In fall 2020, EuHEA launched a new web-based seminar series. Due to the cancellation of the in-person conference in Oslo earlier this year, the chairmen of Oslo’s scientific committee, Oddvar Kaarboe and Pedro Pita Barros, kindly agreed to organize a weekly online seminar, with the support of the countries’ representatives, inviting selected presenters who submitted their papers to the original conference.
Please note that the seminars were not recorded.
17 September 2020, 14.00-15.00 (CET)
Competition, reputation and feedback in health care markets: Experimental evidence
Thomas Rittmannsberger, University of Innsbruck
24 September 2020, 14.00-15.00 (CET)
Reductions in out-of-pocket costs and moral hazard delays in health care
Naimi Johansson, University of Gothenburg
1 October 2020, 14.00-15.00 (CET)
We here address the causal relationship between maternal depression and child human capital using UK cohort data. To do so, we exploit the conditionally-exogenous variation in mothers’ genomes in an instrumental-variable approach, and describe the conditions under which mother’s genetic variants can be used as valid instruments. We show that an additional episode of maternal depression between the child’s birth up to age nine reduces both their cognitive and non-cognitive skills by 20 to 25% of a SD throughout adolescence. Our results are robust to a battery of sensitivity tests addressing, among others, concerns about pleiotropy and genetic inheritance.
8 October 2020, 14.00-15.00 (CET)
This paper analyses the role of medical and non-medical staff in the production of maternity services in the English NHS. Using hospital panel data (2004-2012) and estimating flexible production functions using system GMM estimators, we explore the output contribution of maternity services labour inputs. The results suggest that consultants and doctors have the highest marginal productivities while the productivity of support workers is insignificantly different from zero. Moreover, there is evidence for some degree of complementarity between midwives, support workers and consultants. Moreover, midwives could replace doctors and doctors could replace consultants in the production of maternity services.
15 October 2020, 14.00-15.00 (CET)
Policymakers aim at improving quality of care and the efficiency of health systems. One increasingly popular policy lever is the use of Pay for Performance (P4P) schemes that incentivise the adoption of best practice by financially rewarding process and outcome measures of quality in primary and secondary care. Despite their popularity, the evidence about their effectiveness remains inconclusive with several studies on hospitals suggesting small or mixed improvements in quality, possibly due to the small size of the bonuses or the design of the schemes. This study analyses the effects of a national P4P scheme in the English NHS that incentivises hospitals to achieve best practice in the delivery of hip fracture care.
The Best Practice Tariff (BPT) for hip fracture, introduced in England in 2010, rewards providers based on a care bundle that consists of nine process measures that need to be jointly achieved. The payment for the care bundle therefore implies that the provider receives the financial bonus only if each of the nine measures are met. The development of these measures was clinically driven, formed by consensus with clinicians, informed by evidence and based on the comprehensive National Hip Fracture Database (NHFD). The nine measures include time to surgery within 36 hours, four measures of involvement of orthogeriatricians, the use of a multidisciplinary rehabilitation team, and provision of preventive activities (bone health assessment, falls prevention). In addition to the scheme being evidence-based, the size of the bonus was significant, up to 20% of the baseline tariff. Using patient level data between 2008-2014 on a sample of 275,898 patients with a rich set of covariates, we employ a difference-in-difference (DiD) strategy, with Wales as a control group, to identify the causal effect of this policy.
The policy was successful in increasing the proportion of patients for whom all of the criteria are met by 52 percentage points. However, we find large heterogeneity across different performance measures. The largest improvement is in the measures requiring involvement of geriatricians in the care of patients (between 20 and 65 percentage points). The effect is much smaller in areas in which the achievement was already high in both countries before the introduction of the policy, such as falls prevention and cognitive assessment. Our results further suggest that providers in England focus on achieving all of the criteria, while the number of achieved conditions shows a more uniform distribution in Wales. Overall, we find that a scheme based on care bundle, which is evidence based, and used a sizable bonus can be effective in improving hospital performance.
22 October 2020, 14.00-15.00 (CET)
High and growing prescription drug costs in the United States are a major concern for policy makers. This paper focuses on the extent to which promotional gifts and other transfers made to physicians by pharmaceutical companies causes physicians to prescribe more expensive medicines. In our analysis, we link data from a federal database on the universe of industry payments between 2014 and 2017 to prescribing behavior in Medicare Part D. We develop a novel empirical strategy that uses data on the prescription behavior of physicians in Vermont, where a strict ban on industry payments to physicians is in place, combined with machine learning techniques to construct the counterfactual outcome for physicians who receive payments in the nearby states of New Hampshire and Maine. We find that a gift ban, such as the one implemented in Vermont in 2009, has the potential to result in a 3% decline in the total cost to treat diabetes. We investigate heterogeneity in the treatment effect and find that physicians who have a high share of patients with a low-income subsidy, and thus lower out-of-pocket expenditures, prescribe relatively more brand drugs and expensive drugs in response to industry payments. Our findings illustrate how industry payments interact with insurance to drive up health care costs.
29 October 2020, 14.00-15.00 (CET)
Hospital strikes in the Portuguese National Health Service (NHS) are becoming increasingly frequent. This paper analyses the effect of different health professionals' strikes (physicians, nurses and diagnostic and therapeutic technicians - DTT) on patients' outcomes and hospital activity. Patient-level data, comprising all NHS hospital admissions in mainland Portugal from 2012 to 2018, is used together with a comprehensive strike dataset with almost 130 protests. Pooled OLS is employed to study the impact of strikes on health outcomes. A Hazard model is also used to analyze changes in patients' length of stay. Data suggests that hospital operations are partially disrupted during strikes, with sharp reductions in surgical admissions (up to 54%) and a decline on both inpatient and outpatient care admissions. Controlling for hospital characteristics, time and regional patterns, and differences in patients' composition, results suggest a 6% increase in hospital mortality for patients exposed to physicians' strikes. Urgent readmissions increase for patients exposed to nurses or DTTs' strikes. Results suggest that legal minimum staffing levels defined during strikes, particularly during physicians' strikes, fail to prevent declines in the quality of care provided.
5 November 2020, 14.00-15.00 (CET)
Despite a wide consensus that obesity increases healthcare costs, estimates vary substantially between studies. Since different studies utilize different data sources, it is hard to disentangle how much of this variation that is caused by differences in the analytical approach or the populations under analysis. Our objective is to use different analytical approaches to estimate the size of this relationship, using the same population, and to contrast the findings. We use data from the Health Surveys in Nord Trøndelag (HUNT1-3), which includes 120,000 individuals, who, when linked to the Tuberculosis registry, were followed for 50 years (1960–2008) and where information was measured using self-completed questionnaires, clinical measurements and biological samples. These were linked to national registers including data on total healthcare utilization (2009-2016) and sociodemographic variables. We operationalized a range of analytical frameworks that handle confounding and simultaneity in different manners: multivariate regression analyses using OLS (1), GLM (2) and Two-part models (3), lagged models, when excluding the first years of follow-up and thus inducing a time gap (4) or when using previous measures of BMI as an instrument for current BMI (5), analyses using BMI of the oldest offspring available as an instrument for parental BMI (6) and analyses using genetic polymorphism as instrumental variables for BMI (Mendelian Randomization, MR) (7). Estimates are displayed as the additional healthcare cost (in Norwegian kroner (NOK) 2016 €) of the increase in one BMI point. All analyses were stratified by gender.
For women, the estimated relationship between increasing BMI and the average annual healthcare costs (aHCC) were in the range NOK 373 (95% CI 7 – 740) and NOK 1403.748 (95% CI 578 – 2,229) across all models – with the lowest and highest estimate coming from the MR- and offspring-analyses, respectively. For men, the estimates were higher, and varied more, between NOK 725 (95% CI 453 – 997) and NOK 2,148 (525 – 3,771) – the latter estimated when using offspring BMI as instrument for parents BMI.
In conclusion, our preliminary results indicates that the relationship between obesity and healthcare costs differ between men and women, and that different analytical approaches leads to variation between the estimates—likely because of the complexity surrounding obesity.
Gudrun Bjørnelv, Norwegian University of Science and Technology
12 November 2020, 14.00-15.00 (CET)
Introduction: Severe conditions can lead to health states perceived to be worse than death/being dead. In the conventional approach of calculating quality-adjusted life years (QALYs) states worse than death are assessed by methods such as the time trade-off or the visual analogue scale. Yet, the ability to discriminate states worse than death has been questioned and separate elicitation exercises (for states better and worse than death) can cause a discontinuity of preferences around dead, the gap effect. In addition, as preference weights can extend to minus infinity, the bottom of the valuation scale is usually arbitrarily fixed. The purpose of this study was to re-analyze the necessity of eliciting preference scores for states worse than death.
Methods: This study analyzes four distinct scenarios of providing treatment for patients feeling worse than death. The scenarios differ by whether the average treatment outcome and the lower bound of its 95% confidence interval (CI) are better or worse than death. For patients with outcomes worse than death, a distinction is made with regard to a preference for continuous living.
Results: Given the availability of effective and cheap interventions such as palliative sedation, average treatment outcomes worse than death require an ethical justification, i.e., a preference for continuous living despite feeling worse than death. If patients fulfill this criterion, it allows assigning them a preference score above zero, representing a preference for living, and capturing changes in-between states worse than death above the zero. In agreement, a zero score would be assigned to patients with a preference for death. For treatments with an average outcome better than death but a lower CI bound signifying a preference for death), probabilistic sensitivity analysis could rank states with a preference for death based on the degree of suicidal wish (because tradeoffs between costs and degrees of suicidal wish are ethically questionable). Occasional desires for death that are unresponsive to treatment can be captured by assessing a health profile that includes periods with a desire for continuous living and thus avoids a preference-for-death assessment.
Conclusions: It is possible to define the zero point of the valuation scale as a preference for death and still capture relevant worse-than-death states on the conventional scale when conducting an economic evaluation based on the QALY method. Discrimination between degrees of preference for death appears to be only necessary for the purpose of rank-ordering alternatives in a probabilistic sensitivity analysis and can be captured by the degree of suicidal wish.
19 November 2020, 14.00-15.00 (CET)
This paper investigates the short-run impact of public insurance expansion under the Affordable Care Act on out-of-pocket medical spending (OOP) and risk exposure among low-income, childless households. Using data from the Medical Expenditures Panel Survey (MEPS), I exploit exogenous variation in Medicaid eligibility rules across states, income groups and time. I find that public insurance eligibility reduced risk exposure by 11% of the pre-reform level and mean OOP by 18.2% among targeted households. Medicaid expansion did not increase total expenditures among eligible household, but rather shifted the burden of payment from eligible households and private insurance (20% reduction) to taxpayers in the form of public insurance (45.7% increase). The efficiency of these public funds can be summarized by a MVPF ranging from 0.05 to 0.47 that is highest for households with at least one pre-existing condition.
26 November 2020, 14.00-15.00 (CET)
The present study aims at contributing new insights into the public’s distributional preferences for health care allocation. The present study sets out to compare distributional preferences elicited from a social perspective and an individual ex ante perspective. We applied a web-based stated preference experiment constituting a random sample of 1,104 Danish citizens. Data was collected primo 2019 and merged with administrative data containing individual level information on socio-demographics and health care utilization in order to identify underlying preference determinants.
Comparing individual and social preferences, we find more support for equitable distributions of health care in the individual perspective, suggesting that individuals’ risk preferences are a stronger argument for distributing health care more equally, than social distributional preferences. Underlying determinants of differences in preferences across perspectives are risk aversion, health and age.
Lise Desireé Hansen, The Danish Center for Health Economics
3 December 2020, 14.00-15.00 (CET)
A sudden change in tariffs at a pre-defined point in the treatment can incentivize health care providers to prolong treatments to reach the higher tariff, and then to discharge patients once the higher tariff has been reached. The Dutch reimbursement schedule in hospital rehabilitation care follows a stepwise-function based on treatment duration. We investigated the presence of strategic discharges around the first threshold and assessed whether their occurrence varied by type of provider. Our findings suggest no response to incentives by traditional hospitals (general, academic hospitals, and rehabilitation centers), and strong response by smaller, more profit-oriented independent treatment centers. When examining the variation in response based on the financial position of the organization, we found a higher probability of manipulation among providers in financial distress. Our findings provide multiple insights and possible indicators to identify providers that may be more prone to strategic behavior.
10 December 2020, 9.00-10.00 (CET)
The standard practice with predicting health cost distributions is to employ a regression model with well-established risk factors. However, there is evidence that these models perform poorly in predicting uncommon events (ie, a small proportion of high-user costs) and do not take into account rich health and social administrative data that are increasingly available to health researchers. Data science and – in particular – machine learning, offers the potential improved and more accurate risk prediction, and utilises available information in an era of big data. Therefore the objective of this study was to predict high-cost users among people with an expensive condition (cardiovascular disease [CVD]) using machine learning methods.
We used national linked datasets for a whole high-income country (New Zealand). These included health (eg, hospitalisation, pharmaceutical dispensings, laboratory tests requested, diabetes) and social data (eg, birth country, educational level, income, deprivation etc). The model outcome was the most expensive quintile of cases (top 20%), among people with CVD over a one-year period. We compared the results of several different machine learning models with more traditional regression models. Sensitivity analyses with various thresholds of high-cost users were also conducted.
The machine learning models using administrative data had greater accuracy in predicting high health cost users than the traditional regression models, for application in the same dataset as the training or learning occurred in. In particular, the best machine learning model had a sensitivity of 78.9% vs 0.77% in the traditional regression model, and a harmony score between sensitivity and specificity of 34.5% vs 1.52%, respectively.
This study found that the use of machine learning with administrative datasets increased predictive power (ie, sensitivity) over traditional regression models. This approach may allow for improved high cost risk assessment at the individual level in settings where rich social data are linked to health data. It may also facilitate improved information for policy planning and resource allocation around targeting preventive and treatment efforts, and hence improve health and inform the saving of healthcare costs.
17 December 2020, 14.00-15.00 (CET)
A key challenge facing health systems is determining how best to manage care for high-need, high-cost (HNHC) populations. One meaningful way to identify best practices for managing HNHC populations, who often require care across multiple settings, is to perform a cross-country comparison that evaluates the care trajectory over time. In order to determine if this is feasible, we outline a methodological approach to identify four HNHC patient personas, or vignettes, that reflect priority populations as defined by the National Academy of Medicine. We explored the availability of data for identifying these patient personas across 12 high-income countries: Australia, Canada, England, France, Germany, the Netherlands, New Zealand, Norway, Spain, Sweden, Switzerland, and the United States. Across countries, different patient-level datasets were identified that capture analogous demographics, utilization and costs across seven care domains: primary care, outpatient specialty care, inpatient and emergency care, post-acute rehabilitative care, home health care, long-term care, and outpatient pharmaceuticals. Four countries can reliably examine patient-level linked care across all seven domains and three more countries can examine care across six domains with readily available data. Though our collaborative identified important challenges that limit international comparison of HNHC patients, it is reassuring that many countries can participate in a comparison that may yield important insights on care delivery for these complex populations.