Background: The prevalence of type 2 diabetes is increasing and estimates suggest that by 2030 nearly 1 in 10 adults worldwide will be living with this condition. 10% of the NHS budget is spent on treating diabetes and related complications like heart attacks, strokes and blindness. Improving self-management in people living with type 2 diabetes is crucial in reducing the morbidity and mortality associated with this disease. Uptake of group-based self-management training is low and there is an urgent need for evidence based and effective alternatives. Computer-based interventions have the potential to provide cost-effective self-management training and improve outcomes for people with type 2 diabetes. The eHealth Unit was awarded a 5 year NIHR Programme grant for Applied Research to develop, evaluate and implement an online self-management intervention for adults with type 2 diabetes. This thesis describes my contribution towards developing the intervention. Aims/objectives: To describe the development of an online self-management intervention for adults with type 2 diabetes. Methods: The intervention development process was modelled on the MRC guidelines for developing complex interventions. This started with a systematic review of the literature on computer-based self-management interventions for adults with type 2 diabetes. Suitable theories to underpin the intervention were identified and used to construct a logic model to describe the potential mode of action. Qualitative work with people living with type 2 diabetes and health professionals was used to explore patient and professional defined wants and needs from such interventions. This data was then synthesized to inform the development of a theory-based online intervention called HeLP-Diabetes, designed to improve self-management in people living with type 2 diabetes. Conclusions: The synthesis of previous evidence with new qualitative data from patients and health professionals has helped to create a unique online intervention that will hopefully help bridge current gaps in the delivery of self-management training and improve outcomes for people with type 2 diabetes.

Background: Unexplained physical symptoms (UPS) that lack an organic explanation, even after appropriate investigation, are extremely common amongst UK primary care attenders but knowledge about their outcome is limited. Aim: In a cohort of adult primary care attenders with UPS, this study aims to: 1) Investigate the outcome, in terms of the presence of UPS at six months follow-up and 2) Identify prognostic factors associated with somatic symptom severity, quality of life, anxiety, depression and health care use at six months follow-up. Methods: Screening: Consecutive adults attending nine general practices completed a screening questionnaire to identify those with UPS. Cohort study: Eligible participants completed the baseline questionnaire that enquired about somatic symptoms, quality of life, psychological well-being and past health and social history, and were followed-up after six months. Results: Screening: Questionnaires were completed by 73% (2,826/3,896) of eligible attenders. Over two-thirds were female, median age was 42 years (IQR 30, 55) and median symptom severity score, based on the PHQ-15 was 7 (IQR 4, 11). Most (2,425/2,826 (86%)) had at least one UPS and around half (1,393/2,826 (49%)) had symptoms that were all unexplained (no explanation or diagnosis for any of their symptoms). Just under half (1,248/2,826, (44%)), had an explanation for their symptoms that included functional diagnoses (100/2,826 (4%)), psychological explanations (187/2,826 (7%)), or physical explanations or diagnoses (921/2,826 (33%)). Cohort study: The cohort included 294 participants, were largely female (231/294 (79%)), with a median age of 44 years (IQR 32, 57)) and diverse ethnicity (43% white British). At baseline, the cohort had a high level of morbidity, with moderately severe somatic symptoms (11.5 SD 4.9). Most reported experiencing their symptoms for longer than a year. A third had clinically significant comorbid depression and anxiety. Outcome: There was 245/294 (83%) followed-up at six months; mean PHQ-15 score was 10.5 (SD 5.3). Over a half reported unexplained symptoms (135/245 (55%)), just under half (103/245 (42%)) reported symptoms were still under investigation and only 26/245 (11%) reported that their symptoms had resolved. Options were not mutually exclusive and participants could choose more than one. The predictors of more severe somatic symptoms at follow-up were being female (B=1.31, 95% CI 0.12 to 2.50), higher somatic symptom severity (B=0.53, 95% CI 0.42 to 0.64), experience of childhood physical abuse (B=1.86 95% CI 0.27 to 3.45), perception of poor financial well-being (B=1.90, 95% CI 0.89 to 2.91) and lower physical functioning at baseline (B=-0.10, 95% CI -0.15 to -0.04). Conclusion: Most people with UPS and high symptom severity are unlikely to improve over six months. Historical and current difficulties are associated with higher somatic symptom severity at follow-up. Future work should determine whether these findings are maintained over longer periods. The value of developing prognostic prediction models based on factors identified in this study should be explored.

Background: In 1989 the Saint Vincent declaration set out to reduce the risk for pregnant women with diabetes to those without diabetes. A number of studies, since the declaration found that foetal and neonatal adverse pregnancy outcomes in women with type 1 diabetes remained increased. Literature for women with type 2 diabetes and maternal complications is limited. My aim is to assess whether women with type 1 and 2 diabetes remain at increased risk of pregnancy complications. Methods: Using a primary care database; THIN, I investigated the prevalence of: pregestational diabetes in pregnancy and pregnancy complications. Finally, I examined the risk of pregnancy complications for women with diabetes in pregnancy compared to women without diabetes in pregnancy. Results: The prevalence of type 1 diabetes pregnancy increased from 1.58 to 4.34 per 1,000 pregnancies between 1995 and 2012. The prevalence of type 2 diabetes in pregnancy steadily increased from 2.38 to 4.83 per 1,000 pregnancies between 1995 and 2008; then increased more rapidly until the end of the study period to 10.37 per 1,000 pregnancies in 2012. Women with type 1 diabetes remained at increased risk of caesarean section (RR 2.41 (2.13, 2.72)) and major congenital malformations (RR 2.29 (1.53, 4.85)) compared to women without diabetes after adjusting for maternal characteristics. Women with type 2 diabetes remained at increased risk of caesarean section (RR 1.58 (1.42, 1.75) and perinatal death (RR 2.72 (1.53, 4.85)) when compared to women without diabetes after adjusting for maternal characteristics. Conclusion: Women with type 1 and type 2 diabetes remained at increased risk of experiencing pregnancy complications. There is still substantial work to be done to reduce the adverse outcomes experienced by women with diabetes in pregnancy and meet the recommendations set out in the Saint Vincent declaration nearly thirty years ago.

One challenging prescribing decision in type 2 diabetes mellitus (T2DM) is when clinicians must choose between sitagliptin and sulphonylureas as add-on to metformin based on effectiveness. Evidence on effectiveness of sitagliptin versus sulphonylureas as add-on to metformin was therefore systematically searched and revealed no study evaluating “real-world” comparative effectiveness of these treatments, particularly in older, more comorbid individuals. To address this gap, The Health Improvement Network, UK primary care database was used to extract a cohort of 26,844 individuals with T2DM prescribed these treatments and four cohort studies were undertaken to evaluate their comparative effectiveness. The first two studies demonstrated no difference in HbA1c reduction, approximately 12 months after initiating either treatment as add-on to metformin, however a significant comparative weight reduction with sitagliptin in those aged 18-75 (-2.26kg 95%CI -2.48 to -2.04) and ≥75 (-1.31kg 95%CI -1.96 to -0.66) was found. Two further studies revealed individuals prescribed sitagliptin were 11% more likely to record an undesirable HbA1c >58mmol/mol (Hazard Ratio 1.11 95%CI 1.06-1.16), however nearly twice as likely to record an anti-diabetic treatment change (HR 1.98 95%CI 1.86-2.10) compared to sulphonylurea initiators. This analysis on treatment change also highlighted an underlying inertia in both groups, as 66.4% of those prescribed sitagliptin and 83.7% prescribed sulphonylureas had no treatment change introduced despite recording a HbA1c >58 mmol/mol. This thesis provides “real-world” evidence that both sitagliptin and sulphonylureas are equally effective in lowering HbA1c and achieving glycaemic targets in a population that includes individuals aged ≥75 and with significant comorbidity. Sitagliptin is preferable for weight reduction. There is however, a substantial inertia in changing treatment when targets are not met, which is greater among sulphonylurea initiators. There remains a need to eliminate barriers preventing clinicians changing treatment when these two add-on medications prove inadequate, and further evaluate their longer-term comparative effectiveness.

Multiple imputation (MI) is increasingly used for handling missing data in medical research. The standard implementation of MI assumes that data are missing at random (MAR). However, under missing not at random (MNAR) mechanisms, standard MI might not be satisfactory. When there are external data sources providing population-level information about the incomplete variables, it is desirable to utilise such information in MI. This thesis aims to explore how knowledge about the incomplete covariate's population marginal distribution from an external dataset can be used to improve standard MI under MNAR mechanisms. Two univariate MI methods are proposed for an incomplete binary/categorical covariate to anchor inference to the population: weighted MI and calibrated-δ adjustment MI. Chapter 3 demonstrates how, in weighted MI, the incomplete covariate's population distribution can be incorporated as probability weights in the imputation process to closely match the post-imputation distribution to the population level. Results from analytic and simulation studies of a 2x2 contingency table show that weighted MI can produce more accurate inferences under two general MNAR mechanisms. Weighted MI is also integrated into the multivariate imputation by chained equations (MICE) algorithm for imputing several incomplete covariates, accounting for their population marginal distributions from external data. Chapter 4 develops and evaluates calibrated-δ adjustment MI, which incorporates the incomplete covariate's population distribution as a δ adjustment in the imputation model’s intercept. In a 2x2 contingency table, it is shown analytically and via simulation that appropriately adjusting the imputation model's intercept fully corrects bias when the incomplete covariate is MNAR dependent on its values and the (complete) outcome. An adaptation of the method in the MICE algorithm for multivariate imputation is also explored. Chapter 5 investigates another univariate missing data setting, with a continuous outcome. Under the above MNAR mechanism, the presence of a second sensitivity parameter for the covariate – outcome association in the imputation model is introduced, rendering the calibrated-δ intercept adjustment insufficient. The sensitivity analysis then involves eliciting values of the second sensitivity parameter and deriving the calibrated-δ adjustment in the intercept. Chapter 6 presents two case studies using electronic health records to illustrate the application of the proposed population-calibrated MI methods.

Introduction For some people, ageing is associated with the experience of increased co-morbidity, functional impairment, poor resilience and heightened vulnerability to external stressors, resulting in reduced lifespan as well as health-span. This frailty phenomenon poses challenges to health care systems in the form of increased patient complexity and resource utilisation. The acute care setting, characterised by time-pressure and high patient turn-over, is under strain and struggles to recognise and subsequently reliably intervene, to prevent, reverse or halt the decline of this vulnerable cohort. Methods This mixed-methods study probes existing evidence and ‘real-world’ processes with a systematic review of frailty assessments developed or validated in the acute care setting and a survey of contemporaneous clinical practice in London Acute Medical Units. Content validation and understanding of contextual factors for ideal frailty assessment in acute care is explored using Delphi consensus and Focus Group methodology respectively. The resultant model is developed on existing retrospective national Hospital Episode Statistics data, and prospectively tested on observational data in a local Acute Medical Unit setting. Results Existing frailty scores are preponderantly biophysical in nature, and have poor predictive power for adverse outcomes in the acute care setting. In clinical practice, single-dimension assessment tools predominate. Frailty syndromes and previous high resource utilisation in the form of a simple, clinically relevant tool useful to the multidisciplinary team gain consensus as optimal assessment for the setting. Retrospective testing of the frailty model displays moderate predictive powers for adverse events (inpatient mortality, emergency readmission and institutionalisation) and prospective testing provides concurrent (Frailty Index, Age, Co-Morbidity) and comparative predictive validity (Frailty Index, Co-Morbidity, admission National Early Warning Score) with existing risk stratification models in this setting. Conclusions A risk prediction model based on frailty syndromes and previous high resource utilisation is a valid, feasible and useful for the acute care setting.

Background: It is well established that there are significant disparities in health­care delivery and utilisation. These differences cannot be explained by differences in disease prevalence, evidenced based medicine or patient preference. Similarly, it is acknowledged that there is significant variation in healthcare outcomes for which our understanding is limited. This thesis details the variation in structures and processes of care between, acute NHS hospital trusts in England and studies the association of these factors with healthcare outcomes. Methods: Publicly available data sets largely provided by the Health Social Care Information Centre, were examined to describe the variation in hospital trust level structures and processes. Patient level data on a variety of diagnostic or procedural cohorts was obtained from the Hospital Episode Statistics administrative database. The cohorts studied included all hospital admissions, emergency admissions or pro­cedures, general surgical emergencies, ruptured abdominal aortic aneurysm (AAA) repair and all A A A repairs. Outcomes included mortality, complications and failure to rescue from complications. Resource provision or utilisation between risk adjusted outliers for a given outcome were compared. Multilevel hierarchical binary logistic regression models were used to identify the associations of structure and processes of care with risk adjusted outcomes. Results: There is significant variation in structures and processes of care between hospital trusts. Trusts with the most favourable outcomes had significantly greater levels of staffing, critical care and radiodiagnostic utilisation. Greater medical and nurse staffing levels, radiodiagnostic utilisation, research activity and procedural volume were associated with better healthcare outcomes for a variety of cohorts. Conclusions: This thesis demonstrates that hospital trust level variation in struc­tures and processes of care is associated with the outcome for a broad range of patients. Further research should examine the underlying mechanisms for these associations and, the clinical and cost-effectiveness of resource modification at the trust level.

This thesis assesses the potential public health significance of, sanitation quality and coverage by using microbiological indicator and secondary health outcome data. Sanitation was categorised using The Millennium Development Goal (MDG) definition of improved (private pit latrine with a slab or better) and unimproved (toilets connected to open, pit latrine without a slab, hanging toilet, shared toilets). A spot check of sanitation facilities was conducted in 460 target houses and 1,784 neighbouring houses. Faecal contamination of the household environment was assessed by looking for evidence of contamination with faecal coliforms on children’s hands and on ‘sentinel’ toys (standardised toy balls provided by the study). An analysis of secondary data was conducted on sanitation and reported diarrhoea among children < 5 years of age that had been collected as part of an impact evaluation. Households with private improved sanitation had lower faecal coliform contamination than households with unimproved sanitation [difference in means: -0.31 log10 colony forming units (CFU)/toy ball; 95% CI: -0.61, -0.01]. Access to 100% private improved sanitation coverage in the neighbourhood was associated with a small but statistically insignificant difference in contamination of sentinel toys (difference in means: -0.09 log10 CFU/toy; 95% CI: -0.56, 0.38). Other household sanitary practices such as cleanliness of latrine, wastewater disposal and disposal of animal faeces were important and statistically significant (P value ≤0.06) determinants of household faecal contamination. Children from households with access to private improved sanitation had a similar prevalence of diarrhoea to those with unimproved sanitation (Prevalence Ratio [PR] =1.00; 95% CI: 0.89, 1.13). Children from households with appropriate solid waste disposal systems had lower prevalence of diarrhoea compared to those without (PR=0.78; 95% CI: 0.65, 0.95). 4 Improved sanitation infrastructure quality and coverage may have limited roles in preventing transmission of diarrhoea causing enteric pathogens in the study context in which diarrhoea is endemic. Although in this study, private use and cleanliness of latrine were associated reduction in faecal contamination, but these factors were not associated with reduced diarrhoea prevalence. This may be because, firstly data were collected from slightly different contexts and time, secondly indicator organisms are only weakly associated presence of enteric pathogens and thirdly the population in this study context may have developed some degree of immunity to common circulating pathogens. Findings from this observational studies presented in this thesis adds to the evidence base, which do not support the inclusion of shared facilities as improved. There may be other more important source of children’s exposure to enteric pathogens that onsite sanitation access cannot prevent. Other sanitation related factors like maintenance of sanitation facility, use by all household members including children and faecal sludge management should be considered while defining improved sanitation for international monitoring. We also need to increase research efforts to integrate sanitation, water quality, handwashing and nutritional interventions and to understand better ways to monitor the impact of these interventions.

Background Digital interventions (DIs) provide effective and potentially cost-effective models for improving health behaviour outcomes as they deliver health information and services that are widely disseminated, confidential, and can be tailored to needs of the individual user. DIs have been used successfully for different health behaviours such as smoking cessation and increasing physical activity. However, their effectiveness is limited by low usage rates, with non-engagement a major challenge. Engagement can potentially be optimised through the use of context-specific prompts such as emails and text messages. Aim To inform, develop and assess the potential of using email and text message as prompts to optimise users’ engagement with a DI called HeLP-Diabetes (Healthy Living for People with Diabetes). Context HeLP-Diabetes is a DI targeting self-management for people living with Type 2 diabetes. It aims to provide its users with the tools and support to improve and maintain their health and well-being. Methods The following series of studies with their corresponding objectives were conducted: • A systematic review and meta-analyses to determine the range and effectiveness of technology based engagement prompts, and to identify characteristics of prompts to test in subsequent studies. • Developing, piloting and testing prompts, within the context of HeLP-Diabetes using the Medical Research Council guidance for complex intervention, to identify the methodological and practical challenges of developing engagement prompts and integrating them with HeLP-Diabetes. • A mixed method study to identify the characteristics of prompts, specifically the content and delivery mode, that have the potential to promote engagement with HeLP-Diabetes. The mixed method study used quantitative data from HeLP-Diabetes usage data, and qualitative data from think aloud interviews exploring prompt content with a selected group of users. • Two consecutive randomised controlled trials comparing different prompt content and delivery modes in order to assess the potential impact of prompts on engagement with HeLP-Diabetes and to test the procedures of the pilot trial to identify any potential challenges for future trials.

BACKGROUND: Misuse of drugs is a public health problem which can lead to poor health outcomes. Drug use during pregnancy could potentially harm the unborn baby. Pregnancy usually triggers women to visit their general practitioner (GP) which may provide an opportunity for drug use to be raised and recorded. To date, there are no UK studies with large sample sizes to estimate the burden of drug use during pregnancy. Therefore, my aim was to describe and understand drug use and opioid substitution treatment in and around pregnancy using electronic health records. METHODS: Using a mixed methods design, I firstly, utilized The Health Improvement Network (THIN) to estimate GP recording rates of individuals who use drugs and/or are prescribed opioid substitution treatment in the general population, of women in and around pregnancy and infants with neonatal abstinence syndrome (NAS). Next, I compared rates with national surveys and hospital birth data. Finally, I conducted qualitative interviews to gain GPs’ perspectives regarding their decisions about recording drug-use. RESULTS: GP recording trends for the general population were in keeping with national surveys, but with lower rates. Recording was relatively low in and around pregnancy. GP recording of NAS was similar to hospital data, however rates were lower. Finally, qualitative interview analysis identified that influences on recording drug use were complex and related to pressures at the individual as well as organisational (general practices, Clinical Commissioning Groups) and governmental levels in the shape of government policies. CONCLUSIONS: In conclusion, evidence from the thesis supports the use of THIN as a suitable tool for monitoring trends but not rates of problem drug use in the general population. Electronic primary health records could potentially be used to monitor the impact of problematic drug use in and around pregnancy. The thesis also supports utilising THIN for researching drug use and opioid substitution treatment in the general population.