Medical Decision Making

Chapter 1: Introduction

• • 1.1 How may I be thorough yet efficient when considering the possible causes of my patient’s problems?
• • 1.2 How do I characterize the information I have gathered during the medical interview and physical examination?
• • 1.3 How do I interpret new diagnostic information?
• • 1.4 How do I select the appropriate diagnostic test?
• • 1.5 How do I choose among several risky treatment alternatives?

Chapter 2: Differential diagnosis

• • 2.1 An introduction
• • 2.2 How clinicians make a diagnosis
• • 2.3 The principles of hypothesis-driven differential diagnosis
• • 2.3.1 The first step in differential diagnosis: listening and generating hypotheses
• • 2.3.2 The second step in differential diagnosis: gathering data to test hypotheses
• • 2.3.3 Hypothesis testing
• • 2.3.4 Selecting a course of action
• • 2.4 An extended example
• • 2.4.1 Clinical aphorisms
• • Bibliography

Chapter 3: Probability: quantifying uncertainty

• • 3.1 Uncertainty and probability in medicine
• • 3.1.1 The uncertain nature of clinical information
• • 3.1.2 Definition and key concepts
• • 3.1.3 The meaning of probability: the present state vs. a future event
• • 3.1.4 Odds: an alternative way to express a probability
• • 3.2 How to determine a probability
• • 3.2.1 Probability: a quantification of judgment about the likelihood of an event
• • 3.2.2 Indirect probability assessment
• • 3.2.3 Direct probability assessment
• • 3.3 Sources of error in using personal experience to estimate the probability
• • 3.3.1 Heuristics defined
• • 3.3.2 Heuristic I: representativeness
• • 3.3.3 Heuristic II: availability
• • 3.3.4 Heuristic III: anchoring and adjustment
• • 3.3.5 Correctly using heuristics for estimating probability
• • 3.4 The role of empirical evidence in quantifying uncertainty
• • 3.4.1 Determining probability from the prevalence of disease in patients with a symptom, physical finding, or test result
• • 3.4.2 Determining the probability of a disease from its prevalence in patients with a clinical syndrome
• • 3.4.3 Establishing a probability using a clinical prediction model
• • 3.5 Limitations of published studies of disease prevalence
• • 3.5.1 Caution in using published reports to determine probability
• • 3.6 Taking the special characteristics of the patient into account when determining probabilities
• • Bibliography

Chapter 4: Interpreting new information: Bayes’ theorem

• • 4.1 Introduction
• • 4.2 Conditional probability defined
• • 4.3 Bayes’ theorem
• • 4.3.1 Derivation of Bayes’ theorem
• • 4.3.2 Clinically useful forms of Bayes’ theorem
• • 4.4 The odds ratio form of Bayes’ theorem
• • 4.4.1 The derivation of the odds ratio form of Bayes’ theorem
• • 4.4.2 The likelihood ratio: a measure of test discrimination
• • 4.4.3 Using the odds ratio form of Bayes’ theorem
• • 4.5 Lessons to be learned from using Bayes’ theorem
• • 4.5.1 Further thoughts
• • 4.5.2 The clinical significance of test specificity
• • 4.5.3 The clinical significance of test sensitivity
• • 4.6 The assumptions of Bayes’ theorem
• • 4.7 Using Bayes’ theorem to interpret a sequence of tests
• • 4.8 Using Bayes’ theorem when many diseases are under consideration
• • Bibliography

Chapter 5: Measuring the accuracy of clinical findings

• • 5.1 A language for describing test results
• • 5.1.1 Defining a test result
• • 5.2 The measurement of diagnostic test performance
• • 5.2.1 How to measure test performance
• • 5.2.2 Measures of concordance between index test and disease state
• • 5.2.3 Measures of discordance between index test and disease state
• • 5.2.4 Predictive value
• • 5.3 How to measure diagnostic test performance: a hypothetical example
• • 5.3.1 Description of the study
• • 5.3.2 Description of results
• • 5.3.3 An important limitation of the spleen scan study
• • 5.4 Pitfalls of predictive value
• • 5.5 How to perform a high quality study of diagnostic test performance
• • 5.5.1 The features of a high-quality prospective study of a diagnostic test
• • 5.5.2 Study characteristics that help ensure that the results apply to usual practice
• • 5.5.3 Study characteristics that insure unbiased, reproducible interpretation of the index test and the gold standard test
• • 5.6 Spectrum bias in the measurement of test performance
• • 5.6.1 The first phase of test evaluation: testing the“sickest of the sick” and the “wellest of the well”
• • 5.6.2 The second phase of test evaluation: reluctance to order the gold standard test because of over-confidence in a negative index test result
• • 5.6.3 Effects of spectrum bias
• • 5.6.4 Adjusting for biased estimates of sensitivity and specificity
• • 5.6.5 Heuristics for adjusting published reports for disease severity bias
• • 5.7 When to be concerned about inaccurate measures of test performance
• • 5.8 Test results as a continuous variable: the ROC curve
• • 5.8.1 The distribution of test results in diseased and well individuals
• • 5.8.2 The receiver operating characteristic curve
• • 5.8.3 Using the ROC curve to compare tests
• • 5.8.4 Setting the cut point for a test
• • 5.9 Combining data from studies of test performance: the systematic review and meta-analysis
• • A.5.1 Appendix: derivation of the method for using an ROC curve to choose the definition of an abnormal test result
• • Bibliography

Chapter 6: Decision trees– representing the structure of a decision problem

• • 6.1 Introduction
• • 6.2 Key concepts and terminology
• • 6.2.1 Final outcomes
• • 6.2.2 Branch probabilities and outcome probabilities
• • 6.2.3 Expected value calculations and life expectancy
• • 6.3 Constructing the decision tree for a hypothetical decision problem
• • 6.4 Constructing the decision tree for a medical decision problem
• • 6.4.1 Management of coronary artery disease overview
• • 6.4.2 Simple decision in the management of coronary artery disease
• • 6.4.3 Determining the branch probabilities
• • 6.4.4 Alternate chance node ordering
• • 6.4.5 Computing the life expectancy for the decision alternatives
• • Epilogue
• • Bibliography

Chapter 7: Decision tree analysis

• • 7.1 Introduction
• • 7.2 Folding-back operation
• • 7.2.1 Folding-back operation applied to hypothetical problem
• • 7.2.2 Chance node ordering revisited
• • 7.2.3 Two-stage decision in the management of coronary artery disease
• • 7.2.4 Decision tree for two-stage coronary artery disease management decision
• • 7.2.5 Folding-back operation applied to two-stage coronary artery disease decision problem
• • 7.2.6 Conclusion of the folding-back operation
• • 7.2.7 Comment on number of significant figures used in calculations
• • 7.3 Sensitivity analysis
• • 7.3.1 One-way sensitivity analysis for simple decision problems
• • 7.3.2 Two-way sensitivity analysis for simple decision problems
• • 7.3.3 Sensitivity analysis for problems with two decisions
• • 7.3.4 Sensitivity analysis and clinical policies
• • Epilogue
• • Bibliography

Chapter 8: Outcome utility– representing risk attitudes

• • 8.1 Introduction
• • 8.2 What are risk attitudes?
• • 8.2.1 Risk-tolerant preferences
• • 8.3 Demonstration of risk attitudes in a medical context
• • 8.3.1 Depicting choice of lung cancer treatment as a decision tree
• • 8.3.2 Branch probabilities for the lung cancer treatment decision
• • 8.3.3 von Neumann-Morgenstern utility and the outcome values
• • 8.3.4 Using standard gamble assessment questions to determine outcome utilities
• • 8.3.5 Determining the outcome utilities for the lung cancer decision problem
• • 8.3.6 Computing Patient A’s expected utility for each of the treatments
• • 8.3.7 Risk attitudes matter
• • 8.4 General observations about outcome utilities
• • 8.4.1 Certainty equivalent– providing a tangible meaning for expected utility analysis
• • 8.4.2 Risk attitudes revisited
• • 8.5 Determining outcome utilities– underlying concepts
• • 8.5.1 Lifetime-tradeoff assessment
• • 8.5.2 Survival-tradeoff assessment
• • Epilogue
• • Bibliography

Chapter 9: Outcome utilities– clinical applications

• • 9.1 Introduction
• • 9.2 A parametric model for outcome utilities
• • 9.2.1 What is a parametric model?
• • 9.2.2 The exponential utility model
• • 9.2.3 Scaling exponential utility models
• • 9.2.4 Assumption underlying the exponential utility model
• • 9.2.5 Determining the exponential utility model parameter– first approach
• • 9.2.6 Determining the exponential utility model parameter– alternate assessment approach
• • 9.2.7 Exponential utility model parameter and risk attitudes
• • 9.3 Incorporating risk attitudes into clinical policies
• • 9.3.1 Risk-adjusted clinical policies– underlying concept
• • 9.3.2 Clinical context for illustrating risk-adjusted clinical policy design
• • 9.3.3 Determining the risk parameter threshold
• • 9.3.4 A simpler assessment question
• • 9.3.5 Generalized age- and gender-specific clinical policy
• • 9.3.6 Risk-adjusted clinical policies– what does it all mean?
• • 9.4 Helping patients communicate their preferences
• • Epilogue
• • A.9.1 Exponential utility model parameter nomogram
• • Bibliography

Chapter 10: Outcome utilities– adjusting for the quality of life

• • 10.1 Introduction
• • 10.2 Example– why the quality of life matters
• • 10.3 Quality-lifetime tradeoff models
• • 10.3.1 Parameterizing the quality-lifetime tradeoff model
• • 10.3.2 Quality-lifetime parametric utility model with constant risk attitudes
• • 10.3.3 Quality-lifetime tradeoff models and risk aversion– a fly in the ointment
• • 10.3.4 Quality-lifetime tradeoff modelling and healthcare policy analysis
• • 10.4 Quality-survival tradeoff models
• • 10.4.1 Assessing quality preferences with the quality-survival tradeoff model
• • 10.4.2 Parameterized quality-survival tradeoff model
• • 10.4.3 Parameterized quality-survival tradeoff model and exponential survival
• • 10.5 What does it all mean?– an extended example
• • 10.5.1 Direct approach to outcome utility assessment
• • 10.5.2 Outcome utility assessment based on outcome decomposition
• • Epilogue
• • Bibliography

Chapter 11: Survival models: representing uncertainty about the length of life

• • 11.1 Introduction
• • 11.2 Survival model basics
• • 11.2.1 Survival probabilities
• • 11.2.2 Lifetime probabilities
• • 11.2.3 Lifetime probabilities and the representation of time
• • 11.2.4 Hazard rates
• • 11.2.5 Estimating a survival model from observations
• • 11.2.6 Kaplan–Meier survival model
• • 11.3 Medical example– survival after breast cancer recurrence
• • 11.4 Exponential survival model
• • 11.4.1 Lifetime probabilities with the exponential survival model
• • 11.4.2 Fitting an exponential survival model to observations– first attempt
• • 11.5 Actuarial survival models
• • 11.5.1 Age- and gender-specific actuarial survival models
• • 11.5.2 Further adjustments of the actuarial survival model
• • 11.6 Two-part survival models
• • 11.6.1 Representing observed survival with an exponential survival model– second attempt
• • 11.6.2 Age adjusting a survival model
• • 11.6.3 Computing outcome utilities with the parametric two-part survival model
• • 11.6.4 Limitations
• • Epilogue
• • Bibliography

Chapter 12: Markov models

• • 12.1 Introduction
• • 12.2 Markov model basics
• • 12.2.1 Health states and transition probabilities
• • 12.2.2 Markov model diagrams and notation
• • 12.2.3 Markov independence
• • 12.2.4 Stationarity assumption
• • 12.2.5 Acyclic graph assumption
• • 12.3 Determining transition probabilities
• • 12.3.1 Markov model used to illustrate how transition probabilities are determined
• • 12.3.2 Determining mortality rates
• • 12.3.3 Determining probability for transitions between stages of recurrence
• • 12.3.4 Determining transition probabilities for treatment response
• • 12.4 Markov model analysis– an overview
• • 12.4.1 Direct approach to Markov model analysis
• • 12.4.2 Using Monte Carlo simulation to analyze Markov models
• • Epilogue
• • Bibliography

Chapter 13: Selection and interpretation of diagnostic tests

• • 13.1 Introduction
• • 13.2 Four principles of decision making
• • 13.2.1 Two examples of decision making under uncertainty
• • 13.2.2 The four principles as a framework
• • 13.3 The threshold probability for treatment
• • 13.3.1 The rationale for a treatment threshold probability
• • 13.3.2 Deriving an expression for the treatment threshold probability
• • 13.3.3 Heuristics for setting a treatment threshold probability
• • 13.3.4 Determining the treatment threshold probability for pulmonary embolism– a formal approach
• • 13.4 Threshold probabilities for testing
• • 13.4.1 The criteria for doing a test
• • 13.4.2 A method for deciding when to perform a diagnostic test
• • 13.4.3 Equations for calculating testing thresholds
• • 13.5 Clinical application of the threshold model of decision making
• • 13.5.1 Test selection for suspected pulmonary embolism: an example
• • 13.5.2 Incorporating a clinical prediction model into a probabilistic framework for test selection for suspected pulmonary embolism
• • 13.6 Accounting for the non-diagnostic effects of undergoing a test
• • 13.7 Sensitivity analysis
• • 13.8 Decision curve analysis
• • 13.8.1 Making the plot of net benefit vs. p*
• • 13.8.2 Use of DCA in practice
• • Bibliography

Chapter 14: Medical decision analysis in practice: advanced methods

• • 14.1 An overview of advanced modeling techniques
• • 14.1.1 When are advanced modeling approaches needed?
• • 14.1.2 Types of modeling approaches
• • 14.1.3 Choosing among modeling approaches
• • 14.2 Use of medical decision‐making concepts to analyze a policy problem: the cost‐effectiveness of screening for HIV
• • 14.2.1 The policy question
• • 14.2.2 Steps of the analysis
• • 14.2.3 Define the problem, objectives, and perspective
• • 14.2.4 Identify alternatives and choose the modeling framework
• • 14.2.5 Structure the problem, define chance events, represent the time sequence
• • 14.2.6 Determine the probability of chance events
• • 14.2.7 Value the outcomes
• • 14.2.8 Estimate costs and discount outcomes
• • 14.2.9 Calculate the expected utility, costs, and cost‐effectiveness
• • 14.2.10 Evaluate uncertainty
• • 14.2.11 Address ethical issues, discuss results
• • 14.3 Use of medical decision‐making concepts to analyze a clinical diagnostic problem: strategies to diagnose tumors in the lung
• • 14.3.1 Define the problem, objectives, and perspective
• • 14.3.2 Identify alternatives and choose the modeling framework
• • 14.3.3 Structure the problem, define chance events, and represent the time sequence
• • 14.3.4 Determine the probability of the chance events
• • 14.3.5 Value the outcomes
• • 14.3.6 Estimate costs and discount outcomes
• • 14.3.7 Calculate expected utility, costs, and cost‐effectiveness
• • 14.3.8 Evaluate uncertainty
• • 14.3.9 Address ethical issues, discuss results
• • 14.4 Calibration and validation of decision models
• • 14.5 Use of complex models for individual‐patient decision making
• • 14.5.1 The Alchemist decision support system
• • 14.5.2 Challenges for individual‐patient decision making
• • Bibliography

Chapter 15: Cost-effectiveness analysis

• • 15.1 The clinician’s conflicting roles: patient advocate, member of society, and entrepreneur
• • 15.1.1 Principles of allocating scarce resources
• • 15.2 Cost-effectiveness analysis: a method for comparing management strategies
• • 15.2.1 Using cost-effectiveness analysis to set institutional policy: an exten