Malignant melanoma is a malignant tumour of the pigment forming cells of the skin, the melanocytes, which normally reside within the epidermis abutting the dermo-epidermal junction. Although the cancer is dangerous, its treatment is frequently successful if the condition is detected at an early stage: typically whilst penetration into the dermis is less than 0.7mm. This thesis presents a non-invasive method for assisting in the diagnosis of malignant melanoma by quantifying diagnostically relevant histological information from the analysis of one colour and two infrared images of a lesion. In particular, it is shown that the presence of melanin can be detected when its invasion from the dermo-epidermal junction is 0.02mm. Through development of an optical image formation model of human skin it is shown that all normal skin colours lie on a two-dimensional surface within a three-dimensional colour-space. In contrast, colour co-ordinates corresponding to the penetration of melanin into the dermis deviate from this surface along characteristic paths and thus can be identified. This solution could not be directly applied because the colours resulting from a decrease in the thickness of the top layer of the dermis, the papillary dermis, can occupy the same position in the colour space as colours resulting from melanin descent. This ambiguity can be resolved by transforming the measured colour co-ordinates to compensate for the thickness of the papillary dermis. It is shown that this thickness can be obtained through the analysis of a pair of infrared images. It is further shown, although not experimentally verified, that the amount of epidermal melanin, dermal blood and thickness of the papillary dermis can also be quantified.

Pharmacists have an essential role in intervening upon hospital prescribing errors (PEs). Electronic prescribing (e-prescribing) systems have been proposed to reduce PEs in hospitals. Measuring satisfaction of technology users is important to enhance system utilisation. This four-step programme of work aimed at: 1) developing and assessing the feasibility of a computerised method to collect PEs from a hospital's e-prescribing system, 2) validating the reasons recorded by doctors and pharmacists for discontinuing medication orders at hospital discharge, 3) investigating, using the developed computerised method, the number, types, severity, pharmacists' impact, and predictors of PEs in the context of e-prescribing system at hospital discharge, and 4) assessing the attitudes of pharmacists and doctors towards e-prescribing systems at three teaching hospitals. A pilot validation study was conducted initially that comprised two steps: electronically extracting discontinued orders from the e-prescribing system followed by structured interviews with doctors and pharmacists who made the discontinuation. To characterise pharmacists' interventions upon PEs, a four-week study was conducted; meetings were conducted within a week of data extraction with pharmacists performing their routine clinical work, who categorised the occurrence, type, and severity of their interventions using a scale. An independent senior pharmacist retrospectively rated the severity and potential impact, and subjectively judged whether any error was a computer-related error (CRE). The emerging results, together with the literature findings, were used to adapt a previously published questionnaire for a survey of pharmacists and doctors. The methodological approach was deemed valid, efficient and feasible. The majority of recorded discontinuation reasons were judged accurate (97,93.3%). The incidence of PEs was 8.4% (n= 664/7920; 95% CI: 7.8%-9.0%). Omission (31.0%), drug selection (29.4%) and dosage regimen (18.1%) error types were the most frequent. There were 18 (2.9%) serious, 481 (76.3%) significant and 131 (20.8%) minor erroneous orders. Most erroneous orders (469,74.7%) were rated as being of significant severity and significant impact of pharmacists. CREs (n=279) accounted for 44% of erroneous orders. Drugs with multiple formulations (OR 2.1,95% CI: 1.25-3.37) and prescribing by junior doctors (OR 2.54,95% CI: 1.08-5.90) were significant predictors of PEs. The majority of pharmacists and doctors agreed that the e-prescribing system improved the efficiency of prescribing and reduced dosage regimen errors. Pharmacists were more satisfied with their e-prescribing systems than doctors. Satisfaction was predicted with more efficiency-related attributes of the e-prescribing system, than those related to the quality of patient care. This work showed that PEs commonly occur even with the use of an e-prescribing system. The high rate of CREs raised a substantial question as to the ability of the eprescribing system to facilitate certain types of PEs, or alternatively, that the working system in a hospital has to be changed and made safer. Pharmacists played an important role in efficiently documenting and preventing PEs before they could reach and possibly harm patients. CREs should be understood by pharmacists who need to complement, rather than duplicate, the strengths of the e-prescribing system.

Background: The incidence and prevalence of type 2 diabetes is increasing in all age groups but is particularly high in older people. Older people with type 2 diabetes are a growing but neglected population who have been systematically excluded from most major trials and as a result are currently being treated according to guidelines developed for younger people. They are actually a heterogeneous group consisting both of newly diagnosed patients and of patients who have entered into older age already diagnosed with diabetes. Patients diagnosed with type 2 diabetes aged > 65 years account for more than half of all newly diagnosed cases. Despite this, the effect of age at diagnosis on clinical and non clinical outcomes associated with type 2 diabetes is under-explored. Objective: The main focus of this thesis is to compare the consequences of type 2 diabetes when it is diagnosed among younger and older patients. Design: A large retrospective cohort study using routine data available for Tayside, Scotland was carried out to investigate mortality and morbidity. 10,532 patients newly diagnosed with type 2 diabetes and 21,056 non diabetic patients were followed up between 1993 and 2004. A postal questionnaire study was also administered to those patients newly diagnosed in 2004 to investigate whether the effect of type 2 diabetes on illness perceptions and quality of life also differed by age at diagnosis. Results: The absolute risk of all-cause and cardiovascular mortality increased with age at diagnosis. In contrast, the increased risk of mortality when compared to non diabetic counterparts decreased with age at diagnosis. Similar trends were observed for morbidity. Patients diagnosed in older age had only a marginal increased risk of mortality compared to matched non diabetic patients. The increased risk of morbidity was much higher. The questionnaire study suggested that patients diagnosed in older age were less concerned about xv their diabetes and less affected emotionally than patients diagnosed at a younger age. Impact on quality of life was less marked in patients diagnosed in older age but these results were not statistically significant. Conclusion: The consequences associated with a diagnosis of type 2 diabetes were shown to differ by age at diagnosis. A diagnosis of type 2 diabetes in older age is associated with increased mortality and morbidity, but increased risk of morbidity is more pronounced. There were also observable differences in illness perceptions and quality of life.

This thesis is presented in three chapters each of which is self-contained. Chapter 1 is concerned with a computer project whose main aim was to design and test a record maintenance and retrieval system in a general practice environment, ideally as part of an integrated systern embracing all the health services. The system developed was used 1 - with real patients for a S2 week period. So far two papers describing this system and the results obtained have been published in the International Journal of Biomedical Computing. The first in October 1970, "An on-line record maintenance and retrieval system in general practice", by J. F. Preece, D." Bo Gillings) E. O. Lipmann, and N. G. Pearson; and the second in April 1971, "An analysis of the size and content of medical records used during an on-line record maintenance and retrieval system in general practice", by D. B. Gillings and J. "F. Preece. The presentation here has joined these two papers together. Chapters 2 and 3 are concerned with aspects of a morbidity survey of a community. The Exeter Community Health Research Project was a one year survey of morbidity of a defined community of 75000 people from November 1966 to October 1967. In addition, a private census of the City of Exeter was conducted to obtain personal details such as age, sex, living conditions, social class and smoking habits of the members of the community. Two aspects of this study have been chosen for presentation here. Chapter 2 is concerned with the private census and considers the reliability of the information gathered. Chapter 3 makes some attempt at modelling health services usage. Some discrete distributions are considered, together with their multivariate generalisations. Data is fitted in the univariate and bivariate case .

The increase in obesity prevalence over the past 20 years has become an international public health crisis. The lack of timely, accurate information led to a six year delay in recognising the obesity epidemic in UK children, despite height and weight measurements of children at various ages being routinely collected for many years. The chain of data, analytical methods and public health expertise for child obesity analysis and public health reporting is fragmented. This fragmented workflow leads to delayed or lost intelligence and public health decision-makers have little control of the underlying systems to produce the intelligence they require. This research addressed the specific need for a public health intelligence workflow system to profile child obesity. A web based software system known as Obesity Atlas was developed with and for the Greater Manchester Primary Care Trusts. In both requirements analysis and evaluation it was found that the combination of statistics, charts, and geo-visualisation are more informative if chained together in a workflow than used separately. In order to inform more public health decisions with current resources, there is a need for a common framework for finding, sharing, enacting, and re-using public health intelligence workflows.

Objective: Colorectal cancer survival varies at individual level and also geographically. This study used secondary data to investigate whether hospital organisational factors may explain colorectal cancer survival.;Methods: For 28 acute hospitals treating colorectal cancer in London, data on 15 468 patients first treated between 1996 and 2001 were drawn from the Thames Cancer Registry and their 5-year relative survival was calculated, with standard errors adjusted for clustering. The literature was examined to identify potential hospital organisational predictors of survival. Four English national data sets including measures of hospital organisation were identified and assessed for quality using a standardised method (Directory of Clinical Databases, DoCDat). Variables were assembled relating to the years 2000-2001. A multivariate relative survival model was used to investigate cross-sectional associations between the organisational measures and survival. Effects of missing data were also examined statistically.;Results: The data sets were assessed as of sufficient quality for the study. Most data by hospital were over 80% complete. Missing values for tumour stage and lack of detailed treatment information were the main limitations of the cancer registry data. Hospital organisational variables examined included hospital volume, staffing, waiting times, cancer services standards, and type (teaching/non-teaching). Individual factors in the model included age, sex, deprivation index and stage, but comorbidity could not be tested. There was a significant survival gradient across the hospitals, both before and after adjustment for individual factors. No relationship was found between survival and hospital volume, medical or nurse staffing, or waiting times for referral assessment. However, significant associations were found for teaching status, and for four of the cancer standards.;Discussion: Interpretation is limited by the cross-sectional design, temporal relationships, missing data and the limited number of hospitals. However, the study shows the potential of using hospital datasets to investigate organisational factors in cancer survival, and indicates the possible impact of teaching hospital status and some measures of cancer standards on survival. Further research is indicated to confirm these associations and investigate pathways for the effects.

In this thesis stochastic techniques are used in attempts to understand cancer risk, its relationship to patient age and genotype, as well as its distribution in human populations. The starting point for the thesis is the general observation that cancer incidence grows in approximate proportion to an integer power of age. Quasi-mechanistic mathematical models of cancer incidence have suggested that the integer power in a given case is related to the number of crucial cellular events that must occur for a malignant tumour to evolve from a healthy tissue. This idea and its limitations are explored. Further applications of cancer incidence models are then evaluated and developed. Specifically, a critical examination is presented of the notion that increases in risk associated with a particular predisposing germline gene mutation, can provide information about the disease-associated activity of that gene. Finally, there is a discussion of heterogeneity in liability to cancer. Methods for quantifying this heterogeneity and its effect on incidence patterns are investigated.

The aim of this thesis is to explore pregnancy and perinatal outcomes in diagnosed HIV-infected women receiving antiretroviral therapy (ART) in the UK (United Kingdom) and Ireland. Population-based surveillance data on HIV-infected pregnant women and their children is collected through the National Study of HIV in Pregnancy and Childhood (NSHPC), which includes information on over 8000 pregnancies delivered between 1990 and 2007. The majority of diagnosed infected women now take highly active antiretroviral therapy (HAART) in pregnancy, which reduces the risk of mother-to-child HIV transmission. However, there have been concerns over the potential for maternal and fetal adverse effects, with conflicting findings from European and American studies regarding the association between HAART and premature delivery. In this thesis, trends over time in the demographic characteristics of HIV-infected pregnant women in the UK and Ireland are described, along with changes in the uptake of interventions for preventing mother-to-child transmission. Transmission rates are explored over a period when HAART was routinely available, and subgroups of women managed in the context of regularly updated national guidelines are compared with respect to their risk of transmission. Multivariable logistic regression models are used to assess the association between type of ART exposure in pregnancy and adverse outcomes including pre-eclampsia, prematurity, stillbirth, neonatal death and congenital abnormality. In addition, using data from the European Collaborative Study and the Pediatric Spectrum of HIV Disease project alongside the UK and Ireland data, the effects of differences in populations and methodologies (study design and analytical approach) on the observed association between HAART and premature delivery are investigated, and a pooled analysis of individual motherchild pairs is carried out. Finally, the risks and benefits of ART in terms of adverse pregnancy outcomes and mother-to-child transmission were jointly modelled using Monte Carlo simulation methods, to produce a risk-benefit ratio.

In countries where people of South Asian origin have settled, unexpectedly high coronary heart disease rates have been recorded in South Asian men and women compared with other ethnic groups. In England high CHD mortality is shared by Gujarati Hindus, Punjabi Sikhs and Muslims from Pakistan and Bangladesh. The high CHD rates in these populations are unexplained by levels of smoking, blood pressure, plasma cholesterol or dietary fat intake. To test whether disturbances of haemostatic activity, lipoprotein metabolism or carbohydrate metabolism might underlie the high CHD mortality in South Asians, a population study in east London was undertaken. The results confirmed that the high CHD rates in South Asians compared with the native British population cannot be explained by differences in the distributions of blood pressure or plasma cholesterol. The hypothesis of a disturbance of haemostatic activity was not supported. A pattern of low plasma HDL cholesterol and high triglyceride levels, high serum insulin levels after a glucose load and high prevalence of non-insulin-dependent diabetes was identified in CD Bangladeshis. On the basis of these findings and a review of other recent work it is suggested that: (i) insulin resistance underlies these disturbances of lipoprotein and carbohydrate metabolism in Bangladeshis; (ii) this tendency to insulin resistance is a general pattern in South Asian populations overseas; and (iii) it is a possible underlying mechanism for the high rates of both CHD and diabetes in these populations. The planning of a large study to test this is described. Preliminary results confirm that a syndrome of metabolic disturbances related to insulin resistance, first identified in Bangladeshis, is present also in Gujaratis and Punjabis. This is associated with a striking tendency to central obesity in South Asians. These findings point to the aetiological role of insulin resistance in CHD and suggest possible strategies for prevention in South Asian communities.

Probability distributions over many variables occur frequently in Bayesian inference, statistical physics and simulation studies. Samples from distributions give insight into their typical behavior and can allow approximation of any quantity of interest, such as expectations or normalizing constants. Markov chain Monte Carlo (MCMC), introduced by Metropolis et al. (1953), allows r sampling from distributions with intractable normalization, and remains one of most important tools for approximate computation with probability distributions. I While not needed by MCMC, normalizers are key quantities: in Bayesian statistics marginal likelihoods are needed for model comparison; in statistical physics many physical quantities relate to the partition function. In this thesis we propose and investigate several new Monte Carlo algorithms, both for evaluating normalizing constants and for improved sampling of distributions. Many MCMC correctness proofs rely on using reversible transition operators; this can lead to chains exploring by slow random walks. After reviewing existing MCMC algorithms, we develop a new framework for constructing non-reversible transition operators from existing reversible ones. Next we explore and extend MCMC-based algorithms for computing normalizing constants. In particular we develop a newMCMC operator and Nested Sampling approach for the Potts model. Our results demonstrate that these approaches can be superior to finding normalizing constants by annealing methods and can obtain better posterior samples. Finally we consider 'doubly-intractable' distributions with extra unknown normalizer terms that do not cancel in standard MCMC algorithms. We propose using several deterministic approximations for the unknown terms, and investigate their interaction with sampling algorithms. We then develop novel exact-sampling-based MCMC methods, the Exchange Algorithm and Latent Histories. For the first time these algorithms do not require separate approximation before sampling begins. Moreover, the Exchange Algorithm outperforms the only alternative sampling algorithm for doubly intractable distributions.