vector-borne diseases for the recent period from January 2013 to December 2017 ¹⁴. RRV accounted for 29,843 (63.1%) of these. While the average number of notifications of RRV infections over this five-year duration is around 4,500 per annum, 9,554 cases were reported in 2015 ¹⁴. RRV has been isolated from at least 40 different mosquito species ¹⁵. In addition to its detection in mosquitoes and humans, serological surveys have found widespread infection with RRV among kangaroos, wallabies, bandicoots and other Australian marsupials ^{16, 17}. The virus has also been recovered from horses, cattle, goats, sheep, dogs and birds ^{18, 19}. Therefore, it is likely that RRV is being maintained in zoonotic cycles involving native animals and birds which are putative natural reservoir hosts ²⁰.

Prompt and accurate notification of laboratory-confirmed and probable cases is an important component of arbovirus surveillance as this information enables public health units to initiate a timely and commensurate response to an outbreak ²¹. Serology-dependent case definitions for RRV infection have been plagued with incidences of false-positive reports ^{22, 23, 24}. This has arisen via a combination of reliance solely on IgM results, the persistence of IgM in peripheral blood, antibodies cross-reactive to other alphaviruses such as Barmah Forest virus (BFV), issues relating to original antigenic sin, suboptimal ELISA kits, and irregular inter-laboratory methodologies and diagnostic criteria ²². The inconsistent notification criteria for arbovirus infection by serological determinants recently received criticism within the medical laboratory science profession in Australia ²⁴. On the recommendation of the National Arbovirus and Malaria Advisory Committee, the Case Definitions Working Group (CDWG) of the Communicable Diseases Network Australia undertook a review of surveillance case definitions for RRV infections that was implemented at the start of January 2016 ²⁴. Queensland Health stated that the “historical data prior to the change of case definition will continue to be considered unreliable” ²³.

In order to ensure the collection of reliable and consistent data the CDWG revised the criteria for the categorisation of confirmed or probable RRV infection notifications ²⁴. Infection of RRV is confirmed if any one or more of the following is identified in blood, plasma or serum obtained from the patient: isolation of virus; detection of viral nucleic acid; or IgG seroconversion, as determined by a four-fold or greater rise in anti-RRV IgG antibody. Except for specimens in which IgG is detected ≥ 3 months prior, the presence of both IgM and IgG antibodies is even more suggestive of a probable RRV infection ²⁴.

An average incubation period for RRV, i.e. the time from being bitten to presenting clinical symptoms, is around one week. While patients often seek medical services as soon as they are ill, whereupon their blood is tested, the delay in receiving and handling the sample may mean that it is too late for the virus to be isolated or its RNA to be detected ^{22, 23, 24}. Furthermore, virus isolation is a time-consuming procedure. Hence, there is the need for a more convenient technique, that of detection of anti-RRV antibodies (Table 1).

A combination of serology results that is either IgM-positive and IgG-negative (PN) or positive for both IgM and IgG (PP) has various interpretations, for which the IgM result is key (Table 1). This study proposes that either PN or PP serology from a given locality may be predictive of seroconversions or notifications in the local community in which that patient resides. If by linear regression a significant relationship exists between rates of PN results and infection notifications throughout QLD, a predictive mathematical model for surveillance and subsequent interventional response to RRV outbreaks could be applied across the state.

2. Materials and Methods

From QLD residents suspected of RRV infection by medical examination, 716 anonymised patient sera received in local authority pathology laboratories during the months of December 2015 and January 2016 were tested by commercially available anti-RRV IgM and IgG ELISA (PanBio Ltd., Sinnamon Park, QLD, Australia) (Table 2). Serology data were collated prospectively together with the date of test, age, gender, residential location and postcode. Patients excluded from this study included interstate travellers, those with a previously positive IgG result recorded ≥ 3 months earlier due to confounding implications of prior infection, and those in whose sera IgM to viruses of similar antigenic affinity (e.g. cytomegalovirus or BFV) was detected.

The quantitative values of each serology permutation for RRV were grouped together by QLD region, testing date ranges (1-11, 12-31 December 2015 and 1-15, 16-31 January 2016), gender and age ranges (0-24, 25-49, 50-74, and ≥ 75 years). Projected population data by QLD region, age range and gender were extracted from the Australian Bureau of Statistics 2015 database ²⁵.

Notification data for confirmed or probable cases of infection with RRV for onset dates between 1 December 2015 and 31 January 2016 were extracted and supplied by the Notifiable Conditions System, Public Health Unit, Queensland Health. The samples were collected from 19 regions throughout QLD (examined at statistical area level 4 based on the 2011 Australian Statistical Geography Standard) but principally from Brisbane, Bundaberg, Cairns, Gold Coast, Mackay, Rockhampton, Toowoomba and Townsville (Figure 1).

For prospective IgM and IgG serology results, tests for RRV infection were collated with respective epidemiological data (Table 2). In order to derive equations linear regression model analysis was performed between notification reports and the corresponding rates of IgM & IgG permutations (Table 3). IgM-positive/IgG-negative (PN) and IgM-positive/IgG-positive (PP) serology results from each location were shown by regression analysis to be significantly correlated (P < 0.05), but the correlation value obtained for PP (R² = 0.23) was lower than for PN (R² = 0.53) (Table 3). Therefore, the mathematical equation for PN was subsequently used to predict the number of notification reports during a specified period for each region of QLD.

Serology data and notification rates were normalised between QLD regions by calculating their ratio per 100,000 residents (Table 4), according to the projected 2016 population size ²⁵. The normalised data set was also analysed by multiple regression to determine any statistically significant association between serology results and cases of infection which might indicate a relationship common among QLD regions.

The total number of PN results and notification rates corresponding to onset dates between 1 December 2015 and 31 January 2016 were normalised for each QLD region per 100,000 population and analysed by linear regression. The linear regression equation (line of best fit) derived from the scatter plot of all data was used to estimate case reports corresponding to serology data for the same two-month period (data not shown). The number of cases of RRV estimated using the dependent variable (that for PN) from serology results and actual notified cases for the period 16-31 January 2016 were compared by regression analysis to identify any relationship (Figure 2).

3. Results

A strongly significant relationship was shown to exist between the total number of RRV serology test requests ordered during December 2015 and January 2016 and the total number of notified cases from each QLD region (P = 0.0014) (Table 2). When the serology data were normalised by ratio per 100,000 population, the correlation between PN and regional cases was still significant (P < 0.05). Linear regression model analysis between notification reports and PN serology from each location showed a significant correlation (P < 0.05, R² = 0.53). Hence, the equations obtained from using two different notifications rate, those for RRV confirmed cases (y = 6*x + 8.79) and RRV total cases, i.e. including confirmed and probable cases (y = 9.33*x + 11.7), were used to predict RRV cases in the SA4 region of QLD during the second half of January 2016 (Table 4). The inherent discrepancy between actual results and predicted results (Figure 2) was minimised by adjustment of PN coefficient and y intercept values in order to reduce the total sum of the normalised error (data not shown). The linear regression between normalised predicted case reports and actual case reports was found to be non-significant (P > 0.05) (Figure 2).

4. Discussion

The revision of notification criteria for serology-based RRV infections in Australia that was implemented on 1 January 2016 calls into question the reliability of case records preceding this date ^{23, 24}. The key revision was the implementation of a nationwide collaborative agenda among reporting pathology laboratories to refrain from recognising an acute sample IgM-positive test as a presumptive indicator of infection. A more rigorous laboratory algorithm was to be applied to overcome the occurrence of false positive IgM results ²⁶.

Accordingly, the rationale for this study was to develop a mathematical model predicated on the corrupted data prior to January 2016 and to use this algorithm to predict RRV case reports presently considered valid. A model based on acute IgM-positive serologies should be capable of predicting the rates of currently valid RRV infection notifications based on correlations with case reports that are now known to be unreliable. If the model is accurate this may argue in favour of the presumptive reporting of RRV infection based on IgM-positive results. Yet, a study conducted in Perth, Western Australia, concluded that some 45% of RRV IgM-positive, IgG-negative acute enzyme immune assay results failed to seroconvert whereas around 75% of IgM-positive results tested by either immunoassay or haemagglutination inhibition did seroconvert ²². In light of this apparent anomaly caution is advised as to the validity of presuming RRV infection from IgM-positive ELISA data. Thus, notifications are reserved until convalescent seroconversion is demonstrated ²⁴.

Patient blood was tested prospectively at the headquarters of Queensland Medical Laboratory Pathology (QML Pathology) in Murarrie, south-eastern Brisbane, where all arbovirus serologies received by QML Pathology collection centres throughout the state are processed. Almost all samples would have come from ambulatory outpatients since QML Pathology performs testing primarily on samples from the community and a smaller subset from private hospitals. Samples from the larger public tertiary care public hospitals are not screened by this laboratory. A more comprehensive analysis of data would be achieved by collating results from all QLD commercial medical testing laboratories (i.e. including Sullivan Nicolaides Pathology, Pathology Queensland, Mater Pathology and Medlab Pathology). However, there is no reason to consider that the sample set examined was not representative and, moreover, the restriction served as a convenient method of semi-random statistical sampling.

The linear regression model developed herein using both PN and PP anti-RRV antibody results was able to predict correctly when case numbers were low but failed to predict accurately when notifications increased in size. As more information deepens our understanding of RRV infection this apparent paradox may come to be explained. Future work aims to refine the model to enable the valid predicting of RRV infection over a broad range of data input values.

Accurate predictions of future outbreaks of RRV infection enable federal, state and local government authorities to implement more effective prevention and control measures. Consequently, research is being conducted to investigate novel ways in which to assimilate the abundance of factors associated with outbreak prevalence into surveillance tools capable of effective forecasting ^{8, 10, 27}. Further research is required on RRV seroprevalence in Australia as well as into the seroepidemiological complexities of arboviral outbreaks in general. This may reveal additional novel and exciting statistically resourceful correlations.

5. Conclusion

This study demonstrates an unintuitive application of the examination of seroprevalence data for RRV infection. The linear regression model analysis performed on this limited data set did not validate with sufficient accuracy the use of IgM-positive seroprevalence to predict RRV infection to be of direct benefit to public health stakeholders in its current iteration. However, the predictive analysis did provide qualified evidence for the statistical value of utilising acute sample IgM-positive data for informing best practice in RRV surveillance. Hence, a potential value may exist in gathering pooled seroprevalence data to be harnessed as a means to better inform concurrent surveillance measures, such as for vectors, reservoir hosts and weather conditions, to improve forecasting accuracy and outbreak predictions. We propose that such data could be reported under a third category of possible infection. This information may then be catalogued and archived for the primary purpose of enabling the elucidation of potential statistical correlations in future analyses. Due to the inherent problems associated with interpreting IgM-positive serologies, any presumption of RRV infection based on an acute sample result should not be used directly for patient diagnosis and subsequent treatment.

Acknowledgements

Dr Richard Bradbury and Mr Wayne Pederick (School of Health, Medical & Applied Sciences, Central Queensland University, Rockhampton, QLD, Australia) kindly liaised with QML Pathology, Murarrie, to gain requisite laboratory permissions to undertake this study. Dr Bradbury (now at the Centers for Disease Control and Prevention, Atlanta, USA) critically appraised the first draft of this paper. We thank Mrs Stacey Bronger for providing guidance to one of us, JBS, while on placement at the Serology Department at QML Pathology, and also Mrs Mayet Jayloni and Mrs Mohana Rajmokan for extracting and supplying Notifiable Conditions System-derived arbovirus notification data. The provision and analysis of de-identified human sera in this study was approved by Central Queensland University Human Ethics Committee (application no. H15/03-041).

Funding

NG was in receipt of an International Postgraduate Research Scholarship and an Australian Postgraduate Award administered by Central Queensland University.

Authors’ Contributions

JBS and AWTR conceived the project. JBS carried out notification data analysis. JBS and NG performed modelling predictions and drafted the manuscript. AWTR supervised the project. NG and AWTR performed statistical checks and critically revised various versions of the paper. All authors contributed to preparation of the final version and agreed to its submission.

Competing Interests

The authors have no competing interests to disclose.

List of Abbreviations

arbo, arthropod-borne; BFV, Barmah Forest virus; CDWG, Case Definitions Working Group (of the Communicable Diseases Network Australia); ELISA, enzyme-linked immunosorbent assay; Ig, immunoglobulin; QLD, Queensland; RRV, Ross River virus.

References