From the Publisher:
Probabilistic Reasoning in Intelligent Systems is a complete andaccessible account of the theoretical foundations and computational methods that underlie plausible reasoning under uncertainty. The author provides a coherent explication of probability as a language for reasoning with partial belief and offers a unifying perspective on other AI approaches to uncertainty, such as the Dempster-Shafer formalism, truth maintenance systems, and nonmonotonic logic. The author distinguishes syntactic and semantic approaches to uncertaintyand offers techniques, based on belief networks, that provide a mechanism for making semantics-based systems operational. Specifically, network-propagation techniques serve as a mechanism for combining the theoretical coherence of probability theory with modern demands of reasoning-systems technology: modular declarative inputs, conceptually meaningful inferences, and parallel distributed computation. Application areas include diagnosis, forecasting, image interpretation, multi-sensor fusion, decision support systems, plan recognition, planning, speech recognitionin short, almost every task requiring that conclusions be drawn from uncertain clues and incomplete information.
Probabilistic Reasoning in Intelligent Systems will be of special interest to scholars and researchers in AI, decision theory, statistics, logic, philosophy, cognitive psychology, and the management sciences. Professionals in the areas of knowledge-based systems, operations research, engineering, and statistics will find theoretical and computational tools of immediate practical use. The book can also be used as an excellent text for graduate-level courses in AI, operations research, or applied probability.

Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference

PREAMBLE......e4

APPENDIX 1......e121

APPENDIX 2......e122

APPENDIX 3......e124

REFERENCES......e124

It is important that the medical profession play a significant role in critically evaluating the use of diagnostic procedures and therapies as they are introduced in the detection, management,

/pdf/acc-aha-2006-guidelines-for-the-management-of-patients-with-27qfh9fc7i.pdf

ACC/AHA 2006 Guidelines for the Management of Patients With Valvular Heart Disease

This book presents the definitive exposition of 'prospect theory', a compelling alternative to the classical utility theory of choice. Building on the 1982 volume, Judgement Under Uncertainty, this book brings together seminal papers on prospect theory from economists, decision theorists, and psychologists, including the work of the late Amos Tversky, whose contributions are collected here for the first time. While remaining within a rational choice framework, prospect theory delivers more accurate, empirically verified predictions in key test cases, as well as helping to explain many complex, real-world puzzles. In this volume, it is brought to bear on phenomena as diverse as the principles of legal compensation, the equity premium puzzle in financial markets, and the number of hours that New York cab drivers choose to drive on rainy days. Theoretically elegant and empirically robust, this volume shows how prospect theory has matured into a new science of decision making.

Choices, Values, and Frames

Users of clinical practice guidelines and other recommendations need to know how much confidence they can place in the recommendations Systematic and explicit methods of making judgments can reduce errors and improve communication We have developed a system for grading the quality of evidence and the strength of recommendations that can be applied across a wide range of interventions and contexts In this article we present a summary of our approach from the perspective of a guideline user Judgments about the strength of a recommendation require consideration of the balance between benefits and harms, the quality of the evidence, translation of the evidence into specific circumstances, and the certainty of the baseline risk It is also important to consider costs (resource utilisation) before making a recommendation Inconsistencies among systems for grading the quality of evidence and the strength of recommendations reduce their potential to facilitate critical appraisal and improve communication of these judgments Our system for guiding these complex judgments balances the need for simplicity with the need for full and transparent consideration of all important issues

/pdf/grading-quality-of-evidence-and-strength-of-recommendations-4juchvwoau.pdf

Grading quality of evidence and strength of recommendations.

We discuss the cognitive and the psy- chophysical determinants of choice in risky and risk- less contexts. The psychophysics of value induce risk aversion in the domain of gains and risk seeking in the domain of losses. The psychophysics of chance induce overweighting of sure things and of improbable events, relative to events of moderate probability. De- cision problems can be described or framed in multiple ways that give rise to different preferences, contrary to the invariance criterion of rational choice. The pro- cess of mental accounting, in which people organize the outcomes of transactions, explains some anomalies of consumer behavior. In particular, the acceptability of an option can depend on whether a negative outcome is evaluated as a cost or as an uncompensated loss. The relation between decision values and experience values is discussed. Making decisions is like speaking prose—people do it all the time, knowingly or unknowingly. It is hardly surprising, then, that the topic of decision making is shared by many disciplines, from mathematics and statistics, through economics and political science, to sociology and psychology. The study of decisions ad- dresses both normative and descriptive questions. The normative analysis is concerned with the nature of rationality and the logic of decision making. The de- scriptive analysis, in contrast, is concerned with peo- ple's beliefs and preferences as they are, not as they should be. The tension between normative and de- scriptive considerations characterizes much of the study of judgment and choice. Analyses of decision making commonly distin- guish risky and riskless choices. The paradigmatic example of decision under risk is the acceptability of a gamble that yields monetary outcomes with specified probabilities. A typical riskless decision concerns the acceptability of a transaction in which a good or a service is exchanged for money or labor. In the first part of this article we present an analysis of the cog- nitive and psychophysical factors that determine the value of risky prospects. In the second part we extend this analysis to transactions and trades. Risky Choice Risky choices, such as whether or not to take an umbrella and whether or not to go to war, are made without advance knowledge of their consequences. Because the consequences of such actions depend on uncertain events such as the weather or the opponent's resolve, the choice of an act may be construed as the acceptance of a gamble that can yield various out- comes with different probabilities. It is therefore nat- ural that the study of decision making under risk has focused on choices between simple gambles with monetary outcomes and specified probabilities, in the hope that these simple problems will reveal basic at- titudes toward risk and value. We shall sketch an approach to risky choice that

http://psycnet.apa.org/doi/10.1037/0003-066X.39.4.341

Decision analysis in clinical cardiology : when is coronary angiography required in aortic stenosis ?

Transient Ischemic Attach in a Man with Coronary Artery Disease Two Strategies Neck and Neck

Randomized, double-blind, controlled clinical trials are the gold standard for medical therapy. When a disease is rare or slowly progressive over many years, such a trial may not be feasible. Decision analysis provides a bridge between current studies with short-term surrogate markers and a large, longitudinal clinical trial. With decision analysis, results of current studies can be summarized and the outcome of a long-term study projected under explicit assumptions. Chronic hepatitis C is a disease that is slowly progressive, and the requirements of a longitudinal clinical study could be prohibitive. Therefore, we review the basic steps of decision analysis, apply these steps to two recent decision analyses evaluating the use of interferon-α2b (IFN) in the treatment of chronic hepatitis C, and discuss possible implications of decision modeling for this disease. The estimated marginal cost per year of life gained from IFN therapy ranged from approximately $3000 to $55,000 in these two studies. The wide range is based on different estimates of treatment costs and disease progression. This analysis has identified gaps in the current knowledge regarding the natural progression of hepatitis C and has established criteria to evaluate new developments and their impact on chronic hepatitis C.

Modeling therapeutic benefit in the midst of uncertainty: therapy for hepatitis C.

Background There are no randomized controlled trials to inform the decision of which cranial radiation therapy (CRT) strategy to apply to pediatric patients with T-cell acute lymphoblastic leukemia (ALL). Procedure We performed a decision analysis using a Markov model in which we compared the life expectancy and quality-adjusted life expectancy when administering one of three CRT strategies to a cohort of patients with T-cell ALL: (1) omission of CRT for all patients; (2) CRT only for those with evidence of leukemic involvement in the central nervous system at diagnosis (therapeutic strategy); or (3) CRT for all (prophylactic strategy). Results When considering plausible event-free survival rates and late mortality after cure for groups of pediatric patients with T-cell ALL, the strategies of omitting CRT, administering therapeutic CRT, and administering prophylactic CRT result in similar short-term (7-year) survival. When considering the increased contribution of deaths from late effects, the strategy of prophylactic CRT is associated with lower life expectancy when compared to the other two strategies. The Monte Carlo probabilistic sensitivity analysis demonstrated that the strategy of prophylactic CRT was the preferred strategy only 5% of the time. Conclusions Similar short-term survival may be expected when comparing the strategies of total omission of CRT, therapeutic CRT, and prophylactic CRT for patients with T-cell ALL. Long-term survival is likely inferior for the strategy of prophylactic CRT. The synthesis of nonrandomized trials and the application of decision analysis can help inform complex decision making in pediatric oncology.

Using nonrandomized studies to inform complex clinical decisions: The thorny issue of cranial radiation therapy for T-cell acute lymphoblastic leukemia

As shown in an accompanying article, the benefits of any broadly based screening program for ovarian cancer are limited more by the low prevalence of disease than by the potential benefits of treating early disease or the performance of currently available tests. Screening for such diseases is best restricted to special populations at markedly increased risk. In low-prevalence populations, false-positive results and the costs and risks of the workups they induce are magnified. In determining the marginal cost-effectiveness ratio of such programs, failure to consider even apparently minor factors can markedly affect the conclusions. In the case of screening for ovarian cancer, factoring in the discomfort and potential inconvenience of venipuncture and transvaginal sonography can eliminate the small benefit of the program. Taking the time and energy required to organize relevant data in constructing a decision analytic model can provide important insights into the usefulness of such screening programs. Screening for preclinical disease can lead to early intervention and thereby improve health outcomes and possibly save resources. Because screening is rarely free of cost and often carries risk, either from the test itself or from the subsequent workup it can induce, we must decide when and for whom screening is appropriate. A formal decision analysis can help structure the problem, organize data, elucidate tradeoffs, and estimate benefits and costs. In this issue of Annals, Schapira, Matchar, and Young [1] analyze the effect of screening 40-year-old women for early ovarian cancer. They conclude that screening would, on average, prolong life expectancy of each woman screened by only 14 hours and thus would not be an effective health policy. Although these investigators focused on the characteristics of screening tests and the risks of laparotomy, they correctly identify the crux of the problemthe low prevalence of ovarian cancer. Even assuming perfect test performance and ignoring the risks and morbidities of laparotomy, their model suggests that the benefit of screening remains trivialonly 1.4 daysa toss-up [2]. Recall that modifying lifestyle and risk factors for coronary heart disease is estimated to prolong the life expectancy of a 35-year-old by up to 1 year [3]; annual Papanicolaou smears for cervical cancer could add 3 months to a woman's life [4]; and regular screening for breast cancer may increase the life expectancy of 50-year-old women by 2 months [5]. Even if we required an average gain of only 30 days before endorsing a program, screening for ovarian cancer would be recommended only in populations where disease prevalence exceeds 1400/100 000, 50 times higher than the prevalence in unselected 40-year-old women. The conclusion that population-based screening for ovarian cancer would provide little benefit is all the more surprising because early intervention in ovarian cancer can markedly improve prognosis. A screening program using CA 125 and transvaginal sonography will extend the life expectancy of those patients who have preclinical disease by 11.6 years; a perfect screening test would extend their life expectancy by 26 years. However, the benefits for these women must be balanced against the adverse consequences of screening for the remaining women. Because these tests are imperfect, the number of patients submitted to workup because of false-positive results would far exceed the number of true positives. Screening 40-year-old women for ovarian cancer with CA 125 as a single test (specificity 97.6%) produces 340 false-positive results for each additional early cancer detected. Combining blood CA 125 levels with transvaginal sonography increases the specificity to 99.95% and produces only eight false-positive results for each additional early cancer detected. Clearly, maintaining very high specificity must be a critical goal in any program that screens low-prevalence populations [6]. From another perspective, knowing that a screening program could increase the life expectancy of a patient with occult ovarian cancer by 11.6 years, we might ask how much an unaffected woman would lose. On average, her life expectancy would be shortened by 25 minutes, reflecting the rare complication of laparotomy when results of both CA 125 level and the transvaginal sonogram are falsely positive. Other authors have recently concluded that screening for ovarian cancer is not appropriate unless groups at particularly high risk can be identified, because the performance of current tests is relatively poor [7] and the benefit of treatment is limited [8]. Why did Annals devote space to yet another paper that reaches the same conclusion? Despite assuming better test performance and perhaps best-case efficacy for early treatment, Schapira and coworkers conclude that the benefits would be trivial and would remain so even if better screening tests and better treatments are developed because of the low prevalence of preclinical ovarian cancer. The same logic would apply to screening for many low-prevalence diseases. In adults, no therapy, no matter how efficacious, can improve life expectancy beyond that of a healthy person of the same age. Even when considering diseases that are uniformly fatal in the short-term, the upper bounds of the average benefit of a perfect therapy varies from 5 years (at age 88) to 50 years (at age 20). Using a perfect test and a riskless workup to screen for an immediately fatal disease with a prevalence of 1/1000 could provide a maximum benefit to each person screened of only 2 to 18 days, even if detecting preclinical disease increased survival to full life expectancy. Using imperfect tests and workups with risk further limits the benefit of screening. Toss-ups present a variety of problems for both the clinician and the policy analyst. No analysis, formal or informal, can consider all factors. We explicitly or implicitly edit out minor issues. but when the expected difference between strategies is very small (14 hours in this case), such minor omissions can change our conclusions. For example, the authors of the current analysis ignore the morbidity of the venipuncture, the potential discomfort and embarrassment of the transvaginal sonogram, and the anxiety of waiting for test results. They also ignore the fact that most patients would prefer to defer risks, such as the potential complications of laparotomy, as long as possible. Some of these minor adjustments apply only to the 2.4% of the unaffected population who have a falsely elevated CA 125 level, whereas others apply to everyone screened. Explicit consideration of these factors will markedly diminish the expected 14-hour gain in survival and may wipe it out altogether. Many health policy analyses look at the health benefit of a proposed program first. If that benefit is negative or very small, analysts sometimes do not even consider the economics. Such partial analyses can, however, be misleading because economic factors can be the driving force, particularly if early detection markedly decreases the cost of treating patients with disease (as is the case for many neonatal screening programs). Although Schapira and coworkers have not undertaken a formal cost-effectiveness analysis, an informal back of the envelope analysis may be helpful. The actual variable costs for a CA 125 serologic test and a transvaginal sonogram are approximately $20 and $90, respectively (although charges are higher). If sonograms are performed only in patients with positive serologic results, the direct cost of screening one patient will be approximately $22. If only patients with positive results of both tests undergo laparotomy, the total of direct and induced costs will be approximately $25. Even if shifting patients from late to early ovarian cancer provides no savings in the cost of care, spending $25 to save 14 life-hours, when applied to a population, is equivalent to spending $15 000 to save 1 year of life. If we discount future benefits at 5%/year, the cost effectiveness ratio rises to $30 000 per year of life saved. Comparable ratios are $30 000 to $100 000 per year of life saved for annual breast cancer screening [5] and $15 000 to $30 000 per year of life saved for cervical cancer screening every 4 years [4]. If we could be confident that screening for ovarian cancer would improve life expectancy at a cost of $15 000 to $50 000 per life year saved, it would not be an unreasonable policy. but suggesting a program, no matter how apparently cost-effective, with an expected gain in life expectancy of only 14 hours per patient screened is troubling because no model can ever be complete. If small omissions wipe out the gain, then the cost-effectiveness ratio would become infinitely high. Was it worth performing a decision analysis on a question so well studied, or at least so broadly commented on in recent literature? The two main products of a formal analysis are recommendations and insights. After reading Schapira and coworkers' analysis, our conclusions about ovarian cancer screening should remain unchangedsuch programs cannot be recommended as public policy. but we can now appreciate that the driving force is the low prevalence of disease, not the particular characteristics of the CA 125 serologic analysis or transvaginal sonography. Screening for ovarian cancer should be limited to women at markedly increased risk. The dynamics of screening for ovarian cancer and other low-prevalence diseases are now clearer because Schapira and coworkers have taken the time and energy required to organize the data, estimate the effects of early detection, and provide a model that can be easily explored. The benefits of such careful analyses are clear and well worth the cost.

Stephen G. Pauker

Papers

Decision analysis in clinical cardiology : when is coronary angiography required in aortic stenosis ?

Transient Ischemic Attach in a Man with Coronary Artery Disease Two Strategies Neck and Neck

Modeling therapeutic benefit in the midst of uncertainty: therapy for hepatitis C.

Using nonrandomized studies to inform complex clinical decisions: The thorny issue of cranial radiation therapy for T-cell acute lymphoblastic leukemia

Deciding about screening.