Advances in computing have resulted in many engineering processes being automated. Electrocardiogram (ECG) classification is one such process. The analysis of ECGs can benefit from the wide availability and power of modern computers. This study presents the usage of computer technology in the field of computerised ECG classification. Computerised electrocardiogram classification can help to reduce healthcare costs by enabling suitably equipped general practitioners to refer to hospital only those people with serious heart problems. Computerised ECG classification can also be very useful in shortening hospital waiting lists and saving life by discovering heart diseases early. The thesis investigates the automatic classification of ECGs into different disease categories using Artificial Intelligence (AI) techniques. A comparison of the use of different feature sets and AI classifiers is presented. The feature sets include conventional cardiological features, as well as features taken directly from time domain samples of an ECG. The benchmark AI classifiers tested include those based on neural network, k-Nearest Neighbour and inductive learning techniques. The research proposes two modifications to the learning vector quantisation (LVQ) neural network, namely the All Weights Updating-LVQ (AWU-LVQ) algorithm and the Neighbouring Weights Updating-LVQ (NWU-LVQ) algorithm, yielding an "intelligent" diagnostic heart system with higher accuracy and reduced training time compared to existing AI techniques.

Advances in recent treatments for HIV/AIDS patients have shown dramatic outcomes in extending the incubation period and AIDS survival time, whilst also providing significant improvements in the quality of patients' lives. This thesis establishes a model of the HIV/AIDS epidemic that incorporates the effects of treatments, in particular, the introduction of highly active antiretroviral therapy (HAART), which became widely available during 1996. The technique of compartmental modelling is employed in an attempt to reproduce observed AIDS incidence/prevalence, HIV incidence/prevalence, and deaths from AIDS data. There are movements between compartments (sub-populations affected by the HIV/AIDS epidemic) each with an associated parameter. Each sub-population has a differential-difference equation associated with it. Once these equations have been solved numerically they give a set of steady-state solutions, from which it is possible to estimate HIV/AIDS incidence and prevalence. Some parameter values within the model are obtained from surveys, census results, etc., but others are derived using a maximum likelihood estimation (MLE) procedure. The use of realistic values gives impressive results, creating a remarkable fit with routinely collected data relating to levels of HIV/AIDS incidence and prevalence in the UK homosexual population. Finally, the model is used to project levels of incidence and prevalence over the next few years, and to investigate several possible 'what-if' scenarios, with a brief investigation into the consequent cost implications.

The permanent nature of chronic disease may impair patients' psycho-social and physical well being, may change their attitude towards life goals and influence major life changing decisions (MLCDs) over time. Very little information is available in the literature about the long term impact of chronic diseases, particularly their influence on MLCDs. The aim of this study was to investigate the influence of chronic diseases on MLCDs and to develop a standardised tool for use across all chronic conditions to assess the impact of disease on MLCDs. 675 patients (100 from 6 specialties: cardiovascular, rheumatoid, diabetes, cystic fibrosis, chronic obstructive pulmonary disease, nephrology and 75 from dermatology) from the University Hospital of Wales, Cardiff and University Hospital Llandough, Llandough were invited to take part. The data was obtained through surveys, interviews and focus group discussions and this was reduced to core items through "content analysis". In a further study, 225 patients from the seven specialties were asked to complete the MLCDP (version la) and, factor analysis was applied to confirm the breadth and depth of the allocated domains and to determine construct validity of the MLCDP. 385 (57%) patients of 675 took part in the first phase of the study. Themes and statements generated from 316 (82%) evaluable responses (postal survey=258, individual interviews=50, focus group=8) were used for the development of the MLCD Profile. The most frequently reported MLCDs concerned early retirement, having children, job, career choice, relationships, housing, moving abroad and education. The correlation between the patients' age and the total number of reported MLCDs was significant (rs = -0.46, p=<0.001, n=308) showing negative relationship. In total, 41 affected MLCD themes were identified and grouped into 15 core MLCD categories. The working definition of health-related "Major Life Changing Decision" was also developed. The 45-item draft profile was grouped into six MLCD domains. 19 clinicians took part in the "content validation" stage and there was good agreement among the panel members for their ratings of language clarity, relevance, completeness and scaling. (Interclass correlation coefficient=0.71, p=<0.0001, CI=0.61-0.78, kappa coefficient=0.81, p=<0.0001, 0=0.69-0.93). This led to a new 41-item version of the MLCDP (version la), covering five MLCD domains: education, job/career, family/relationships, social and physical. 210 patients (30 from each of the seven specialties) were recruited into the second phase completed the MLCDP and data were analysed using factor analysis. The Cronbach's alpha value of 0.8 indicated good reliability. Several items were made redundant as a result of factor analysis this analysis supported the evidence of construct validity. Item prevalence ranking helped to retain conceptually important items at this stage. This profile was easy to complete for most patients (n=131, 97%) and mean completion time was 5.7 minutes. A 32-item version of the MLCDP (version 2) was finally developed, which requires future examination of its other psychometric properties. The MLCDP is potentially of benefit in alerting clinicians to the long term impact of a chronic disease on patients, and as a tool to assess the true burden of chronic diseases on individuals' long term quality of life. Clinicians' knowledge about the influence of chronic diseases on MLCDs is important to provide better and timely guidance to patients, to support better treatment decisions and eventually to lead to better health outcomes. This study also proposed strategies for patients to consider when taking MLCDs. Such support might result in more appropriate decision-taking and improved health outcomes.

The recent increased availability of geographically linked individual level health outcome data and improvements in exposure mapping techniques, which furnish point exposure estimates, motivate the development of spatial statistical methodology that takes full advantage of individual level data. Kernel density estimation is a powerful tool for mapping the risk of a health outcome that uses individual level data. Development of kernel density methodology has provided a global significance test for regions of elevated relative risk and a test for the spatial association between a health outcome and environmental exposure. Comparisons with some existing spatial statistical techniques highlight the strengths of the kernel density based methods. Moreover, simulation exercises indicate that the kernel density test for spatial association is a more powerful testing procedure than the most popular standard test proposed by Stone. Kernel density estimation and the global significance test for regions of elevated relative risk are illustrated for congenital malformations around a landfill site and sex ratios in Cardiff and the Vale of Glamorgan. The application of these methodologies revealed that both birth outcomes had a statistically significant heterogeneous spatial pattern over the relevant study regions even after adjustment for known confounders. Good quality, high resolution environmental exposure data was unavailable and prevented a direct application of the kernel density test for spatial association with a health outcome. However, the test can be applied to any two relative risk/density surfaces and was used to compare the spatial patterns of chromosomal and non-chromosomal anomalies in the region of the Nant y Gwyddon landfill site. It was concluded that the spatial patterns for the two sets of anomalies were different. The test was also used to assess the quality of the adjustment for confounders when producing expected risk surfaces and the adjustment was found to be adequate.

The explorative power of high throughput technologies in cancer research has become well established in recent years, exemplified by diverse gene microarray studies. However, development of the necessary biomedical data analysis tools has historically been confined to a commercial environment, while comprehensive, user-friendly analysis approaches are still needed. Availability of freely-available software, notably the 'R' project statistical programming language, allowed development of a user-friendly multivariate statistics application - Informatics Tenovus (I-10) - in this project. I-10 provides a platform through which powerful existing and future 'R' project statistical analysis methodologies can be applied, without prior programming knowledge. The new system was tested in the context of exploring antihormone resistance in breast cancer, analysing microarray datasets from in vitro models of acquired Tamoxifen (TAMR) or Faslodex resistance (FASR) versus endocrine responsive MCF-7 cells. The analysis not only revealed known de-regulated genes, but also further potential future markers/targets for endocrine response/resistance. The advantages of the 'R' programming environment together with Microsoft Visual Basic.net technology for producing user-friendly biomedical analysis tools facilitated subsequent development of a tool which could explore SEER cancer patient datasets. This new cancer query survival tool - Superstes -allows detailed statistical modelling of the impact that multiple patient attributes (in this instance derived from the SEER breast and colorectal cancer datasets) have on patient survival. The versatility of 'R' was additionally demonstrated in further exploring classifiers, where it was able to interface with the sophisticated, freely available machine learning application 'Weka'. Using 'R' and Weka, breast cancer patient survival was modelled using equivalent patient attributes to the Nottingham Prognostic Index and a 10 year survival subset of the SEER breast cancer dataset. Several machine learning methodologies were compared for their ability to accurately model survival, with their value in routine clinical use for prediction of patient survival then critically evaluated.

Gut damage, resulting in maldigestion or malabsorption of food and stimulation of the immune system, has been linked to growth faltering in young children in the developing world. Gut damage occurs along a spectrum, with only the more severe damage resulting in visible symptoms such as diarrhoea; most gut damage in young children is sub-clinical but chronic, and over time it can have a significant impact on a child’s growth rate. Hand-washing with soap has been found to reduce the risk of diarrhoea by 42-47%. Would this simple intervention also reduce the sub-clinical yet chronic form of gut damage associated with childhood growth faltering? Framed within the bio-cultural research paradigm, and theoretically informed by insights from Critical Medical Anthropology, this study used a mixed-method, longitudinal approach in order to investigate this question. Eighty-eight children aged 3-12 months were recruited from eight slum communities in Kathmandu, Nepal. Each community was allocated to a control or intervention group (n=43 and 45 children, respectively). In intervention areas, a community-based hand-washing with soap programme was devised and implemented for six months; in control areas, mothers continued their normal practices. The intervention was evaluated by comparing five outcomes: rates of maternal hand-washing, levels of child morbidity, gut damage, immune stimulation and growth faltering in the two groups. Hand-washing rates increased amongst intervention mothers: by the end of the intervention, mothers living in hand-washing areas were more likely to report hand-washing with soap after cleaning the baby’s bottom and before cooking, eating or feeding the child (for all, P<.01). As a result, children in the intervention areas experienced a decrease in both the number of diarrhoeal episodes (3.0 vs. 4.3 episodes, P=.049) and the number of days with diarrhoeal symptoms over the period of study (9.67 vs. 16.33 days, P=.023). Yet, despite reducing diarrhoeal morbidity, hand-washing had no impact on the biochemical or growth status of the children: there was no significant reduction in levels of gut damage or immune stimulation in children from intervention areas over the period of the study. Consequently there was no improvement in growth rates for these children, as measured by height-for-age, weight-for-age and weight-for-height z-scores. This study concludes that when children live in highly contaminated, over-crowded environments, with poor access to clean water and sanitation, selective interventions focusing on one small behavioural change are unlikely to have an impact. In such highly contaminated environments, faecal contamination of hands is just one of the many pathways by which these children are exposed to pathogenic organisms. The biggest threat to the health of these children is not poor hygiene behaviour, but life in the slum. Comprehensive strategies to provide basic services and raise general standards of living in the slums are the best way in which to have a significant impact: piecemeal interventions focusing on single issues risk being ineffective both in terms of health impact and cost-effectiveness. This point is situated within the literature on effective and sustainable health interventions and the wider social and political debates surrounding global public health policy and practice in the 21st Century.

1 Abstract 1.1 Title Functional problems: prevalence in secondary care and perceptions of doctors 1.2 Aims (a) To evaluate the prevalence of functional problems in general medicine, gastroenterology, gynaecology and psychiatry outpatient specialist clinics in a hospital setting in Sri Lanka and (b) To assess the perceptions of doctors in these specialties around functional problems in the UK and in Sri Lanka. 1.3 Methodology (a) Prevalence study: Functional problems were defined as those for which a cause could not be ascertained after clinical evaluation and investigations. The prevalence of functional problems was assessed over a three month period in each specialty clinic. The patients‘ records were reviewed at three and six months to confirm the diagnosis of a functional problem. (b) Perceptions of clinicians: Qualitative methodology was used to ascertain the perceptions of doctors in the two settings covering the three specialties. A grounded theory approach was used and sixty interviews were carried out. Emphasis was placed on identifying socio-cultural implications around perceived causations and the management of these functional problems. 1.4 Results (a) Prevalence study: The prevalence study ascertained that functional problems were the commonest diagnoses in the general medicine/gastroenterology and gynaecology clinics accounting for almost a fifth of patients. In psychiatry, functional problems were the fourth common diagnosis and accounted for nearly ten percent. Patients of all consultations with functional symptoms tended to be younger; there were long delays in iv making the diagnosis and a substantial proportion of patients were subjected to iatrogenic harm from invasive investigations and inappropriate therapeutic measures. (b) Perceptions study: The perceptions study revealed divergent views by doctors about what functional problems were, how they could be categorized and how best to manage them. Socio cultural factors were thought to be intricately linked to causation and outcomes. Nonetheless, most doctors tended to isolate the clinical presentation and management from the cultural context in the way they dealt with their patients. 1.5 Conclusions Functional problems were commonly seen and diagnosed in hospital outpatient clinics. The Sri Lankan prevalence was similar to that reported from the UK. In both cultural settings the doctors who were sensitive to socio-cultural factors used tools beyond pharmacotherapy and those who held improved quality of life as the goal of treatment as opposed to cure, reported greater success in managing people with these problems. v 1.6 Acknowledgements The School of Medicine and Health, Durham University, for granting a scholarship through Project Sri Lanka for the PhD, for fulfilling research training needs and providing accommodation during my stay in the UK The Faculty of Medicine, University of Ruhuna, for granting me study leave for three years Professor Pali Hungin, my supervisor in the School of Medicine and Health, Durham University, for helping convert the research idea to a PhD thesis Professor Martyn Evans, my second supervisor in the School of Medicine and Health, Durham University, for guidance and support Professor David Petley, Deputy Head (Research), Faculty of Social Sciences and Health, Durham University, for research support To Professor Susirith Mendis, Vice Chancellor and Professor Ranjith Senaratne, former Vice Chancellor of the University of Ruhuna, Professor Thilak Weerasooriya, Dean and Professor P.L. Ariyananda, former Dean, Faculty of Medicine, University of Ruhuna, for support in obtaining study leave and a travel grant Dr Chandanie Hewage, Head, Department of Psychiatry at the Faculty of Medicine, University of Ruhuna, for constant support and encouragement over the study leave period My colleagues in the Department of Psychiatry, Faculty of Medicine, for covering the additional work during my study leave period Mrs Alex Motley, for help with correspondence and formatting Dr Simon Stockley, General Practitioner, for assistance with the pilot study Dr Amanda Gash, Consultant Psychiatrist and Dr Suresh Babu, Consultant Psychiatrist at the Tees Esk and Wear Valleys NHS Trust, Dr Deepak Dwarakanath, Consultant Gastroenterologist and Dr Anne Ryall, Consultant Gynaecologist at North Tees and Hartlepool NHS Trust, who were the co investigators for the UK arm of the qualitative study My co-investigators from Sri Lanka for the prevalence study, Dr Gamini Jayawardene, Dr Gayani Punchihewa and Dr Ajith Jayasekera, Consultant Psychiatrists, from Teaching Hospital Karapitiya Galle, Professor Thilak Weerarathna and Dr Arosha Dissanayake, Consultant Physicians from Teaching Hospital Karapitiya Galle, Professor Malik Goonawardene and Dr Dammika Jayasooriya, Consultant Obstetricians from Teaching Hospital Mahamodera, Galle vi Dr Rachel Casiday, Dr Helen Hancock, Dr Helen Close, Dr Sharyn Maxwell, Dr Eileen Scott, Durham University, for helping with the methodological aspects of the research Professor James Mason, Director of Research, School of Medicine and Health, Durham University, for advice on statistical matters Dr J. Howse, Doctoral Researcher, Durham University, for second coding of transcripts Mrs Judith Walsh, Mrs Susan Williams, Durham University, for assistance with administrative matters Members of the NHS National Research Ethics Committee, Durham and Tees Valley Research Ethics Committee 2, and Research Ethics Committee, Faculty of Medicine, University of Ruhuna, for helping conduct an ethically sound research study Staff at the North Tees and Hartlepool NHS Trust, Research and Development Office for guidance on methodological and ethical aspects of the research Directors of Teaching Hospital Karapitiya, Galle and Teaching Hospital, Mahamodera, Galle, for permitting the analysis of patient records for the prevalence study All the participants in the research Mrs Barbara Hungin for encouragement and support Sri Lankan expatriate community for assistance with logistical aspects Staff at the Keenan House, where I was accommodated during my stay in the UK My family members for their constant encouragement and support during three long years of researching

People of South Asian ethnicities in the UK are at a high risk of obesity and related illnesses. This thesis tests predictions derived from the developmental origins literature regarding adiposity in British Pakistanis in middle childhood. Based on previous research, it was predicted that British Pakistani children would be more adipose and have lower birth weights compared to white British children. It was also predicted that second generation British Pakistani children would be more adipose and have lower birth weights than the third generation. White British children (n=211) and British Pakistani children (n=137), including second generation (n=82) and third generation (n=51), aged 7-11 years were measured for body mass index (BMI), waist circumference, triceps and subscapular skinfold thicknesses, blood pressure, and resting heart rate. Birth weight data were collected from one hospital in Middlesbrough (n=184). In addition, a pilot study was conducted on lifestyle, which explored potential differences in British Pakistani children’s lifestyle that may affect adiposity. Dietary data (n=30) were gathered by multiple-pass recall interviews. Physical activity data were collected by accelerometry (n=27) and also by multiple-pass recall interviews (n=30). A questionnaire was developed for the lifestyle pilot to gain an understanding of parents’ (n=24) knowledge and practices of healthy lifestyles. Focus group interviews were conducted in one school with children (n=18), which explored children’s knowledge and practices of healthy lifestyles. The same focus groups also explored the issue of child participation in the study to understand different motivations between ethnicities and sexes. Compared to white British children, British Pakistani children were significantly fatter by standard deviation scores for triceps (p=0.003) and subscapular skinfolds (p<0.000), but not by BMI (p=0.599) or waist circumference (p=0.253). British Pakistani children had significantly lower birth weights (p<0.001), and were more frequently classified as low birth weight (p=0.01) and small-for-gestational age (p<0.001). These results may support the foetal origins hypothesis, which is that early life influences can adversely affect later health, by linking foetal development with adiposity in childhood. There was a higher proportion of overfat by subscapular skinfold thickness (p<0.001) in second generation British Pakistani boys compared to the third generation. The two generation groups did not differ significantly in any measure of birth weight. The lifestyle pilot sub-study suggests that differences in lifestyle patterns may exist between the British Pakistani and white British families who participated, and it could provide a basis for a full study on this topic. Methods implemented in public health research should reconsider using the body mass index alone as a predictor of body fatness, especially in populations including British Pakistanis. Qualitative methodologies should be used to inform study design as a way of illuminating complex and interrelated issues such as obesity and ethnicity.

I am greatly indebted to The NovoNordisk Research Foundation for sponsoring a three year clinical fellowship which enabled me to complete both ADDITION and PACE studies.

This thesis investigates methods to combine and infer information from genetic epidemiological studies. Three issues are explored, each in a distinct and self-contained chapter. Chapter 1 investigates how best to incorporate treatment information in genetic analyses of blood pressure. Different approaches to adjusting for treatment are compared in a number of simulated scenarios, and the approaches that utilise all the observed data are generally shown to perform best. One particular condition, however, causes these approaches to suffer bias. This is where a genetic variant (or some other factor) interacts with treatment. This chapter therefore urges caution in the interpretation of results from these studies, and suggests some possible approaches to identifying existing interactions with treatment. Chapter 2 concerns participant privacy in genome-wide association studies (GWAS). Recent methods claim to be able to infer whether an individual participated in a study, using only aggregate statistics from the study such as allele frequencies. In the past, these statistics have been freely published online. This chapter explores the full implications of these methods, by investigating their true capabilities and limitations. In addition, some modifications are proposed to one particular method, to demonstrate how it can be adapted for use in practice. This work finds that participant identification is possible in ideal conditions, but common characteristics of real studies may prevent any reliable application of these methods in practice. Chapter 3 proposes a new approach to synthesising data between studies. This approach – named “DataSHIELD” – guarantees identical results to an individual-level meta-analysis, while offering greater flexibility than the studylevel meta-analysis. DataSHIELD also potentially circumvents some of the laws that restrict data use, because it does not involve sharing any individual-level data between studies. This chapter outlines the principles underpinning DataSHIELD, and demonstrates its use in a simulated data example.