RESEARCH ACTIVITIES IN BIOSTATISTICS

There are several areas of ongoing research in Biostatistics at our department. You can also look at a list of our Biostatistics faculty and their individual specialism.

Bayesian Hierarchical and Mixed Modeling
Bo Cai Ph.D.
James Hardin Ph.D.
Monir Hossain Ph.D.
James Hussey Ph.D.
Hongmei Zhang Ph.D.
Jiajia Zhang Ph.D.
This area of specialism is quite general and covers modeling where random effect components are thought to be relevant. This could simply be where unobserved confounders are present in the data, which give rise to extra variability, or where special effects must be included (such as temporal or spatial correlation effects). A non-Bayesian or Bayesian approach can be taken to this modeling and it is applicable widely in biostatistical applications (such as clinical trials, cohort studies, case-control studies or cross-sectional studies).

Measurement Error Modeling
Matteo Bottai Sc.D.
Hongmei Zhang Ph.D.
This focus is a special area of mixed modeling and involves the direct modeling of errors in the measurement of covariates. Measurement error arises frequently in biostatistical applications and its accommodation is a major area of concern. Dietary self-assessment questionnaires are a good example where errors in response are common (over- or under-estimating foods, for example).

Semi-parametric Modeling
Matteo Bottai Sc.D.
The use of parametric models is pervasive in much of biostatistics. Often these are assumed to be linear or have links to covariates that are linear. However it is often the case that relations with covariates are more complex and the real underlying relation is more complex. Spline models are commonly used to replace linear models to allow such complexity and these lead to generalized additive models.

Longitudinal Analysis
Bo Cai Ph.D.
When measures are made over time on an observational unit this yields longitudinal data. This type of data is widespread in biostatistical applications. Intervention or clinical trials are based on repeated measurements on the same unit. This set of methods involves modeling time dependence of individual outcomes. Often Bayesian or generalized estimating equation methods are used in these applications.

Variable Selection
Bo Cai Ph.D.
Often in biostatistical applications a dependent outcome variable is related to more than one predictor or covariate. When multiple covariates are to be included in a model the issue of which to include becomes important. The topic of variable selection is a major concern across many area of biostatistics. It is particularly important when assessing the inclusion of fixed and also random effects within models.

Singular Information
Matteo Bottai Sc.D.
When parameters are close to boundary their behavior can be changed considerably. The usual models describing the probability distribution of parameters do not operate well close to this boundary. For example, a disease map is clustered. If the clustering reduces to zero then this is a boundary of the clustering parameter space. This can be used to set up correct hypothesis tests and models.

Spatial Biostatistics
Monir Hossain Ph.D.
This field concerns the use of spatial statistical methods within biostatistical applications. For example, applications in epidemiology can include relative risk estimation on maps, disease cluster assessment, evaluation of putative health hazards, prospective public health surveillance. This is often called spatial epidemiology. In addition, the inclusion of geographical location in the analysis of survival data in clinical trials or longitudinal analysis, where the residence address is deemed important, is also included. This field is closely related to the application of geographical information system (GIS) methods in public health, but differs from this in that it emphasizes statistical/inferential aspects. It is becoming increasingly important in the emerging area of syndromic surveillance and public health preparedness.