Health Economists’ Study Group (HESG) Summer Meeting, Aberdeen, 28 – 30 June 2017. Draft Programme List of accepted abstracts Paper Title No 1 The Preference Structure of Dentists Job Search and Choice Abstract Author(s) Heath workforce supply decisions involve multiple stages, including searching for a job and then evaluating and choosing whether to accept a job offer. Current practice when modelling supply decisions focuses on the final evaluation or job offer stage but this potentially misses important information that can be learnt at the search stage. We address this methodological issue by conceptualising the problem as separate decision stages and provide a framework to operationalise and test the conceptualisation. Two novel Discrete Choice Experiments were presented to a sample of 275 Australian Dentists: the first mimicked an online job search site in which respondents decide which job postings they would apply for and the second presented respondents with a job offers which they could accept/reject. Data were analysed applying data pooling and using latent class choice models. Dentists appear to make similar trade-offs across job characteristics in the job search and offer evaluation stages but the propensity to select a job at the search stage differs to the offer evaluation stage for the job type (Employee, Associate Partner, Owner, Locum or Public jobs) differs. As the dentist job market becomes more corporatized and dentists become more like employees and less like small business owners, job switching/retention will become more of a focus in the marketplace. Understanding the different stages of supply decisions is important in developing models, such as job structures or incentive structures, to improve the job attraction and retention of dentists. Huynh, E., Lancsar, E., Swait, J. 3 Too broad to be sensitive? An exploration of the responsiveness of the ICECAP-O capability measure compared to the EQ-5D-3L to the change of clinical and QoL aspects in people with Parkinson's 4 The impact of informal caregiving on mortality and health: evidence from the British Household Panel Survey Background The vision of the ICECAP-O measure is to enable economic evaluations to incorporate wellbeing in a broader sense. This scope, however, may generate concerns over its sensitivity to capture specific changes in a narrower context focused on health. Objective: This study aims to assess the responsiveness of the ICECAP-O to the clinical and quality of life (QoL) changes in people with Parkinson's over two years. Methods: The ICECAP-O, along with EQ-5D-3L, PDQ-39 (a Parkinson's specific QoL measure), demographics and clinical characteristics were collected from 1,023 participants with various severity of Parkinson's at intervals two years apart in the PD MED trials. The Hoehn and Yahr (H&Y) motor measure, PDQ-39 summary score and its eight dimension scores were used as external indicators to classify participants to five change groups: largely/slightly improved/deteriorated, and no change. Correlation coefficients, and effect size (ES) statistics of the change were used to evaluate the responsiveness of ICECAP-O and EQ-5D-3L. Results: The ICECAP-O change score was slightly more strongly correlated with the PDQ-39 (r = 0.526) than EQ-5D-3L (with PDQ-39: r = 0.483), which is also supported by the ES statistics. The ICECAP-O was less responsive to the large deterioration of H&Y than EQ-5D-3L. Overall, there was no statistically significant difference in the ES statistics between ICECAP-O and EQ-5D to any of the external indicators. Conclusion: The broad scope of ICECAP-O could provide rich information on the capability wellbeing without compromising its sensitivity to the clinical and specific QoL change in the Parkinson's population. This paper estimates mortality effects of informal care provision in England, using a combination of genetic matching and survival modelling, applied to data from the British Household Panel Survey 1991-2010. A stream of recent publications have found lower mortality rates among informal carers, contrary to the general understanding that caregiving places strain on physical and mental health and would therefor lead to increased mortality. A key issue often not well accounted for in the literature is self-selection into caregiving by relatively healthy individuals. To better account for the self-selection and other unobservable differences between caregivers and non-caregivers, we use genetic matching, following Diamond and Sekhon (2013) combined with survival models. The genetic matching resulted in an improvement on the balance on key variables between the treated and control group, compared to conventional propensity score matching. We however still find a positive effect on longevity (i.e. lower mortality rate) for individuals providing informal care and this applies to both co-residential and non co-residential care. The matching process however deals well with possible self-selection, as evidenced by a placebo test using health outcomes. Unmatched samples identify an improvement in health for caregivers, whereas using the matched samples we find a negative, however insignificant effect. Xin, Y., Lewsey, J., McIntosh, E. Fernandez, J-L., Zigante, V. 7 The Economics of the Positive List for Innovative and Costly Drugs in Hospitals 8 Socioeconomic Status and Body Mass Index among South African Children and Adolescents: A Longitudinal Study In French hospitals, the expensive and innovative pharmaceuticals are funded through a system known as the outside GHS list. This funding scheme enables hospitals to be fully reimbursed by the state for their purchase of the listed pharmaceuticals, in order to guarantee equity of patient access to pharmaceutical innovation. However, this 'exceptional' funding scheme has substantially weighted on the health care expenditures, with a fast-growing list in the first years of its implementation. As the recent years have shown a willingness to make the management of the list more dynamic - thus entailing more frequent delistings - this paper seeks to study the effect of a drug delisting from the outside GHS list, and subsequent integration in the general funding scheme, on the quantities prescribed. The focus is on the anticancer drugs delisted in 2010 after entry of generic competitors. Two administrative French hospital-level datasets on drug consumption were combined to trace the evolution of hospital prescriptions from 2008 to 2012. The estimation strategy chosen is a Differences-in- Differences that makes use of the existence of several delisting waves over the period. Results suggest that the 2010 delisting has a negative impact on the quantity per pharmaceutical, as compared to the still listed pharmaceuticals. However, when looking at the consumption of all the therapeutically-equivalent pharmaceuticals per hospital, we find no significant impact of the delisting on the average quantity. A possible interpretation of these two results is that delisting enhances substitutions across pharmaceuticals of the same therapeutic class. The immediate and long-term health consequences of obesity as well as the contribution of poor socio-economic status is now being recognized as an important public health concern. In crosssectional research poor household socio-economic status have been associated with obesity, however little is known of the influence in developing country's contexts especially from a longitudinal perspective. Thus, in this work, we exploit the panel nature of a nationally representative South African dataset to assess the trajectories in childhood and adolescent weight from a life course perspective as well as the contributions of socio-economic status to gradients in body mass index (BMI) of children up to the age of 18 years by gender. The study used South Africa's National Income Dynamic Survey waves 1 and 4 for the years of 2008 and 2014/15 respectively. Following individuals over time and using a panel fixed effects model, we account for time-varying heterogeneity. Initial results indicated an inverse relationship between household income which was used as proxy for socioeconomic status and body mass index. Rachet Jacquet, L., Toulemon, L. Mirelman, AJ. 9 Valuing shared decision-making in end-stage knee osteoarthritis: a discrete choice experiment 10 Probabilistic sensitivity in health economic models; how many simulations should we run? Background: Patient-centred care and shared decision-making (SDM) are priorities for health systems, and will require resource commitments. However, cost-effectiveness analysis (CEA) using quality-adjusted life-years (QALYs) will likely fail to capture all benefits of SDM. Objective: To elicit societal preferences for the trade-off between SDM and health outcomes, to allow the value of SDM to be incorporated in a CEA. Methods: A discrete choice experiment (DCE) in a sample of Canadians aged 60 and above. Participants were asked to imagine they had knee osteoarthritis (OA) and were then presented with 10 hypothetical choices of consultations, each with two different specialists, to support treatment decision-making. Consultations were described using three attributes: waiting time, level of SDM experienced, and the chance of improvement in pain or discomfort one-year following the consultation. A condition logit model was estimated to determine the marginal rates of substitution. Results: A total of 115 individuals completed the survey. All attributes had a statistically significant impact on the choice, and coefficient signs met a priori expectations. Having a specialist who made every effort (compared to no effort) to engage in SDM was equivalent to an 8% increase in the chance of improving to have no pain or discomfort. Conclusions: Previous studies have shown that individuals value SDM. This study provides evidence that people are willing to forego potential health improvements for greater SDM, and discusses how this evidence can be incorporated within a CEA which uses QALYs at the measure of benefit. Introduction: Probabilistic sensitivity analysis (PSA) addresses parameter uncertainty inherent in a decision problem. Literature on the number of simulations required suggests a "sufficient number", or until "convergence" which is seldom defined. In this study, we aim to define convergence, using empirical data to propose simulation numbers for different outcomes. Methods: Thirty models were used to produce 250,000 individual simulations each, assumed to be reflective of the 'true' model result. Random samples were drawn to simulate 1,000 PSAs of 25,000 simulations. We identified the number of simulations required for the mean values to come within percentages or fixed amounts of 'true' results (arithmetic and jackknifed), and the edges of the distribution. Results: For the mean results, costs and QALYs were within £1,000 and 0.05 of the mean by 1,700 simulations. The incremental cost-effectiveness ratio took the longest to stabilise, requiring a mean of over 6,000 simulations to stabilise to within 1%. The edges of the distributions took longer to stabilise in all cases, not stabilising with high precision until over 10,000 simulations.Conclusions: Health economic models have such complexity that low numbers of samples cannot guarantee an accurate representation. Ideally convergence would be assessed for each model (and we provide the tools to do this). However, as a general rule we recommend 10,000 simulations be used for decision making. This gives a high degree of precision on the mean values and acceptable accuracy at the edges of the Trenaman, L.Bryan, S., Payne, K., Stacey, D., Bansback, N. Hatswell, A., Bullement, A., Paulden, M. et al distribution, without excessive computational burden. Conclusions also held using the jackknifed mean. 11 Preferences of patients, health professionals and the general public for centralising specialist cancer surgery services: a discrete choice experiment Background: The centralisations of specialist cancer surgical services across London Cancer and Greater Manchester Cancer aim to improve patient outcomes, but are likely to increase travel demands on patients and families. The aim of this study was to evaluate patients', health professionals', and the general public's preferences for the changes associated with centralising specialist cancer surgery services in England. Methods: We conducted a discrete choice experiment including the following attributes: patients' travel time to hospital; risk of serious complications; risk of death; annual number of operations at the specialist centre; access to a specialist multidisciplinary team (MDT) to determine treatment; and, availability of specialist surgeon cover 24/7 after the operation. Data were collected by postal questionnaires and online surveys. Results: We obtained 444 responses (206 from cancer patients, 111 from health professionals involved in the treatment of cancer patients and 127 from members of the public). Preferences were influenced by risk of complications, risk of death and access to specialist MDT. Travel time was the least important factor, with participants willing to travel 75 minutes longer to reduce their risk of complications by 1%, and over 5 hours longer to reduce risk of death by 1%. Findings were similar across the three groups. Implications: Planners should measure the impact of the reorganisations on the factors identified as important in this study. They should disseminate information about these factors to stakeholders when deciding whether or not and how to centralise services. Vallejo-Torres, L., Melnychuk, M., Ramsay, A., Vindrola, C., Boaden,R.,Hunter, R., Clarke, C., Wood, V., Darley, S., Fulop, N., Morris, S. 14 Framework for a cross-sectoral and cross-temporal economic evaluation of Public Health interventions The complex nature of public health interventions raises methodological and technical issues in their evaluation. Focusing the analysis only on health costs and consequences potentially underestimates the full impact of the intervention: it may be necessary to assess a broader range of non-health costs and effects falling on other sectors (e.g. education, criminal justice, employment), with the impacts often far in the future. Moreover, additional concerns about equity become critically important. Our aim is to define appropriate methods for the economic evaluation of public health interventions. We propose a framework which extends a 'traditional' cost-effectiveness analysis to include inequality analysis, evaluation across sectors and over time. To capture cross-sectoral impacts, outcomes Ramponi, F., Walker, S., Richardson, G., Kanaan, M. 17 Unintended consequences of professional regulation? The case of the medical workforce in England falling on different sectors are assessed and trade-offs among them are estimated. A compensation scheme across decision makers in order to address resource allocation problems, generated when costs fall on a sector and benefits on another is proposed. To reflect intertemporal issues, extrapolation techniques are employed to estimate long-term consequences and the impact of budgetary policies is assessed. Indeed, rigidity of budget constraints over time may limit the scope to implement interventions which would be cost-effective if decision makers had more control over inter temporal budgets. To show the important implications of our work for future evaluations, the framework will be explored using a hypothetical brief alcohol intervention. Regulation can have positive or negative effects on labour markets depending on perceived costs Gutacker, N., Bojke, and benefits. The UK General Medical Council introduced medical revalidation in December 2012, C., Bloor, K., requiring all licensed doctors to demonstrate every 5 years that they are up to date and fit to Walshe, K. practise. It is intended to incentivise continuous investment in human capital but is time consuming and may limit clinical autonomy, thereby imposing costs on doctors. Preliminary qualitative evidence suggests that revalidation may have driven doctors out of the profession, depriving the NHS of essential labour inputs. This study analysed activity data from Hospital Episode Statistics for all consultants in English hospitals from April 2009 to March 2016 (n=19,334). We estimate semiparametric Cox models to test whether consultants became more likely to stop clinical activity after the introduction of mandatory revalidation. Crucially, consultants underwent revalidation at different times and we differentiate periods when they were a) not subject to revalidation, b) preparing for their first or subsequent (if previously deferred) revalidation meeting, and c) after they had received a positive recommendation. After adjusting for demographic factors and specialty, we find that consultants who have not yet revalidated are at an increased hazard of exit (HR: 2.38, 95% CI: 2.16 to 2.61) compared with pre-policy levels. The risk of exit remains elevated after a positive recommendation (HR: 1.93, 95% CI: 1.73 to 2.15) but is statistically significantly lower (p<0.001), suggesting that revalidation has had a causal negative effect on retention over and above other contemporaneous influences. 19 Modelling the cost-effectiveness of diagnostic tests in the absence of a gold standard: a case study of cardiac surgery clotting tests 22 Clinical decision making in augmentative and assistive alternative communication technology for children: A conjoint analysis approach Introduction: Assessing the cost-effectiveness of diagnostic tests can often be more difficult than evaluating pharmaceuticals due to 'imperfect' or even non-existent gold standards. Without reliable gold standards, it can be unclear how to incorporate diagnostic accuracy (sensitivities cannot be used), which could lead to inaccurate cost-effectiveness estimates. For example, there is no gold standard test to detect coagulopathy (clotting abnormalities). Patients undergoing cardiac surgery are at risk of coagulopathy, and early diagnosis of specific clotting defects would enable more timely, targeted treatment. Aims: To explore if patient level data can provide a solution to diagnostic costeffectiveness analyses without a robust gold standard using an exemplar of new point-of-care (POC) testing of clotting abnormalities in cardiac surgery. Methods: Decision-analytic models were developed to simulate the costs and effects of introducing POC testing compared to current practice. In the POC test arm, patients with a positive test result were assumed to receive prophylactic treatment for coagulopathic bleeding. Many model parameters were estimated from individual patient data from an observational study of POC testing in cardiac surgery patients, including the failure rate of prophylactic treatment, inpatient costs and survival. Incremental cost-effectiveness ratios were calculated, and deterministic and probabilistic sensitivity analyses conducted. Results: There was very little difference in costs/ effects between POC testing and current practice and great uncertainty around cost-effectiveness results, generating no evidence to support testing. Conclusions: Individual patient data was confirmed as a potential solution to address a lack of a gold standard when assessing the cost-effectiveness of diagnostic tests. Background: Children with significant speech and language difficulties related to a range of conditions may rely on communication aid technology to support language acquisition and communication. Augmentative and alternative communication (AAC) systems often use graphic symbols to represent language for pre-literate children. The diverse characteristics of these children and the proliferation of technology means selecting the optimum system for a child is challenging. AAC device prescription has long lasting consequences for children and their development but little is known about the decision making of clinicians in this context. Aim: To conduct a preference elicitation exercise with prescribing clinicians to determine what aspects of children and of devices are important in decision making. Methods and Results: A list of child and device attributes (approximately 20 each) were generated following a systematic review of the literature, focus groups with 30 AAC clinical specialist stakeholders and discussions with 20 AAC field experts. For the child these included: level of learning ability and mobility, and for the device: type of vocabulary organisation and portability. Due to the large number of attributes we decided a two-part bestworst scaling (BWS) object case survey was the most appropriate data-gathering instrument. The Stokes, E., Wordsworth, S., Leal, J., Harris, J., Mumford, A., Reeves, B., Murphy, G. Webb, E., Meads, D., Lynch, Y., Randall, N., Judge, S., Goldbart, J., Meredith, S., Moulam, L., Hess, S., Murray, J. survey was conducted online with clinicians using a D-efficient fractional factorial design. We describe the results and how these feed into a future DCE combining both attributes of children and devices. This will both examine the trade-offs clinicians make and crucially, given the large heterogeneity in the patient population, the interactions between child-related and device-related factors. 23 Impact of a tax on sugar-sweetened beverages on the demand for nonalcoholic drinks among households with children in the UK from stated and revealed preferences Fiscal policies have surfaced as a policy option to tackle the growing burden of obesity, diabetes, cardiovascular diseases and several cancers associated with unhealthy diets. Critical in setting such policies is the likely own- and cross-price elasticity of demand that determines overall impact. Price elasticity is typically assessed using observational studies of expenditures and market driven price changes, and concentrates on own-price elasticity. The limitations to such studies are well known. An alternative is to assess stated preferences related to price changes introduced explicitly as a fiscal policy. This study conducts a Discrete Choice Experiment (DCE) to analyse the likely demand response to a hypothetical tax on sugar-sweetened beverages (SSBs) in the UK. Using a controlled design, the study focuses on testing the effects of a) market versus policy driven price changes, b) signalling the tax as a health-related measure and b) substitution between categories of drinks and branded and non-branded products SSB prices change. The DCE was conducted as an online survey among a randomly selected sample (n=600) of UK households (who have children and have purchased at least 2 litres of SSBs in the previous month) from a large nationally representative panel (Kantar Worldpanel). Using this panel allows us to assess the stated preference against actual purchases of foods and beverages via home-scan technology, understand preferences of different types of consumers, control for respondents usual behavior as well as compare the responsiveness to changes in prices estimated from both types of data. Quaife, M., Lagarde, M., Smith, R.,Cornelsen, L., 26 Heterogeneous treatment effects in mental health: a Malawian case study 27 The effects of retirement on health behaviour, health and wellbeing: Evidence from the English Longitudinal Study of Ageing Background: Mental health is widely ignored in research and policy for lower income countries (LICs) and yet one of the main contributor to the burden of disease in LICs. We have hardly any knowledge about financial incentives and their general and distributional effects on mental health of the general adult population living in LICs. Aim: To explain the causal effect of a conditional cash transfer (CCT) on mental health in a randomized control trial with adults living in rural Malawi. Methods: We use data on 937 individuals from two waves of the Malawian Longitudinal Study of Family and Health to analyse the causal effect of the CCT on general mental health measured by the SF12 scale. We use principal component analysis to compute item weights for the Malawian population and test its reliability. Ordinary least square regression and bootstrapped quantile regression are applied to identify the effect of the CCT programme on SF12. Results: Preliminary findings suggest significant positive effects of CCT receipt along the distribution of the SF12 measure, decreasing in magnitude with increasing percentiles q0.1(2.602) to q0.75(0.262). The OLS estimates shows that cash transfer receipt has a positive average treatment effect on the treated on mental health with size 1.026. Implications: Financial incentives offered to individuals living in a rural LICsetting cause on average better mental health however the effect is not constant over the full distribution. The effect-size varies and is most effective and strongest for the lowest percentile(s) as predicted by theoretical models. Retirement from paid work is a major transitional point in later life for many people, and can have a large impact on lifestyle choices relating to health behaviour and subsequent health and well-being. Existing studies show that changes in health behaviour in later life are associated with changes made by their partner. Another body of evidence has assessed the causal effects of retirement on health behaviour, health and well-being. This paper aims to bring these two sources of evidence together to estimate new impacts of both own and partner's retirement on health behaviour, health and wellbeing. The English Longitudinal Study of Ageing (ELSA), which is a bi-annual survey from 2002/03 to 2012/13 (wave 1 - 6) are utilised. We restrict our sample to persons who retired immediately from paid work and estimate fixed effects regression models. Preliminary results suggest that the retirement has a positive association with the level of one's physical activity while other measures of health, well-being or health behaviour do not show significant correlation to one's retirement status. Surprisingly, the retirement of one's partner does not seem to influence his/her health or health behaviour. Further analysis using fixed IV effects will be reported, using state pension age as an instrumental variable (IV) for the reported retirement status. Ohrnberger, J., Fichera, E., Sutton, M., Anselmi, L. Leckcivilize, A., McNamee, P. 28 Eliciting public preferences for access to cognitive behavioural therapy by employment status 30 The decision to invest in health: the role of future health and wealth expectations The Department for Work and Pensions (DWP) is considering an intervention that will identify jobseeker benefit claimants who are at high risk of long-term unemployment due to mental health difficulties, and will refer them to access cognitive behavioural therapy (CBT) outside of NHS provision, funded by the DWP. The aim of this study is to examine public preferences regarding such an intervention. Since the target group is facing multiple difficulties and arguably has higher needs, the public may support the intervention; or, the public may find it unfair that the unemployed are 'fast-tracked'. There are two research questions. First, are public preferences inequality-averse with respect to access to CBT? For example, if people are inequality-averse, then they may prefer a longer total wait that is evenly distributed across people over a shorter total wait that is unevenly distributed across people. Second, are the relevant public preferences symmetric (or satisfy anonymity) across employment groups? For example, people may want to give higher priority to those waiting for CBT if they are also unemployed, or if they are also holding down a job. The paper reports on an online survey of the UK voting public (n=1000), using binary choice questions from a societal perspective. Preliminary results indicate that around three quarters of respondents are inequality averse; and slightly over half are asymmetric, in favour of the unemployed. However, when the two considerations are combined, the modal preference appears to be inequality averse and symmetric. Further analyses are ongoing. Previous studies that have explored the role of expectations on health investments have predominately focused on longevity as a predictor; individuals with high expected mortality rates, who are less likely to benefit from future consumption, are less likely to save. However, the Health Capital model suggests that future wealth might also contribute by influencing people's expectations on the return of health investments. Healthy days, the product of health investments, are not inherently demanded; rather, healthy days provide an individual with the ability to produce income, which is used to increase utility, and the ability to enjoy leisure time, which may be the product of past financial investments. As such, individuals who expect to enjoy wealth in the future may be more motivated to invest in their health in the present, in order to maximize their lifetime utility, whilst individuals who expect a decline in wealth in the future will be less motivated to invest. This paper reports preliminary analyses that explores the role of future health and wealth expectations on the decision to invest in health, using data from The English Longitudinal Study of Ageing. Longevity and future financial expectations are used to predict smoking cessation, moderate or no alcohol consumption, and physical activity participation. Panel data analysis controls for unobserved endogeneity in the form of personality traits and genetic determinants. Findings from this study will contribute to disentangling the relationship between health investments, wealth, and expectations. Langdon, M.,Powell, P. Tsuchiya, A. Dysart, L., McNamee, P., van der Pol, M. 33 Valuing end of life QALYs: individual and societal preferences 34 Comparative effectiveness of non-vitamin K antagonist oral anticoagulants (NOACs) and warfarin in the Scottish atrial fibrillation population: the value of real world evidence Criteria used by the Scottish Medicines Consortium (SMC) and National Institute for Health and Care Excellence (NICE) to assess end-of-life (EoL) technologies can enable their provision when the cost per quality-adjusted life year (QALY) is relatively high. However, empirical evidence of public preferences is equivocal for EoL health gains and limited for the value given to different types of EoL health benefits - quality-of-life (QoL) improvement and life-extension (LE). In this study, we examine individual and societal preferences for QALYs gained at the EoL relative to those from non-terminal health problems (Non-EoL) and for different types of EoL QALYs. Eight health scenarios were designed depicting i) QoL improvements for Non-EoL temporary and chronic health problems and ii) QoL and LE improvements for EoL health problems. Preferences were elicited using Person Trade-Off (PTO) and Willingness to Pay (WTP) techniques via Computer Assisted Personal Interview (CAPI) to a quota sample of 901 nationally representative individuals in Scotland between March and October 2016. PTO results suggest a preference for Non-EoL health gains over EoL health gains and for QoL improvements at EoL. WTP results are less clear but at the aggregate level are suggestive of a preference for EoL health gains over Non-EoL health gains and there is no clear preference for QoL improvements or LEs at EoL. Overall our results indicate no clear preference for EoL (particularly LE) health gains raising implications for current policy, which we will discuss alongside methodological issues around individual vs. societal preferences. Introduction: Different methods are available to correct for the absence of randomisation when estimating average treatment effects (ATE) using observational data. This study explores these methods, assessing whether RCT findings are generalizable to Scottish clinical practice. Methods: Propensity score (PS) matching, Inverse Probability Weighting and PS regression were explored utilising linkage of the Prescribing Information System (PIS), Scottish Morbidity Records 01 and death records for newly anticoagulated patients, with a diagnosis of atrial fibrillation (AF) or atrial flutter. AF related outcomes were compared for four treatment groups: Warfarin (79.1%), Apixaban (9.7%), Dabigatran (1.3%), Rivaroxaban (9.9%). Patients were followed from first oral anticoagulant prescription to first clinical event or death. Censoring was applied for treatment switching and various follow-up times. Using a continuous treatment approach, hazard ratios were estimated, adjusting for age, sex and comorbidities. Results: At 2 years follow-up PS matching showed no difference in risk of ischaemic stroke for Apixaban (HR: 0.94 (CI: 0.63-1.38)), Dabigatran (HR: 0.71 (CI: 0.34-1.47)), and Rivaroxaban (HR: 1.09 (CI: 0.82-1.45)) compared with Warfarin. A reduced risk of death was observed for Apixaban (HR: 0.44 (CI: 0.34-0.58)) and Rivaroxaban (HR: 0.65 (CI: 0.54-0.79). Hazard ratios were consistent across methods for these two drugs. The risk of death for Dabigatran differed substantially depending on the method used. Conclusion: For each NOAC the risk of McHugh, N., Pinto Prades, J.L., Baker, R., Mason, H., Donaldson, C. Ciminata, G., Geue, C., Deidda, M., Kreif, N., Wu, O., Langhorne, P. ischemic stroke was different from RCT findings. Results were generally consistent regardless of the method used. For small sample sizes, methods may yield different results and PS matching may not be appropriate. 36 Can variation in hospital mortality rates be explained by variation in admission threshold? Background: Mortality rates amongst the admitted patient population are used to measure hospital quality. Recent research has shown that variations in mortality rates on different days of the week can be explained by variations in admission thresholds not previously captured by standard riskadjustment methods. Variations in mortality rates across hospitals may also reflect variations in admission thresholds. Aim: To examine whether variations in hospitals' propensities to admit are associated with standardised mortality amongst the admitted population. Data:12.9million A&E attendances and 3.4million emergency admissions to all 140 non-acute Trusts in England for which the standardised hospital mortality indicator is calculated, covering 2013/14. Methods:Regression of risk-adjusted mortality on risk-adjusted admission probabilities between Trusts and over time. Results:Mean admission rates across Trusts are 0.26 (SD 0.054). Mortality rates are negatively associated with admission probabilities (-0.04, p=0.001), after standard case-mix adjustment. At the mean mortality rate of 4%, a 1 percentage point increase in the risk-adjusted admission rate is associated with a 10% relative reduction in the risk-adjusted mortality rate. Propensity to admit explained an additional 9% of the variation in hospital mortality rates over that explained by variations in case-mix. Conclusions: Hospital mortality rates are used as performance indicators, and have triggered public enquiries. Variations in Trusts' threshold for emergency admission explain variation in mortality rates. Hospitals seeking to reduce their standardised mortality rate could do so artificially by lowering their admission threshold. Failure to adjust for admission propensity across hospitals could result in misleading conclusions about quality. Meacock, R., Parkinson, N., Sutton, M. 37 Impact of adherence to diabetes mellitus medical follow-up on hospital admissions: Panel data evidence from France 38 Social preferences and acceptability of standard and behavioural economic inspired policies designed to reduce and prevent obesity Diabetes is associated with a large economic burden, and the major cost driver is hospitalization. That's why policy planners have developed strategies to prevent diabetes complications and limit hospital use. In particular, the French authority for health has issued eight guidelines, including regular controls of blood pressure and lipids to reduce cardiovascular risk, and regular screening for damage to the eyes, kidneys and feet. However, less is known on the efficacy of that prevention strategy. We aim to analyze the impact of diabetes follow-up care on the probability of being hospitalized. We use six waves (2010-2015) of an administrative dataset of detailed medical records from a major French social security provider. Our study sample is a balanced panel sample of 52,218 adults with diabetes. We construct a score of medical follow-up representing the quality of adherence to the eight current guidelines. We control for patient's socioeconomic features, diabetes severity, ambulatory care consumption, and geographic variables measuring the supply of care providers. We estimate a dynamic panel data model, which accounts for the impact of previous period hospital admissions, previous adherence to medical recommendations, and potential confounders. We use a random effect model with a Mundlak correction to deal with the presence of potential biases associated with patients' unobserved heterogeneity. Our results indicate that higher adherence to medical guidance is associated with a lower probability of being hospitalized. Optimizing adherence to diabetes follow-up guidelines may contribute to prevent the risk of hospitalization and avoid costs. The obesity epidemic is a significant public policy issue facing the international community with a number of associated adverse outcomes, including increased risk of chronic disease and death, resulting in substantial associated costs. A range of policies have been suggested to reduce and prevent obesity including those informed by standard economics and those harnessing behavioural economics to nudge individuals to change their behaviour to improve their health. What is not known is which policy interventions taxpayers find acceptable and would prefer to fund via their taxes. In this paper we present a study which used a best-best discrete choice experiment to investigate social acceptability of eight policies. The alternatives between which respondents were asked to choose were described by three attributes: policy type, effectiveness in terms of impact on the obesity rate and cost in higher taxes. The experimental design allowed for testing of the impact on choice of each attribute in isolation and in combination. Data were collected from an online panel of 1000 respondents representative of Australian taxpayers in age and gender. Discrete choice analysis accounted forheterogeneity in both preferences and scale. Predicted probability analysis was used to explore social acceptability of the eight policies while welfare analysis was undertaken to investigate willingness to pay higher taxes for particular policies in isolation and accounting for Bussiere, C., Sirven N., Rapp T., SevillaDedieu C. Lancsar, E., Ride, J., Au, N., Burgess, L., Peeters, A. effectiveness in terms of the impact on obesity rates. Policy implications in relation to informing optimal investment and targeting of policy to prevent and reduce obesity are highlighted as are methodological insights. 39 Modelling implausible EQ5D-5L states: prevalence in the general public and its effect on health state valuation 40 How should policy account of patient preferences when consolidating hospital services? The case of colon cancer surgery in Piedmont, Italy The EQ-5D (and other health state classification systems) assumes the level in a dimension can vary independently of other dimensions. However, some combinations look less 'plausible' than others. If 'implausible' states are used in health state valuations, then respondents may have difficulty imagining them, causing measurement error. There is currently no standard solution: some valuation studies exclude or 'back off' from such states, whilst others leave them in. This study aims to address two gaps in the literature: (1) to model the frequencies of self-reported EQ-5D-5L states in the general public to propose an evidence-based criterion for plausibility of EQ-5D-5L states; and (2) to re-model Discrete Choice (DCE) valuation data by contrasting between i) models that drop all implausible states (by our definition), and ii) models that drop the same number of randomly selected plausible states, to quantify the effect of implausibility of states in valuation designs. For the first aim, we pool across five secondary datasets from UK online surveys, where respondents were asked to report their health in EQ-5D-5L. The 17,955 respondents are distributed across 827 (26%) unique EQ-5D-5L states with a very long tail. Different count data models using combinations of different dimension levels are fitted to the frequency data of the 3,125 possible EQ-5D-5L states, controlling for severity. For the second aim, we use a subset of the above dataset, which used DCE with duration to value EQ-5D-5L (n=8,619). Analysis is currently ongoing. As a result of regional policy, the number of hospitals performing colon cancer surgery in Piedmont fell from 65 in 2004 to 52 in 2014. The policy has changed where patients seek care, the proportion treated at the nearest hospital falling from 68% in 2004 to 62% in 2014. Our objective is to understand the trade-off among distance and quality (measured in terms of waiting times) expressed by colon cancer patients in relation to the reconfiguration of cancer care services. The study is based on the Hospital Discharge Data over the years 2004-2014, on a sample of more than 26000 patients. We applied the conditional logit, which allows quantification of the impact of hospital-specific characteristics on patients' probability of preferring one hospital over alternatives. We analyse changes of preferences over time, highlighting the policy effect of concentrating services Bansback, N., Marten, O., Mulhern, B., Tsuchiya, A. Listorti, E., Street, A., Alfieri, A. in more specialized facilities. Preliminary results confirm the negative impact of distance and positive impact of quality on patients' choice. The concentration policy is associated with a decrease in the disutility of both distance and waiting time, perhaps due to the restricted choice set available to patients. By considering the context in which patients choose, such as changes in territorial configuration, we provide insights into how changes in policy can impact upon patient preferences. 41 The effect of house prices on long term care market: Evidence from England 42 Impact of GP turnover on patient outcomes The English housing market has experienced the greatest growth of prices among the OECD countries since the last decades. In this paper we focus on the consequences of this inflationary trend and analyse the effects of high house prices on an industry that typically works with low margins, the care homes that provide long term residential care services. The effect associated with high prices is a priori uncertain. On the one hand it may suppose an important barrier that can disincentivise the entry of care homes in certain markets. Alternatively, areas with high house prices may represent a business opportunity. Our work aims at disentangling these effects. In order to identify potential endogeneity concerns related to the effect of house prices on the care homes entries, we use an instrumental variables approach that exploits the variability in the restrictiveness of planning regulations across English districts. For doing that we construct a unique dataset that merges information from several sources in order collate information regarding the characteristics of the dynamics in the care homes market, the housing markets and the planning regulations. To our knowledge, this is the first paper that provides causal evidence with regards to the effects of housing prices in the context of entries in the market of care homes. Furthermore, this research also makes contributions to other strands of the literature. It provides further evidence to the literature that analyses the market of care homes in England and the literature on the effects of house prices. Background: General Practitioner (GP) turnover influences workload and care continuity. However, the impact of GP turnover on patient outcomes has been under researched. We sought to identify which aspects of patient outcomes are most sensitive to GP turnover. Methods and Data: Using complete career histories of all GPs in England, we calculated annual turnover rates for each practice. We matched these to data from Hospital Episode Statistics (admitted patient care, accident and emergency and outpatient data), GP Patient Survey, Quality and Outcomes Framework (QOF), GP prescribing, and other GP practice level datasets from NHS Digital for GP practice characteristics. Our sample contains 54,627 practice-year observations over the period 2010-2016. Results: On average, 9% of GPs leave their practice each year. Patient outcomes improve when new GPs join practices and deteriorate when GPs leave. When a GP leaves a practice with replacement, overall Gonzalo Almorox, E., Braakmann, N., Bilotkach, V., Wildman, J. Lau, Y-S., Sutton, M., Urwin, S. patient satisfaction decreases by 0.5%, A&E attendances increases by 3.3%, antibiotic prescriptions increase by 3.1%, and overall QOF achievement remains the same. Discussion and Conclusion: GP turnover is important for a range of patient outcomes. Government initiatives to reduce GP turnover are supported by the impact on patient outcomes. 44 french GPs' willingness to delegate tasks: when risk aversion meets financial incentives 46 Matching methods for estimation of treatment effect using observational data: A case study in breast cancer using data from the SEER registry Within hospital settings, delegation to paramedics is fairly recent in France. Whether General Practitioners are likely to follow hospital consultants is unknown. A 2012 survey of 2,000 GPs might help foresee GPs willingness to do so. This paper tests whether a more favourable funding system might help increasing GPs willingness. We implement a quasi-experimental design wherein GP's are randomly selected to form three groups of equal size, each of them being exposed to a different funding scheme when declaring their willingness to delegate tasks to nurses: fully funded (FF) by the social security administration, self-funded by GPs' revenues (Self Funded, SF) and half-funded by both the social security administration and GPs (Half Funded, HF). GP's likelihood to be in favour of task delegation is estimated with a Probit model which especially considers GP's attitude toward risk (aversion or preference), among a set of covariates such as age, gender, rural/urban area, GP's density and funding scheme. This article shows that, firstly GPs are more likely to favour delegation where they share a lower proportion of the cost. Secondly, the effect of risk aversion on the likelihood of being in favour of delegation is not altered by the funding scheme. Background: Several methods for estimating treatment effects from observational data have been described in Technical Support Document 17 by the National Institute for Health and Care Excellence Decision Support Unit. One technique described is matching patients based on observed characteristics. Previous studies using matching with observational data have concluded it produces clinically implausible treatment effects, and therefore researchers should be cautious about using observational data for this purpose. Aims: To provide generalisable lessons building on currently available literature, and provide a worked example of matching for use in an economic evaluation. Methods: A literature review searched for studies conducting matching for economic evaluations. Data from the Surveillance, Epidemiology, and End Results program were used to run several matching algorithms for comparison of surgery and radiotherapy with surgery in breast cancer patients. Methods tested were exact and propensity score nearest neighbour (simple; with calliper; with replacement). Treatment effects were compared with the relevant clinical literature. Results: Videau, Y., Combes, J-B., Paraponaris, A. Sly, I, Bell Gorrod, H., Gray, L. The literature indicated little consideration was given to how matching affected results of the economic evaluations and found several areas for improvement. Results were in line with existing recommendations; working iteratively to guide subsequent method choice produced better results, and method selection should be guided by dataset properties. Methods which worked best in reducing variance and bias gave the most clinically plausible treatment effects. Conclusions: Analysis of observational data can reliably estimate treatment effects when implemented correctly, with assumptions appropriately tested. More worked examples of conducting matching would aid future researchers. 49 Waits in A&E Departments of the English NHS Background: A core performance target for the English NHS is that at least 95% of patients attending Gaughan, J., Accident and Emergency (A&E) departments should be transferred, admitted or discharged within Kasteridis, P., four hours of arrival. This target has been breached with increasing frequency in recent years. Mason, A., Street, A. Potential explanations include: increased A&E attendances, staff shortages and a lack of substitutes such as social care. We investigate the impact of potential drivers of four-hour A&E waits. Methods: We employ a panel of quarterly data for all English NHS Trusts treating patients in a Type 1 A&E department from 2011/12 to 2013/14. To analyse determinants of waits over four hours, we use count models and adjust for variations in supply (A&E medical staffing, care home beds), demand (local population characteristics, A&E attendances) and hospital characteristics (bed occupancy rates, delayed transfers of care (DTOC)). Results: Bed occupancy rate is strongly and positively associated with waits over four hours in A&E. This finding is robust to a range of model specifications. We also find some evidence that fewer waits are associated with more care home beds, while more waits are associated with more DTOCs and larger local populations aged 65 and over. There was also some evidence that waits over four hours are more likely in A&E departments with more doctors. Conclusions: Emerging findings indicate that bed occupancy is a key determinant of waits over four hours. DTOCs and supply-side factors within and outside of hospitals may also play a role. 52 The medical "sorting hat": Do gender and ethnicity affect training doctors' chances of being selected into the preferred specialties? There is a strong interest in ensuring that the medical profession is representative of the society it serves and yet the distribution of doctors across specializations is highly unequal in terms of gender, ethnicity and socio-economic background. It is important to understand whether this is a consequence of preferences and choices of doctors or of the selection procedures. We examine whether, other things being equal, doctors' characteristics are correlated with their interview scores in the selection process. We focus on the selection that takes place after training doctors have made their speciality application choices. The interview is a crucial element of this selection stage. Using a unique linked dataset (UKMED) we study individuals who started medical studies in the United Kingdom in 2007 and 2008. We analyse the role of demographic characteristics in determining interview scores. We control for shortlisting score (objective pre-interview measure of doctors' achievements), medical school and other relevant characteristics. We find a positive statistically significant effect for women and a negative effect for Black and minority ethnic (BME) doctors, ceteris paribus. We then apply an Oaxaca decomposition and find that those differences cannot be explained by differences in the distribution of covariates between groups. Contrariwise, these reflect differences in the estimators between demographic groups, meaning that the same covariates may be valued differently by selectors. Our results indicate that further investigation is required to understand the observed differences and in particular to understand why BME doctors are scored less favourably in the assessment process. Rodriguez Santana, I., Chalkley, M. 53 How Does Participation in Community Assets Affect HealthRelated Quality of Life and Health Care Utilisation? Community assets - including community centres, libraries, markets and pubs - are being promoted Jones, A., Munford, as a way of increasing health and reducing demand for formal healthcare services. However, little L., Sutton, M. causal evidence on their effectiveness is known. Using bespoke data on 3470 individuals aged 65years and older with a chronic condition collected in the north-west of England, we estimate the impact of community asset participation on three outcome measures: health-related quality-of-life (EuroQol-5D-5L); the costs of health care utilisation; and the net-benefits of participation. We obtained the geo-location of all community assets and create a range of potential instruments based on (i) the number of community assets within given distances of each individual's place of residence and (ii) the minimum distance to the nearest community asset. Using linear regression models with endogenous treatment effects, we show that in the participation equation distance to the nearest community asset has a substantial effect on participation. A one mile increase in distance to the nearest community asset is associated with a 66 percentage point reduction in participation. There is a large, positive, and significant effect of community asset participation on health-related quality of life; a 0.137 point gain in EuroQol5D-5L scores. The effect on healthcare utilisation is negative but not statistically significant. Using the current NICE threshold values of £20,000 per QALY gives a net-benefit estimate per participant per year of £3624.20. Therefore the social value of developing community assets is potentially substantial. Work is ongoing applying more sophisticated spatial econometric methods to the data. 55 Health Insurance Contracts and Healthcare Utilization - An Experiment on the Role of Forward Looking Behavior Aron-Dine et al. (2015) find that individuals do not only react to the spot price of health insurance contracts with a deductible. Yet, given field data it is difficult to keep the spot price among individuals constant while systematically varying the future price. For this, we use a controlled laboratory experiment in which subjects go through a cycle of periods. In each period they face probabilistic health events and have to choose between treatment and non-treatment. Treatment is free if subjects accumulate costs beyond their deductible, but prior to that non-treatment might be less costly than treatment. We vary the likelihood of hitting the deductible, i.e. the future price, through two channels: the number of periods and the height of the deductible. We also vary whether subjects receive regular updates on their remaining deductible. By eliciting individual risk and time preferences, we are able to link them to the observed utilization behavior. In line with Aron-Dine et al. (2015), we find that subjects do not only react to the spot price. Moreover, we show that varying the future price has a significant effect on spending behavior, regardless whether the same future price is reached by changing the deductible or the number of periods. Risk-averse individuals tend to display stronger forward-looking behavior when they have sufficient information. Our results contribute to further understanding health care utilization under dynamic incentives. They emphasize the importance of acknowledging forward-looking behavior, risk preferences and information in this context. Han, J., KairiesSchwarz, N. 56 Socioeconomic inequality in postnatal depression: a potential early-life root of disadvantage 57 Accounting for treatment adherence in health economic model for sleep apnoea This paper investigates socioeconomic inequality in postnatal depression, and the role it might play in intergenerational transmission of inequality. Infants' development is thought to be particularly sensitive to mothers' mental health at this time, so that such early life exposure might be a root of later inequalities. Postnatal depression is common, and heightened contact with health services during this period presents opportunities for intervention. However, lower uptake among the disadvantaged might be widening inequalities, with potential consequences for mothers, children and the health and care systems. The concentration index is used to explore inequality across the full range of income, rather than focussing on the extremes of the distribution, as done in previous studies of inequalities in postnatal depression. Regression-based decomposition of the concentration index is used to examine factors associated with inequality in postnatal depression, and to explore the association with later inequality in their children's outcomes. Data are taken from the UK Millennium Cohort Study. Along with marked socioeconomic inequality in postnatal depressive symptoms, the key finding from the study is an association between this inequality and inequality of outcomes for children at age 11, even after accounting for a range of other factors that might explain such an association. These findings highlight the importance of identifying and investing in interventions that reduce inequalities as well as improving postnatal mental health. Such interventions could have an impact on socioeconomic gradients in both postnatal mental health and downstream outcomes. It is important that adherence with an intervention is reflected in cost-effectiveness models to replicate real-life use. Despite this, it is frequently implicitly assumed adherence is the same in both the short and long term. Adherence is rarely modelled, often due to a lack of good data. This work uses a case study of treatment for mild-moderate sleep apnoea. There is evidence the optimal treatment decision is sensitive to assumptions on adherence with two interventions - Mandibular Adjustment Devices (MADs) and Continuous Positive Airway Pressure (CPAP). Literature was searched, finding 13 and 5 studies presenting proportions of patients still using CPAP and MADs, respectively, at a range of times. Meta-analyses of these data were carried out, with ceasing use of the device as a time-to-event outcome. By using the results of the meta-analysis in the costeffectiveness model, we found the decision uncertainty around the optimal treatment was reduced. The value of further research on long-term adherence has also been explored. Recent nonparametric regression methods have been used to calculate the Expected Value of Perfect Partial Information (EVPPI) finding value in gaining more information on long-term adherence with both interventions. Extending recent work on the calculation of the Expected Value of Sample Information (EVSI) we calculated the value of a study collecting information on adherence at Ride, J. Simons, C., Jackson, C. multiple time points. We found value in collecting information on adherence with both interventions at short time periods, at a small number of time points and for small populations. 59 Eliciting Patients' Preferences in Kidney Transplantation: A Discrete Choice Experiment Kidney transplantation provides an expected survival advantage over dialysis treatment for patients with end--stage renal disease. Still, due to the disparity between large number of transplant candidates and scarcity of organs patients may face the trade-off between a long waiting list for a high quality kidney, or a "marginal'' organ transplanted immediately. Current allocation protocols do not explicitly take into consideration patients' preferences. We study patients' time and risk preferences for kidney transplantation by means of a discrete choice experiment (DCE). We have the unique opportunity of running the experiment on the entire population of individuals actually waiting for a kidney transplant in an important hospital in Italy. We find heterogeneity in time and risk preferences. Differences are not limited to mean values of willingness to wait (WTW) for better kidneys: we employ a mixture logit model to retrieve individual WTW and compare the entire distribution of preferences across different subgroups. We find that younger candidates are willing to wait longer than older candidates for extra year of graft survival and to give up augmented infectious and neoplastic risks. Moreover, patients with longer time on dialysis are willing to wait longer than the other patients to give up augmented infectious and neoplastic risks, and for an extra year of graft survival. The implication for transplant practice is that accounting for individual preferences in kidney allocation algorithm would improve patients' satisfaction and efficiency of the donor receiver matching process. Genie, M., Nicolo, A., Pasini, G. 60 Public health insurance coverage and child health 62 To what extent do preferences for public health interventions differ and whose preferences should count? A case study using weight loss maintenance In recent years the importance of expanding health insurance coverage in countries without Tonei, V., Nolan, A. universal healthcare access has been at the centre of political debate. Public health insurance programs are mainly justified by the aim of reducing inequalities in access to health care, and, in a final instance, improving public health. This goal is particularly important for children because early health interventions have been shown to be critical for future health and life outcomes. The paper investigates the relationship between public health insurance coverage, child health and utilisation of health care services in the Republic of Ireland. To identify the effects of interest we employ a variety of econometric methods to estimate reduced form health production functions which include child fixed effect models, difference-in-difference propensity score matching and value added models. We take advantage of the variety of information from Growing Up in Ireland (GUI) - a nationally representative child cohort study - by focusing on different measures of child health (physical health, psychological well-being) and alternative measures of health care utilisation (visits to the GPs and to hospital Accident and Emergency departments). Our preliminary results show that children whose families benefit from public health insurance coverage have higher healthcare use, but the effects on health are ambiguous. Further research is needed to understand whether this evidence is due to a lower level of health endowment at birth for low-income children (compared to children from high-income families), an inefficient utilisation of health care services or other factors. Patient preferences are increasingly elicited using discrete choice experiments (DCEs). The adoption Mott, D., Ternent, of patient-centred approaches to healthcare decision-making has increased interest in the L., Vale, L. incorporation of quantitative patient preference information. Patient preferences have been generated from various sources such as randomised controlled trials (RCTs), registries and online panels. However, little is known about the influence that the sample source and the relative experience of respondents might have on the results. A DCE containing embedded rationality tests was completed by four different groups: one recruited via a RCT and three from an online panel. Three of the groups, the RCT group and two groups from the panel, could be described as patients with the fourth described as non-patients. Preferences were elicited and compared for a public health intervention aimed at assisting individuals with weight-loss maintenance. The results suggest that preferences differ across all groups. The largest differences occurred when comparing the group from the RCT with the three groups from the panel. In contrast, the differences between the three online panel groups were relatively minor. The RCT group were also 23% more likely to pass all of the rationality tests. The results of this study suggest that preferences could vary depending on the type of patient sample recruited. Additionally, whilst respondents from a RCT may be more engaged with DCE tasks than individuals recruited from elsewhere, they may not be representative of the population of patients. This raises concerns surrounding the generalisability of the results of DCE studies that elicit patient preferences. 63 Non-marginal budget impacts in health technology assessment: a theoretical presentation Conventional decision-making for health technology assessment relies upon entirely accepting or rejecting new interventions based upon their incremental cost-effectiveness ratio (ICER) relative to some benchmark threshold (k). More recently, attention has focused more on a fuller consideration of the likely opportunity costs of such interventions based on the size of their budgetary impact, centring on the empirical estimation of the health production function (HPF) (Claxton, Lomas, Palmer & Soares, under review). Given the likely diminishing marginal efficiency of healthcare spending, larger budgetary impacts are likely to imply larger health losses per pound spent. While a decision rule comparing k to the ICER is appropriate for small budgetary impacts, partial acceptance may prove cost-effective where the budgetary impact is greater, due to the non-linearity of the HPF. Previously, this has been held to be irrelevant for two reasons. The first (absence of an estimate of the HPF beyond the margin) no longer holds in the light of recent empirical estimation of the HPF. Second, the partial funding of services has been held to contravene horizontal equity. This paper presents a four-quadrant model of decision making under conditions of non-marginal budgetary impact, consistent with a reduced form of the full mathematical programming problem, argues that problems of contraventions of horizontal equity are not unique to the case of partial funding of treatments with large budgetary impacts, and demonstrates that any apparent such problem is potentially obviated when the price of treatment is endogenous rather than assumed exogenous. Howdon, D., Lomas, J. 64 Examining variation among acute trusts in patient delayed discharge 66 Using deliberative methods to establish a sufficient level of capability wellbeing for use in decision-making in the contexts of public health and social care Background: Delayed discharge adversely affects patient care and increases costs to England's National Health Service. Patient care pathways inform patients on expected length of stay and discharge time providing a benchmark against which patients report their experience. We present an analysis of delayed discharge with two novel characteristics: (1) delayed discharge is defined according to patient experience from the Adult Inpatient Surveys 2007 to 2014, (2) trust differences are interpreted as unwarranted variability. Methods: Our empirical approach utilises a first stage probit model to explain patient reported delayed discharge as a function of age, gender, and admission characteristics related to severity. The probit includes trust fixed effects which estimate trust-specific probability of delayed discharge. In the second stage, trust fixed effects are related with trust internal factors argued to be drivers of delayed discharge: the natural variation of the patient mix and artificial (unwarranted) variation in the organisation and delivery of health care (staff, processes, resources). Socio-economic regional characteristics are also included. Results: Being male, emergency admissions, critical care, and ward transfers showed statistically significant increases in patient-specific probability of delayed discharge. Regarding trust-specific probability of delayed discharge, larger delays were observed in hospitals admitting older patients, performing more procedures, having a higher ratio of admissions to non-medical staff, and having more specialties. A larger delay in regions with higher income deprivation hints social care needs. Conclusions: Our findings suggest reducing unwarranted systemic variability would reduce delayed discharge. Two opportunities for improvement are non-clinical staff capacity and economies of scope. Background: In 2013 NICE extended its remit to include social care. NICE permits the use of a small number of well-being instruments as alternatives to EQ-5D, acknowledging that "people using social care services and the workforce providing them have different priorities and needs". One such alternative is ICECAP-A, a capability wellbeing instrument with five dimensions, each with four levels.Aim: To establish a decision-rule for use with ICECAP-A, based upon the concept of short-fall sufficiency. Methods: A sufficient level of capability wellbeing was established through public deliberation, core principles of which are that participants be informed and that there be free, open and reflective debate. A series of one-day/6.5 hour citizen's workshops were conducted with recruitment from within purposively selected local authority areas. Workshops consisted of a mixture of background information, individual tasks, group discussion and voting. Representatives from each workshop were then invited to attend a 'consensus workshop'. Follow-up interviews facilitated evaluation of deliberation. Results: 62 participants took part in deliberative work, across eight workshops. Participants actively engaged and provided positive feedback about their DeVolder, R., SerraSastre, V., Zomora, B. Kinghorn, P. experience. Key considerations for participants included: the realistic ability of public services to enhance some areas of well-being; not removing incentives for self-help; signalling the type of society participants wished to be part of; compensating for a suspicion that policy-makers would cut-back on what had been agreed. The resulting sufficient level of well-being (defined by ICECAP-A) was state 3,3,3,3,3. Conclusion: Future research will elicit a monetary threshold for a year of sufficient capability wellbeing. 69 Choice certainty and deliberative thinking in discrete choice experiments. A theoretical and empirical investigation Background: Post-decision certainty scales are used to decrease hypothetical bias in responses to contingent valuation or choice experiment (CE) surveys. In general, researchers re-code uncertain respondents' answers or re-weight more certain respondents in the econometric model. No conceptual framework has been proposed and tested to explain the rationale behind either strategy. We propose that rather than response certainty, what matters to identify reliable responses is certainty variability across choice tasks. Respondent's choice certainty should differ across the different choice sets if they are engaged in the task. Aims and methods: This paper aims to better understand CE certainty using insights from Kahneman's theory of dual processing. First, we map respondents' certainty patterns to fast-intuitive (System 1) versus rational-deliberative (System 2) thinking. We test our assumptions using data from a CE investigating women's preference for breast cancer screening. Second, we compare different calibration techniques in terms of their impacts on parameter precision and welfare estimates in two different settings in France and in Canada. Results: We find respondents with higher certainty variability are more likely to use System 2 (they have higher response time, seldom use decision heuristics, and have lower response errors) compared to constantly certain respondents who more often use decision heuristics. Re-weighting variably certain respondents improves the precision of welfare estimates up to 69%, whereas reweighting constantly certain individuals reduces statistical efficiency. Conclusion: Certainty variability plays a moderating role in the relationship between choice certainty and choice consistency. Re-weighting constantly certain respondents may be counterproductive. Regier, D., Sicsic, J., Watson, V. 71 Exploring the Validity of Incremental Willingness to Pay: Application to Dental Interventions 72 Moral Hazard in Prevention and Treatment: A Reference Dependent Model Background: When willingness-to-pay (WTP) is used in health economics concerns are frequently raised about the validity of the values elicited. In the standard approach participants are asked for an absolute WTP value but when eliciting values for multiple interventions from each respondent results are often non-discriminatory and contain preference reversals (implicit and explicit rank of options directly contradict each other). The incremental approach imposes an exogenous framework which theoretically eliminates preference reversals and increases discriminatory power by ascertaining an absolute value for the participants least preferred option then asking how much more they are willing to pay to secure their next preferred option. Method: Using a representative sample of 787 people living in England (450 participants in the standard approach; 337 in the incremental) we gathered WTP values for five societal dental interventions using a questionnaire. Taxation was the payment vehicle and preventable bias was controlled for. Initially eliminating protest responses the values are used to address preference reversals and discriminatory power for each approach. Results: Initial results indicate preference reversals are eliminated for the incremental approach with 92% of values corresponding exactly with the explicit rank (3% of responses do this in the absolute approach). However when we consider the occurrence of protest responses in the incremental approach we find possible evidence of preference reversals. The values associated with the interventions are non-discriminatory for both approaches at the median. Both of these factors lead us to believe that incremental approach does not offer a significant advantage in this context. Moral hazard describes the changes of behaviors in prevention and treatment caused by health insurance. Cheaper treatments may discourage preventive efforts (ex-ante moral hazard), and encourage patients to spend more (ex-post moral hazard). The Von Neumann-Morgenstern utility function, adopted by orthodox economic analyses leads to inconsistent predictions with what people actually behave regarding prevention and treatment. People could be driven bankrupt by chronic diseases, and have extraordinarily strong preferences over expensive medical products. This paper endeavors to apply the reference dependent model to analyze people's preventive decisions and the demand for treatment. It also examines the effect of health insurance, and sheds lights on the design and regulations of health insurance. In the basic model where preventive efforts are costly and affect the future probability of illness, optimal level of preventive effort is determined solely by one's concerns over his future health, thus a pooling insurance that reduces the cost of treatment universally will not create ex-ante moral hazard. As for treatment, people with different income levels have the same desired level of treatments, and those with insufficient income will go bankrupt. When prevention also contributes to improving treatment efficiency and post-treatment Carr, K., Donaldson, C., Wildman, J. Vernazza, C. Dong, Y. prognosis, demand for treatment and optimal level of prevention become interdependent, and it is possible to influence people's preventive behavior through the design of health insurance. As for optimal insurance, this paper favors coinsurance policy over deductibles, and the optimal coinsurance rate depends on the distribution of income. 74 Inequality in LongTerm Care public Benefits 75 Market structure, patient choice, and hospital quality for elective patients. We investigate the evolution of inequity in public Long-Term Care (LTC) use in Spain during the period 2011 to 2014. At the start of this period, the universal system introduced in 2006 was fully developed, while a set of reforms motivated by the financial crisis were implemented in 2012. We use rich individual administrative information on health status, sociodemographic characteristics and use of services for the universe of public LTC recipients. This allows a detailed analysis of inequity in different types of services and their provision. Our findings suggest that the system is not equitable as the type of LTC benefits are distributed differently across socioeconomic groups. We find striking differences in the type of provision: while publicly provided LTC services are concentrated amongst the worse-off, the better-off receive a voucher to cover LTC expenses from their preferred provider. Our results also show that cash benefits to cover informal care costs are pro-rich distributed, especially after the reform. However, the better-off are more likely to combine informal caregiving with formal services, while the poor have disproportionally more informal caregiving as a unique benefit. We find that the use of nursing homes is concentrated among the poor. Yet, there are large differences on whether the service is publicly provided (pro-poor) or via vouchers (pro-rich). We show that this different allocation is probably driven by the waiting time for nursing homes given capacity constraints of public nursing homes, especially after the costcontainment reforms introduced by the 2012 reform. Research question. The effect of competition on quality of care is contentious. Almost all studies have examined the effect of competition on quality of care for patients admitted with an emergency, most often patients with AMI. But if hospitals compete they can only do so for elective patients, not emergency. We therefore examine the effect of competition on quality for elective hip and knee replacement and coronary artery bypass graft patients. Data. Hospital Episode Statistics for 2002/3 to 2010/11 for 455,000 hip replacement, 486, 000 knee replacement, 114,000 CABG patients. Quality is measured as emergency readmissions within 28 days and deaths within 30 days. Methods. We test whether the effect of market structure changed after the relaxation of Hernández-Pizarro, H., Garcia-Gomez, P., LópezCasasnovas, G. Moscelli, G., Gravelle, H., Siciliani, L. 77 Does Better Health Insurance Coverage Impact The Demand For Health Care? A French Natural Experiment 78 Exploring the impact of new medical technology on workforce planning constraints on patient choice of elective provider in 2006. We allow for time-varying endogeneity due to the effect of unobserved patient characteristics on patient choice of hospital by using Two Stage Residual Inclusion. Results. We find that the change in the effect of market structure due to the 2006 choice reforms was to reduce quality by increasing the probability of a post-operative emergency readmission for hip and knee replacement patients. There was no effect of the choice reform on hospital quality for coronary bypass patients. We find no evidence of self-selection of patients into hospitals, suggesting that a rich set of patient-level covariates controls for differences in casemix. In Europe, there are few studies on the link between health insurance and health care consumption, the reason of it might be the existence of a statutory national health insurance (NHI) in most European countries. However, voluntary health insurance (VHI) covers health care expenses not (fully) financed by NHI. Interestingly, in France, a major VHI company decided to replace its unique basic level of coverage by a range of offers, allowing us to use a natural experiment to test some theoretical assumptions about consumption patterns. Reimbursement claim data from 36,524 insurees was analysed over the period 2009-2015. We used the propensity score matching (PSM) technique to compare the insurees having opted for an extended coverage (EC) to those still covered by the basic one (BC). Difference-in-differences (DD) models were used to estimate the impact of a change for EC on health care consumption assessed in both euros and quantity. Surprisingly enough, EC insurees began increasing their consumption before change. DD models showed a strong significant rise after change but for the first year only. When considering categories of care, we found strongly significant positive effects for dental prostheses and vision, and a few negative effects, especially for hospital. Better coverage tends to increase health care consumption immediately after change, particularly for costly care. But the rise in consumption before change seems to contradict the pent-up demand theory and suggests possible determinants of coverage extension, which are unplanned expenses, as the rise in hospital care before change appears to indicate. This paper contributes to the existing literature on the diffusion of medical technologies. The objective of this research is twofold. First of all, we examine the substitution or complementarity effects across different types of new technologies introduced into the NHS. In particular, the analysis is looking at the introduction of PTCA for the treatment of cardiovascular disease and the introduction of laparoscopic procedures for Bariatric surgery (i.e. gastric band, gastric bypass and sleeve gastrectomy). Secondly, we compute estimates of the degree to which the workforce reacts to the introduction of new technology, based on elasticity of supply measures. Data is combined Sevilla-Dedieu, C., Billaudeau, N., Paraponaris, A., Maynou, L., McGuire, A., SerraSastre, V. 79 Cost-utility analysis using EQ5D utility values: how does the choice of value set or method matter? from different sources to analyse these relationships: mainly, the Hospital Episodes Statistics (HES) and the NHS Electronic Staff Records (ESR). We apply panel data techniques to determine the manner in which technology is diffused across the NHS, with a particular emphasis on the impact that technology has on the workforce composition. Our analysis is at the trust level and the empirical specification explores the relationship between volume and workforce also controlling for trust and at risk population characteristics. Given the lack of quantitative evidence on the degree of substitution across different forms of input in treating surgical cases, such analysis gives indicative estimates of productivity gains attributable to flexible workforce planning and technology uptake Objectives: Multiple value sets and methods are available for deriving EQ-5D utility values in costutility analysis (CUA). We aimed to investigate whether the use of different EQ-5D value sets and methods would produce same cost-effectiveness results. Methods: The investigation was based on a CUA of hemodialysis (HD) and peritoneal dialysis (PD) for patients with end-stage renal disease (ESRD). For calculating quality-adjusted life years (QALYs), utility values were estimated from EQ-5D5L and SF-12 data using seven EQ-5D-5L value sets, seven EQ-5D-3L value sets, and four mapping functions from SF-12 to EQ-5D. Markov models were used to assess the cost-effectiveness for nondiabetic and diabetic patients separately, by comparing incremental cost-effectiveness ratios (ICERs) of HD compared to PD with a maximum willingness-to-pay (WTP). Probabilistic sensitivity analysis (PSA) was performed. Results: For non-diabetic patients, ICER was lower than WTP using four 5L value sets (UK, Canada, China, and Singapore) and three 3L value sets (UK, South Korea, and Singapore) while it was higher than WTP using other 5L or 3L value sets; for diabetic patients, ICER was higher than WTP using all EQ-5D value sets. PSA showed HD is more cost-effective for nondiabetic patients but PD is optimal for diabetic patients using all EQ-5D value sets. SF-12-mapped EQ-5D showed very different results compared to EQ-5D value sets. Conclusions: Different value sets and methods for deriving EQ-5D values would generate different cost-effectiveness results, highlighting the importance of using utility values directly generated from EQ-5D data using a country's own EQ-5D value sets in CUA. Yang, F., Luo, N.
© Copyright 2026 Paperzz