This thesis presents results of an investigation of the variations in health in the elderly Thai between 2002 and 2007. The analyses are based on the Surveys of Elderly in Thailand in 2002 and 2007. Health at old age is one of the key concerns about population ageing in Thailand because older people are more frails. The differences in health status at old age between areas of residence, individual characteristics and time periods were investigated in this study. The variations of health in old age between areas of residence were measured using multilevel models and the results showed that the differences of health in old age between areas of residences are lower than the differences from individual characteristics. The rise of expected life years lead to the concern about whether these extra years will be spent in healthy or unhealthy life. To investigate trends of health in old age Thai, this study adopted healthy life expectancy calculated by Sullivan‟s method as the health measure. Because health has many dimensions, this study calculated healthy life expectancy based on self-rated health, self-care disability and mobility disability to represent different aspects of general health and disability in later life. The results showed that trends in healthy life expectancy varied by age, gender and the health indicators applied. A population projection for Thailand for 2000-2050 was calculated using the cohort components method. The results showed that based on the assumptions that fertility and mortality continue to decline as recently observed, the number and proportion of old people aged 60 and over will increase rapidly particularly the older old people aged 80 and over and old age women. The disability projection for Thailand in 2000-2050 also showed a large increase in the absolute number and percentage of disabled old age people. The trends in numbers of old age people and their health in the future result in rises of health expenditure in old age and in the demands for health care services, especially for long-term care and social security. The results from this study then inform policy making and plans for care of the elderly in Thailand in the future.

Introduction: In biomedical systems repeated measurements are often collected, thus presenting a statistical challenge due to high temporal correlation. This research investigates the potential utility of two distinct statistical methodologies in their application. Application: Two clinically diverse biomedical systems, linked by the common methodological interest of assessing control and performance, are considered. (i) An application to renal anaemia aims to investigate the stability of haemoglobin levels, measured monthly for 13 months within 151 patients, with the ultimate goal of improving patient control; (ii) the second an application concerns cerebral autoregulation (a stable cerebral blood flow over a range of arterial blood pressure), to maintain patient safety during a surgical procedure to prevent stroke. Repeated measurements of cerebral blood flow and arterial blood pressure were collected on 36 patients, yielding a total of 4519 cerebral blood flow and 4574 arterial blood pressure measurements (note that the number of observations vary between patients). Statistical methodology: Functional data analysis and multilevel modelling are utilised in the investigation of these two biomedical systems. Functional data analysis considers observations as a function rather than a highly correlated sequence of measurements. Multilevel modelling assumes that measurements are clustered and that within clusters, measurements are scattered about a trend in an uncorrelated manner. Results: Assessment of control within the renal anaemia system and knowledge of the relationship within the cerebral autoregulation system, has been achieved through the successful application of functional data analysis. Loess curves were used as means of exploring the cerebral blood flow – arterial blood pressure relationship in the cerebral autoregulation application. B-splines and phase plots were used to explore haemoglobin control in the renal anaemia system. Further, multilevel modellingincorporating autoregressive correlation structures appropriately models the dependency amongst model residuals due to temporal correlation. Both functional data analysis and multilevelmodelling have demonstrated their utility in the application to model control in biomedical systems. Conclusions: The novel application of these statistical methodologies has successfully provided contemporary insight into these biomedical systems and shows strong prospects for further applications.

Quantitative bio-analytical techniques that enable parallel measurements of large numbers of biomolecules generate vast amounts of information for studying and characterising biological systems. These analytical methods are commonly referred to as omics technologies, and can be applied for measurements of e.g. mRNA transcript, protein or metabolite abundances in a biological sample. The work presented in this thesis focuses on the application of multivariate prediction models for modelling and analysis of biological data generated by omics technologies. Omics data commonly contain up to tens of thousands of variables, which are often both noisy and multicollinear. Multivariate statistical methods have previously been shown to be valuable for visualisation and predictive modelling of biological and chemical data with similar properties to omics data. In this thesis currently available multivariate modelling methods are used in new applications, and new methods are developed to address some of the specific challenges associated with modelling of biological data. Three closely related areas of multivariate modelling of biological data are described and demonstrated in this thesis. First, a multivariate projection method is used in a novel application for predictive modelling between omics data sets, demonstrating how data from two analytical sources can be integrated and modelled to- gether by exploring covariation patterns between the data sets. This approach is exemplified by modelling of data from two studies, the first containing proteomic and metabolic profiling data and the second containing transcriptomic and metabolic profiling data. Second, a method for piecewise multivariate modelling of short timeseries data is developed and demonstrated by modelling of simulated data as well as metabolic profiling data from a toxicity study, providing a new method for characterisation of multivariate bio-analytical time-series data. Third, a kernel-based method is developed and applied for non-linear multivariate prediction modelling of omics data, addressing the specific challenge of modelling non-linear variation in biological data.

Research objectives. 1) To create an original and useful software application; 2) to investigate the utility of dyna-linking for teaching upper limb anatomy. Dyna-linking is an arrangement whereby interaction with one representation automatically drives the behaviour of another representation. Method. An iterative user-centred software development methodology was used to build, test and refine successive prototypes of an upper limb software tutorial. A randomised trial then tested the null hypothesis: There will be no significant difference in learning outcomes between participants using dyna-linked 2D and 3D representations of the upper limb and those using non dyna-linked representations. Data was analysed in SPSS using factorial analysis of variance (ANOVA). Results and analysis. The study failed to reject the null hypothesis as there was no signi cant di fference between experimental conditions. Post-hoc analysis revealed that participants with low prior knowledge performed significantly better (p = 0.036) without dyna-linking (mean gain = 7.45) than with dyna-linking (mean gain = 4.58). Participants with high prior knowledge performed equally well with or without dyna-linking. These findings reveal an aptitude by treatment interaction (ATI) whereby the effectiveness of dyna-linking varies according to learner ability. On average, participants using the non dyna-linked system spent 3 minutes and 4 seconds longer studying the tutorial. Participants using the non dyna-linked system clicked 30% more on the representations. Dyna-linking had a high perceived value in questionnaire surveys (n=48) and a focus group (n=7). Conclusion. Dyna-linking has a high perceived value but may actually over-automate learning by prematurely giving novice learners a fully worked solution. Further research is required to confirm if this finding is repeated in other domains, with different learners and more sophisticated implementations of dyna-linking.

The aims of this thesis were to determine if there is an association between damp housing and asthma and to investigate whether damp housing adversely influences asthma severity. Asthmatic subjects reported more damp in both their current (Odds Ratio 4.1, 95%CL 2.3 to 7.6) and previous (Odds Ratio 1.9, 95%CI 1.1 to 3.2) dwellings than control subjects. The surveyor confirmed 112 (51%) dwellings to have evidence of damp and 57 (26%) evidence of visible mould growth. Dampness was detected in 58/90 (64%) dwellings of asthmatic subjects compared with 54/132 (41%) dwellings of control subjects (Odds Ratio 2.62, 95%CI 1.50 to 4.55). There was an increasing prevalence of damp in the dwelling with increasing severity of asthma. This association could not be explained by potential bias in study design and persisted after controlling for socio-economic and other confounding variables (adjusted odds ratio 3.03, 95% CI 1.65 to 5.57). Asthma severity scores correlated statistically with measures of total damp (r=0.30, p=0.006) and visible mould growth (r=0.23, p=0.035) in the dwelling. Patients living in homes with evidence of damp had a lower FEV<SUB>1</SUB> (mean difference 10%, 95% CI 1.0 to 20) and a lower FEV<SUB>1</SUB>/FVC ratio (mean difference 5.4%, 95% CI -0.1 to 10.9) than patients living in dry dwellings. These associations persisted after controlling for unemployment, household income and cigarette smoking. Asthma is significantly associated with living in damp housing. Measures of asthma severity, disability and airflow obstruction are higher in patients living in damp, mouldy dwellings. Effective measures to reduce the risk of damp and condensation occurring in dwellings are required to be incorporated into future housing design.

Any system, however trivial, is subjected to data analysis on the signals it produces. Over the last 50 years the influx of new techniques and expansions of older ones have allowed a number of new applications, in a variety of fields, to be analysed and to some degree understood. One of the industries that is benefiting from this growth is the medical field and has been further progressed with the growth of interdisciplinary collaboration. From a signal processing perspective, the challenge comes from the complex and sometimes chaotic nature of the signals that we measure from the body, such as those from the brain and to some degree the heart. In this work we will make a contribution to dealing with such systems, in the form of a recent time-frequency data analysis method, the Hilbert-Huang Transform (HHT), and extensions to it. This thesis presents an analysis of the state of the art in seizure and heart arrhythmia detection and prediction methods. We then present a novel real-time implementation of the algorithm both in software and hardware and the motivations for doing so. First, we present our software implementation, encompassing realtime capabilities and identifying elements that need to be considered for practical use. We then translated this software into hardware to aid real-time implementation and integration. With these implementations in place we apply the HHT method to the topic of epilepsy (seizures) and additionally make contributions to heart arrhythmias and neonate brain dynamics. We use the HHT and some additional algorithms to quantify features associated with each application for detection and prediction. We also quantify significance of activity in such a way as to merge prediction and detection into one framework. Finally, we assess the real-time capabilities of our methods for practical use as a biosignal analysis tool.

The identification of risk loci in the Human Leukocyte Antigen (HLA) region using single-SNP association tests has been hampered by the extent of linkage disequilibrium (LD). Penalised regression via Least Absolute Shrinkage and Selection Operator (LASSO) can be used as a method for selection of variables in multi-SNP analysis, and to deal with the problem of multi-collinearity among predictors. This method applies a penalty that shrinks the estimates of the regression coefficients towards zero. This is equivalent to applying a double exponential (DE) prior distribution to the coefficients with a mode at zero, corresponding to the prior belief that most of the effects are negligible in a Bayesian approach. Parameter inference is based on the posterior mode, with non-zero values indicating marker-disease associations. Single-SNP, stepwise regression and the LASSO approach were applied to case-control studies of rheumatoid arthritis, a disease which has been associated with markers from the HLA region. A generalisation of the LASSO called the HyperLasso (HLASSO), which uses the normal-exponential-gamma prior in place of the DE, was also investigated. These approaches were applied to data from the Genetics of Rheumatoid Arthritis (GoRA) study. Genotype imputation was used as a means to jointly analyse the GoRA and the Wellcome Trust Case Control Consortium (WTCCC) HLA SNPs. The North American Rheumatoid Arthritis Consortium (NARAC) study was used to validate the findings. After controlling for type-I error, the penalised approaches greatly reduced the number of positive signals compared to single-SNP analysis, suggesting that correlation among SNP loci was better handled. The HLASSO results were sparser but similar to the LASSO results. One SNP in HLA-DPB1 was replicated in the NARAC study. In both models, the robustness of the retained variables was verified by bootstrapping. The results suggest that SNP-selection using LASSO or HLASSO shows a substantial benefit in identifying risk loci in regions of high LD.

Studies investigating exposure to disinfection by-products (DBPs) via chlorinated waters during pregnancy and adverse fetal growth outcomes have been limited by potential exposure measurement error, lack of exposure assessment validation and potential residual confounding. Factors driving DBP exposure are poorly understood, making it difficult to target resources appropriately in order to improve exposure assessment. These issues were investigated through DBP exposure assessment and validation for a new investigation of DBPs and fetal growth within the Born in Bradford (BiB) cohort study. Analysis of individual water use in the BiB cohort found that water consumption, showering, bathing and swimming varied by demographic and lifestyle factors. Sampling, analysis, and modelling of trihalomethanes (THMs) in tap water showed that THM concentrations exhibited clear seasonal variation, but spatial variability was limited across the study area. Various metrics of exposure to THMs during pregnancy were created, including ‘personalised’ semi-individual metrics. Analysis of these metrics revealed individual water use to be the main driver of THM exposure in this cohort, with spatial and temporal variability having little influence. Compared with a fully integrated THM exposure metric (incorporating ingestion, showering/bathing and swimming), metrics based only on THM concentrations or THM ingestion misclassified over 50% of women. A nested validation study was conducted using a 7-day water diary and urinary trichloroacetic acid (TCAA) biomarker. This found error in self-reported water use and TCAA ingestion estimates to vary by employment status - error being greater for employed women. Urinary TCAA was not correlated with TCAA in tap water, reinforcing that individual water use is the most influential driver of DBP exposure in this cohort. Recommendations for future research include improved individual water use assessment covering more activities and time-points in pregnancy, stratified analysis of questionnaire validation studies, and use of urinary TCAA as a main exposure measure in epidemiological studies.

The modern biological sciences are fraught with statistical difficulties. Biomolecular stochasticity, experimental noise, and the “large p, small n” problem all contribute to the challenge of data analysis. Nevertheless, we routinely seek to draw robust, meaningful conclusions from observations. In this thesis, we explore methods for assessing the effects of data variability upon downstream inference, in an attempt to quantify and promote the stability of the inferences we make. We start with a review of existing methods for addressing this problem, focusing upon the bootstrap and similar methods. The key requirement for all such approaches is a statistical model that approximates the data generating process. We move on to consider biomarker discovery problems. We present a novel algorithm for proposing putative biomarkers on the strength of both their predictive ability and the stability with which they are selected. In a simulation study, we find our approach to perform favourably in comparison to strategies that select on the basis of predictive performance alone. We then consider the real problem of identifying protein peak biomarkers for HAM/TSP, an inflammatory condition of the central nervous system caused by HTLV-1 infection. We apply our algorithm to a set of SELDI mass spectral data, and identify a number of putative biomarkers. Additional experimental work, together with known results from the literature, provides corroborating evidence for the validity of these putative biomarkers. Having focused on static observations, we then make the natural progression to time course data sets. We propose a (Bayesian) bootstrap approach for such data, and then apply our method in the context of gene network inference and the estimation of parameters in ordinary differential equation models. We find that the inferred gene networks are relatively unstable, and demonstrate the importance of finding distributions of ODE parameter estimates, rather than single point estimates.

Genome-wide association studies (GWAS) have recently shown their potential in the discovery of genetic factors associated with common diseases. Genetic association studies including GWAS can be used to explore the role of genetic variation underlying the associations between birth size, growth and metabolic phenotypes such as adiposity, lipid and glucose levels and hypertension. The aim of this thesis was to 1) review methods for genetic association analyses, 2) fit models for growth measurements, and to investigate prenatal predictors of early growth and associations between early growth and adult metabolic phenotypes, and 3) to examine genetic variation underlying birth size, postnatal growth and adult metabolic phenotypes. The primary study population comprised Northern Finland Birth Cohort 1966 (NFBC1966) members with DNA (N=5,753). Phenotypes included height/weight throughout childhood and adult metabolic phenotypes. Parametric growth curves were fitted to obtain peak growth velocities and timings of peaks and nadirs. These growth parameters were analysed in relation to birth and adult metabolic phenotypes and genetic variation. Meta-analyses of GWAS included samples with similar data. Shorter babies grew faster in length immediately after birth. Faster postnatal growth was associated with higher adult blood pressure and adiposity, independently of birth weight. Risk alleles at type 2 diabetes locus (ADCY5) were inversely associated with birth weight in a GWAS meta-analysis. Variants near BMI candidate genes LEPR and PCSK1 were associated with infant BMI. The established obesity locus (FTO) had a strong association with BMI after age 5 years. A GWAS meta-analysis of metabolic phenotypes suggested distinct pathways leading to the development of a metabolic syndrome. Adult height variants were associated with infant and/or pubertal height growth. The results suggest that foetal programming, growth acceleration and genetic susceptibility contribute to the associations between growth and metabolic phenotypes, and that some of the genetic effects are age-dependent.