ORAL ABSTRACTS: APPLIED HEALTH ECONOMICS
The economic evaluation, performed from a health care perspective, took the form of a cost-effectiveness analysis in which the costs and effects were estimated for a planned birth in a birth centre (n=1668), hospital (n=701) and at home (n=1086). Separate analyses were performed for different types of birth centres, based on location and integration profile. The primary clinical outcomes were the optimality index (OI) a tool to measure ‘maximum outcome with minimal intervention’, containing both process and outcome items and a composite adverse outcome (CAO) score, which is a combined measure of adverse outcomes (including mortality and admission to the neonatal intensive care unit).
Multiple regression was used to estimate the differences in total cost and clinical outcomes and to adjust for potential confounders. Non-parametric bootstrapping was used to calculate uncertainty around all costs and effects estimates.
No clinically relevant differences in clinical outcomes were found between planned births in a birth centre, hospital and at home. Within the types of birth centres only the OI score of nulliparous women with a planned birth in a freestanding birth centre was clinical relevantly better (p<0.001) compared to a planned birth in an alongside birth centre. The total adjusted mean costs for births planned in a birth centre, hospital and at home were respectively €3326, €3330 and €2998. Focusing on the different types of birth centres, the total adjusted mean costs for births planned in a freestanding birth centre were €3531, alongside €3342 and on-site €3399 and costs ranged from €3283 to €3385 dependent on integration profile.
There were no differences in costs and effects for women at low-risk of complications with a planned birth under supervision of a community midwife in a birth centre and in a hospital. For nulliparous and multiparous women at low-risk of complications, planned birth at home was the most cost-effective option compared to planned birth in a birth centre and in a hospital. There were no differences in costs and effects among the types of birth centres.
Method(s): The 4 Pillars intervention uses primary care practice-chosen components, including immunization standing orders, vaccination access improvements, tracking tools, and a practice immunization champion to improve vaccination rates. This intervention was compared to control in 2 US cities among diverse populations and practices. In <65-year-old adults, 2 vaccines were targeted: influenza and Tdap (tetanus, diphtheria, acellular pertussis). A decision tree model was used to estimate intervention cost-effectiveness compared to control, with outcomes as costs/QALY gained. Trial data supplied changes in vaccination rates and intervention implementation/maintenance costs. US databases and literature data were used to model vaccine effectiveness, illness rates, and costs with/without vaccination over a 10 year time horizon. Future costs and effectiveness were discounted at 3% per year.
Result(s): Total vaccination and illness costs with the intervention (cost $1.78 per eligible patient per year), were $46.09 higher compared to control while gaining 0.001 QALYs, or $43,590/QALY gained. The intervention was not favored, at a $100,000/QALY threshold, when varying these influenza-related parameters: yearly attack rate <2.5% (base case 6.6%), case-fatality <0.048% (base 0.134%), influenza vaccine effectiveness <26.2% (base 59%), or program-related absolute increase in influenza vaccination <2.0% (base 8.6%). Results were insensitive to plausible individual variation of all other parameters, including absolute improvement in pertussis vaccination rate (11.8%). In a probabilistic sensitivity analysis, the 4 Pillars intervention was favored in 54.4% of model iterations at a $50,000/QALY acceptability threshold and in 90.8% at $100,000/QALY. In a separate scenario analysis, the intervention became cost saving if the total economic burden of influenza was >$1967 per case (base $265).
Conclusion(s): The 4 Pillars Immunization Toolkit was an economically reasonable intervention to improve vaccination rates in adults <65 years old.
Method(s): We propose a framework for describing co-dependent technologies that consists of three tests (genotypic diagnosis, phenotypic expression and therapy responder status) and a treatment. Based upon the presence of the condition of interest, the second and third tests characterize the ability to respond to therapy and the phenotypic expression – which places a limit on the ability to benefit from therapy – respectively. Three decision variables are identified – the cut-point for the probability of responding to therapy, the cut-point for the phenotypic expression that leads to treatment and the willingness to pay for health gain. The effectiveness of the therapy in responders and non-responders is determined exogenously.
Result(s): Our analysis shows that for a given probability of response, the optimal cut-point for the phenotypic expression is identified as the point at which the benefits for a responding patient means the patient is indifferent between the new treatment and standard care. We present a series of analyses exploring the relationship between the distributions of the probability of responding to therapy, phenotypic expression and the net benefit from the new technology.
Conclusion(s): Our analyses demonstrate that the benefit from the adoption of precision medicine technologies can be optimized by treating response probability and phenotypic expression as decision variables not exogenously determined parameters.
Method(s): Linear programming, a mathematical optimisation technique, was used to identify the optimal parental information approach to as part of the NBSP informed consent process. Responses from a discrete choice experiment (n = 702) comprising 4 attributes (3 process and capability to make an informed decision) were used to populate an objective function that aimed to maximise capability to make an informed decision. The optimal solution was constrained so that monetary and time costs of information provision can be no greater than those of current practice.
Result(s): The mathematical optimisation results suggested that the types of information given to parents should differ from current practice. Information should be provided during a woman’s pregnancy in an individual discussion supported by a leaflet, rather than after the baby is born, as in the current programme. Adopting this approach could increase parents’ capability to make an informed decision by 145% whilst saving the NHS £50,000 per year and 5097 hours of midwife time.
Conclusion(s): Optimisation techniques, when used alongside cost-effectiveness analysis, have the potential to enhance medical decision making, specifically, when health system capacity constraints are important.
Method(s): The Cost-Effectiveness Model Output (CEMO) tool was developed in Microsoft Excel. Analysts must enter information about their model, including the strategies, cost perspectives considered (e.g., health sector, societal), effectiveness outcomes considered (e.g., QALYs, life years), number of Markov cycles, time horizon, parameters varied in sensitivity analyses, and the number of Monte Carlo simulations used in probabilistic analysis. The tool uses this information to create a customized ‘Model Results’ worksheet that is unique to the analyst’s model. Analysts can use their preferred software to develop their model, before exporting the raw results into the Model Results worksheet. The tool then uses these raw results to calculate present values (if necessary – the tool supports differential and/or non-constant discounting if required), estimates of net health benefit (NHB) and net monetary benefit (NMB), the expected value of perfect information (EVPI), and estimates of the expected value of partial perfect information (EVPPI) for each parameter. The tool also automatically generates a standardized set of tables and figures for the analyst to report to decision makers.
Result(s): The CEMO tool provides the following outputs: (1) for deterministic and probabilistic analyses, separate tables of costs and effects, incremental costs and effects, incremental cost-effectiveness ratios (ICERs), NHB, NMB, the ranking of strategies by cost-effectiveness, and the probability that each strategy is cost-effective (probabilistic analysis only); (2) plots on the cost-effectiveness plane for all analyses; (3) results tables and ‘tornado’ graphs for one-way sensitivity analyses; (4) results tables for two-way sensitivity analyses; (5) cost-effectiveness acceptability curves (CEACs) and the cost-effectiveness acceptability frontier (CEAF) for probabilistic analyses; (6) tables and figures reporting EVPI and EVPPI.
Conclusion(s): The CEMO tool reduces the burden on analysts who conduct CEAs and improves the consistency of the data considered by decision makers.
Method(s): We use a simple model to simulate a set of hypothetical interventions, each with a given level of costs, health effects (which may be quantified in quality-adjusted life-years [QALYs] or some other measure) and a second beneficial attribute. A budget constraint is assumed such that not all available interventions can be funded. We identify the set of interventions that forms a production possibilities frontier (PPF) of treatment combinations that maximises combinations of QALYs and the second attribute for the budget constraint. This frontier determines the optimal set of interventions for the constrained budget depending on the rate the decision maker chooses to trade-off QALYs against the second attribute. We then apply an MCDA approach that attaches weights to the level of the second attribute and the net health benefit of each intervention according to a range of cost-effectiveness threshold values. For each set of weights and thresholds we identify the interventions that maximise the MCDA score. These interventions are described as the MCDA-preferred interventions. We compare the MCDA-preferred interventions to the PPF. The model is complemented with a brief survey of the literature on the methodology and application of MCDA in healthcare resource allocations.
Result(s): We find that the MCDA approach generally does not select interventions that lie on the PPF. Accordingly, the MCDA approach will not always maximise desirable outcomes.
Conclusion(s): There is growing interest in using MCDA to guide healthcare resource allocation. Although the methods for MCDA as a guide to healthcare resource allocation have yet to be fully described, it has been suggested that MCDA should replace cost-effectiveness analysis. The results presented here show that MCDA can result in sub-optimal resource allocation. This does not necessarily imply that MDCA is not useful. Indeed, it can be helpful in clarifying decision makers' objectives. Nevertheless, advocates of MCDA should carefully consider its limitations before recommending its use. Further exploration of MCDA's limitations will inform how attributes other than costs and QALYs can be appropriately integrated in healthcare resource allocation.