The objective of this thesis was to explore and contribute to improve the use of studies reporting survival outcomes in cost-effectiveness analysis and evidence syntheses. A review of the literature was undertaken to compare and contrast methods commonly used to assess effectiveness and cost-effectiveness for survival outcomes. A method that re-creates sufficient statistics from published Kaplan-Meier curves was developed to allow re-analysis of published survival results. Finally, a new approach to modelling that allows external information on various different survival metrics to be combined with trial data were developed and illustrated with an example. The literature review highlighted that in numerous health technology assessments, survival outcomes were not analysed in a single, coherent analysis, but instead by two separate models: a parametric model was fitted to the standard treatment and a hazard ratio estimated from another model was then applied to this baseline in order to generate the treatment arm. To improve this current state of methodology, we found that KM data for each arm of the RCT could be reliably reconstructed from the published survival curves, by using a simple algorithm based on the inverted KM equations and the information on numbers at risk reported alongside KM curves and/or the total number of events reported. The KM data of one RCT on patients with head and neck cancer were re-created and then used to illustrate the new methods developed for the extrapolation of RCT data using external data. A US cancer database (SEER), mortality statistics from the general population and expert opinion were used to impose constraints on overall survival, conditional survival, and hazard ratio. We obtained fitted survival curves consistent with both the RCT and external evidence, therefore increasing our confidence in the extrapolations and resulting estimated mean survival differences between the arms.

It is widely recognised that our knowledge of the nutrient intake of pre-school children aged 2-5 years and of their nutritional requirements for growth, is poor. Consequently, opinion is divided on whether modified diets, aimed primarily at the prevention of future adult diseases, adequately support growth in pre-school children. Between May 1988 and April 1990 the nutrient intake and growth of 153 pre-school children from Edinburgh, aged 2-5 years, was assessed. 54 children repeated the study after an interval of 12 months to give a total of 207 assessments. Nutrient intake was determined by the 7 day weighed inventory method. Anthropometric measurements included height, weight and skinfold thicknesses. Supplementary information and social details are recorded by questionnaire. The data was grouped and analysed by age and gender of the children, and also by socioeconomic group. For each group of children a low mean energy intake of 80-85% of the current UK Estimated Average Requirement of energy was found (Department of Health 1991). The intake of other nutrients ranged widely, with group mean intakes at or above values of Recommended Nutrient Intakes, except for the intake of vitamin D which was very low and of iron in 2 year old children which was also low. The % of energy from fat, sugar, starch and dextrin varied considerably but no correlations were found with energy intake. Thus low fat or high sugar diets did not affect the average daily intake of energy. Such diets, however, did significantly affect the quality of the diet in terms of mineral and vitamin intakes per 1000 kilocalories. Also, highly significant correlations were found between intake of nutrients during the first and second survey of children studied twice.

Introduction: The increased use of meta-analysis in systematic reviews of healthcare interventions has highlighted several types of bias that can arise during the completion of a randomised controlled trial (RCT). Publication bias has been recognised as a potential threat to the validity of meta-analysis and can make the readily available evidence unreliable for decision making. Until recently, outcome reporting bias (ORB) has received less attention. ORB occurs when outcomes are selectively reported for publication based on their results. This can impact upon the results of a meta-analysis, biasing the pooled treatment effect. Methods: Empirical evidence from a series of cohort studies that have assessed study publication bias and ORB in RCTs is reviewed and summarised. Methods for identifying ORB in a review and trial reports are discussed using several motivating examples. Guidelines are obtained and assessed from worldwide organisations and UK based charities that fund RCTs; with regards to trial registration, protocol adherence and trial publication. Statistical, educational and policy solutions are reviewed. A novel weighted linear regression is used to predict missing outcome data when bias is suspected in a review containing structurally related outcomes. A maximum bias bound previously proposed is further assessed using a simulation study to test the sensitivity of the method when heterogeneity is present. The outcome reporting bias in trials (ORBln study is introduced and the impact ORB has on this cohort of reviews from the Cochrane Library is assessed using the bound for maximum bias. Results: Four empirical studies that examined the association between outcome reporting bias and statistical significance found that statistically significant outcomes were more likely to be completely reported than non-significant outcomes (range of odds ratios: 2.2 to 4.7). Guidelines for 73 organisations and charities were reviewed; only eleven of these organisations or charities mentioned the publication of negative as well as positive outcomes. The simulation study indicates that the bound for maximum bias is a useful approach for reviewers to apply to assess the robustness of the conclusions of a review to ORB and can be applied in the presence of heterogeneity. Assessing the impact of ORB in the cohort of reviews in the ORBIT study indicates that ORB had an impact in 15% of the reviews considered. Conclusions: This work summarises the direct empirical evidence for the existence of study publication bias and ORB. Researchers need to be aware of the problems of both types of bias and efforts should be concentrated on improving the reporting of trials. There is a need to provide more detailed guidance for those conducting and reporting RCTs to help prevent the selective reporting of results. There are several complementary initiatives to either prevent ORB or to reduce the occurrence of ORB. Therefore, with the implementation and development of these solutions it is hoped that in the future ORB will become less of a problem. However, methods that detect ORB and adjust results for ORB can be useful while the initiatives are given time to make an impact.

The steps involved in the medication process for instance prescribing, dispensing and administering medicines (the "medication process") have always posed some risk to individual patient health. Such "medication errors" were sufficiently common indeveloped countries by the 1960s to warrant formal research. Moreover, the frequency of errors appears not to have declined with the introduction of electronic systems. Several technological approaches has been used to reduce medication errors such as Computerized Physician Order Entry (CPOE), Electronic Patient Record (EPR), drug record system in Denmark and PRESGUID project in France, based on models and theories of errors in healthcare and electronic systems that were drawn from different countries including the UK. Examples of the systems and their evaluation methodology are examined, some of them have already been implemented and others are still being assessed. Consequently, an addition of Clinical Decision Support Systems in current prescribing systems (CDSSs) is necessary. The research considers the development of a medication management model including all the steps of the medication process that involves a clinical decision support system starting from medication prescribing to administration. A necessary preliminary step is to understand the types and contexts of the risks involved. The design of the model was based on the systematic review and meta-analysis outcomes that analysed journal papers, addressing individual themes from definitions, the source of risks and the consequences of errors to comparisons of the medication errors rates between simple prescribing systems and electronic prescribing systems that include CDSSs. The English National Programme For IT is also given prominence as it is a test case for many developed countries and the context for much of my practical work. The new model demonstrates a notable reduction in the frequency of medication errors and the number of patients with Adverse Drug Events. A system dynamics methodology is used to design the model; the first step is the development of Causal Loop Diagrams (CLDs) which are used as an alternative summary of systematic review finding. They show connections between human and technological factors in the medication process and suggest points for support and intervention potentially addressed by the model that includes CDSSs. The second step is the development of system dynamic models of hospital medication flow in order to detect the effect of CDSSs in reducing the rate of patients with medication errors and application of the effect data of CDSSs to one large hospital to identify the consequence of the model with CDSSs on patient’s rate. An evaluation is performed based on focus groups and discussions with clinical pharmacists to demonstrate how the drug safety model can be used. Furthermore an exploitation of pharmacist’s opinions on CLDs and drugsafety model by using a short survey has been presented to improve the consequences of medication errors

In general, medical students are unenthusiastic about working in geriatric medicine (GM). Variables potentially related to this problem were identified and questionnaires designed to measure them: attitudes toward the work-of GM, toward GM patients, to old people, amount of prior contact with old people, discussions with other doctors and respondents' gender. Fishbein and Ajzen's (1975) Theory of-Reasoned Action guided a study of how these variables related to final year students' enthusiasm for working in GM, i. e. their behavioural intention (BI). The model explained 60% of the variance in BI. Aact (attitude toward working in GM) and SN (subjective normative pressure) together explained 50% of the variance in BI. Multiple regression showed'that attitude toward GM work most influenced BI. Correlations between the components of this attitude and BI were examined. A method for identifying primary variables is described. Evaluations rather than beliefs emerged as primary. The more students valued social and caring aspects of medicine'(e. g. dealing with social problems and treating chronic cases), the less unenthusiastic they were about GM. The more they valued its financial rewards and status aspects the more unenthusiastic they were. Attitude toward GM patients related weakly to`BI. Attitude to old people was not convincingly related to BI. Women's mean rating on BI was slightly higher than men's. Women students who had worked with old people had the highest BI scores. Cluster analysis showed that students may be grouped according to their values and also-supported the findings of the correlational analysis. A longitudinal study traced change and stability in students' primary values during medical training. Only the value students attached to dealing with social problems appeared to change. It declined during the clinical period. Students appeared to enter medical school with the value systems relevant to an interest in GM already formed. Ways by which GM's'image may be improved are discussed

The studies in the thesis are divided into two broad sections. The first section is descriptive. Data were collected prospectively over a five-and-one-half year period for approximately 4000 fractures. Validation of the data is performed. The data are used to describe the epidemiology of the fracture in the Lothian Region, and the anatomical outcome of the fracture. Multiple logistic regression analysis of the data is performed to identify those factors (recordable at patient presentation) that are prognostic of outcome. The statistical method used provides weighted significance for each of these factors, and thus mathematical formulae predictive of outcomes are constructable. A number of formulae are produced, depending on the displacement of the fracture at presentation (minimally displaced or displaced), and on the outcome measure (early and late instability, the risk of malunion, and carpal malalignment). The second section is validative. The studies in this section are an assessment of the performance of the mathematical formulae in the clinical setting. In the first study, data are collected prospectively for 139 patients, and outcomes recorded. In the second study, a group of clinicians involved in fracture management are asked to predict fracture outcome using first clinical experience and then the predictive formula. Results. The distal radius fracture occurred predominately in the older female patient following a simple fall. The fracture in this typical patient was usually unstable. The most consistently important predictors of fracture outcome were patient age, fracture displacement, comminution and ulnar variance. The mathematical formulae were able to correctly predict anatomical outcome in approximately 7/10 patients in the validative study. This was a significant improvement upon the predictive accuracy of the clinicians using experience alone. Use of the predictive formula also significantly reduced inter-observer variation in the assessment of fracture stability. Conclusion. Use of the predictive formula in the Accident and Emergency setting could improve decision-making in fracture management. By promoting an assessment of fracture stability rather than fracture displacement, appropriate management choices are facilitated. The unstable fracture can be referred for operative management, an ineffective closed reduction avoided. The thesis also demonstrates the potential value of the method employed. Multiple logistic regression analysis may provide a guide to treatment where the management of the condition is dependent upon the natural history.

We have developed' a multiple - state stochastic Markov model incorporating the factors hypertension, hypercholesterolaemia, diabetes, body mass index and the events ischaemic heart disease, stroke and death. Smoking is treated deterministically in our model. The model is specified by transition intensities obtained from Framingham Heart Study data and adjusted using Health Survey for England data to produce results that are consistent with the English population. We can calculate the expected future lifetime for an individual using the model. 'In terms of the expected future lifetime we have estimated the effect of treatment with a statin, a cholesterol lowering drug that reduces the intensity of having myocardial infarction and stroke. Statins can increase expected future iifetime by more than a year for certain classes of individuals. Also current smokers have a substantially reduced lifetime compared to those who never smoke and expected lifetime increases on giving up smoking. We have used the model to project future levels of obesity which can increase to very high levels if present trends continue, but this is unlikely to have a substantial effect on expected future lifetime. Due to unavailability of suitable longitudinal data from the UK, the model was parameterised using data from USA and adjusted to obtain results for the English population. As a result we have been unable to measure the uncertainty in the results for the English population directly. We have used parameter simulation to quantify the uncertainty in the results obtained. The results obtained in this thesis are based on the Framingham data sets and some of the results may be affected by characteristics specific to this population.

No description available.

An accurate diagnosis is a crucial part of an effective treatment. Diagnostic errors cause unwanted side effects for healthy individuals and witheld treatments for diseased patients. Meta-analysis techniques allow the accuracy of diagnostic tests to be estimated using all the available sources of evidence. The most common measures of diagnostic accuracy are sensitivity (true positive rate) and specificity (true negative rate). As part of this thesis, current methods developed for synthesising data from diagnostic test studies are reviewed and critiqued, and then applied to estimate the accuracy of the Ddimer test for diagnosing Deep Vein Thrombosis (DVT). The fit of the different models is assessed via the Deviance Information Criterion and the Residual Deviance and the most complex synthesis models are found to provide the best fit to the data. When covariates are added to these models, only the incorporation of study setting sensitivity is found to improve the fit of the model. Diagnostic tests are rarely used in isolation and consideration of multiple tests in combination may also require evaluation. In this thesis, a multiple equations with shared parameters approach is proposed which estimated the accuracy of a combination of tests in two stages: i) estimate the conditional accuracy of the tests; and ii) estimate the accuracy of possible combinations of tests as functions of the conditional accuracies. Such a modeling approach allows the inclusion of different sources of evidence to be used simultaneously. The final part of the thesis evaluated the cost-effectiveness of different strategies for diagnosing DVT by incorporating the results from the aforementioned evidence synthesis models into an economic decision analytic model. In conclusion, the assumption of conditional independence can affect the analyses of the effectiveness and the cost-effectiveness of combinations of diagnostic tests, thus leading to potentially wrong decisions if the dependence is not explicitly modelled.

Meta-analyses usually combine published studies, omitting those that for some reason have not been published. If the reason for not publishing is other than random, the problem of publication bias arises. Research into publication bias suggests that it is the ‘interest level’, or statistical significance of findings, not study rigour or quality, that determines which research gets published and subsequently publicly available. When the results of the scientific literature as a whole are considered, such publication practices distort the true picture, which may exaggerate clinical effects resulting in potentially erroneous clinical decision-making. Therefore, meta-analyses (as well as other more complex evidence synthesis models) based on the published literature should be seen as ‘at risk’ of publication bias, which has the potential to bias conclusions and thus adversely affect decision-making. Many methods exist for detecting publication bias, but this alone is not sufficient if results from meta-analyses are going to be used within a decision-making framework. What is required in the view of this thesis is a reliable way to adjust pooled estimates for publication bias. This thesis explores different novel and existing approaches to publication bias adjustment, including frequentist and Bayesian approaches with the aim to identifying those with the most desirable statistical properties. Special attention is given to regression-based methods commonly used to test for the presence of publication bias (and other ‘small-study effects’). The regression-based approach is seen to produce very encouraging results in a case study for which gold standard data exists. The incorporation of external information about the direction and strength of the bias is also explored in the hope of improving the methods’ performance. Ultimately, the routine estimation of the bias-adjusted effect is recommended as it improves the overall results compared to standard meta-analysis.