Psychology of Medical Decision Making: Psychology of Medical Decision Making Heuristics and Biases


Impediments to Diagnostic Accuracy: Impediments to Diagnostic Accuracy Heuristics Biases Cognitive Errors


Representativeness: Representativeness Does “A” belong to class “B”? “A” resembles “B” The judged probability that A is in class B is influenced by the degree that A resembles B. Probability of a diagnosis is judged by the similarity of the case to a diagnostic category.


Representativeness: Consequences: Representativeness: Consequences Errors in diagnosis due to representativeness: Ignoring base rate and assigning too high or too low an initial probability to a diagnosis Insensitivity to sample size Overestimating probability of rare diseases Underestimating probability of common diseases Misconception of chance or law of small numbers Regression toward the mean


Availability: Availability The probability of an event is a function of the ease of recall: recenency and retrievability of highly familiar. Thus: Probability of recent, salient events is overestimated Probability of remote, less memorable events is underestimated Probability of rare but vivid events is overestimated Probability of common, ordinary events is underestimated


Availability: Availability Errors in diagnosis due to availability Inappropriate possibilities on the differential diagnosis Excluding appropriate possibilities Excessively high or low estimates of prior probability of diagnosis Tversky and Kahneman, 1973, Cognitive Psychology, 5, 207-232.


Anchoring and Insufficient Adjustment: Anchoring and Insufficient Adjustment How should we make probability estimates when information comes in one piece at a time? Ideally, base accurate estimate on initial info Adjust estimate appropriately based on new info Frequently, the process has two flaws Inaccurate initial estimate Insufficient weight given to the new information Hence the final probability estimate depends too much on the initial estimate Conjunctives are overestimated and disjunctives are underestimated.


Anchoring and Insufficient Adjustment: Anchoring and Insufficient Adjustment Estimates of patient prognosis changed very little from day 1 to day 3 in the ICU despite new information. When two physicians gave estimates for the same patient, they often disagreed widely. However, two days later, each having received the same new information, neither would have adjusted the probability very much. Poses et al., Med Decis Making 1990;10:6-14


Anchoring and Insufficient Adjustment: Consequences: Anchoring and Insufficient Adjustment: Consequences Errors in diagnosis due to anchoring and adjustment: Persisting in inappropriately low or high probability estimates after receiving new information


Ego Bias: Ego Bias Self-serving probability estimates We tend to overestimate our own performance and underestimate the performance of others.


Ego Bias: Ego Bias High Low Mortality Mortality Mortality Rate Specialty Specialty Estimate for personally 2.35% 0.25% performed operations: Estimate for operations 3.87% 0.93% performed by colleagues: Actual mortality rate: 2.42% 0.44% Detmer et al., J Med Educ 1978;53:682-683


Confidence and the Illusion of Validity: Confidence and the Illusion of Validity Experts and non-experts have great confidence in their judgments even when these judgments can be shown to be incorrect. Why such confidence?


Confidence and the Illusion of Validity: Confidence and the Illusion of Validity Bushyhead and Christensen-Szalanski, Med Decis Making 1981;1:115-23


Confidence and the Illusion of Validity: Confidence and the Illusion of Validity Bushyhead and Christensen-Szalanski, Med Decis Making 1981;1:115-23


Confidence and the Illusion of Validity: Confidence and the Illusion of Validity Med Decis Making 1981;1:115-23


Confidence and the Illusion of Validity: Confidence and the Illusion of Validity Dawson et al, Med Decis Making 1993;13:258-266.


Confidence and the Illusion of Validity: Consequences: Confidence and the Illusion of Validity: Consequences Physicians commonly have incomplete feedback in the accuracy of their judgments and their judgment policies. This will often lead to overestimation of the accuracy of judgments which can lead to diagnostic and therapeutic errors.


Value-induced Bias: Consequences: Value-induced Bias: Consequences Value induced bias leads to distorted estimates of probabilities. However, because they are distorted in a predictable direction it may be possible to anticipate these errors.


Hindsight Bias: Hindsight Bias When people know the outcome of an event, they exaggerate the extent to which they could have predicted the event beforehand. People who actually have to predict an event beforehand find the task more difficult than the hindsight subjects claim it would be.


Hindsight Bias or Outcome Bias: Consequences: Hindsight Bias or Outcome Bias: Consequences A diagnosis or a bad outcome appears more predictable after the fact than it actually was. This can lead to incorrect conclusions in peer review or malpractice evaluations. Evaluations of another’s judgments should be based on the information available to the other, without information of the eventual diagnosis or the outcome of the patient.


Corrective Procedures: Corrective Procedures Many biases stem from two tendencies: 1. People tend to give insufficient weight to the base-rate frequency of outcomes. 2. People tend to overestimate how predictable an outcome is given the available clinical cues (i.e., signs and symptoms).


Corrective Procedures: Corrective Procedures Thus, to improve estimates, consider: 1. Increasing the weight given to base rate information, and 2. Regressing your intuitive estimate back toward the base rate, since the cues are not as predictive as your estimate implies.


Avoid Associative Errors: Avoid Associative Errors Search for neglected information Search for information contrary to the decision Learn to live with a healthy dose of uncertainty


Avoid Perceptual Biases: Avoid Perceptual Biases Reframe decisions and consider multiple descriptions of the same situation Consider multiple other reference points Consider how the heuristics and biases you know about might impact the decision


Normative Principles: Valuation: Normative Principles: Valuation Expected Utility (EU): Determine utility for each outcome (For multi-attribute outcomes, this may involve weighting and combining attribute-wise utilities (For future outcomes, this may involve discounting) Weight utility by probability of outcome Add up to obtain EU Decision rule: Choose option with highest expected utility


Expected Utility: Expected Utility


Framing: Loss Aversion(Thaler, 1980): Framing: Loss Aversion (Thaler, 1980) “Losses loom larger than gains” Can cause violations of task invariance: Assume you’ve been exposed to a disease which leads to a quick and painless death within a week. The probability you have the disease is 0.001. What’s the most you’d pay for a cure? ($200) Suppose volunteers were needed for research on the above disease. You’d have to expose yourself to a 0.001 chance of contracting the disease, and couldn’t purchase the cure. What’s the least you’d accept to be a volunteer? ($10,000)


Reflection Effect: Risk Avoidance and Risk Seeking: Reflection Effect: Risk Avoidance and Risk Seeking When outcomes are described as gains (e.g. survival), people tend to avoid risk. When outcomes are described as losses (e.g. mortality), people tend to seek risk. This violates the normative principle of descriptive invariance: Decisions should not depend on outcomes, not their descriptions.


Prospect Theory(Kahneman & Tversky, 1979): Prospect Theory (Kahneman & Tversky, 1979)


Preference Reversal: Preference Reversal Presentation formal Single Presentation Side-by-side presentation


Particular topics of this section.: Particular topics of this section. Pattern recognition Use of medical knowledge is analogous to visual recognition. Uncertainty, decisions, and error Recognition primed decision making: Familiar and unfamiliar situations Judgment helps physicians apply strategies flexibly Development of medical competence and expertise Illness scripts, collections of cases Medical education’s assumptions about physician’s thinking Example of an intervention designed to teach residents not to make a diagnostic error


Physicians’ Intuitive Decision Making. : Physicians’ Intuitive Decision Making. Experienced physicians can make accurate judgments and decisions in conditions of uncertainty. Often can’t explain how This ability is based upon their extensive knowledge, residing in memory and trained to perform well in clinical situations. (Abernathy, 1995; Schmidt, 1990) Elements of their ability: Knowledge activated through pattern recognition processes Judgments of extent or degree Problem solving (Lesgold, 1988; Greeno, 1988; Patel, 1986; Simon, 1995) Scripts that are sensitive to context. (Anderson, 1990) Adaptiveness: more effortful strategies can produce more accuracy. (Payne, 1993)


“Getting” medicine is not easy: “Getting” medicine is not easy


Pattern recognition is easier if you have experience: Pattern recognition is easier if you have experience


Slide35: Experience: A problem of perception Slide from Slawson, Shaughnessy, Becker, 1999. Do you see the Dalmation in the picture? Moral: Clinical experience sometimes helps see, sometimes prevents seeing the right picture Now that you see it, can you try to not see it? Moral: Experience can result in ideas that are difficult to change


One learns the basic patterns: One learns the basic patterns


One sees them in new situations.: One sees them in new situations.


Then one can see the pattern where before it had been confusing.: Then one can see the pattern where before it had been confusing.


Physicians’ knowledge of medicine is similarly accessed through pattern recognition.: Physicians’ knowledge of medicine is similarly accessed through pattern recognition. Assertion: Just as you learned to see the object in the ambiguous picture, physicians learn to see the diseases in situations that may be difficult for others to interpret.


Case example: Case example Patient is a 30-year old man with a gunshot wound to the chest, in shock, with distended neck veins and diminished breath sounds. (Abernathy, 1995) What is it?


Pattern recognition, hidden cues: Pattern recognition, hidden cues In OR, the surgeon opened the chest The shock was due to a tension hemothorax rather than a tension pneumothorax. The bullet had cut a major artery, not the lung, A needle in the chest in the ER would not have been much help. In his initial examination the surgeon had tapped the patient’s chest. It sounded like a watermelon (full of fluid) rather than a drum (full of air). In this example of pattern recognition, everyone in the room knew the patterns only the attending surgeon had the defining cue.


Additional information: Additional information 30-year-old man, with gunshot wound to chest, in shock Distended neck veins, diminished breath sounds. Looks like a tension pneumothorax: air released into the pleural space not only makes it difficult for the patient to breathe but also cuts off the return of blood from the body to the heart. Life-threatening situation due to decreased cardiac output Immediate easy solution: stick a needle in the chest to drain off the air pressure. After a quick inspection, the attending surgeon instructed that patient be taken to OR for a full thoracotomy. Objections were raised. “His brain is starving for oxygen while you scrub!”


Potential contradiction:The key to rational medical decision making is probability.: Potential contradiction: The key to rational medical decision making is probability. How do we deal with probabilities, using our pattern recognizing brains? Patterns are seen, or not seen; probabilities have degree


Analogy: Medical knowledge is like knowledge of visual patterns.: Analogy: Medical knowledge is like knowledge of visual patterns. A large store of visual patterns/medical knowledge structures are in memory. Activation of the physician’s knowledge by the patient (pattern recognition) can be very rapid. It may be hard to explain what features define a medical concept, how a patient’s disease was recognized. One can suddenly “see” a case differently, rapid reorganization. Others can help one “see” a concept.


Is there too much medical decision making error?: Is there too much medical decision making error? The amount of medical error is small, given the demands of the task. (McDonald, 2000; Gigerenzer, 1996) This is because of the physician’s ability to learn, to bring knowledge to mind when needed, to adjust the knowledge in novel situations, and to revise the knowledge in the face of experience or new information. This is not to minimize the effects of physician cognition, complex systems, inadequate staffing, and so on, in the production of medical error. (Leape, 2000)


Poor Medical Outcomes and Physician Decision Making : Poor Medical Outcomes and Physician Decision Making Some poor medical outcomes may be caused by errors in the physician’s ways of dealing with uncertainty in decision making. Reasoning strategies Behavior strategies Understanding the psychology of physician decision making may offer ways to reduce errors Errors due to poor pattern recognition. Errors due to physician ignorance of information that has been published. Errors due to inappropriate strategies for handling uncertainty. (Swets, 2000; Kahneman, 1982)


Ways in which judgment may be inaccurate: Ways in which judgment may be inaccurate Comparison of individuals’ use of information in judgment with a correct model (Kahneman, 1982) Indicates they often use fewer cues than are relevant Indicates they put different weight on cues than they ought Comparison of different physicians’ judgments indicates variability between individuals (Gillis, 1981), between regions (Tape, 1991) Comparison of individuals’ use of information with their self report Not accurate self understanding Though may recognize own Hence, communication about how much attention to pay to different cues (in training) may be inaccurate, confusing.


Complex Knowledge Structures for MedicineSchmidt, Norman, and Boshuizen, 1990: Complex Knowledge Structures for Medicine Schmidt, Norman, and Boshuizen, 1990 Expertise is not superior reasoning skills knowledge of pathophysiological states and mechanisms Expertise is cognitive structures that describe the features of prototypical patients containing clinically relevant information about the disease its consequences the context it develops in Experts have several types of cognitive structure, and can use any of them, as needed


Development of Medical Expertise: Four Stages of Knowledge Structures: Development of Medical Expertise: Four Stages of Knowledge Structures Schmidt, Norman, Boshuizen, 1990 Stage 1. Elaborated causal networks Learned during the basic science years Facts and relationships Nodes and links Causal models of disease processes


Development of Medical Expertise: Development of Medical Expertise Stage 2. Compilation of abridged networks Starts when exposed to real patients Knowledge gets compiled (rewritten, automated) simplified causal models explain signs and symptoms associated with diagnostic labels “Diagnosing a first clinical case requires quite a lot of mental effort and involves extensive reasoning based on the elaborate causal networks available to the student, but when he sees his second or third similar case, shortcuts will emerge. He will no longer have to activate all possibly relevant knowledge in order to understand what is going on in his patient; only knowledge pertinent to understanding the case will be activated" (p 614).


Development of Medical Expertise: Development of Medical Expertise Stage 3. Illness Scripts. Based on repeated experience with patients Illness scripts are sufficient to diagnose and treat diseases. Lists of features that characterize the disease Specification of what to do Information about context Information about temporal features of disease


Development of Medical Expertise Generic Illness Script (Table 1, p 615, Schmidt et al, 1990): Development of Medical Expertise Generic Illness Script (Table 1, p 615, Schmidt et al, 1990)


Development of Medical Expertise: Development of Medical Expertise Stage 4. Cases. Patient encounters stored as instance scripts. Based on long experience Physician remembers many individual patients Each has a different variant of the disease New (or newly sick) patients are recognized as “similar to Patient X” treated as Patient X was treated


Stages and Models of Physician Knowledge and Learning: Stages and Models of Physician Knowledge and Learning


Clinical Presentation Based approach (University of Calgary, 1991): Clinical Presentation Based approach (University of Calgary, 1991) Correctly managing a case needs both relevant knowledge a task specific reasoning process For curriculum, identify universe of clinical presentations For primary care, 120 different presentations For each, identify expert’s problem solving process and related basic science knowledge Typical educational sequence: Present schema, demonstrate its use Lectures on relevant basic science and clinical science knowledge Small group problem solving with many cases, selected because of the distinctions between them, and they cover the possibilities.


Example of a medical error, Pseudodiagnosticity,and a program to eliminate it.: Example of a medical error, Pseudodiagnosticity, and a program to eliminate it. Pseudodiagnosticity: (Green, 1995) Use of cues in diagnosis that are not related to the present diagnostic task, though they are related to the disease in general. Distinction between factors that predict: Development of Coronary Artery Disease Acute Cardiac Ischemia (Jayes, 1992)


Pseudodiagnosticity (continued) : Pseudodiagnosticity (continued) Empirically, what predicts Acute Cardiac Ischemia in the acute setting? Nature and location of patient's symptoms Chief complaint of chest pain History of ischemic disease Certain specific electrocardiographic findings ST segment elevation or depression Q waves T wave changes


Pseudodiagnosticity (continued): Pseudodiagnosticity (continued) What risk factors predict the development of coronary artery disease? Family history of premature coronary artery disease Age Male gender Smoking Diabetes mellitus Increased serum cholesterol Hypertension


Pseudodiagnosticity (cont): Pseudodiagnosticity (cont) What cues do ER doctors use in deciding whether to admit? Those predicting acute cardiac ischemia, or those predicting the development of coronary artery disease? 787 patients evaluated for suspected acute cardiac ischemia Method: hospital record review


Pseudodiagnosticity (cont): Pseudodiagnosticity (cont)


Pseudodiagnosticity (cont): Pseudodiagnosticity (cont) These ER physicians used low-relevance cues (hypertension, diabetes history, age) as much as they used some relevant cues (ST-segment, MI history) and more than they used other relevant cues (Q-waves, T-wave changes, chief complaint). These are "pseudodiagnostic" cues.


Pseudodiagnosticity (cont): Pseudodiagnosticity (cont) Why do physicians use this pseudodiagnostic information? Associative memory -- what comes to mind? Scripts -- How has the physician learned to handle suspected MI? Experience or Education? Most literature does not make the distinction between CAD risk factors and acute predictors. CAD risk factors are discussed as crucial information related to heart disease.


If use of pseudodiagnostic cues is education-based, how prevent pseudodiagnostic errors?: If use of pseudodiagnostic cues is education-based, how prevent pseudodiagnostic errors? Raise Consciousness: Physicians could review their own decision making, aware of this common error, so they can catch it and avoid it. (Kahneman, 1982; Wolf, 1985; Chapman, 2000) Doing so, they will revise their scripts. Practice: Medical schools could “drill” the accurate predictors. This will form physicians' clinical scripts. Medical schools could teach sets of simulated cases which include the accurate predictors. (Wigton, 1990; Tape, 1992) Decision Aid: Provide decision support tool on site.


Conclusion: These psychological concepts of MDM explain physicians ability to make decisions under uncertainty:: Conclusion: These psychological concepts of MDM explain physicians ability to make decisions under uncertainty: Pattern recognition Like seeing cows. Medical concepts come quickly to mind. Scripts, strategies for handling patients with particular illnesses Uncertainty, decisions, and error Illness scripts incorporate strategies for diagnosis, for hedging bets Judgment helps physicians apply strategies flexibly Development of medical competence and expertise Illness scripts, collections of cases Medical education’s assumptions about physician’s thinking Example: Training residents to manage suspected MI without relying on pseudodiagnostic cues. Providing insight can be as effective as providing a decision aid.