THEME 4: INFORMED DECISIONS

A primary purpose of data collection, mathematical and statistical modelling, and data analysis is to learn more about the complex systems that characterise the world we live in and make the best possible decisions about their operation. Examples of these real-world systems include computer networks, electricity grids, health systems, ecosystems, economic markets, agricultural supply chains, smart cities and societal networks. Importantly, components in these systems rarely act as independent entities: for example, in a typical ecosystem consisting of plants and animals pursuing their own survival as individuals, a modification to one component influences many other components.

The aim of the research in the Informed Decisions theme is twofold. The first aim is to develop new methodologies for optimal decision-making strategies, taking into account multiple influencing factors and timescales. These new methods constitute frontier research in their own right. The second aim is to apply these methods to a range of problems.

Under this theme we develop new decision-making methodologies, and exploit the outcomes from the Challenging Data, Multiscale Models, and Enabling Algorithms themes to make decisions that benefit society, by characterising optimal decision-making strategies over a range of timescales.

Due to the nature of the research priorities, the projects that fall under the Informed Decisions theme engage strongly with our Partner and Industry Affiliate Organisations to translate the outcomes of this theme into real world outcomes with impact.

People

Key Achievements

In 2019 ACEMS researchers developed a wide range of targeted research outputs and outcomes to solve important problems in the three domains of Healthy People, Sustainable Environments and Prosperous Societies. Some of these achievements are described below.

General Theory:

CI Peter Taylor, with AI Jing Fu and colleague Bill Moran, have continued their work on a major project to propose asymptotically-optimal policies for a very general class of distributed resource allocation problems. Requests arrive with a task that could be completed using different sets of resources. For example, a customer that requires connection from node A to node F in the diagram of a simple communication network could be allocated a route via nodes B, C and E, or via nodes B and D. Given a knowledge of the requests that are already being served by the system, the task is to decide which route to use in order to optimise a long-term rate of earning revenue. The team’s idea is to use a generalisation of the well-known Whittle Index policy that takes into account the finite capacity constraints of the system.

CI Kate Smith-Miles has been working with Research Fellow Mehdi Foumani to improve decision making tools for government when trying to incentivise manufacturing companies to reduce their carbon emissions. Working with the assumption that manufacturers will try to maximise their profitability within the constraints of their economic environment, the ACEMS team have developed mathematical models to help government determine optimal policies to drive behaviour change, exploring how the parameters of various schemes - namely carbon emission trading schemes, taxes on emissions, and baseline penalty constraints - can be tweaked to help manufacturers simultaneously reduce carbon emissions while remaining economically competitive in a global market.

Healthy People:

Farzana Jahan presenting at the 2019 Young Statisticians Conference in Canberra

ACEMS researchers, led by AIs Earl Duncan and Susanna Cramb, CI Kerrie Mengersen and Industry Collaborator Peter Baade from the Cancer Council Queensland, continued work on the Australian Cancer Atlas which was released in 2018. The Atlas is a novel interactive web-based map of cancer incidence and survival rates, based on Bayesian spatial models that provide robust small-area estimates, preserve privacy and enable probabilistic comparisons with the national average. Significant enhancements of the models, visualisations and the web-based tool were developed in 2019. The project won the 2019 Queensland Spatial Enablement Award. There has also been strong international interest in the Atlas. The ACEMS team has developed a short course to help colleagues in New Zealand to develop their own Atlas and is in similar discussion with the Netherlands.

The innovations in visualisation for the Australian Cancer Atlas were developed by Masters student Stephanie Kobakian. Stephanie was co-supervised across two ACEMS nodes, QUT and Monash, and worked closely with ACEMS Industry Affiliate Cancer Council Queensland. The challenge addressed by Stephanie is that the Atlas employs around 2000 statistical small areas (SA2) to convey cancer incidence and survival rates. However, most of these SA2 areas are highly concentrated in city areas, so they can be very difficult to see using common (chloropleth) mapping techniques, especially when compared with the very large geographic areas representing the more sparse rural and remote populations. The alternative visualisation developed by Stephanie takes spatial areas in the form of polygons and creates a map of tessellated hexagons, arranging them to replicate spatial relationships of geographic areas in each city. The new map can provide new insight into spatial patterns of cancer. An R software package, sugarbag (Kobakian and Cook, 2019) was developed to implement the approach.

A number of other projects on visualisation of health outcome data have been secured by ACEMS researchers. For example, one of the principal researchers in the Australian Cancer Atlas, Susanna Cramb (who was named a 2019 STEMM Superstar) is an AI on a recently announced successful NHMRC Ideas Grant on spatial visualisation to align national health outcome data with regional health policy objectives.

Masters student Jessie Roberts and AI Nicole White have continued to build collaborations with our Industry Affiliate, the Australian Red Cross Lifeblood (formerly the Australian Red Cross Blood Service) around influenza forecasting and associations with blood collection outcomes. The industry partner for this project was originally ex-ACEMS member Stephen Wright; recently his role has been taken over by Helen Faddy and Elvina Viennet.

In another project with Red Cross, AIs Mark Fackrell and Joyce Zhang and PhD student Achini Eranda have been analysing data related to the waiting processes in Donor Blood Centres. They have access to timestamp data from four different Donor Centres serving communities with different demographic characteristics. The aim is to propose procedures to minimise waiting times of donors with a view to ensuring that as many as possible continue to donate.

As part of another ongoing collaboration with the Australian Red Cross Lifeblood, PhD student James Yu is working under the guidance of CI Matt Wand to build statistical models to help characterise the effect of plasma donation on donors’ IgG levels. This project is important since it will guide Red Cross on the development and implementation of guidelines for how often it is safe to have repeat plasma donations.

With colleagues from Wayne State University in Michigan and the University of Waterloo in Canada, ACEMS members (led by CI Louise Ryan) are developing an analytical framework to integrate data from multiple longitudinal cohorts measuring child developmental outcomes in order to understand the impact of pre-natal alcohol exposure. The results will be used to establish new clinical criteria for identifying children with Fetal Alcohol Spectrum Disorder. Advanced analytical methods are required since the data are very complex and high dimensional.

ACEMS researchers have also engaged in public debate concerning clinical and public health issues. For example, based on a systematic review and meta-analysis, a research team led by CI Nigel Bean found that clinical guidelines should be revised in order to use Free T4 levels to diagnose sub-clinical thyroid conditions rather than the current practice of using TSH levels.

CI Peter Taylor, with AIs Mark Fackrell, Malgorzata O’Reilly and Mojtaba Heydar, have been looking at the problem of assigning patients to beds in large hospitals. Hospitals are often left in a situation where a bed with ideal facilities is not available when a patient has to be admitted. Such patients are often accommodated in beds with second-choice facilities. Questions then arise as to what to do when a first-choice bed becomes available. The team proposed a novel and sophisticated dynamic programming approach to solving this problem.

Sustainable Environments:

A successful project involving cross-node, cross-Centre of Excellence researchers has led to the acceptance of an article in a prestigious journal, Ecology Letters. The project which involved the ARC Centre of Excellence for Environmental Decisions (CEED) and ACEMS, and nodes at UQ, QUT and UNSW, was led by Research Fellow Matthew Adams, who was jointly funded and supervised by UQ and QUT nodes. The research addressed the common problem of well-intentioned environmental management decisions backfiring due to unforeseen perverse outcomes. For example, a decision to eradicate feral cats in order to protect a threatened species of bird might lead to an explosion in rats which eat the birds’ eggs. To avoid such outcomes, managers and ecologists seek accurate predictions of the ecosystem-wide impacts of interventions. However, they usually only have limited data so such predictions are incredibly difficult. The project team developed a method to generate noisy time series based on these data to forecast whether a species’ future population will be positively or negatively affected by rapid eradication of another species. This approach can substantially support management decisions to improve complex environmental outcomes.

Another approach to quantitatively supporting conservation decisions was examined by QUT CI Kerrie Mengersen in collaboration with colleagues in the UK. The decision-analytical approach uses a measure of the Value of Information (VoI) to select management alternatives where the effectiveness of the intervention is uncertain. The work has motivated a call for common reporting standards to leverage this powerful tool and suggested promising areas for new research.

ACEMS CI Louise Ryan engaged with the NSW Office of the Chief Scientist and Engineer to advise on methods and strategies for handling decision-making under uncertainty. With other ACEMS researchers, she contributed to reports through the analysis of time-series related to water levels in groundwater aquifers and developed innovative RShiny tools that allow decision makers to undertake their own interactive analysis in a simple, fast manner.

ACEMS researchers have used a combination of virtual reality, aerial thermal-imaging and ground surveys to build a better statistical model for predicting the location of koalas and, ultimately, protecting their habitat. The study used the mashup of high-tech 360-degree imagery and heat-seeking drone cameras along with traditional techniques of ground surveys to develop a model that could be used to identify areas most likely to be home to koalas, which are facing population decline. This project was undertaken before the devastating 2019-2020 bushfires and is even more relevant afterwards as koalas are more endangered. This research, led by AIs Erin Peterson and Cath Leigh and CI Kerrie Mengersen, involved a number of undergraduate members of ACEMS, as well as PhD student Jacinta Holloway and a range of external partners including koala NGOs, citizen science groups and other university research groups.

Stemming from previous work on modelling the effect of dredging on seagrass, ACEMS researchers, led by AI Paul Wu and CI Kerrie Mengersen, have commenced a new collaboration with IFREMER (France), University of Pau and Australian universities that aims to model seagrass resilience under climate change scenarios. The new approach combines deterministic simulation models with statistical DBNs.

In another project, ACEMS researchers investigated: How healthy are our recreational waterways? When should they be closed to the public for health and safety reasons? This was a problem addressed by an ACEMS team, led by AI Paul Wu and CI Kerrie Mengersen, in collaboration with Healthy Land and Water. The team developed a predictive model and early warning system, with an online dashboard for a number of city councils in south-east Queensland.

ACEMS researchers, led by AIs Erin Peterson and Julie Vercelloni and CI Kerrie Mengersen, continued work on the Virtual Reef Diver online citizen science and statistical analysis system for improved modelling of the Great Barrier Reef, in collaboration with ACEMS Partner Organisation the Australian Institute of Marine Science (AIMS) and others. This project was a finalist for the annual, prestigious Eureka Awards. It also won the “people and community” category of the Asia-Pacific Spatial Excellence Award and is a finalist for the Oceanic Asia-Pacific Spatial Excellence Awards.

According to AIMS, the recovery ability of the Great Barrier Reef (GBR) has declined by an average of 84% over the past 30 years. The delay in recovery appears to be driven by a combination of cumulative chronic pressures and the legacy effect of acute disturbances. In 2019, AIMS agreed to fund a new postdoc, David Warne, to work with AIMS Partner Investigator Juan Ortiz and QUT researchers Paul Wu, Jin Wang, Mat Simpson and Kerrie Mengersen to use a combination of statistical and mechanistic modelling approaches to characterise the early recovery of GBR reefs. This will improve managers’ ability to inform the spatio-temporal prioritisation of management actions.

In collaboration with scientists at the Queensland Department of Environment and Science, ACEMS researchers, led by AIs Erin Peterson and Cath Leigh and CI Kerrie Mengersen, developed statistical predictive tools that could lead to the deployment of many more low-cost sensors in the rivers and streams that flow into the Great Barrier Reef. These sensors provide the potential for being part of a cost-effective monitoring system.

Our undergraduate research interns in ACEMS are doing amazing work! One such project was led by student Abhishek Varghese, with supervision from ACEMS researchers Kerrie Mengersen and Chris Drovandi, Professor Antonietta Mira from USI in Switzerland and collaborators from Queensland Biosecurity. This project aimed to describe the disease dynamics of banana bunchy top virus (BBTV), one of the most economically important vector-borne banana diseases throughout the Asia-Pacific Basin that presents a significant challenge to the agricultural sector. Instead of the current models of BBTV, which are largely deterministic and limited by an incomplete understanding of interactions in complex natural systems, a new stochastic model was developed. This model improves on the current deterministic models, providing a much more complete understanding of interactions, inspection accuracy, temperatures and vector (aphid) activity. The model can be used for monitoring and forecasting of various disease management strategies.

Invasive species have also been the topic of other research in ACEMS. A project led by Insha Ullah and Kerrie Mengersen involved the use of Bayesian mixture models and their Big Data implementations, with application to invasive presence-only data about fire ants in Brisbane. Bayesian spatial and spatio-temporal approaches to modelling dengue fever have also been developed by PhD student Aswi Aswi in collaboration with the QUT Public Health research group led by Professor Wenbiao Hu.

Prosperous Societies:

ACEMS’ work with the United Nations to enhance the use of novel data sources by official statistics agencies around the world has continued in 2019. An ACEMS team led by PhD students Jacinta Holloway and Brigitte Colin, together with AI Kate Helmstedt, CI Kerrie Mengersen and international colleague Michael Schmidt, developed new statistical machine learning methods for analysing satellite data to inform about vegetation cover and crops. These fast, accurate methods will contribute to the Sustainable Development Goals (SDGs) set by the United Nations and World Bank for countries to reach in order to improve quality of life and environment globally by 2030.

Paul Wu attended the Australian trials for the swimming world championships as part of ACEMS work with QAS and SAL

ACEMS has developed a strong portfolio of research projects and collaborations in sport. A joint postdoc was funded by ACEMS, Queensland Academy of Sport and Swimming Australia to enhance the outcomes of physical training in athletes by integrating data and developing statistical and machine learning methods. This project also involved a group of three undergraduate students funded under ACEMS’ Industry Collaboration Support Scheme. Another collaboration with the Australian Institute of Sport, Queensland Academy of Sport, the Australian Football League and other sporting organisations aims to contribute to the improved performance of Australia’s athletes and sporting teams. An international collaboration involving Sports Science experts at University of Technology Sydney, the Institute of Sports and Preventive Medicine, Saarland University in Germany, and the German Football Association has been established to develop innovative new methods for analysing training trajectories in young players, using similar methodologies to the ones that had been previously used to model growth patterns in young children. In addition to having important practical impact for coaches and athletes, the novel methodological solutions developed in these projects make a contribution to the fields of mathematics and statistics themselves and are also applicable to a wider range of applied fields.

A topic of major interest for business, government and the community is the analysis of online data sources such as social media to identify patterns, describe profiles of individuals and groups, and predict events. An ACEMS research team led AIs Jonathan Tuke and Lewis Mitchell and CI Nigel Bean has been investigating ways to combine data from Twitter, Flickr and Facebook with machine learning and Bayesian approaches to predict events such as protests or social unrest in Australia. Importantly, they have focused on quantifying the uncertainty of such predictions. Their method is termed “Pachinko Prediction” since it involves sorting pieces of evidence into bins for different days and location, like coloured marbles being sorted into jars.

Plans for 2020

Plans for 2020 are already well underway, and the following is a summary of what ACEMS researchers across Australia are planning to do.

In 2020 this research theme will continue to develop and apply new mathematical, statistical and machine learning methodology and computational algorithms that are targeted to solution of real world challenges. The following projects are a sample of objectives for 2020 that relate to projects that are being undertaken by ACEMS members at multiple nodes. They are mainly collated under the three ACEMS Domains of Healthy People, Sustainable Environments and Prosperous Societies. Many involve continuation and extension of projects from previous years, though several new ones are listed.

General Theory:

• Continue to investigate Whittle index methods for large resource allocation problems.

Healthy People:

• Research approaches to emulation of complex systems and modelling of high dimensional problems related to environmental associations with health.
• Continue to improve mathematical, statistical and machine learning models for human health, including national-level mapping of spatio-temporal patterns in cancer, state-level modelling of health systems and patient-specific understanding of key characteristics of atrial fibrillation.
• Develop statistical strategies for analysing complex, high dimensional data related to child development outcomes.
• Use statistical modelling and machine learning to automate analysis of Electroencephalograms (EEG) in order to more efficiently and reliably evaluate data collected in sleep studies.
• Develop innovative statistical models to support individualised decision making for people engaged in sport, integrating heterogeneous data of varying quantity and quality for complex, uncertain systems.
• Continue to improve statistical models for the analysis of medical images and radiotherapy treatment management.

Sustainable Environments:

• Research approaches for extending complex systems and Bayesian network modelling methods for inference under parameter uncertainty.
• Continue to improve methods for managing and utilising novel data sources for improved environmental management and prediction.
• Develop decision tools that community groups can use to help understand the complexities of uncertainty associated with mechanistic models and use those tools to make informed decisions about their local environments.
• Continue to improve environmental-friendly methods for reduction on carbon footprint in process models under uncertainty.
• Expand the ability of proposed frameworks for improved water and air quality, as well as decision making in the face of uncertainty about future climate.
• Investigate improved decision making around logistics and transportation, particularly with use of drones in combination with trucks for deliveries, and constraints related to power consumption while trying to deliver goods in the fastest and cheapest way.

Prosperous Societies:

• Research and develop prototype methods for data exploration, integration, modelling and visualisation for complex sets of athlete and sports data.
• Continue to improve wind and energy system modelling.
• Deliver on the consultancy for the new Women’s and Children’s Hospital in Adelaide.
• Investigate the role of treatment, environmental exposures and other influences on geographical disparities observed in the Australian Cancer Atlas.
• Develop spatio-temporal models of other health outcomes such as diabetes hospitalisation, dengue, infectious diseases and trauma.
• Work with other countries to develop a local equivalent of the Australian Cancer Atlas, starting with New Zealand and the Netherlands.
• Develop new methodologies for inference on the effect of pollution on human health.

The following figure is extracted from Stephanie Kobakian’s Masters thesis on “New algorithms for effectively visualising Australian spatio-temporal disease data”.

Model uncertainty, and difference between the model and true underlying data, quantified in several different ways, for a 20-year monitoring program.

This graphic was developed during a successful project involving cross-node and cross-Centre of Excellence researchers. The project team developed a method to generate noisy time series to forecast whether a species’ future population will be positively or negatively affected by rapid eradication of another species. This approach can substantially support management decisions to improve complex environmental outcomes.