Diagnostics For Mixture Regression Models: Applications To Public Health
Funder
National Health and Medical Research Council
Funding Amount
$128,250.00
Summary
In many public health studies, finite mixture regression models are often used to analyse data arising from heterogeneous populations. It is important to assess the stability of parameter estimates and the validity of statistical inferences when the underlying assumptions appear to be violated, but appropriate diagnostics are lacking in the literature. This research aims to develop effective diagnostic methods for assessing the adequacy of mixture regression models and the sensitivity of accompa ....In many public health studies, finite mixture regression models are often used to analyse data arising from heterogeneous populations. It is important to assess the stability of parameter estimates and the validity of statistical inferences when the underlying assumptions appear to be violated, but appropriate diagnostics are lacking in the literature. This research aims to develop effective diagnostic methods for assessing the adequacy of mixture regression models and the sensitivity of accompanying test statistics. The methodology developed will enable health care professionals to focus on substantive issues and to draw accurate and valid conclusions inferred from correlated and over-dispersed outcomes. In the presence of anomalous observations, the influence diagnostics can provide insights into the source of heterogeneity and the apparent over-dispersion, while accommodating the inherent correlation due to the longitudinal study design or nested data structure. Significance of the research lies in its scientific novelty and the breadth of its practical applications. The benefits to public health will accrue both nationally and internationally. For the empirical studies that motivated and are linked to this research, evaluation of health outcomes has significant implications in the prevention and control of recurrent urinary tract infections, hospital strategic planning, and post-stroke care and rehabilitation management. Moreover, appropriate assessment of a physical activity intervention for older adults is pertinent to falls prevention and reduction of musculoskeletal disorders among sedentary seniors.Read moreRead less
Hierarchical Finite Mixture Modelling Of Health Outcomes: A Risk-adjusted Random Effects Approach
Funder
National Health and Medical Research Council
Funding Amount
$117,000.00
Summary
In medical and health studies, finite mixture regression models have been used to analyze data arising from heterogeneous populations. Traditionally, the application of mixture models is mainly concerned with finite normal mixtures. Recent computational advances and methodological developments have enhanced the extension of the method to non-normal finite mixtures, such as the modelling of discrete responses in finite mixture of generalized linear models and overlapping phases of failure time da ....In medical and health studies, finite mixture regression models have been used to analyze data arising from heterogeneous populations. Traditionally, the application of mixture models is mainly concerned with finite normal mixtures. Recent computational advances and methodological developments have enhanced the extension of the method to non-normal finite mixtures, such as the modelling of discrete responses in finite mixture of generalized linear models and overlapping phases of failure time data in the context of survival analysis. However, due to the hierarchical study design or the data collection procedure, the inherent correlation structure and-or clustering effects present may contribute to extra variations and violation of the independence assumption, resulting in spurious associations and misleading inferences based on the finite mixture model. This project aims to present a unified approach to accommodate both heterogeneity and dependency of observations, by incorporating random effects into finite mixture regression models. The new methodology will provide an integrated framework to analyze heterogeneous and correlated health outcomes. Three empirical studies are considered, namely, evaluation of an occupational injury reduction intervention, length of hospital stay modeling, and analysis of survival times of patients after cardiac surgery. The long term benefits to bioscience are accurate and valid conclusions inferred from medical and health studies, as well as the correct identification of high-risk subgroups. For the three application areas of this project, the improved analyses will specifically enable the evaluation of a participatory ergonomics intervention, the assessment of hospital efficiency and factors influencing length of hospitalization, and the determination of effectiveness of treatments prescribed pre- and post- operation, respectively.Read moreRead less
Prof Speed is a statistician specializing in bioinformatics and computational biology, applying my skills in support of basic research in molecular and cell biology and genetics.
I am a statistician specializing in bioinformatics and computational biology, applying my skills in support of basic research in molecular and cell biology and genetics.
NONPARAMETRIC STATISTICS. Nonparametric statistical methods are techniques that implicitly choose statistical models from exceptionally large and highly adaptive classes. The project aims to develop innovative and practicable nonparametric methods in four areas: Statistical Smoothing, Data Mining, Mixture Methods and Robust Inference. The significance of the work lies in its novelty, the breadth of its practical motivation, and its position at the leading edge of contemporary work in statisti ....NONPARAMETRIC STATISTICS. Nonparametric statistical methods are techniques that implicitly choose statistical models from exceptionally large and highly adaptive classes. The project aims to develop innovative and practicable nonparametric methods in four areas: Statistical Smoothing, Data Mining, Mixture Methods and Robust Inference. The significance of the work lies in its novelty, the breadth of its practical motivation, and its position at the leading edge of contemporary work in statistics. Expected outcomes include new technologies for data analysis.Read moreRead less
Statistical Methods For The Analysis Of Trends In Coronary Heart Disease
Funder
National Health and Medical Research Council
Funding Amount
$112,747.00
Summary
Coronary heart disease is a leading cause of mortality, morbidity and medical costs in Australia. During the 1950's and 1960's, rates of coronary disease increased rapidly, then in the late 1960's they started to decline. This decrease has continued steadily for 30 years. While some other westernised countries have had this same experience, in Eastern Europe and in many developing countries coronary disease is increasing. There is a huge amount of evidence from experimental studies in animal and ....Coronary heart disease is a leading cause of mortality, morbidity and medical costs in Australia. During the 1950's and 1960's, rates of coronary disease increased rapidly, then in the late 1960's they started to decline. This decrease has continued steadily for 30 years. While some other westernised countries have had this same experience, in Eastern Europe and in many developing countries coronary disease is increasing. There is a huge amount of evidence from experimental studies in animal and human subjects and population studies in many countries that the major determinants of coronary disease are high blood pressure, cigarette smoking and high cholesterol (and other lipids) as well as dietary factors, obesity and physical inactivity. Recently several large multicentre studies have found unexpectedly weaker associations between heart risk factors and disease rates. It is hypothesised that this is due to inappropriate analyses in which data from populations at different stages of the coronary epidemic have been combined. The aim of this study is to develop improved statistical methodology to help understand recent findings from large scale studies, such as the World Health Organization's MONICA Project, the US ARIC study and the Seven Countries study. It will provide new theoretical results and statistical software for their implementation. From a public health perspective the most important outcome will be clarification of recent apparently anomalous findings about the importance of established risk factors and effective treatments in reducing coronary disease at the population level.Read moreRead less
Dynamic prediction models in Australian rules football using real time performance statistics. The study is a collaborative venture with Champion Data, the Australian leader in the collection and transmission of real time sporting data, and official provider of the Australian Football League (AFL) statistics. The aim is to develop a real time on line predictive model for AFL football. The model will use the statistics Champion Data collect as the match progresses as inputs to continually updat ....Dynamic prediction models in Australian rules football using real time performance statistics. The study is a collaborative venture with Champion Data, the Australian leader in the collection and transmission of real time sporting data, and official provider of the Australian Football League (AFL) statistics. The aim is to develop a real time on line predictive model for AFL football. The model will use the statistics Champion Data collect as the match progresses as inputs to continually update estimates of the probabilities of various outcomes of interest such as the winner of the match and the margin of victory. The project will assist Champion in their strategic aim to provide an on line form guide.Read moreRead less
Bayesian Statistical Inference for Implicitly defined Probability Models. Bayesian statistics has recently been used to provide solutions for a large number of hitherto intractable problems in science and technology. The success of Bayesian statistics has mainly been due to the application of so-called Markov chain Monte Carlo computational techniques. We aim to improve these algorithms, by providing fast, simple and efficient computational implementations. We will use the results to give ins ....Bayesian Statistical Inference for Implicitly defined Probability Models. Bayesian statistics has recently been used to provide solutions for a large number of hitherto intractable problems in science and technology. The success of Bayesian statistics has mainly been due to the application of so-called Markov chain Monte Carlo computational techniques. We aim to improve these algorithms, by providing fast, simple and efficient computational implementations. We will use the results to give insight by carefully quantifying and modelling uncertainty for such topics as the transmission rate of infectious diseases, the spatial distribution of plant and animal species, investigating biological theory for the genome of a virus, and changes in human fertility.Read moreRead less
A Program Of Methodological And Collaborative Research In Biostatistics And Population Health
Funder
National Health and Medical Research Council
Funding Amount
$264,081.00
Summary
Biostatistics is a critical component of health and medical research, especially for studies in population health. However, there is an increasing gap between supply and demand for high-level biostatistical input. This proposal combines novel methodological research into methods for analysing incomplete data, with collaborative research applying new ideas and complex analyses to important health problems. The fellowship will facilitate my development as a future leader in this key area.
Discovery Early Career Researcher Award - Grant ID: DE140100993
Funder
Australian Research Council
Funding Amount
$293,520.00
Summary
Mathematics of importance: The optimal importance sampling algorithm for estimating the probability of a black swan event. Rare event simulation and modelling is critical to our understanding of high-cost hard-to-predict events such as nuclear accidents, natural disasters, and financial crises. Quantitative analysis of such high-impact events demands the accurate estimation of the probability of occurrence of such rare events. In realistic models this probability is very difficult to estimate, ....Mathematics of importance: The optimal importance sampling algorithm for estimating the probability of a black swan event. Rare event simulation and modelling is critical to our understanding of high-cost hard-to-predict events such as nuclear accidents, natural disasters, and financial crises. Quantitative analysis of such high-impact events demands the accurate estimation of the probability of occurrence of such rare events. In realistic models this probability is very difficult to estimate, because exact simple analytical formulas are not available and the existing estimation methods fail spectacularly. There is an urgent need for new efficient methodology. This project develops a new Monte Carlo method that will be able to estimate reliably and accurately rare-event probabilities. Read moreRead less