Background Liver disease is a serious problem both in the UK and globally. While the incidence and mortality from several chronic diseases are decreasing, mortality from liver disease is increasing. As well as the medical sequelae for an individual with liver disease, in the UK the increase in chronic liver disease poses particular problems with respect to increasing hospital admissions, mortality and significant costs to the public both in terms of treatment and in loss of productivity. The increase in society of several risk factors for chronic liver disease, notably alcohol intake, obesity and type 2 diabetes, mean that these problems are likely to increase in the future. Despite these apparent problems there are surprisingly few reliable sources of data on the occurrence of chronic liver disease (cirrhosis) in the general population of the UK and the rate and consequence of disease progression particularly among ambulatory patients. Nor are their robust estimates of the prevalence of abnormal liver function tests (which may represent undiagnosed liver disease) and their associations with mortality. This thesis utilises two distinct datasets to examine separate areas of interest in the epidemiology of liver disease in the UK. The first three studies contained within this thesis are concerned with the epidemiology of cirrhosis in the general population of the UK. The second group of three studies focuses on the prevalence of elevated liver function tests in a population of older people in the UK, the demographic, clinical and lifestyle factors associated with such and the mortality following an elevated liver function test. Objectives 1. To estimate the incidence and prevalence of cirrhosis in the population of the UK 2. To describe the mortality associated with cirrhosis compared with the general population and the disease progression of cirrhosis 3. To estimate the prevalence of elevated liver function tests among people aged 75 and over in the UK 4. To describe the association between elevated liver function test and demographic, lifestyle, clinical characteristics and mortality among people aged 75 and over. Methods To examine objectives 1 and 2 I utilised the General Practice Research Database (GPRD) constructing a population based cohort of 4537 subjects with cirrhosis and 44,403 age, sex and practice matched controls. I used Poisson regression to estimate incidence rate ratios and describe trends in alcoholic and non-alcohol-related cirrhosis. Using Cox regression within an historical matched cohort design I estimated the absolute excess mortality rates and hazard ratios for mortality in people with cirrhosis compared to the general population. I described the probability of progressing from one disease state to another. To examine objectives 3 and 4 I accessed data from one arm of the Medical Research Council (MRC) Trial of Assessment and Management of Older People in the Community, a representative sample of community dwelling people aged 75 and over, totalling 15,308 participants. The prevalence of abnormal liver function was described as the proportion of study participants with elevated aspartate transaminase, alkaline phosphatase or serum bilirubin. Associations between elevated liver function and demographic, lifestyle and clinical factors were examined using multivariable logistic regression. I determined the absolute mortality rates and hazard ratios for all-cause and cause-specific mortality using a Cox proportional hazards model. Findings Epidemiology of cirrhosis (GPRD) These studies have shown an increasing trend in both the incidence and prevalence of cirrhosis in the UK with an estimated 45% increase in incidence of cirrhosis in the 10-year period studied. I estimate that 76 per 100,000 people were living with cirrhosis in 2001. Just over half of all cirrhosis was associated with alcoholism. Disease progression with cirrhosis among this mainly ambulatory population was rapid with a rate of decompensation in people with compensated disease of 5% per year and 1 in 10 dying in the first year following diagnosis. This figure increased to 25% of people dying within one year for those with decompensated disease. Mortality in subjects with compensated and decompensated cirrhosis was 93.4 and 178.0 per 1000 person years compared with only 19.2 per 1000 person years in the general control population. Following adjustment for age and sex people with compensated and decompensated disease were respectively 5 and 10 times more likely to die than the general population. Epidemiology of abnormal liver function tests (MRC cohort) Abnormalities in liver function were common with roughly 1 in 6 people aged 75 and over having at least one elevated liver enzyme, although most of these elevations were mild. A single elevated measurement of aspartate transaminase was associated with an increased consumption of alcohol and a lower age in contrast with that of a single measurement of alkaline phosphatase which showed an association with higher age and lower alcohol consumption. An elevated bilirubin measurement was strongly associated with being male. Having a single elevated liver function test was associated with a modest increase in the hazard of death compared with people with normal liver function tests (adjusted hazard ratio for death 1.27 (95% CI[1.19, 1.36]). As well as an unsurprising increase in the hazard ratio for death from liver disease, elevated aspartate transaminase or alkaline phosphatase were both associated with modest increases in the hazard of death from cancer (adjusted hazard ratios of 1.56 (95%CI[1.21, 2.01]) and 1.61 (95%CI[1.39, 1.86]) respectively). Elevated alkaline phosphatase was additionally associated with increases in the hazard of death from respiratory disease (adjusted hazard ratio 1.58 (95%CI[1.32, 1.90])) and cardiovascular disease (adjusted hazard ratio 1.34 (95%CI[1.17, 1.55])). Conclusions From my work on the incidence and prevalence of cirrhosis I estimate that a minimum of 31,000 people in the UK are living with cirrhosis, a figure which is likely to rise given increasing trends in the incidence of cirrhosis described in this thesis. The significant mortality and disease progression associated with cirrhosis means that more needs to be done to combat both the incidence and progression of this disease both on an individual and population level. Elevations in enzymes regarded as reflecting liver function are common in people aged 75 and over and in most people these abnormalities are less than 2x the upper limit of normal for the assays used. These elevations I observed are associated with both a modest increase in all-cause mortality and also with an increase in death due to specific causes. Rather than simply a marker of liver function the investigation of people with elevated liver function tests, particularly those with severely elevated tests, may lead to the identification of potentially treatable conditions that underlie death.

Background Sulaymaniyah is one of the three provinces of the Kurdish region in northern Iraq with a population of 1,700,000. Burn injuries remain a major concern for health authorities in this region where published data on the nature and size of the problem are scarce. The objectives of this PhD project were to investigate the epidemiology of burn injuries, burn mortality, intentional self-harm burns and risk factors for burns in pre-school children. Methods This project involved three main studies; an incidence and outcome study, a three-year admissions study and a case-control study. In the incidence and outcome study which was undertaken prospectively from 3rd November 2007 to 2nd November 2008 at the only burns centre in Sulaymaniyah, all patients attending for a new burn injury were included whether admitted or treated as an outpatient. Patients admitted for intentional self-harm within this study were separately analysed. In the three-year admissions study, all acute burn admissions of 2006-2008 were included. The case-control study investigating risk factors for burns in children aged 0-5 years, involved incident burn cases and controls admitted for other conditions. The risk factors for death, for self-harm and for childhood burns were analysed using multiple logistic regression. Results The incidence and outcome study: A total of 2975 patients were recruited (male 52%, female 48%; median age 18 years). The all-age incidence of burns was 389 per 100,000 per year and the highest incidence was in preschool children (1044 per 100,000 per year). The mechanisms of injury included scalds (53%), flame (37%), contact (7%), chemical (1%), electrical (1%) and explosives (1%). Most burns occurred at home (83%; male 68%, female 96%). There were 884 admissions during the year amounting to an admission rate of 40.4 (males 34.6, females 46.2) per 100,000 per year with the highest rate being in preschool children (82.3 per 100,000 per year). Flame injuries accounted for most women admissions (91%) and scalds for most child admissions (84%). The mortality rate was 9.1 (males 2.5, females 15.6) per 100,000 per year. The median total body surface area (TBSA) burnt was 18% and median hospital stay was 8 days. In¬hospital mortality was 28%. Adjusted odds ratios for death were 36.4 (95% confidence interval 15.9-83.3) for TBSA burnt ≥ 40%; 5.4 (1.7-18.5) for age of 60 and over; 3.6 (1.7-7.3) for inhalation injury; 5.6 (2.5-12.9) for self-inflicted burns and 3.0 (1.3-6.8) for autumn season. Regarding self-harm burns, there were 197 cases of intentional self-harm burns during the year (male 6%, female 94%) amounting to an incidence rate of 8.4 (male 1.2, female 15.5) per 100,000 per year. The median age of patients was 20 years, the median TBSA burnt was 74%, the median hospital stay was 4 days and in-hospital mortality was 88%. The adjusted odds ratios for the risk factors for self-harm were 13.8 (6.9-27.4) for female sex; 3.9 (2.2-7.0) for young age of 11-18 years; 2.5 (1.2-5.5) for lower levels of education; 2.4 (1.3-4.4) for spring season; and 2.7 (1.4-5.2) for small family size of 1-3 members. The three-year admissions study: There were 2829 acute burn admissions from 1st January 2006 until 31st December 2008 with an in-hospital mortality rate of 27%. There was similar number of patients in each year with no significant differences in terms of sex, median age, median TBSA burnt, and in-hospital mortality. The case-control study: The case-control study included 248 cases & 248 controls. 79% of cases were scalds, 17% contact and 4% flame injuries. Burns most commonly occurred in sitting rooms (53%) and in the kitchen (36%) and were most commonly caused by tea utensils (42%) and kerosene stoves (36%). The adjusted odds ratios for risk factors for burns were 5.4 (2.6-11.7) for poor living standard; 5.3 (3.4-8.5) for child activity score; 2.8 (1.5- 5.2), for family history of burns; 1.3 (1.0- 1.7) for a one unit increase in presence of home hazards; 0.4 (0.2- 0.7) for presence of a second carer; and 0.14 (0.03-0.6) for presence of disabilities. Conclusion Burns are an important public health problem with high incidence and mortality rates. Morbidity is highest in pre-school children and mortality is highest in young females. Suicide by self-burning is common and mostly affects adolescents and young women. Collective action is required from the health authorities and their partners to address these issues through developing prevention strategies incorporating further research to the situation, improvement of service delivery to those affected and preventive interventions particularly addressing burns in pre-school children and intentional self-harm burns in women.

A further assumption made by these methods is that of assuming knowledge of the error variance. Theoretical results are presented with respect to estimating all the model parameters from a single study dataset. An investigation into the practical elements of this theoretical work is presented, concluding that there is not enough information within the data to practically estimate all the model parameters.

Introduction: The main pharmacovigilance system in the United Kingdom is the ‘Yellow Card’ spontaneous reporting system which suffers from low reporting rate, and long lag time between drug launch and ADR recognition. Aim: The aim of this study is to develop a pharmacovigilance system to permit the early detection of adverse drug reactions using routinely acquired NHS health data with minimal cost and resources. Methods: There are 2 methods for this study; Phase 1: The extraction of drug persistence data from routinely acquired NHS health data, and Phase 2: Identifying the exact reason(s) for patient discontinuation of drug therapy within 6 months of the index prescription. Results: Phase1: During the study period 4243 patients were initiated on ramipril, 8849 patients on simvastatin, 3242 patients on ARBs, 3646 patients on amlodipine and 269 patients on lercanidipine. The 1, 2-3 and 4-6 month discontinuation rates were 9.9%, 4.9% and 4.2% respectively for ramipril, 9.5%, 3.4% and 3.2% for simvastatin, 8.7%, 2.9% and 2.5% for ARBs, 16.2%, 6.3% and 4.8% for amlodipine, and 17.8%, 3.7% and 3.7% for lercanidipine. Drug discontinuation rates determined agree closely with published data from trials and post marketing surveys in terms of the peak time at which ADRs and discontinuations occur (1 month), the populations most frequently affected (females and the young or elderly depending on drug), and the relationship between the frequency of ADRs and discontinuations relative to the drug of interest, especially for antihypertensive (CCBs>ACEIs>ARBs). Phase 2: Six (20%) of 30 participating primary care practices, contributing to the PTI database, agreed to be approached directly. Completed data was returned for 98% of patients whom discontinued amlodipine due to a specific ADR. Conclusions: Drug discontinuation rates obtained from health care databases is a good surrogate for ADR/E rates. Specific reasons for discontinuation, such as adverse drug reactions, can be identified directly from such electronic databases or more effectively from the primary care medical records held in primary care practices.

Simple randomisation is the easiest method for allocating participants to treatment groups in clinical trials. In the long run it balances all features of participants across the groups but may not be suitable for small to medium sized trials. If important prognostic factors are identified at the design stage then stratified randomisation or minimisation can help to balance these features. Aim: To examine the relative benefits of different randomisation algorithms and determine guidelines for which randomisation design is advisable for a given trial. For a trial of known size with a specified number of important prognostic factors, and levels within these, it will be possible to identify the most appropriate randomisation technique for that trial. Methods: A review of methods of randomisation was first conducted followed by a survey of trialists into the current use of randomisation methods in clinical trials. Using simulations the following comparisons were made; simple randomisation compared with minimisation, whether to stratify or minimise by centre and predictability versus balance when using minimisation. The recommendations resulting from the simulations were used to design a prototype generic randomisation program. Results: The review and the survey both highlighted the probability of imbalance using simple randomisation. Minimisation was seen to be superior in producing balanced groups but the method was criticised for being more complex and unpredictable. The simulations showed that several factors influence imbalance including size of trial, the number of prognostic factors and the number of categories within these. Optimal algorithms for maintaining balance while reducing predictability were presented for varying trial parameters. Conclusions: Minimisation is a suitable method of randomisation for most clinical trials. Several strategies can be employed to address the conflicting issues of predictability and imbalance without resorting to complex mathematical algorithms.

Introduction: Intensive Care Unit (ICU) medical personnel, in an ongoing process termed ventilation management, utilize patient physiology and pathology data to define ventilator apparatus settings. Aims: The aim of the research is to develop and evaluate in comparison hybrid ventilation advisor systems, that could support ventilation management process, specific to lung pathology for patients ventilated in control mode. Methodology: A questionnaire was designed and circulated to Intensivists. Patient data, as defined by the questionnaire analysis, were collected and categorized into three lung pathologies. Three ICU doctors evaluated correlation analysis of the recorded data. Evaluation results were used for identifying models basic architecture. Two custom software toolboxes were developed for developing hybrid systems; namely the EVolution Of Fuzzy INference Engines (EVOFINE) and the FUzzy Neural (FUN) toolbox. Eight hybrid systems developed with EVOFINE, FUN, ANFIS and ANN techniques were evaluated against applied clinical decisions and patient scenarios. Results: Seventeen (17) models were designed for each of the eight (8) modeling techniques. The modelled process consisted of twelve physiology variables and six ventilator settings. The number of models’ inputs ranged from single to six based on correlation and evaluation findings. Evaluation against clinical recommendations has shown that ANNs performed better; mean average error as percentage for four of the applied techniques was 0.16%, 1.29% & 0.62 for ANN empirical, 0.05%, 2.23% & 2.30% for ANFIS, 0.93%, 2.33% & 1.89% for EVOFINE and 0.73%, 2.63% & 6.56 for FUN NM, in Normal, COPD and ALI-ARDS categories respectively. Additionally evaluation against clinical disagreement SD has shown that 70.6% of the NN empirical models were performing in 90% of their suggestions within clinical SD, while the percentages were 53%, 53% and 59% for the EVOFINE, ANFIS and NN Normalized models respectively. The EVOFINE and ANFIS produced Fuzzy Systems whose architecture is transparent for the user. Visual observation of ANFIS architectures revealed possibly hazardous advices. Evaluation against clinical disagreement has shown that the NN empirical was not producing hazardous advices, while EVOFINE, ANFIS and NN Normalized were shown to produce potentially hazardous advice in 17.6%, 23% and 5.8% of the developed models.

The methodology of meta-analysis of diagnostic studies is underdeveloped compared to that of randomised controlled trials. However, summarising evidence from diagnostic studies is important, and interest is growing in diagnostic study meta-analysis. Current recommendations are to use summary receiver operating characteristic (SROC) curve methods, in particular the bivariate or Hierarchical SROC (HSROC) methods, which share many statistical properties. These methods have been little used as they are recent developments requiring statistical expertise. To test these methods, a systematic review on the diagnosis of carotid stenosis by four noninvasive tests was undertaken to provide data for the SROC curve methods. Both the bivariate and HSROC methods failed for all four tests. Further investigation of the behaviour of the SROC curve methods, in particular why they did not work for a large proportion of datasets, was undertaken in a simulation study. This found a failure rate of 50%. Failure of the SROC curve methods was more likely when the individual studies were small and the average sensitivity or specificity was high, and these are characteristics of real studies. The DiagMeta package was developed for use in R, a freely available software package —for use by reviewers who may lack methodological expertise —with guidance on when the SROC curve model fails and alternative analyses. Finally, an empirical comparison was made between individual studies' receiver operating characteristic (ROC) curves and their SROC curve. Even when ameta-analysis can be successfully performed, the resulting SROC curve is difficult to interpret and may not lie close to the individual studies' ROC curves. The SROC curve model is difficult to fit and can limited by the data. Reviewers should therefore try to obtain as much data as possible prior to meta-analysis.

Motorcyclists that have no protective structures while motorcycling as other occupants of automobiles do can be particularly vulnerable to accident injuries (i.e., motorcycles are not as crashworthy as automobiles). Motorcyclists' susceptibility to accident injuries in nature may act synergistically with the complexity of conflicting manoeuvres between motorcycles and other motor vehicles to increase their injury severities in accidents that take place at junctions (e.g., T-junction or crossroad). Previous studies have applied crash prediction models to investigate influential factors on the occurrences of different crash configurations among automobiles but statistical models of motorcyclist injury severity resulting from different motorcycle-car crash configurations have rarely been developed. This current research attempts to develop the appropriate statistical models of motorcyclist injury severity by various crash configurations conditioned on crash occurrence at T-junctions in the UK. T-junctions are selected in this study because such junctions represent the single greatest danger to motorcyclists - for junction-type accidents, the statistics from the UK Stats19 accident injury database over the years 1991 and 2004 suggested that T-junctions were ranked the highest in terms of injury severity (Le., accidents at T-junctions resulted in approximately 65% of all casualties that sustained fatal or serious injuries) and accident occurrence (i.e., accidents at Tjunctions accounted for 62% of all motorcyclist casualties). This may be in part because there is a comparatively large number ofT-junctions in the UK. Although the author was unable to take into account the exposure factor due to the lack of such data (Le., the total number of T-junctions, and the number of motorcycles travelling on these locations), it remains true that more severe accidents happen at T-junctions than any other type of junction. In this present study, motorcycle-car accidents at Tjunctions were classified into several crash configurations based on two methods that have been widely used in literature. The first method is based on the conflicts that arise from the pre-crash manoeuvres of the motorcycle and car. The second method is on the basis of first points of impact of the motorcycle and car. The crash configurations that are classified in this current study based on the mixture of these two methods include (a) accidents involving gap acceptance (i.e., approach-turn crash and angle crash), (b) head-on crash, and (c) same-direction crash (i.e., sideswipe crash and rear-end crash). Since injury severity levels in traffic accidents are typically progressive (ranging from no injury to fatal/death), the ordered response models have come into fairly wide use as a framework for analysing such responses. Using the accident data extracted from the Stats19 accident injury database over 14-year period (1991~2004), the ordered probit (OP) model of motorcyclist injury severity were estimated because the dependent variable (i.e., no injury, slight injury, KSI: killed or seriously injured) is intrinsically discrete and ordinal. A set of the independent variables were included as the predictor variables, including rider/motorist attributes, vehicle factors, weather/temporal factors, roadway/geometric characteristics, and crash factors. The current research firstly estimated the aggregate OP model of motorcyclist injury severity by motorcycle-car accidents in whole. Additional disaggregate models of motorcyclist injury severity by various crash configurations were subsequently conducted. It appears in this current research that while the aggregate model by motorcycle-car accidents in whole is useful to uncover a general overview of the factors that were associated with the increased motorcyclist injury severity, the dis aggregate models by various crash configurations provide valuable insights (that may not be uncovered by an aggregate crash model) that motorcyclist injury severity in different crash configurations are associated with different pre-crash conditions. For example, the preliminary analysis by conducting descriptive analysis reveals that the deadliest crash manner in approach-turn crashes and angle crashes was a collision in which a right-turn car collided with an approaching motorcycle. Such crash patterns that occurred at stop-/give-way controlled junctions appear to exacerbate motorcyclist injury severity. The disaggregate models by the deadliest crash manners in approachturn crashes and angle crashes suggest that injuries tended to be more severe in crashes where a right-turn motorist was identified to fail to yield to an approaching motorcyclist. Other disaggregate crash models also identified important determinants of motorcyclist injury severity. For instance, the estimation results of the head-on crash model reveal that motorcyclists were more injurious in collisions where curves were present for cars than where the bend was absent. Another noteworthy result is that a traversing motorcycle colliding with a travelling-straight car predisposed motorcyclists to a greater risk of KSIs. These findings were clearly obscured by the estimation of the aggregate model by accidents in whole. In the course of the investigation of the factors that affect motorcyclist injury severity, it became clear that another problem, that of a right-turn motorist's failure to yield to motorcyclists (for the deadliest crash patterns in both approach-turn crash and angle crash), needs to be further examined. The logistic models are estimated to evaluate the likelihood of motorist's right-of-way violation over non right-of-way violation as a function of human attributes, weather/temporal factors, roadway/geometric factors, vehicle characteristics, and crash factors. The logistic models uncover the factors determining the likelihood of motorists' failure to yield. Noteworthy findings include, for instance, teenaged motorists, elderly motorists, male motorists, and professional motorists (Le., those driving heavy goods vehicles and buses/coaches) were more likely to infringe upon motorcycle's right-of-way. In addition, violation cases appeared to be more likely to occur on non built-up roadways, and during evening/midnight/early morning hours This present research has attempted to fill the research gaps that crash prediction models focused on analysing motorcyclist injury severity in different crash configurations have rarely been developed. The results obtained in this current research, by exploring a broad range of variables including attributes of riders and motorists, roadway/geometric characteristics, weather/temporal factors, and vehicle characteristics, provide valuable insights into the underlying relationship between risk factors and motorcycle injury severity both at an aggregate level and at a disaggregate level. This research finally discusses the implications of the findings and offers a guideline for future research.

The move to electronic discharge summary systems was anticipated to solve the longstanding problems associated with poor data quality and reduce delay in the production and transmission of discharge summaries between secondary and primary care health care providers in the UK National Health Service. A consequence of investment in a national IT infrastructure for electronic health records has focused attention on template design and the IT system requirements. The routine practices of doctors involved in discharge summary construction, and other factors that contribute to the problems of delay and data quality, have been less well explored. This study aimed to gain an understanding of paper-based discharge summary construction in a secondary care context in order to identify and analyse the implications for improving electronic discharge summary systems, and potentially avoid inadvertent transfer of inherent problems. A mixed method case study design was used to examine the patient discharge process and the construction of discharge summaries in one NHS Hospital Trust. Data was collected through semi-structured interviews with hospital doctors (n=10) and simulated discharge summary production (n=10). A syntactic analysis was also performed on discharge summaries (n=11) and proformas (n=3). The data was analysed thematically and inductively in order to identify the factors that contribute to the twin problems of data quality and delay associated with discharge summaries. The pragmatic, semantic, syntactic conceptual framework (Morris, 1938), and Speech Act (Austin, 1962) and Mental Frame (Minsky,1981) theories, were used to analyse how information contained in discharge summaries was represented, interpreted and used. This study found that moving from a paper based to an electronic discharge summary system will not necessarily resolve the problems of poor data quality and delayed production of discharge summaries. More comprehensive solutions are required in order to facilitate more effective discharge summary communication between secondary and primary care health professionals, and to address entrenched custom and practice in current hospital practice. These include uni-professional (medical) orientation of discharge summaries, attitude of senior doctors, inadequate preparation of junior doctors, inconsistent data entry including absence of common usage of short forms and abbreviations, and little accountability for quality control. Recommendations include training for junior doctors, regulating the use of shortened forms, improving the features of data entry systems, structuring the clinical coding data and introducing systems to ensure greater organizational accountability for effective discharge communication. More comprehensive change related to the introduction of multidisciplinary contribution discharge summary construction and integration of discharge summary standards in care pathways may improve overall discharge summary quality.

Summary Background Elevated blood lipids (particularly cholesterol and sub-fractions) contribute to the risk of developing cerebral, peripheral and cardiovascular disease and associated complications which are leading causes of morbidity and death. Statins reduce the risk of suffering vascular events, with or without decreasing cholesterol levels. Statin prescribing continues to increase but there is scope to improve prescribing and dosing, particularly in primary care. However, there is insufficient empirical evidence to inform approaches to quality improvement. Methods Following pilot work, we designed a new model of primary care based pharmacist-led intervention for General Practitioners (GPs) and nurses. The aim of the intervention (called Statin Outreach Support, SOS) was to improve statin prescribing by GPs, in line with recent evidence, targeting patients at highest risk of suffering a vascular event. Eleven trained pharmacists worked in SOS allocated practices one day per week for a year. During this period, the pharmacist met three times with all GPs, all nurses and other practice staff. Between meetings, pharmacists used patient level clinical and prescribing data to identify eligible patients and help practices initiate, up-titrate the dose or switch to simvastatin 40mg where indicated. The effectiveness of SOS was tested in a prospective single blind cluster randomised controlled trial. Usual care (UC) practices received no pharmacist support during the study. With a mean of 1.7 years follow up, the study had over 90% power (at 5% significance) to detect a difference of 12% in the proportion of patients with controlled cholesterol after practices had received the SOS intervention. Results Thirty one practices were recruited from the UK’s largest Health Board area. At randomisation, 16 practices were allocated to the SOS intervention and 15 to UC with 4,040 patients included at baseline. Recruited practices showed few differences compared with invited, non participating practices. Practices and patients randomised to each arm of the study had similar distributions with respect to age, complications, cholesterol levels and statin prescribing. The mean age was 68 years; 53% male, 45% ischaemic aetiology. Fifty nine percent had no statin prescribed at baseline; only 51% had cholesterol controlled. Follow up included 7586 patients in 29 practices (one practice had disbanded between recruitment and randomisation and another practice dropped out). Compared with UC, the SOS intervention achieved the primary endpoint of increasing the proportion of patients prescribed Simvastatin 40mg with controlled cholesterol (SOS 44.9% vs. UC 27.9%; odds ratio 1.79 (95% CI: 1.61, 1.98), p< 0.001). Secondary endpoints were also improved in the SOS arm practices. The intervention effect was strong and consistent across most subgroups including a positive impact on patients from practices in areas of greater socioeconomic deprivation. Conclusion A pragmatic, new, complex intervention was developed, tested and shown to be effective in a cluster randomised controlled trial with good internal and external validity. If implemented on a wider scale, in practices with comparable characteristics and baseline prescribing, the SOS intervention has the potential to reduce the burden of vascular events for patients with vascular disease. This work provides a convincing evidence base for the role of pharmacists collaborating with primary care practices, to improve statin prescribing and drug based cholesterol management, for patients at highest risk of suffering vascular events.