- Email: [email protected]
Cognitive Errors
C o g n i t i v e E r ro r s Thinking Clearly When It Could Be Child Maltreatment Antoinette L. Laskey,
MD, MPH
KEYWORDS Cognitive errors Child maltreatment Confirmation bias Anchoring Implicit biases KEY POINTS Cognitive errors occur in every profession. Dozens of cognitive errors have been demonstrated to happen in medicine, many leading to poor patient outcomes. The diagnosis of child maltreatment can be susceptible to cognitive errors because of conditions of high stress, limited or questionable quality of data, and the subtlety of some diagnoses. Errors may be avoided through deliberate efforts to consider alternative diagnoses, avoiding premature closure, seeking objective input without social cues, and multidisciplinary collaboration.
Medical decision making is an exercise in process management: health care providers must gather data from multiple data sources, sometimes simultaneously, interpret signs and symptoms, sometimes without adequate history to fully understand the information provided, and direct patient management including further diagnostics and treatment, often without complete data. Most of the time the process seems to work adequately, that is, the patient is appropriately diagnosed and treated. Yet there continue to be studies that demonstrate failures in diagnosis and treatment related to factors that seem to be unrelated to the medical condition and more related to the patient’s specific demographic characteristics.1–4 It is important that the diagnosis of child maltreatment be made accurately. Failure to correctly diagnose an abused child as abused (a false negative) could result in that child being returned to a potentially dangerous environment. Conversely, the overdiagnosis of child abuse (a false positive) could result in a child being removed from an environment wherein no one has caused harm to the child. The potential for errors
Disclosures: None. Department of Pediatrics, Primary Children’s Hospital, University of Utah School of Medicine, 675 East 500 South, Suite 300, Salt Lake City, UT 84102, USA E-mail address: [email protected] Pediatr Clin N Am 61 (2014) 997–1005 http://dx.doi.org/10.1016/j.pcl.2014.06.012 pediatric.theclinics.com 0031-3955/14/$ – see front matter Ó 2014 Elsevier Inc. All rights reserved.
998
Laskey
in decision making related to child abuse extend far beyond the health care arena and include law enforcement, social welfare systems, prosecutorial decisions, and judge or jury decisions. Understanding the potential pitfalls in decision making is especially important for process improvement. Social psychologists have spent considerable effort in understanding how humans think, but only relatively recently have physicians looked to the literature to understand what clinicians can do better. Humans have evolved to process an astonishing amount of information, often dealing with large simultaneous sensory inputs assaulting all of their senses. How is it that we are able to take in information, filter out what is not immediately relevant, and arrive at a reasonable conclusion? Are there shortcuts we can use to improve our processing speed without sacrificing the quality of the decision making? Perhaps most importantly, is it possible to avoid cognitive errors that we make and are compounded when community professionals from other disciplines, such as law enforcement and child protection services, are involved in a decision process? COGNITIVE PROCESSES: AN OVERVIEW
Humans routinely make decisions in an overwhelming, complex environment. There are also limited cognitive resources available at any given moment to process all of the information available. Given the inability to completely process all of the data, humans have adapted by using cognitive shortcuts, which can be both hardwired and derived. One easy-to-appreciate example of a hardwired shortcut is that of recognizing potential danger. It does not take a prior experience with a potential threat, such as an obviously hostile animal, to realize that care must be taken. Derived shortcuts also are common occurrences that increase over the course of a lifetime. Driving represents a classic example of a derived shortcut. Memories of learning to drive often will conjure the stress associated with the myriad of data bombarding the novice driver. It is only after years of practice that the process becomes nearly fully automated. What was once a difficult, stressful, and complex effort changes into a background process that frees the mind to handle other tasks. However, this automation can be an impediment to rapid responses to the unexpected. Deviations from the norm will take longer to recognize and process when one is not fully engaged in the task at hand. Just as driving or tying one’s shoes, or doing any number of complex learned tasks becomes less of an elaborate mental juggling act, so too does one’s ability to complete tasks associated with the work. In the early years of a person’s career, every part of the job requires thought and deliberation. As one gains experience and feedback on process and decision making, the process becomes more refined and automated. However, just as automation improves efficiency in tasks such as driving, it can also decrease the ability of an individual to recognize and respond to deviations from the norm. Decision theory is the study of identifying what is known, what is unknown, and the other factors involved in arriving at an optimal (and therefore correct) decision. Social psychologists use multiple techniques to understand the processes involved in decisions. Decisions are very often reliant on heuristics, which are the shortcuts used to aid in decision making sometimes referred to as rules of thumb; new, incoming data are constantly incorporated and compared with previously gathered information. If they match previously acquired data, a shortcut in the thought process may be used to arrive at the answer. Heuristics can certainly be helpful in improving efficiency, but also may lead one astray if they compound biases in the thought process. Biases may be either explicit or implicit. An implicit bias is one that an individual holds without being aware. Implicit biases can be difficult for an individual to
Cognitive Errors
acknowledge or control, given that they are happening without any real understanding or conscious effort. It has been clearly demonstrated that even people who profess to be egalitarian in their beliefs, and act outwardly in egalitarian ways, hold implicit biases that can influence decision making.5 It is especially important to draw a distinction between implicit and explicit biases. Explicit biases are very much part of a person’s conscious awareness and can be considered less socially desirable. Examples of explicit biases include overt racism, sexism, or ageism. Much is now known about how humans process complex information. Research has shown that it is possible to improve thinking and reduce errors in judgment. Multiple studies have demonstrated that cognitive errors may lead to incorrect medical diagnoses. It is therefore important that professionals engaged in these high-stakes decisions, such as the diagnosis of child abuse, make efforts to understand how and when these errors might occur and how best to mitigate their effects. COGNITIVE ERRORS AND CHILD MALTREATMENT
There are many common assumptions made by professionals that might not apply when the case involves child abuse. First, health care professionals assume that parents will be open and honest when seeking medical care for their child. However, if the perpetrator is a parent, he or she may not share the relevant information leading up to the symptoms that brought the child to medical attention, or worse, this person may lie and obfuscate. Second, there is a generally held belief that health care professionals use science and logic alone to reach a correct diagnosis, yet there is evidence to the contrary that “the art of medicine” is actually antithetical to science in some cases.6 There are situations wherein a better, more scientifically sound approach to medical care is available to clinicians who resist implementation of the new protocol because they “know what is best.”7 Third, primary care providers often feel they know their patients’ families and can judge who is at risk and who is not. Finally, primary care providers often feel that they would be able to detect problems in parent-child relationships by how a child interacts with their parent (eg, the child seems well bonded to the mother so there is no reason to think the mother could be abusing him).8,9 None of these factors is meant as an indictment of a profession or group of individuals; rather, it is an acknowledgment of the complexities of the human thought process in light of difficult decisions. The problems inherent in the human thought process are abundant. Many cases of potential child maltreatment are subtle and there are few “a-ha” moments that bring absolute clarity. Sometimes, despite best efforts, professionals are not fully engaged in the work of the day and rely more readily on derived shortcuts without realizing this is occurring. Often, parents and caregivers do not know what information is relevant or most important to share when talking to medical personnel or investigators, causing them to offer information that could be construed as misinformation or attempts at deception. Moreover, in other circumstances parents or caregivers may lie deliberately. A closer look at common cognitive errors can help illustrate how they can confound arrival at an optimal answer. An in-depth exploration of all possible cognitive errors is beyond the scope of this article (Table 1). Implicit Stereotypes
Stereotypes have a negative connotation, suggesting willful discrimination. Implicit stereotypes, however, are those that are typically not in the conscious awareness of the holder. Stereotypes are derived shortcuts that allow rapid categorization of people
999
1000
Laskey
Table 1 Common errors and possible solutions Type of Cognitive Error
“Symptoms”
Possible Solution
Implicit stereotypes
Relying on generalizations to describe a family or caregiver. Examples: “bad parent,” “nice family”
Using another colleague to staff the case because he or she will not have the same initial impression of the family
Anchoring
Difficulty considering alternative diagnoses even if all the information does not fit as one would expect
Devil’s advocate (either a colleague or simply proposing countertheories to one’s self), intentionally proposing alternative diagnoses to see if the working diagnosis is still the best fit
Triage cueing
Sending a patient to a specialist based on a very specific symptom or finding (which could be an anchor)
Multidisciplinary approach to diagnosis allows multiple viewpoints given identical case details, prevents “thinking in your box”
and naturally develop over a lifetime of exposures to information from the environment. When asked to conjure a mental image of an elderly person or a nurse or a homeless person, it is likely that this is not a difficult task. Clearly the image called to mind does not accurately represent all who are in that category, being a mental shortcut, but it does serve a purpose. Likewise, if a medical provider interacts with a family in a clinical setting, it is not difficult to understand that a subconscious categorization occurs, putting them into a “good” or “bad” family category without objective data to support this segregation. Some data that are used to categorize people are present immediately on first contact: the style of clothes or body art, the age, the speech patterns, and the color of the skin all are seen, processed, and sorted without conscious effort. Just as stereotypes are heuristics that allow us to categorize people easily, if not necessarily accurately, biases also exist and are frequently implicit. Social psychologists have demonstrated repeatedly that in-group preferences, that is, preferring those who are like as opposed to those who are different, are more natural than out-group preferences. For this reason, it is more natural to give the benefit of doubt to those who are similar to the observer and hold those who are not to a different standard. In perhaps the most powerful demonstration of implicit biases, the Implicit Association Test developed by Nosek and Banaji (Available at https://Implicit.Harvard.edu) has been used to demonstrate people’s natural tendency to easily categorize consistent pairings while having more difficulty categorizing inconsistent pairings. In the classic example, the time lag in a rapid categorization test is measured when a subject is required to associate a negative word with a white person’s face or a positive word with a black person’s face. It has been shown to be demonstrably slower than when the pairing is a positive word and a white face or a negative word and a black face. Because the positive or negative association is implicit, people are often surprised to learn their results and frequently express frustration because they perceive themselves to be egalitarian. The relevance of stereotypes and implicit biases is found throughout the child maltreatment literature. Hampton and Newberger10 found that hospitals were more likely to report a family to child protective services (CPS) if the patient was black,
Cognitive Errors
the parents were young, or the family was poor. In a study of patients presenting to the emergency department, Jenny and colleagues11 found that abusive head trauma was missed nearly one-third of the time on initial presentation to medical care, and cases were more commonly missed if the patient was white or from an intact family. In another study of children admitted to an urban academic medical center with skeletal injuries, children with skeletal injuries who were from minority or low-income families were more likely than white or insured children (a proxy for socioeconomic status) to be evaluated for abuse and to be reported for abuse.4 In an attempt to create an experimental model for the influence of race and socioeconomic status on the diagnosis of child abuse, Laskey and colleagues12 used a methodology adapted from the psychology literature, and showed a difference in the frequency of the diagnosis of abuse among low-income white children in comparison with other categories. In this study, practicing primary care pediatricians were asked to rate the likelihood that an injury of ambiguous etiology (ie, it could potentially have been either accidental or inflicted) was abusive. The results demonstrated that socioeconomic status had more of an effect on the diagnosis of abuse than did race, despite identical clinical histories.12 In the law review literature, stereotypes have been shown to influence behavior and memories in investigators, prosecutors, and juries. The concept of misremembering, that is, creating a new memory from pieces of information that are reassembled into a new memory, has been shown to occur in stereotypical ways. In one study, a story of a crime is presented and the race of the characters is changed. The results showed that people tended to forget mitigating factors for black characters, and hostile actions were more likely to be attributed to them, even if they were not the aggressors in the story. Similarly, hostile actions by white characters were often forgotten. Of note, research subjects were often very confident in their memories, even when those memories were wrong.13 It should be apparent, therefore, that implicit stereotypes and implicit biases play an important role in how information is processed, particularly when a quick decision is required or a decision that relies on less than optimal information is necessary. How might this affect the accuracy of determinations when it comes to child maltreatment? Implicit beliefs may lead one to accept a story at face value because the alternative (eg, the diagnosis of child abuse) does not fit an implicit belief. If presented with a “nice” family and a child with injuries possibly related to child abuse, it might be an easier path to follow to consider alternative diagnoses. It is not uncommon that a bleeding condition, metabolic or genetic bone disease, nutritional deficiency, previously unrecognized congenital anomaly, or an accident is more aggressively pursued as the potential cause rather than to consider the unfortunately more common explanation of abuse. This is not to say that these conditions should be dismissed from the differential diagnosis, but neither should abuse be ignored in the evaluation. Alternatively, an ascertainment bias (also known as self-fulfilling prophecy), that is, looking for something only where it is expected to be found, may occur. If one only believes that “bad parents” (defined in whatever subconscious way that a health care provider believes) abuse their children, people who fit such a model of bad parenthood will be much more likely to be evaluated for abuse than those who do not. Of course cases are more likely to be found where one looks, leading to a circular fulfillment of the bias. Anchoring
Anchoring is a problem of premature closure. When first presented with information, both explicit information and implicit data that are perceived but are not explicitly presented, the human tendency is to begin an immediate sorting and processing of the
1001
1002
Laskey
data. Data about a patient and family are available for this on first contact with the chart. Names often are markers of social position and frequently have racial or ethnic clues.14 Race is commonly assigned by appearance and not by a parent’s report. A clinical interaction begins with questions by a provider and answers from a caregiver or parent. From the outset, the clinician is building and testing hypotheses to explain the chief complaint and the associated findings. Sometimes these initial impressions form an anchor from which it is challenging to change course. Once the basic categorizations are made, further information is processed in that context. Anchors can be descriptors (eg, “difficult parent,” hypochondriac), or a working diagnosis or historical piece of data (eg, fever several days ago but not currently, illness in household). Just as stereotypes can lead to ascertainment bias, anchors may result in a confirmation bias. Humans like to be correct and will often work to find support for a theory. Examples abound of people’s willingness to disregard information that is contradictory to a held belief and then embracing supporting information, even if it is dubious. Confirmation biases are especially problematic in accurate decision making because once a hypothesis has been generated, the tendency is to abandon opposite hypothesis testing. This action can result in the inability to accurately process new historical information or new test results. When doctors perform testing, the analysis of the new data is often in the context of whether it fits with the working theory. Although it is often said that something is being “ruled out,” research has shown that what is happening more often is a “rule in.” As one builds a mental case for a given condition, information that does not support the theory is more difficult to “see,” or be incorporated into the overall picture. It is not unusual that the information is not even truly processed at the same level as the supporting data. It is the avoidance of cognitive dissonance that leads to a cycle of seeking confirmation of one’s theories and finding support in data that are obtained, even if there may be more than one way to interpret the new information.15 Triage Cueing
The adage “to the hammer, the world is a nail” is an apt description of triage cueing. When a patient is referred to a specialist, a specific question is often posed that the specialist would be ideally suited to answer. However, this begs the question of whether the information used to arrive at the referral to the specialist was correct or whether any cognitive errors may have occurred along the way. If a physician is evaluating an infant with bruising, referral to a hematologist will result in an evaluation of whether the patient has a bleeding disorder. If such a workup fails to demonstrate abnormality, the question has been answered and the case is returned to the hands of the primary care physician (PCP). But was the question answered? Whose responsibility is it to pursue the cause of the bruising? This closing of the loop sometimes does not occur. It is the natural inclination of a specialist to answer the question that is asked, not answer the unasked question. Triage cueing, that is, the referral based on an assumption of diagnosis, could result in missing the ultimate diagnosis if the referral was to the wrong specialist. In child abuse pediatrics, the nature of the evaluation is inherently multidisciplinary. When a PCP consults with a child abuse pediatrician, the effects of triage cueing can be mitigated by having it be the responsibility of the child abuse pediatrician to engage other specialists while sharing the broad base of information gathered in the course of the medical evaluation. A corollary to triage cueing is diagnosis momentum. In cases of caregiver-fabricated illness (previously known as Munchausen syndrome by proxy and medical child abuse), patients often have exhaustive problem lists with many diagnoses. When a
Cognitive Errors
parent seeks care with a new clinician, and provides what appears to be a detailed medical history and even supporting documentation for the child’s diagnoses, it is easy to start with these data and move forward with the evaluation and management. The problem lies in the complexity of the case and the disinclination to “start from the beginning.” Diagnosis momentum uses the preassigned diagnosis as a given fact. Unfortunately, the disorder inherent in caregiver-fabricated illness relies on this flaw in thinking. It is not uncommon to discover on thorough review of the medical documentation that many of the diagnoses carried by these patients are in fact conditions that were in a differential diagnosis and then ruled out. Similar to anchoring bias, labels are sticky and will build over time, placing these children at extreme risk of morbidity and mortality.16 COUNTERING ERRORS IN THINKING
It is not possible to avoid all errors in a process that has so many unknowns. The goal of safe, effective medical care should be the minimization of errors wherever those errors may arise. Psychologists have shown that there are processes that can be used to improve decision making. The following are some easy-to-apply strategies to improve decision making. 1. An important countermeasure is the use of the devil’s advocate who, either as an individual working on a case or as a member of a broader team of professionals, can pose the question, “what are the alternative explanations for the information currently available?” When this question is asked as a member of a diverse team, it is possible to consider alternative hypotheses and propose tests and theories while minimizing the effects of a confirmation bias. 2. Avoid subjective or emotive descriptors. Many biases are compounded inadvertently by the way people naturally communicate. Descriptors are important to building a mental image. In medicine, handoffs between professionals, such as occur during shift changes or between teams in the hospital setting or even between professions when a case is passed from the medical team to CPS are frequently filled with descriptors. These details influence the receiver of the information by coloring how new incoming information is processed. If a parent is framed in a negative light, the new provider will interact with that parent from the anchor of this negative information. If a diagnosis is suggested as the working hypothesis, it will form an anchor against which new incoming data will be compared. Clearly it is important to relay information gathered to promote effective communication about a patient. The key to minimizing the cognitive errors in the system is to consider how that information is relayed. Data presented in handoffs should be evaluated to determine if it is necessary to the care of the patient, objective or subjective, and known versus hypothesized. 3. Seek objective input from colleagues on challenging cases. Another strategy that can be used with socially charged situations, such as when a particular situation is triggering a visceral reaction (eg, “this is a good family” or “this person is really frightening”), is to present purely objective data to a colleague. Although clearly a diagnosis cannot be made in a vacuum of information, by stripping away some of the social cues that might trigger an implicit bias, the diagnostic process can be improved. The history, medical presentation, physical examination findings, and test results can be objectively interpreted by someone who has not had his or her implicit biases activated. It is important when using this approach to present the data as cleanly as possible without any extraneous commentary to avoid swaying the interpretation.
1003
1004
Laskey
4. Multidisciplinary thinking. Child abuse pediatricians often work as part of a larger multidisciplinary and interdisciplinary team. A key strategy to avoiding errors such as anchoring or triage cueing is to think broadly, the “think outside of the box” approach. Because the diagnosis of child abuse is a difficult one to make with serious consequences for a wrong diagnosis in either direction, it is especially important to use the collective wisdom of a team. Because each provider on the medical team will interpret the information through the lens of his or her specialty, open discussion of supporting information in addition to refuting information will improve the quality of the diagnosis, avoiding a singular focus. SUMMARY
The diagnosis of child abuse is one that no provider wants to make, including those in primary care. It is known that cognitive errors happen, but they are not inevitable and steps can be taken to minimize them. It is through introspection that process improvement can be achieved. Diagnosis and medical management can be optimized through collaboration, discussion, and effective teamwork. Mistakes can lead to a better understanding of how humans think. As James Joyce said, “mistakes are the portals of discovery.” REFERENCES
1. Loring M, Powell B. Gender, race, and DSM-III: a study of the objectivity of psychiatric diagnostic behavior. J Health Soc Behav 1988;29(1):1–22. 2. McKinlay JB, Potter DA, Feldman HA. Non-medical influences on medical decision-making. Soc Sci Med 1996;42(5):769–76. 3. Feldman HA, McKinlay JB, Potter DA, et al. Nonmedical influences on medical decision making: an experimental technique using videotapes, factorial design, and survey sampling. Health Serv Res 1997;32(3):343–66. 4. Lane WG, Rubin DM, Monteith R, et al. Racial differences in the evaluation of pediatric fractures for physical abuse. JAMA 2002;288(13):1603–9. 5. Nosek BA, Greenwald AG, Banaji MR. The implicit association test at age 7: a methodological and conceptual review. Social psychology and the unconscious: the automaticity of higher mental processes. East Sussex (UK): Psychology Press; 2007. p. 265–92. 6. Suarez-Almazor ME, Russell AS. The art versus the science of medicine. Are clinical practice guidelines the answer? Ann Rheum Dis 1998;57(2):67–9. 7. Bohmer R, Edmondson A, Feldman L. Intermountain healthcare. Harvard Business Review School 2002. Case 603, 066. 8. Flaherty EG, Sege RD, Griffith J, et al. From suspicion of physical child abuse to reporting: primary care clinician decision-making. Pediatrics 2008;122(3):611–9. 9. Sege R, Flaherty E, Jones R, et al. To report or not to report: examination of the initial primary care management of suspicious childhood injuries. Acad Pediatr 2011;11(6):460–6. 10. Hampton RL, Newberger EH. Child abuse incidence and reporting by hospitals: significance of severity, class, and race. Am J Public Health 1985;75(1):56–60. 11. Jenny C, Hymel KP, Ritzen A, et al. Analysis of missed cases of abusive head trauma. JAMA 1999;281(7):621–6. 12. Laskey AL, Stump TE, Perkins SM, et al. Influence of race and socioeconomic status on the diagnosis of child abuse: a randomized study. J Pediatr 2012; 160(6):1003–8.e1001.
Cognitive Errors
13. Levinson JD. Forgotten racial equality: implicit bias, decision-making and misremembering. Bepress Legal Series. 2006:1630. 14. Bertrand M. Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. National Bureau of Economic Research; 2003. (Working Page 9873). 15. Croskerry P. Achieving quality in clinical decision making: cognitive strategies and detection of bias. Acad Emerg Med 2002;9(11):1184–204. 16. Flaherty EG, MacMillan HL, Christian CW, et al. Caregiver-fabricated illness in a child: a manifestation of child maltreatment. Pediatrics 2013;132(3):590–7.
1005
MD, MPH
KEYWORDS Cognitive errors Child maltreatment Confirmation bias Anchoring Implicit biases KEY POINTS Cognitive errors occur in every profession. Dozens of cognitive errors have been demonstrated to happen in medicine, many leading to poor patient outcomes. The diagnosis of child maltreatment can be susceptible to cognitive errors because of conditions of high stress, limited or questionable quality of data, and the subtlety of some diagnoses. Errors may be avoided through deliberate efforts to consider alternative diagnoses, avoiding premature closure, seeking objective input without social cues, and multidisciplinary collaboration.
Medical decision making is an exercise in process management: health care providers must gather data from multiple data sources, sometimes simultaneously, interpret signs and symptoms, sometimes without adequate history to fully understand the information provided, and direct patient management including further diagnostics and treatment, often without complete data. Most of the time the process seems to work adequately, that is, the patient is appropriately diagnosed and treated. Yet there continue to be studies that demonstrate failures in diagnosis and treatment related to factors that seem to be unrelated to the medical condition and more related to the patient’s specific demographic characteristics.1–4 It is important that the diagnosis of child maltreatment be made accurately. Failure to correctly diagnose an abused child as abused (a false negative) could result in that child being returned to a potentially dangerous environment. Conversely, the overdiagnosis of child abuse (a false positive) could result in a child being removed from an environment wherein no one has caused harm to the child. The potential for errors
Disclosures: None. Department of Pediatrics, Primary Children’s Hospital, University of Utah School of Medicine, 675 East 500 South, Suite 300, Salt Lake City, UT 84102, USA E-mail address: [email protected] Pediatr Clin N Am 61 (2014) 997–1005 http://dx.doi.org/10.1016/j.pcl.2014.06.012 pediatric.theclinics.com 0031-3955/14/$ – see front matter Ó 2014 Elsevier Inc. All rights reserved.
998
Laskey
in decision making related to child abuse extend far beyond the health care arena and include law enforcement, social welfare systems, prosecutorial decisions, and judge or jury decisions. Understanding the potential pitfalls in decision making is especially important for process improvement. Social psychologists have spent considerable effort in understanding how humans think, but only relatively recently have physicians looked to the literature to understand what clinicians can do better. Humans have evolved to process an astonishing amount of information, often dealing with large simultaneous sensory inputs assaulting all of their senses. How is it that we are able to take in information, filter out what is not immediately relevant, and arrive at a reasonable conclusion? Are there shortcuts we can use to improve our processing speed without sacrificing the quality of the decision making? Perhaps most importantly, is it possible to avoid cognitive errors that we make and are compounded when community professionals from other disciplines, such as law enforcement and child protection services, are involved in a decision process? COGNITIVE PROCESSES: AN OVERVIEW
Humans routinely make decisions in an overwhelming, complex environment. There are also limited cognitive resources available at any given moment to process all of the information available. Given the inability to completely process all of the data, humans have adapted by using cognitive shortcuts, which can be both hardwired and derived. One easy-to-appreciate example of a hardwired shortcut is that of recognizing potential danger. It does not take a prior experience with a potential threat, such as an obviously hostile animal, to realize that care must be taken. Derived shortcuts also are common occurrences that increase over the course of a lifetime. Driving represents a classic example of a derived shortcut. Memories of learning to drive often will conjure the stress associated with the myriad of data bombarding the novice driver. It is only after years of practice that the process becomes nearly fully automated. What was once a difficult, stressful, and complex effort changes into a background process that frees the mind to handle other tasks. However, this automation can be an impediment to rapid responses to the unexpected. Deviations from the norm will take longer to recognize and process when one is not fully engaged in the task at hand. Just as driving or tying one’s shoes, or doing any number of complex learned tasks becomes less of an elaborate mental juggling act, so too does one’s ability to complete tasks associated with the work. In the early years of a person’s career, every part of the job requires thought and deliberation. As one gains experience and feedback on process and decision making, the process becomes more refined and automated. However, just as automation improves efficiency in tasks such as driving, it can also decrease the ability of an individual to recognize and respond to deviations from the norm. Decision theory is the study of identifying what is known, what is unknown, and the other factors involved in arriving at an optimal (and therefore correct) decision. Social psychologists use multiple techniques to understand the processes involved in decisions. Decisions are very often reliant on heuristics, which are the shortcuts used to aid in decision making sometimes referred to as rules of thumb; new, incoming data are constantly incorporated and compared with previously gathered information. If they match previously acquired data, a shortcut in the thought process may be used to arrive at the answer. Heuristics can certainly be helpful in improving efficiency, but also may lead one astray if they compound biases in the thought process. Biases may be either explicit or implicit. An implicit bias is one that an individual holds without being aware. Implicit biases can be difficult for an individual to
Cognitive Errors
acknowledge or control, given that they are happening without any real understanding or conscious effort. It has been clearly demonstrated that even people who profess to be egalitarian in their beliefs, and act outwardly in egalitarian ways, hold implicit biases that can influence decision making.5 It is especially important to draw a distinction between implicit and explicit biases. Explicit biases are very much part of a person’s conscious awareness and can be considered less socially desirable. Examples of explicit biases include overt racism, sexism, or ageism. Much is now known about how humans process complex information. Research has shown that it is possible to improve thinking and reduce errors in judgment. Multiple studies have demonstrated that cognitive errors may lead to incorrect medical diagnoses. It is therefore important that professionals engaged in these high-stakes decisions, such as the diagnosis of child abuse, make efforts to understand how and when these errors might occur and how best to mitigate their effects. COGNITIVE ERRORS AND CHILD MALTREATMENT
There are many common assumptions made by professionals that might not apply when the case involves child abuse. First, health care professionals assume that parents will be open and honest when seeking medical care for their child. However, if the perpetrator is a parent, he or she may not share the relevant information leading up to the symptoms that brought the child to medical attention, or worse, this person may lie and obfuscate. Second, there is a generally held belief that health care professionals use science and logic alone to reach a correct diagnosis, yet there is evidence to the contrary that “the art of medicine” is actually antithetical to science in some cases.6 There are situations wherein a better, more scientifically sound approach to medical care is available to clinicians who resist implementation of the new protocol because they “know what is best.”7 Third, primary care providers often feel they know their patients’ families and can judge who is at risk and who is not. Finally, primary care providers often feel that they would be able to detect problems in parent-child relationships by how a child interacts with their parent (eg, the child seems well bonded to the mother so there is no reason to think the mother could be abusing him).8,9 None of these factors is meant as an indictment of a profession or group of individuals; rather, it is an acknowledgment of the complexities of the human thought process in light of difficult decisions. The problems inherent in the human thought process are abundant. Many cases of potential child maltreatment are subtle and there are few “a-ha” moments that bring absolute clarity. Sometimes, despite best efforts, professionals are not fully engaged in the work of the day and rely more readily on derived shortcuts without realizing this is occurring. Often, parents and caregivers do not know what information is relevant or most important to share when talking to medical personnel or investigators, causing them to offer information that could be construed as misinformation or attempts at deception. Moreover, in other circumstances parents or caregivers may lie deliberately. A closer look at common cognitive errors can help illustrate how they can confound arrival at an optimal answer. An in-depth exploration of all possible cognitive errors is beyond the scope of this article (Table 1). Implicit Stereotypes
Stereotypes have a negative connotation, suggesting willful discrimination. Implicit stereotypes, however, are those that are typically not in the conscious awareness of the holder. Stereotypes are derived shortcuts that allow rapid categorization of people
999
1000
Laskey
Table 1 Common errors and possible solutions Type of Cognitive Error
“Symptoms”
Possible Solution
Implicit stereotypes
Relying on generalizations to describe a family or caregiver. Examples: “bad parent,” “nice family”
Using another colleague to staff the case because he or she will not have the same initial impression of the family
Anchoring
Difficulty considering alternative diagnoses even if all the information does not fit as one would expect
Devil’s advocate (either a colleague or simply proposing countertheories to one’s self), intentionally proposing alternative diagnoses to see if the working diagnosis is still the best fit
Triage cueing
Sending a patient to a specialist based on a very specific symptom or finding (which could be an anchor)
Multidisciplinary approach to diagnosis allows multiple viewpoints given identical case details, prevents “thinking in your box”
and naturally develop over a lifetime of exposures to information from the environment. When asked to conjure a mental image of an elderly person or a nurse or a homeless person, it is likely that this is not a difficult task. Clearly the image called to mind does not accurately represent all who are in that category, being a mental shortcut, but it does serve a purpose. Likewise, if a medical provider interacts with a family in a clinical setting, it is not difficult to understand that a subconscious categorization occurs, putting them into a “good” or “bad” family category without objective data to support this segregation. Some data that are used to categorize people are present immediately on first contact: the style of clothes or body art, the age, the speech patterns, and the color of the skin all are seen, processed, and sorted without conscious effort. Just as stereotypes are heuristics that allow us to categorize people easily, if not necessarily accurately, biases also exist and are frequently implicit. Social psychologists have demonstrated repeatedly that in-group preferences, that is, preferring those who are like as opposed to those who are different, are more natural than out-group preferences. For this reason, it is more natural to give the benefit of doubt to those who are similar to the observer and hold those who are not to a different standard. In perhaps the most powerful demonstration of implicit biases, the Implicit Association Test developed by Nosek and Banaji (Available at https://Implicit.Harvard.edu) has been used to demonstrate people’s natural tendency to easily categorize consistent pairings while having more difficulty categorizing inconsistent pairings. In the classic example, the time lag in a rapid categorization test is measured when a subject is required to associate a negative word with a white person’s face or a positive word with a black person’s face. It has been shown to be demonstrably slower than when the pairing is a positive word and a white face or a negative word and a black face. Because the positive or negative association is implicit, people are often surprised to learn their results and frequently express frustration because they perceive themselves to be egalitarian. The relevance of stereotypes and implicit biases is found throughout the child maltreatment literature. Hampton and Newberger10 found that hospitals were more likely to report a family to child protective services (CPS) if the patient was black,
Cognitive Errors
the parents were young, or the family was poor. In a study of patients presenting to the emergency department, Jenny and colleagues11 found that abusive head trauma was missed nearly one-third of the time on initial presentation to medical care, and cases were more commonly missed if the patient was white or from an intact family. In another study of children admitted to an urban academic medical center with skeletal injuries, children with skeletal injuries who were from minority or low-income families were more likely than white or insured children (a proxy for socioeconomic status) to be evaluated for abuse and to be reported for abuse.4 In an attempt to create an experimental model for the influence of race and socioeconomic status on the diagnosis of child abuse, Laskey and colleagues12 used a methodology adapted from the psychology literature, and showed a difference in the frequency of the diagnosis of abuse among low-income white children in comparison with other categories. In this study, practicing primary care pediatricians were asked to rate the likelihood that an injury of ambiguous etiology (ie, it could potentially have been either accidental or inflicted) was abusive. The results demonstrated that socioeconomic status had more of an effect on the diagnosis of abuse than did race, despite identical clinical histories.12 In the law review literature, stereotypes have been shown to influence behavior and memories in investigators, prosecutors, and juries. The concept of misremembering, that is, creating a new memory from pieces of information that are reassembled into a new memory, has been shown to occur in stereotypical ways. In one study, a story of a crime is presented and the race of the characters is changed. The results showed that people tended to forget mitigating factors for black characters, and hostile actions were more likely to be attributed to them, even if they were not the aggressors in the story. Similarly, hostile actions by white characters were often forgotten. Of note, research subjects were often very confident in their memories, even when those memories were wrong.13 It should be apparent, therefore, that implicit stereotypes and implicit biases play an important role in how information is processed, particularly when a quick decision is required or a decision that relies on less than optimal information is necessary. How might this affect the accuracy of determinations when it comes to child maltreatment? Implicit beliefs may lead one to accept a story at face value because the alternative (eg, the diagnosis of child abuse) does not fit an implicit belief. If presented with a “nice” family and a child with injuries possibly related to child abuse, it might be an easier path to follow to consider alternative diagnoses. It is not uncommon that a bleeding condition, metabolic or genetic bone disease, nutritional deficiency, previously unrecognized congenital anomaly, or an accident is more aggressively pursued as the potential cause rather than to consider the unfortunately more common explanation of abuse. This is not to say that these conditions should be dismissed from the differential diagnosis, but neither should abuse be ignored in the evaluation. Alternatively, an ascertainment bias (also known as self-fulfilling prophecy), that is, looking for something only where it is expected to be found, may occur. If one only believes that “bad parents” (defined in whatever subconscious way that a health care provider believes) abuse their children, people who fit such a model of bad parenthood will be much more likely to be evaluated for abuse than those who do not. Of course cases are more likely to be found where one looks, leading to a circular fulfillment of the bias. Anchoring
Anchoring is a problem of premature closure. When first presented with information, both explicit information and implicit data that are perceived but are not explicitly presented, the human tendency is to begin an immediate sorting and processing of the
1001
1002
Laskey
data. Data about a patient and family are available for this on first contact with the chart. Names often are markers of social position and frequently have racial or ethnic clues.14 Race is commonly assigned by appearance and not by a parent’s report. A clinical interaction begins with questions by a provider and answers from a caregiver or parent. From the outset, the clinician is building and testing hypotheses to explain the chief complaint and the associated findings. Sometimes these initial impressions form an anchor from which it is challenging to change course. Once the basic categorizations are made, further information is processed in that context. Anchors can be descriptors (eg, “difficult parent,” hypochondriac), or a working diagnosis or historical piece of data (eg, fever several days ago but not currently, illness in household). Just as stereotypes can lead to ascertainment bias, anchors may result in a confirmation bias. Humans like to be correct and will often work to find support for a theory. Examples abound of people’s willingness to disregard information that is contradictory to a held belief and then embracing supporting information, even if it is dubious. Confirmation biases are especially problematic in accurate decision making because once a hypothesis has been generated, the tendency is to abandon opposite hypothesis testing. This action can result in the inability to accurately process new historical information or new test results. When doctors perform testing, the analysis of the new data is often in the context of whether it fits with the working theory. Although it is often said that something is being “ruled out,” research has shown that what is happening more often is a “rule in.” As one builds a mental case for a given condition, information that does not support the theory is more difficult to “see,” or be incorporated into the overall picture. It is not unusual that the information is not even truly processed at the same level as the supporting data. It is the avoidance of cognitive dissonance that leads to a cycle of seeking confirmation of one’s theories and finding support in data that are obtained, even if there may be more than one way to interpret the new information.15 Triage Cueing
The adage “to the hammer, the world is a nail” is an apt description of triage cueing. When a patient is referred to a specialist, a specific question is often posed that the specialist would be ideally suited to answer. However, this begs the question of whether the information used to arrive at the referral to the specialist was correct or whether any cognitive errors may have occurred along the way. If a physician is evaluating an infant with bruising, referral to a hematologist will result in an evaluation of whether the patient has a bleeding disorder. If such a workup fails to demonstrate abnormality, the question has been answered and the case is returned to the hands of the primary care physician (PCP). But was the question answered? Whose responsibility is it to pursue the cause of the bruising? This closing of the loop sometimes does not occur. It is the natural inclination of a specialist to answer the question that is asked, not answer the unasked question. Triage cueing, that is, the referral based on an assumption of diagnosis, could result in missing the ultimate diagnosis if the referral was to the wrong specialist. In child abuse pediatrics, the nature of the evaluation is inherently multidisciplinary. When a PCP consults with a child abuse pediatrician, the effects of triage cueing can be mitigated by having it be the responsibility of the child abuse pediatrician to engage other specialists while sharing the broad base of information gathered in the course of the medical evaluation. A corollary to triage cueing is diagnosis momentum. In cases of caregiver-fabricated illness (previously known as Munchausen syndrome by proxy and medical child abuse), patients often have exhaustive problem lists with many diagnoses. When a
Cognitive Errors
parent seeks care with a new clinician, and provides what appears to be a detailed medical history and even supporting documentation for the child’s diagnoses, it is easy to start with these data and move forward with the evaluation and management. The problem lies in the complexity of the case and the disinclination to “start from the beginning.” Diagnosis momentum uses the preassigned diagnosis as a given fact. Unfortunately, the disorder inherent in caregiver-fabricated illness relies on this flaw in thinking. It is not uncommon to discover on thorough review of the medical documentation that many of the diagnoses carried by these patients are in fact conditions that were in a differential diagnosis and then ruled out. Similar to anchoring bias, labels are sticky and will build over time, placing these children at extreme risk of morbidity and mortality.16 COUNTERING ERRORS IN THINKING
It is not possible to avoid all errors in a process that has so many unknowns. The goal of safe, effective medical care should be the minimization of errors wherever those errors may arise. Psychologists have shown that there are processes that can be used to improve decision making. The following are some easy-to-apply strategies to improve decision making. 1. An important countermeasure is the use of the devil’s advocate who, either as an individual working on a case or as a member of a broader team of professionals, can pose the question, “what are the alternative explanations for the information currently available?” When this question is asked as a member of a diverse team, it is possible to consider alternative hypotheses and propose tests and theories while minimizing the effects of a confirmation bias. 2. Avoid subjective or emotive descriptors. Many biases are compounded inadvertently by the way people naturally communicate. Descriptors are important to building a mental image. In medicine, handoffs between professionals, such as occur during shift changes or between teams in the hospital setting or even between professions when a case is passed from the medical team to CPS are frequently filled with descriptors. These details influence the receiver of the information by coloring how new incoming information is processed. If a parent is framed in a negative light, the new provider will interact with that parent from the anchor of this negative information. If a diagnosis is suggested as the working hypothesis, it will form an anchor against which new incoming data will be compared. Clearly it is important to relay information gathered to promote effective communication about a patient. The key to minimizing the cognitive errors in the system is to consider how that information is relayed. Data presented in handoffs should be evaluated to determine if it is necessary to the care of the patient, objective or subjective, and known versus hypothesized. 3. Seek objective input from colleagues on challenging cases. Another strategy that can be used with socially charged situations, such as when a particular situation is triggering a visceral reaction (eg, “this is a good family” or “this person is really frightening”), is to present purely objective data to a colleague. Although clearly a diagnosis cannot be made in a vacuum of information, by stripping away some of the social cues that might trigger an implicit bias, the diagnostic process can be improved. The history, medical presentation, physical examination findings, and test results can be objectively interpreted by someone who has not had his or her implicit biases activated. It is important when using this approach to present the data as cleanly as possible without any extraneous commentary to avoid swaying the interpretation.
1003
1004
Laskey
4. Multidisciplinary thinking. Child abuse pediatricians often work as part of a larger multidisciplinary and interdisciplinary team. A key strategy to avoiding errors such as anchoring or triage cueing is to think broadly, the “think outside of the box” approach. Because the diagnosis of child abuse is a difficult one to make with serious consequences for a wrong diagnosis in either direction, it is especially important to use the collective wisdom of a team. Because each provider on the medical team will interpret the information through the lens of his or her specialty, open discussion of supporting information in addition to refuting information will improve the quality of the diagnosis, avoiding a singular focus. SUMMARY
The diagnosis of child abuse is one that no provider wants to make, including those in primary care. It is known that cognitive errors happen, but they are not inevitable and steps can be taken to minimize them. It is through introspection that process improvement can be achieved. Diagnosis and medical management can be optimized through collaboration, discussion, and effective teamwork. Mistakes can lead to a better understanding of how humans think. As James Joyce said, “mistakes are the portals of discovery.” REFERENCES
1. Loring M, Powell B. Gender, race, and DSM-III: a study of the objectivity of psychiatric diagnostic behavior. J Health Soc Behav 1988;29(1):1–22. 2. McKinlay JB, Potter DA, Feldman HA. Non-medical influences on medical decision-making. Soc Sci Med 1996;42(5):769–76. 3. Feldman HA, McKinlay JB, Potter DA, et al. Nonmedical influences on medical decision making: an experimental technique using videotapes, factorial design, and survey sampling. Health Serv Res 1997;32(3):343–66. 4. Lane WG, Rubin DM, Monteith R, et al. Racial differences in the evaluation of pediatric fractures for physical abuse. JAMA 2002;288(13):1603–9. 5. Nosek BA, Greenwald AG, Banaji MR. The implicit association test at age 7: a methodological and conceptual review. Social psychology and the unconscious: the automaticity of higher mental processes. East Sussex (UK): Psychology Press; 2007. p. 265–92. 6. Suarez-Almazor ME, Russell AS. The art versus the science of medicine. Are clinical practice guidelines the answer? Ann Rheum Dis 1998;57(2):67–9. 7. Bohmer R, Edmondson A, Feldman L. Intermountain healthcare. Harvard Business Review School 2002. Case 603, 066. 8. Flaherty EG, Sege RD, Griffith J, et al. From suspicion of physical child abuse to reporting: primary care clinician decision-making. Pediatrics 2008;122(3):611–9. 9. Sege R, Flaherty E, Jones R, et al. To report or not to report: examination of the initial primary care management of suspicious childhood injuries. Acad Pediatr 2011;11(6):460–6. 10. Hampton RL, Newberger EH. Child abuse incidence and reporting by hospitals: significance of severity, class, and race. Am J Public Health 1985;75(1):56–60. 11. Jenny C, Hymel KP, Ritzen A, et al. Analysis of missed cases of abusive head trauma. JAMA 1999;281(7):621–6. 12. Laskey AL, Stump TE, Perkins SM, et al. Influence of race and socioeconomic status on the diagnosis of child abuse: a randomized study. J Pediatr 2012; 160(6):1003–8.e1001.
Cognitive Errors
13. Levinson JD. Forgotten racial equality: implicit bias, decision-making and misremembering. Bepress Legal Series. 2006:1630. 14. Bertrand M. Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. National Bureau of Economic Research; 2003. (Working Page 9873). 15. Croskerry P. Achieving quality in clinical decision making: cognitive strategies and detection of bias. Acad Emerg Med 2002;9(11):1184–204. 16. Flaherty EG, MacMillan HL, Christian CW, et al. Caregiver-fabricated illness in a child: a manifestation of child maltreatment. Pediatrics 2013;132(3):590–7.
1005