Formal Decision Analysis Research Articles

Decision analysis for seabird recovery: Navigating complexity across ecosystems, balancing competing values and bridging spaces between research and implementation

AbstractConservation management decisions are challenging due to multiple, competing values (objectives) held by various parties, complex management options (alternatives), potential irreversible consequences, ubiquitous uncertainty and opaque governance structures. These complications are exacerbated when target species are highly mobile, utilise aquatic and terrestrial ecosystems and interact with human interests. Even when decisions are made, these challenges may still inhibit management implementation. Decision analyses offer solutions, but are rarely applied in conservation management, especially for highly mobile species across ecosystems.We applied a formal decision analysis for the conservation management of a highly mobile seabird (Kuaka; Whenua Hou Diving Petrel;Pelecanoides whenuahouensis) across ecosystems. Together with Māori (Indigenous Peoples of Aotearoa New Zealand), government staff and industry representatives, we identified seven fundamental objectives and 11 potential alternatives. We predicted consequences of alternatives across objectives using state‐of‐the‐art statistical models, including integrated population models, and expert elicitations hosted through Shiny apps. We then used various decision‐analytical tools to weigh trade‐offs and identify the preferred alternative.The preferred alternative across objectives consisted of a combination of mitigation of vessel‐based light pollution, plant and competition management and inter‐island translocations. This alternative was predicted to provide the most effective conservation outcomes across values and ecosystems and was rapidly implemented in a stepwise fashion following this decision analysis.Synthesis and application. We illustrate that decision analyses provide inclusive environments for all participants despite diverse backgrounds and views, enable rational navigation of complex decision landscapes and facilitate transparent identification of preferred management options across all fundamental values, including Māori value systems and industry concerns. Additionally, we show that decision analyses can rapidly bridge the spaces between research and implementation. Therefore, we argue that the application of decision analyses has great potential to improve the field of conservation.

To go or not to go in phase 2/3 oncology trials, a critical question with a unified answer

The decision on how fast to go from single-arm Phase 1 after efficacy single-detection to randomized-controlled Phase 2/3 in oncology drug development is multifaceted and complex, if not often contentious. To facilitate the process, we have incorporated all viable options including the under-utilized adaptive Phase 2/3 designs into a comprehensive decision matrix for completeness and transparency. Illustrated with the 2-in-1 adaptive Phase 2/3 design, a formal decision analysis that explicitly optimizes the tradeoff between cost and benefit reveals a surprisingly robust efficacy bar for Go-No Go transition. It implies that similar robust bars exist for other adaptive Phase 2/3 designs, which remove a major hurdle for their wider adoption in practice.

Decisions, uncertainty and spatial information

In this paper we review how the uncertainty in spatial information has been characterized. This includes both continuous predictions of spatial variables, and thematic maps of landcover classes. We contend that much work in this area has failed to engage adequately with the decision processes of the end-user of information, and that the engagement of spatial statisticians is essential to achieve this. We examine generalized measures of uncertainty, and those focussed on particular decision models. We conclude that the latter are likely to be the most fruitful, particularly if they emerge from a formal decision analysis. We outline the principles of value of information theory, and suggest that this represents an ideal framework in which to develop measures of uncertainty which can support both the rational collection of data and the interpretation of the resulting information.

Assessing the risks of changing ongoing management of endangered species

AbstractRecovery programmes for endangered species can become increasingly demanding over time, but managers may be reluctant to change ongoing actions that are believed to be assisting recovery. We used a quantitative risk assessment to choose support strategies for a reintroduced population of Mauritius olive white‐eyesZosterops chloronothos. Facing increasing costs, managers considered changing the ongoing supplementary feeding strategy, but at the same time worried this could jeopardize the observed positive population trend. We used a feeding experiment to compare the current feeding regime and a cheaper alternative (a simple sugar/water mix). Results suggested the cheaper alternative would only marginally reduce population vital rates. We assessed the influence of these results and the associated uncertainty on population recovery and management costs using two decision‐analytic criteria, incremental cost‐effectiveness ratio and stochastic dominance. The new feeding regime was expected to be, on average, more cost‐effective than thestatus quo. Moreover, even negative outcomes would only likely mean a slower growing population, not a declining one, whereas not changing feeding regime actually entailed greater risk. Because shifting from the current regime to a cheaper sugar/water mixture was both a risk‐averse and a cost‐effective choice, we decided to implement this change. Four years after the experiment, the population continues to grow and costs have been contained, matching predictions almost exactly. In this case, the field experiment provided useful empirical information about prospective actions; the risk analysis then helped us understand the real implications of changing the feeding regime. We encourage managers of recovery plans facing a similar situation to explicitly recognize trade‐offs and risk aversion, and address them by combining targeted research and formal decision analysis.

Coherent combination of probabilistic outputs for group decision making: an algebraic approach

Current decision support systems address domains that are heterogeneous in nature and becoming progressively larger. Such systems often require the input of expert judgement about a variety of different fields and an intensive computational power to produce the scores necessary to rank the available policies. Recently, integrating decision support systems have been introduced to enable a formal Bayesian multi-agent decision analysis to be distributed and consequently efficient. In such systems, where different panels of experts independently oversee disjoint but correlated vectors of variables, each expert group needs to deliver only certain summaries of the variables under their jurisdiction, derived from a conditional independence structure common to all panels, to properly derive an overall score for the available policies. Here we present an algebraic approach that makes this methodology feasible for a wide range of modelling contexts and that enables us to identify the summaries needed for such a combination of judgements. We are also able to demonstrate that coherence, in a sense we formalize here, is still guaranteed when panels only share a partial specification of their model with other panel members. We illustrate this algebraic approach by applying it to a specific class of Bayesian networks and demonstrate how we can use it to derive closed form formulae for the computations of the joint moments of variables that determine the score of different policies.

Predicting reintroduction outcomes for highly vulnerable species that do not currently coexist with their key threats.

Predicting reintroduction outcomes before populations are released is inherently challenging. It becomes even more difficult when the species being considered for reintroduction no longer coexists with the key threats limiting its distribution. However, data from other species facing the same threats can be used to make predictions under these circumstances. We used an integrated Bayesian modeling approach to predict growth of a reintroduced population at a range of predator densities when no data are available for the species in the presence of that predator. North Island Saddlebacks (Philesturnus rufusater) were extirpated from mainland New Zealand by exotic mammalian predators, particularly ship rats (black rats [Rattus rattus]) but are now being considered for reintroduction to sites with intensive predator control. We initially modeled data from previous saddleback reintroductions to predator-free sites to predict population growth at a new predator-free site while accounting for random variation in vital rates among sites. We then predicted population growth at different rat-tracking rates (an index of rat density) by incorporating a previously modeled relationship between rat-tracking and vital rates of another predator-sensitive species, the North Island Robin (Petroica longipes), and accounted for greater vulnerability of saddlebacks to rat predation based on information on historical declines of both species. The results allowed population growth to be predicted as a function of management effort while accounting for uncertainty, which means formal decision analysis could be used to decide whether to proceed with a reintroduction. Similar approaches could be applied to other situations where data on the species of interest are limited and provide an alternative to decision making based solely on expert judgment.

Quantifying patient preferences for symptomatic breast clinic referral: a decision analysis study

ObjectivesDecision analysis study that incorporates patient preferences and probability estimates to investigate the impact of women’s preferences for referral or an alternative strategy of watchful waiting if faced with symptoms...

A Bayesian Nonparametric Approach to Testing for Dependence Between Random Variables

Nonparametric and nonlinear measures of statistical dependence between pairs of random variables are important tools in modern data analysis. In particular the emergence of large data sets can now support the relaxation of linearity assumptions implicit in traditional association scores such as correlation. Here we describe a Bayesian nonparametric procedure that leads to a tractable, explicit and analytic quantification of the relative evidence for dependence vs independence. Our approach uses Polya tree priors on the space of probability measures which can then be embedded within a decision theoretic test for dependence. Polya tree priors can accommodate known uncertainty in the form of the underlying sampling distribution and provides an explicit posterior probability measure of both dependence and independence. Well known advantages of having an explicit probability measure include: easy comparison of evidence across different studies; encoding prior information; quantifying changes in dependence across different experimental conditions, and the integration of results within formal decision analysis.

Comparative cost-effectiveness of non-invasive imaging tests in patients presenting with chronic stable chest pain with suspected coronary artery disease: a systematic review.

Coronary artery disease (CAD) remains one of the leading causes of morbidity and mortality globally. The most cost-effective imaging strategy to diagnose CAD in patients with stable chest pain is however uncertain. To review the evidence on comparative cost-effectiveness of different imaging strategies for patients presenting with stable chest pain symptoms suggestive for CAD. Systematic review. Studies performing a formal economic evaluation or decision analysis in the English language published between January 1995 and December 2015 were identified using PubMed, Medline (OvidSP), Embase, Web of Science, Cochrane economic evaluations library, and EconLit. Reviews and meta-analyses were excluded. Two independent reviewers assessed titles and abstracts. Of the 4498 titles identified, 70 met our selection criteria. One reviewer used a modified version of the CHEERS checklist to assess study quality. One reviewer extracted data on study details, which were checked by a second reviewer. There is a major heterogeneity between the available cost-effectiveness studies included in this study. The included studies compared very different testing strategies in very different ways and provided mostly short-term results. Strategies of no-testing and xECG were underrepresented. Nonetheless, the findings from this systematic review suggest that for patients with a low to intermediate prior probability of having obstructive CAD, computed tomography coronary angiography (CTCA) may be cost-effective as an initial diagnostic imaging test in comparison with CAG or other non-invasive diagnostic tests. If functional testing is required, stress echocardiography (SE) or single-photon emission computed tomography (SPECT) are suggested to be cost-effective initial strategies in patients with intermediate prior probability of CAD. Yet, other functional testing strategies such as xECG and positron-emission tomography (PET) scanning have not been studied as intensely. Immediate CAG is suggested to be a cost-effective strategy for patients at a high prior probability of having obstructive CAD whom may benefit from revascularization. The study emphasizes the inextricable link between clinical effectiveness and economic efficiency. Evidence suggests that the optimal diagnostic imaging strategy for individuals suspected of having CAD is CTCA for low and intermediate disease probability, followed by SE or SPECT as necessary, and invasive CAG for high disease probability. Further studies are needed to evaluate the cost-effectiveness of alternative non-invasive tests, including a no-testing strategy.

The CAR Method for Using Preference Strength in Multi-criteria Decision Making

Multi-criteria decision aid (MCDA) methods have been around for quite some time. However, the elicitation of preference information in MCDA processes, and in particular the lack of practical means supporting it, is still a significant problem in real-life applications of MCDA. There is obviously a need for methods that neither require formal decision analysis knowledge, nor are too cognitively demanding by forcing people to express unrealistic precision or to state more than they are able to. We suggest a method, the CAR method, which is more accessible than our earlier approaches in the field while trying to balance between the need for simplicity and the requirement of accuracy. CAR takes primarily ordinal knowledge into account, but, still recognizing that there is sometimes a quite substantial information loss involved in ordinality, we have conservatively extended a pure ordinal scale approach with the possibility to supply more information. Thus, the main idea here is not to suggest a method or tool with a very large or complex expressibility, but rather to investigate one that should be sufficient in most situations, and in particular better, at least in some respects, than some hitherto popular ones from the SMART family as well as AHP, which we demonstrate in a set of simulation studies as well as a large end-user study.

Inferior vena cava filter retrieval provides no advantage in the average patient.

Inferior vena cava (IVC) filter placement is performed to mitigate the risk of pulmonary embolism (PE) when anticoagulation is contraindicated or ineffective. Technical advances now allow catheter-based filter retrieval. Many believe the benefits of retrieval are self-evident, yet retrieval carries an inherent complication risk and cost. The purpose of this study was to quantitatively weigh the risks and benefits of IVC filter retrieval using formal decision analysis. A Markov state-transition model was used to simulate two clinical scenarios: to leave a previously placed IVC filter or to retrieve it. Analysis was performed during the lifetime of the individual, and outcomes were expressed in quality-adjusted life-years (QALYs). The base case is a 60-year-old man with a filter placed within 3 months who no longer requires mechanical thromboprophylaxis. Potential events included PE, filter complications, and death from all other causes during each cycle. Tolls were used to incorporate the disutility of short-term treatment for PE and filter complications. For the base case and sensitivity analyses, we used utilities and probabilities derived from the literature. In the base case scenario, leaving the filter in place was preferred to filter retrieval, yielding 22.3 vs 21.9 QALYs. One-way sensitivity analysis demonstrated that filter retrieval may be preferable if the utility of living with a filter is <0.98. For all probabilities of retrieval and PE mortality, leaving the filter in place is preferred. Leaving a previously placed IVC filter provides a 0.4 QALY benefit over retrieving the filter for the average patient. This decision is sensitive to the utility of living with the IVC filter.

Endovascular Stent Graft Exclusion of Thoracic Arteriovenous Malformation in a Patient With Cobb Syndrome

this study was to quantitatively weigh the risks and benefits of IVC filter retrieval using formal decision analysis. Methods: A Markov state-transition model was used to simulate two clinical scenarios: to leave a previously placed IVC filter or to retrieve it. Analysis was performed over the lifetime of the individual, and outcomes were expressed in quality-adjusted life years (QALYs). The base case is a 60-year-old man with a filter placed within 3 months who no longer requires mechanical thromboprophylaxis. Potential events included PE, filter complications, and death from all other causes during each cycle. Tolls were used to incorporate the disutility of short-term treatment for PE and/or filter complications. For the base-case and sensitivity analyses, we used utilities and probabilities derived from the literature. Results: In the base-case scenario, leaving the filter in place was preferred to filter retrieval, yielding 22.3 vs 21.9 QALYs. One-way sensitivity analysis demonstrated that filter retrieval might be preferable if the utility of living with a filter is <0.98. For all probabilities of retrieval and PE mortality, leaving the filter in place is preferred. Conclusions: Leaving a previously placed IVC filter provides a 0.4 QALY benefit over retrieving the filter for the average patient. This decision is sensitive to the utility of living with the IVC filter and underlying PE risk.

PS180. Open Surgical Reconstruction for Benign Superior Vena Cava Syndrome

(PE) when anticoagulation is contraindicated or ineffective. Technical advances now allow catheter-based filter retrieval. Many believe the benefits of retrieval are selfevident, yet retrieval carries an inherent complication risk and cost. The purpose of this study was to quantitatively weigh the risks and benefits of IVC filter retrieval using formal decision analysis. Methods: A Markov state-transition model was used to simulate two clinical scenarios: to leave a previously placed IVC filter or to retrieve it. Analysis was performed over the lifetime of the individual and outcomes were expressed in quality-adjusted life-years (QALYs). The base case is a 60-year-old man with a filter placed #3 months who no longer requires mechanical thromboprophylaxis. Potential events included PE, filter complications, and death from all other causes during each cycle. Tolls were used to incorporate the disutility of short-term treatment for PE and/or filter complications. For the base case and sensitivity analyses, we used utilities and probabilities derived from the literature. Results: In the base-case scenario, leaving the filter in place was preferred to filter retrieval, yielding 22.3 vs 21.9 QALYs. One-way sensitivity analysis demonstrated that filter retrieval may be preferable if the utility of living with a filter is <0.98. For all probabilities of retrieval and PE mortality, leaving the filter in place is preferred. Conclusions: Leaving a previously placed IVC filter provides a 0.4 QALY benefit over retrieving the filter for the average patient. This decision is sensitive to the utility of living with the IVC filter and underlying PE risk.

Resolving coastal conflicts using marine spatial planning

We applied marine spatial planning (MSP) to manage conflicts in a multi-use coastal area of Kenya. MSP involves several steps which were supported by using geographical information systems (GISs), multi-criteria decision analysis (MCDA) and optimization. GIS was used in identifying overlapping coastal uses and mapping conflict hotspots. MCDA was used to incorporate the preferences of user groups and managers into a formal decision analysis procedure. Optimization was applied in generating optimal allocation alternatives to competing uses. Through this analysis three important objectives that build a foundation for future planning of Kenya's coastal waters were achieved: 1) engaging competing stakeholders; 2) illustrating how MSP can be adapted to aid decision-making in multi-use coastal regions; and 3) developing a draft coastal use allocation plan. The successful application of MSP to resolve conflicts in coastal regions depends on the level of stakeholder involvement, data availability and the existing knowledge base.

From the Editors

From the Editors

Clinical model to estimate the pretest probability of malignancy in patients with pulmonary focal Ground-glass Opacity.

Effective strategies for managing patients with pulmonary focal Ground-glass Opacity (fGGO) depend on the pretest probability of malignancy. Estimating a clinical probability of malignancy in patients with fGGOs can facilitate the selection and interpretation of subsequent diagnostic tests. METHODS : Data from patients with pulmonary fGGO lesions, who were diagnosed at Sun Yat-sen University Cancer Center, was retrospectively collected. Multiple logistic regression analysis was used to identify independent clinical predictors for malignancy and to develop a clinical predictive model to estimate the pretest probability of malignancy in patients with fGGOs. One hundred and sixty-five pulmonary fGGO nodules were detected in 128 patients. Independent predictors for malignant fGGOs included a history of other cancers (odds ratio [OR], 0.264; 95% confidence interval [CI], 0.072 to 0.970), pleural indentation (OR, 8.766; 95% CI, 3.033-25.390), vessel-convergence sign (OR, 23.626; 95% CI, 6.200 to 90.027) and air bronchogram (OR, 7.41; 95% CI, 2.037 to 26.961). Model accuracy was satisfactory (area under the curve of the receiver operating characteristic, 0.934; 95% CI, 0.894 to 0.975), and there was excellent agreement between the predicted probability and the observed frequency of malignant fGGOs. We have developed a predictive model, which could be used to generate pretest probabilities of malignant fGGOs, and the equation could be incorporated into a formal decision analysis.

PRM208 - Using Multi-Criteria Decision Analysis to Support Allocation Decisions in Large Translational Research Projects

Large translational research projects often have abstract objectives, such as reducing the burden of disease and health care costs from type-2 diabetes (T2DM). Such an abstract objective entails: i) a very large number of possible strategies to reach the objective and ii) a lack of detailed data and high levels in uncertainty. Currently, no methods to support project selection and resource allocation decisions in such a setting are available. As a case study, we supported a resource allocation decision for the remaining funds in a large Dutch translational research consortium with the aforementioned objective, and compared the results to the decision made at the start of the project. We used the problem structuring, model building methods from multi-criteria decision analysis to identify four different alternative research strategies, and a set of evaluation criteria. Consequently, we used a combination of judgment from experts involved in the project and previously published data on the burden of disease and health care costs to evaluate the alternatives. Finally, a decision analysis was performed using Stochastic Multicriteria Acceptability Analysis for ordinal data (SMAA-O), which allows for the combined use of quantitative and qualitative (ranked) data. Using our method, it was decided to allocate remaining resources to the identification of biomarkers and development of technologies that can be used in the prevention of macrovascular complications in T2DM patients. This decision differed from the one made at the start of the project, which was not supported by any formal decision analysis. Our study shows that our method using SMAA-O can be a practical and valuable tool to support decisions on the allocation of research funds within large translational research consortia.

Demographics of reintroduced populations: Estimation, modeling, and decision analysis

Reintroduction canbe necessary for recovering populations of threatened species. However, the success of reintroduction efforts has been poorer than many biologists and managers would hope. To increase the benefits gained from reintroduction, management decision making should be couched within formal decision-analytic frameworks. Decision analysis is a structured process for informing decision making that recognizes that all decisions have a set of components—objectives, alternative management actions, predictive models, and optimization methods—that can be decomposed, analyzed, and recomposed to facilitate optimal, transparent decisions. Because the outcome of interest in reintroduction efforts is typically population viability or related metrics, models used in decision analysis efforts for reintroductions will need to include population models. In this special section of the Journal of Wildlife Management, we highlight examples of the construction and use of models for informing management decisions in reintroduced populations. In this introductory contribution, we review concepts in decision analysis, population modeling for analysis of decisions in reintroduction settings, and future directions. Increased use of formal decision analysis, including adaptive management, has great potential to inform reintroduction efforts. Adopting these practices will require close collaboration among managers, decision analysts, population modelers, and field biologists. 2013 The Wildlife Society.

Color-Flow Duplex Screening for Upper Extremity Proximity Injuries: A Low-Yield Strategy for Therapeutic Intervention

Although the incidence of injury to the upper extremity screened with angiography as a result of proximity penetrating trauma is similar to that of the lower extremity, intervention rates seem to be higher. However, studies evaluating the incidence of injury as a result of proximity penetrating trauma have primarily focused on the lower extremity. This study shows the incidence and clinical significance of vascular injury as a result of proximity trauma to the upper extremity in a large cohort of patients screened with color-flow duplex. A retrospective study was conducted from January 1, 2005 to January 1, 2012 on all patients undergoing color-flow duplex as a result of proximity penetrating trauma to the upper extremity. Data on injury location, mechanism, associated extremity and nonextremity injuries, and use and results of color-flow duplex were recorded and analyzed. A total of 341 patients were identified who underwent color-flow duplex because of proximity penetrating trauma to the upper extremity. Injuries occurred in 370 extremities, with 253 located in the upper arm and 117 in the forearm. Overall, 18 (4.9%) injuries were identified on screening duplex ultrasound, of which 12 (3.2%) were arterial and 5 (1.4%) were venous. The therapeutic intervention rate for detected injuries to the upper arm was 1.6% (4/253), whereas no injuries of the forearm were identified that necessitated intervention. Although color-flow duplex is an inexpensive and noninvasive means of detecting injuries as a result of proximity penetrating trauma, screening upper extremity wounds is unlikely to detect clinically significant arterial injuries in need of therapeutic intervention. Venous injuries in the form of deep venous thromboses were detected in only 1.4% of patients. These findings suggest that screening for proximity penetrating trauma of the upper extremity is unlikely to detect injuries at a rate that would prove cost-effective on formal decision analysis.

Advancing the Guánica Bay (Puerto Rico) Watershed Management Plan

Consideration of stakeholder values in watershed planning and management is a necessity, but sufficiently eliciting, understanding, and organizing those values can be daunting. Many studies have demonstrated the usefulness of formal decision analysis to integrate expert knowledge with stakeholder values. Here, some decision analysis structuring tools were used to advance a proposed watershed management plan toward a broader recognition of values in the watershed and adjacent coastal and marine systems. The U.S. Interagency Coral Reef Task Force selected Guánica Bay, Puerto Rico as a site to explore watershed management for protection of coral reefs. A Watershed Management Plan (WMP) was introduced in 2008 that led to subsequent activities of several federal and commonwealth agencies and nongovernmental organizations. Using information from the WMP and these other organizations, qualitative decision analysis structuring methods were applied to gain insight into desired and undesired prospects. The steps included selection of a decision context and overall objective, identifying fundamental objectives, and exploring means to achieve them. This initial organization, to be better informed by stakeholder deliberation and review, provided an approach to consider watershed and marine coastal issues simultaneously.