Chapter 1. An Internist’s Approach to the Neuropsychiatric Patient

Diagnosis traditionally has been viewed as an analytic reasoning process akin to the scientific method. Early clinical observations or patient complaints trigger hypotheses. A clinician tests these hypotheses by obtaining additional historical and physical examination data to determine the merits of the competing hypotheses. The physician synthesizes the relevant data to create a composite of the patient’s clinical findings, called a problem representation (Bowen 2006). A clinician finds the best match for this problem representation from the mental representations of diseases that he or she has developed through experience and study. If no diagnosis reaches a probability of disease above which the clinician believes that the benefits of treatment outweigh the risks of further testing (i.e., threshold to treat), the clinician obtains additional collateral history and laboratory or radiological data until the data reach the threshold to treat.

The cognitive psychology literature indicates that the analytic reasoning approach of the scientific method is only one of two possible pathways to diagnosis. Kahneman (2009) referred to these two pathways as slow (i.e., analytic reasoning) and fast thinking (i.e., nonanalytic or intuitive reasoning; Kahneman 2009). Nonanalytic clinical reasoning typically occurs through rapid, unconscious pattern recognition, with the diagnosis emerging before the scientific method can be applied. For example, when a clinician sees a patient’s skin covered with “dew drops on a rose petal” in a dermatomal distribution, the diagnosis of shingles immediately comes to mind. In such cases, the scientific method is set aside because of experience, diagnostic confidence, and efficiency. When no obvious pattern is recognized, a clinician typically combines nonanalytic and analytic reasoning, with certain patterns triggering a nonanalytic differential diagnosis followed by an analytic consideration of each of the potential diagnoses.

Nonanalytic reasoning is a black box; an observer can infer its nature only through the statements and decisions of the observee. However, even direct questioning about a doctor’s clinical reasoning will provide insight only into conscious cognitive processes, whereas subconscious processes (e.g., implicit biases) will remain obscure to both observer and observee. Based on psychological research, script theory has emerged as a conceptual framework that can help explain some of the observed phenomena of nonanalytic reasoning. Scripts have been defined as

For example, a restaurant script includes waiting to be seated, ordering drinks and then your meal, tipping the waiter, and so on. One follows this routine or script without thinking. Script theory was transferred to the diagnostic literature through the term illness scripts (Clancey 1983). Illness scripts describe clinicians’ mental representations of “the stories of diseases”: their clinical findings, associated risk factors, and pathophysiology. For example, a psychiatrist may rapidly recognize the script or pattern of a depressive disorder in a sad, anhedonic patient with a flat affect and a family history of depression. Each clinician has idiosyncratic illness scripts for a given disease based on experience and “book knowledge.” Illness script theory provides valuable insights into nonanalytic rapid diagnostic reasoning.

Accurate, comprehensive knowledge of illness scripts distinguishes expert and experienced clinicians from novices (Elstein et al. 1978). This knowledge forms the basis for heuristics (i.e., mental “rules of thumb”) and pattern recognition, in which a clinician will mentally try to match a patient’s presenting clinical findings with the appropriate illness script. Given the complexity of medical diagnosis, the accuracy of diagnostic reasoning is remarkable, and errors occur because of bias or misapplication of heuristics. The literature describes more than 30 different types of heuristics and biases (e.g., availability, representativeness), but the specifics are less important than the recognition that a clinician should seek ways to minimize these errors. The most common suggested “antidote” to nonanalytic reasoning–based diagnostic errors is for a clinician to force a “double-check” of the diagnosis with analytic reasoning, via a cognitive forcing strategy. For example, a clinician feels confident or certain about a diagnosis of major depressive disorder but chooses to review the medical differential diagnosis of depression and orders a thyroid-stimulating hormone level test to evaluate for hypothyroidism. The notion of a double-check on nonanalytic reasoning leads directly to a consideration of analytic reasoning. Given the tendency of many clinicians to discount medical causes of psychiatric symptoms and, conversely, psychiatric causes of general medical symptoms, an awareness of the limitations of implicit scripts is particularly important in this area.

Six traditions of medical analytic reasoning have been described: anatomic, pathophysiological, descriptive, criteria-based, probabilistic, and biopsychosocial (Richardson 2007). Anatomic analytic reasoning is based on finding pathology consistent with a given disease in a relevant specimen. Pathophysiological, or causal, analytic reasoning emphasizes coherence of the pathophysiological findings with the most likely diagnosis. Descriptive analytic reasoning is an explicit, conscious process of matching a patient’s presenting clinical findings with typical or atypical descriptions of a given disease. Criteria-based analytic reasoning extends descriptive reasoning by using explicit criteria with a scoring system for determining a diagnosis within a taxonomy. These latter two models—descriptive and criterion-based—are forms of reasoning closest to the current diagnostic model in DSM-5 (American Psychiatric Association 2013) and in clinical psychiatry, where we have, at present, a more limited number of anatomic or pathophysiological models. Probabilistic (i.e., Bayesian) analytic reasoning focuses on revising probability estimates of diseases based on clinical findings, using prediction rules or likelihood ratios when available, until the threshold to treat has been reached. Finally, biopsychosocial analytic reasoning—an approach that has been explicitly used in clinical psychiatry since the classic writings of George Engel (1977)—is a holistic process that focuses not only on a biological diagnosis but also on psychological and social aspects of a case to apprehend the causes and the patient’s experience of an illness.

Although anatomic analytic reasoning sets the gold standard, for some diagnoses the risks of a pathological approach outweigh the benefits. For example, a biopsy may entail a significant probability of morbidity or mortality or may lack adequate sensitivity (i.e., too many false-negative results) or specificity (i.e., too many false-positive results). In these cases, internists fall back on the other traditions, and diagnosis becomes a probabilistic science to an even greater degree. A clinician must infer the diagnosis by seeking the disease that best explains the clinical findings within a given context (Lipton 2000).

If one considers psychiatry in the context of the six traditional approaches to medical analytic reasoning, it can be said that descriptive reasoning served as the primary approach to diagnosis until the period of neuroanatomic and infectious disease–related investigation in the late nineteenth century (e.g., the work of psychiatrists Alois Alzheimer and Solomon Carter Fuller in dementia-related illness), when classification schemes emerged for the purpose of collecting census data on mental illness. These classification efforts began with the 1952 publication of DSM-I but were more fully realized with the 1980 publication of DSM-III, which ushered in an emphasis on criteria-based diagnosis in psychiatry (American Psychiatric Association 1952, 1980). The predominance of this analytic tradition in psychiatry, in the philosophical sense, stems from the limited pathophysiological knowledge of psychiatric conditions, the lack of “objective” gold standards for the diagnosis of most conditions (Kapur et al. 2012), and the historic tendency for diagnoses with clearer pathophysiology or etiologies (e.g., the psychoses associated with tertiary syphilis) to be subsumed by other branches of medical practice. In addition, many of the descriptive criteria for psychiatric illnesses include physical symptoms but are not normed in populations that are medically ill or complex. Such an approach is not limited to psychiatry. For example, the American College of Rheumatology has created a set of diagnostic criteria (i.e., American College of Rheumatology–endorsed criteria for rheumatic diseases) for diseases within the subspecialty’s purview, because rheumatologists face the same diagnostic challenges that psychiatrists do (American College of Rheumatology 2017).

Psychiatric diagnosis, especially when dependent on experience-based implicit scripts, is even more difficult if we consider the behavioral disturbances of some psychiatric illnesses, such as those that occur with somatic symptom disorder and some personality disorders. The emotional intensity of some forms of psychiatric illness, sometimes described with the terms transference and countertransference, can create diagnostic and management issues that can be quite complex in both psychiatric and general medical practice (Groves 1978; Kahana and Bibring 1964). Cognitive psychologists refer to the negative influence that these emotional phenomena can have on diagnostic accuracy as affective bias (Croskerry 2005). Objectivity and emotional neutrality toward patients are often considered best practices in medical diagnosis. Yet the emotions that patients engender in psychiatrists and other physicians can serve as key guides to clinical care and as clues that increase or decrease the likelihood of accurate identification of a disease. These emotional responses are idiosyncratic, and each physician, based on his or her own experiences over years of practice, will have different opinions about the value of these responses in aiding the diagnostic process. Thus, affective bias can be a double-edged sword in neuropsychiatric diagnosis.

As if neuropsychiatric diagnosis were not difficult enough, medical disorders that can present as psychiatric illness compound the challenge. Clinicians must distinguish among three possibilities: 1) a primary medical disease with psychiatric symptoms not related to a known psychiatric disease, 2) a primary psychiatric disease with physical symptoms not caused by a known medical disease, or 3) primary medical and psychiatric diseases in which physical and mental symptoms are caused or exacerbated by both diseases (Figure 1–1). The incidence of medical misdiagnosis in patients with psychiatric symptoms (i.e., a psychiatrist or nonpsychiatrist physician diagnosing a primary psychiatric disease when the cause is a medical disorder) is controversial, with estimates ranging dramatically (from 19% to 80%) across studies of varying methodological rigor (Carlson et al. 1981; Hall et al. 1978, 1981; Han et al. 2009; Koranyi 1979; Olshaker et al. 1997; Riba and Hale 1990; Summers et al. 1981; Zun et al. 1996).

Regardless of the incidence of misdiagnosis, improving diagnostic accuracy is desirable, given the potential for reduced morbidity through appropriate treatment of medical disorders or for more effective care of both medical and psychiatric disorders when they co-occur.

The causes of these errors are multifactorial. A lack of skill and practice in both medical history taking and physical examination performance may increase the risk for diagnostic error. In one study, 32% (24/73) of faculty and private practice psychiatrists reported that they felt less than competent to perform a basic medical examination (McIntyre and Romano 1977). Frequent performance of an initial physical examination for inpatients and outpatients was reported by only 13% (10/73) and 8% (6/73) of these psychiatrists, respectively, although these percentages likely vary significantly depending on the setting (i.e., inpatient vs. ambulatory) and on how recently psychiatrists were trained. Regardless of setting, failure to perform an initial intake physical examination increases the risk for error due to nonanalytic bias. The availability heuristic (“Common things are common”) and the representativeness heuristic (“If it looks, quacks, and waddles like a duck, then it’s a duck”) are powerful diagnostic strategies that can lead psychiatrists or other physicians astray when they are dealing with medically confusing presentations. In addition, confirmation bias (i.e., overvaluing data that support one’s leading hypothesis and undervaluing data that do not) can lead a clinician to ignore or even fail to collect data relevant to a possible medical diagnosis.

Improving psychiatrists’ and other physicians’ knowledge about medical disorders that present psychiatrically and promoting development of clinical skills are obvious ways to reduce diagnostic error. This approach requires significant changes in physician awareness of the causes of these types of medical disorders, exposure to complex cases within residency training, and postgraduate professional development. Because medical students learn about medical and psychiatric illnesses in a siloed way, they may find it difficult to develop medical diagnostic hypotheses for psychiatric problems. Overcoming this cognitive bias is essential in clinical training. Knowing the contexts within which errors are most likely to occur and understanding the cognitive psychology literature may increase clinicians’ suspicion for medical causes of psychiatric symptoms. Although some have advocated routine use of comprehensive examinations and extensive laboratory testing to rule out medical diagnoses, little evidence exists to support either the diagnostic utility or the cost-effectiveness of such a strategy. A targeted approach seems more appropriate, given the low yield of comprehensive laboratory testing.

An extensive literature review conducted by the American Association for Emergency Psychiatry (AAEP) Task Force on Medical Clearance of Adult Psychiatric Patients (Anderson et al. 2017) yielded a list of patient characteristics suggesting higher risk both for medical mimics of psychiatric disease and for coexisting medical illness (Table 1–1; Wilson et al. 2017). Patients with psychiatric presentations who have any of these risk factors should receive a comprehensive medical examination.

Approaching from a different angle the dilemma posed by mixed physical and psychiatric symptoms, Shah et al. (2012) developed a screening algorithm designed to identify patients with psychiatric manifestations of a medical disease who could be directly referred for psychiatric evaluation without further medical workup or laboratory testing. This algorithm consisted of five criteria, all of which needed to be met for “medical clearance” and discharge to a psychiatric receiving facility:

In a retrospective chart review of 485 emergency department patients who were believed to have a primary psychiatric disorder, only 1.2% of the patients who met all five of these “medical clearance” criteria prior to transfer to psychiatric care required further medical evaluation (Shah et al. 2012). Despite the circularity existing in the criteria (i.e., no evidence of an acute medical condition), their excellent negative predictive value suggests that clinicians could use them effectively to rule out medical complexity. Although the screening algorithm of Shah et al. (2012) has yet to be prospectively validated, its five “medical clearance” criteria may—together with additional risk factor criteria derived from the AAEP Task Force recommendations (Wilson et al. 2017)—yield a reasonably accurate patient risk profile that can help both psychiatric and nonpsychiatric clinicians decide whether to perform additional laboratory testing across various settings.

In addition to the risk factor checklist (see Table 1–1), the cognitive psychology literature describes cognitive forcing strategies as another means of improving diagnostic accuracy and reducing error. A cognitive forcing strategy (CFS) is a reasoning approach in which the clinician “forces” him- or herself to use analytic reasoning before making a final diagnosis, even when it seems unnecessary. In terms of the best strategy, a recent meta-analysis found that a reflection-based CFS focused on verification of an initial diagnostic hypothesis, through either a protocol or a differential diagnosis checklist, improved accuracy on paper- or Web-based clinical vignettes, whereas a reflection-based CFS focused on general reflection (e.g., “Reconsider your diagnosis”) or cognitive “debiasing” (e.g., metacognitive self-reminders to “Be aware of your biases”) did not (Norman et al. 2017). A typical diagnostic verification CFS protocol is presented in Table 1–2 (Mamede et al. 2008). In two studies in which internal medicine residents used a diagnostic verification protocol to check their initial diagnoses in paper-based simulated diagnostic problems, investigators found that this type of CFS protocol increased diagnostic accuracy by up to 40% (Mamede et al. 2008, 2010). This improvement occurred primarily with difficult cases, when initial diagnostic accuracy was less than 50% (Mamede and Schmidt 2017).

On the basis of the literature reviewed in the previous section, we suggest a best-practice strategy for reducing medical misdiagnosis in psychiatry and medicine. Initially, a clinician can use the “medical clearance” criteria of Shah et al. (2012) (see previous section, “Strategies for Minimizing Medical Misdiagnosis”) to determine the need for a comprehensive medical evaluation in settings other than emergency departments and inpatient units (in which a comprehensive medical examination is standard practice). If a nonpsychiatrist performs the initial evaluation and the patient meets all five of the Shah et al. (2012) criteria as assessed by this physician, then the patient can be considered “medically cleared” and the psychiatrist can focus attention on psychiatric diagnoses, assuming that the medical evaluation is considered adequate (i.e., no evidence of an acute medical condition). If the psychiatrist feels competent to perform the initial evaluation and the patient does not meet all five criteria, then the psychiatrist may choose either to perform a comprehensive medical examination him- or herself (if he or she is comfortable with those clinical skills) or to request a medical consultation. If the psychiatrist chooses to conduct the examination him- or herself, a diagnostic verification CFS should be used with some type of clinical decision support (e.g., a checklist of common medical masqueraders) for every encounter, because clinicians often “don’t know when they don’t know” (Meyer at al. 2013). For example, in one study that examined the relationship between physicians’ diagnostic accuracy and their confidence (rated on a scale of 1 to 10) in that accuracy (Meyer at al. 2013), internists’ self-rated confidence remained similar regardless of the accuracy of their diagnoses (confidence ratings of 7.2 and 6.4 for diagnostic accuracies of 55% and 5%, respectively). Both overconfidence and cognitive “miserliness” (i.e., the tendency of the brain to minimize its effort [Stanovich 2018]) make the use of diagnostic verification CFS protocols challenging to implement for individual clinicians. Incorporation of such protocols into the electronic medical record—or into clinical decision support systems that generate differential diagnosis checklists highlighting medical presentations for common psychiatric problems—could facilitate their use. Finally, psychiatrists, like all physicians, need to be prepared to question even established diagnoses when patients fail to respond as expected or their clinical course is atypical. Obtaining a second opinion from a colleague in the same or another specialty can be important. This is especially true when the patient has an acute onset of new illness, subtle neurological signs, an atypical presentation, or an unexpected response to treatment. For all physicians, maintaining clinical humility, being aware of the high rate of error in all medical diagnostic processes, and appreciating the complexity of clinical decision making are important.

The following case examples illustrate the recommended approach:

This case illustrates the potential value of a careful assessment of risk factors (as described in Table 1–1) to reduce the risk of medical misdiagnosis.

This case illustrates the atypical presentations of thyroid-related disorders and the importance of thyroid screening—with the relatively inexpensive thyroid-stimulating hormone level test—for most, if not all, patients with psychiatric presentations.

Neuropsychiatric diagnosis is a particularly challenging task because of a relative paucity of definitive, “objective” signs or laboratory findings for many psychiatric illnesses. Diagnostic cognitive strategies in psychiatry, descriptive and criterion-based, also tend to be normed so as not to consider the role of other medical disorders as causes of psychiatric presentations. Although future research is likely to identify valuable diagnostic biomarkers, psychiatrists need strategies that they can use now to reduce the risk of medical misdiagnosis.

In this chapter, we provided two suggested strategies—a risk factor checklist (see Table 1–1) and a diagnostic verification CFS protocol (see Table 1–2)—for reducing error and enhancing accuracy in the diagnosis of medical symptoms presenting as psychiatric illness. Although some evidence supports these strategies, future research is needed to confirm their utility.