The importance of cognition for improving diagnostic safety: Salerno redux?

The medical school at Salerno

Much of modern medicine is focused on the attainment of declarative knowledge, or knowing what (typically the essential facts of anatomy, physiology, biochemistry and pathophysiology) while placing less emphasis on procedural knowledge or knowing how (how to think, reason, and make decisions about clinical data). In contrast, at the Schola Medica Salernitana, the medical school at Salerno in Southern Italy, founded in the 9th century and considered the oldest medical school of modern civilization, prospective medical students initially undertook three years of training in philosophy, logic, and dialectical reasoning (scientia logicalis), followed by five years of medical studies, in order to receive their medical certificate. Thus, an early emphasis on logic and reasoning was established to emphasise empirical observation and action based on rational thought. Medical students at Salerno would use these skills, what are now sometimes referred to as ‘soft’ skills, to identify the nature of the illness based on symptoms, environmental factors, and patient history, an early form of what now would be considered evidence-based medicine, and engage in rigorous discussion with colleagues to optimise their decisions [1] (Figure 1).

Figure 1:

A miniature depicting the Schola Medica Salernitana from a copy of Avicenna’s Canons (from Wikipedia).

Cognition may be defined as the mental action or process of acquiring and using knowledge and understanding through thought, experience, and the senses. Cognition is essential to the practice of medicine and the two have co-existed for thousands of years. The study of human cognition is Psychology. It is difficult to think of an area of clinical practice that does not depend on cognitive factors, yet these are not generally discussed in recent times in health professions education (or in pre-medical education at universities). Indeed, most healthcare providers have limited exposure to the whole realm of cognition and what it means for their behavior. Unquestionably, cognition plays a critical role in clinical medicine. This is especially true for the process of establishing a diagnosis, and efforts to improve clinical reasoning are now front and center in the quest to reduce the harm associated with diagnostic errors (Table 1).

The history of how cognition has risen to become the central challenge in efforts to improve diagnosis retraces the origins of the diagnostic error movement itself. Although the early years of the patient safety movement focused largely on system-related aspects of performance, the work of psychologist James Reason, and his famous ‘Swiss Cheese’ model of accident causation, emphasized the critical role that cognitive and human factor elements play in accident causation. His classic book Human Error [14], published in 1990, recounted the work of Herbert Simon on human decision-making [15], and how heuristics and biases can influence cognition, the work of Kahneman and Tversky [16], 17]. Simon received the Nobel Prize for his work in behavioral economics in 1978 and Kahneman received his in 2002 (Tversky died in 1996 and Nobel Prizes are not awarded posthumously).

The new interest in cognition and decision-making was revolutionary, and was especially relevant to patient safety, where the consequence of faulty-decision making translated directly to patient harm and lives lost. The patient safety movement solidified a relationship between Psychology and Medicine that didn’t exist before, and modern efforts to improve diagnosis are squarely focused on issues at this interface. Cognition is relevant to essentially every aspect of diagnosis and medical practice generally.

Impact of the Halifax series

Each country had its own initiatives for the emergence of their respective patient safety movements. A major feature in Canada was a series of 10 annual conferences from 2001 to 2010, held at different venues across the country, in which a variety of safety issues were explored, one of which was diagnostic failure. The series was initially developed by two emergency physicians, Pat Croskerry and Sam Campbell, and an anesthesiologist Jan Davies. Diagnostic error emerged as a topic of interest in 2005 [3], 18] although its magnitude was not fully appreciated at that time and it was not an immediate focus. In fact, medication errors were seen as a greater threat to patient safety getting 70 mentions in the 1999 IOM report “To Err is Human”, compared with 2 for diagnostic error [19], 20].This was not altogether surprising. Medication errors were proximal in nature i.e. typically visible, tangible, and measurable, whereas diagnostic errors, due to their inherently cognitive nature, were invisible.

At the same time, the meetings of the National Patient Safety Foundation dramatically increased interest in patient safety in the United States, building on “To Err is Human” [19]. Juxtaposed to the emerging interest in cognition and clinical reasoning, the relevance to the problem of diagnostic errors was glaring. It was this mix that led to the creation of the diagnostic error in medicine conference series, the formation of The Society to Improve Diagnosis in Medicine, and ultimately to the landmark report on Improving Diagnosis in Health Care from the National Academies of Science, Engineering, and Medicine (NASEM) [21]. The chronology and key intermediaries that culminated in publishing this remarkable document are illustrated in Figure 2.

Figure 2:

Chronology of events and organizations relevant to producing the NASEM report improving diagnosis in health care. TWA – Trans World Airlines; ASRS – Aviation Safety and Reporting System; IHI – Institute for Healthcare Improvement; NPSF – National Patient Safety Foundation; IOM – Institute of Medicine; DEM – Diagnostic Error in Medicine; SIDM – Society to Improve Diagnosis in Medicine; CIDM – Community Improving Diagnosis in Medicine.

Impact of improving diagnosis in health care

Improving Diagnosis in Health Care, for the first time in the patient safety movement, called out the central role of clinical reasoning as the basis for safe diagnosis: “Timely, accurate, and patient-centered diagnosis relies on proficiency in clinical reasoning, which is often regarded as the clinician’s quintessential competency.” The report emphasized a co-equal role of system-related and cognitive contributions to diagnostic error, and included clear direction in its conclusions and recommendations to improve clinical reasoning.

The report solidified the central role of Kahneman and Tversky’s ‘dual processing’ framework and relied extensively on the work of Croskerry and colleagues to illustrate how System 1 and System 2 processes constituted the essential elements of diagnosis throughout clinical practice (Figure 3) [22], [23], [24]. The particulars of this model, and its many implications for clinical decision making, lie at the heart of understanding the role cognition plays in every healthcare interaction today.

Figure 3:

The dual process model of clinical reasoning. The model runs from left to right, describing two systems (1 and 2), and their respective processes (type 1 and type 2). The patient initially presents with symptoms or signs that may form a pattern that is immediately recognised and the diagnosis can be made in a fraction of a second, activating type 1 processing. The four channels in system 1 characterise the four principal origins of type 1 described in more detail in Croskerry et al. [22] If the illness presentation is not recognised, system 2 may be activated typically resulting in a slower response that may take minutes to hours to days. If system 1 is activated and system 2 doesn’t approve, it can override it (executive override) signifying an executive function of the cognitive control that is achieved by the prefrontal cortex. The T represents a toggle function, ability to move back and forth between the two systems. System 2 can be overridden by a system 1 response. Overall, there is a strong tendency to default to system 1 where less cognitive effort is required – referred to as cognitive miserliness.

“Improving Diagnosis in Health Care” has had enormous impact on patient safety by leading to the emergence of a new field, focused on diagnostic error. Addressing diagnostic error requires attention to both the system-related and the cognitive elements of diagnosis, and thus a major imperative in this new area was the need to understand and improve diagnostic reasoning. Although landmark work on diagnostic reasoning had already been published [25], [26], [27], [28], [29], the NASEM report and the ensuing funding that it enabled has led to an explosion of interest and research on the cognitive aspects of diagnosis, and how it can be improved to avoid error. Clinicians, cognitive scientists, researchers, patients, and educators have all become interested and engaged in work to improve diagnostic reasoning, and their publications are appearing regularly in DIAGNOSIS and other leading patient safety journals.

Adjacent issues

Though not addressed in any detail in the report per se, the interest in improving diagnosis has stimulated work in several related areas:

The importance of knowledge: The best clinicians have both a comprehensive knowledge base and well-honed skills in clinical reasoning. Both are critical; in some respects it is certainly true that “knowledge is king” [30], but in practice most diagnostic errors reflect defects in clinical reasoning, not knowledge deficits [3], 31]. Interventions to improve knowledge [32], and interventions to improve clinical reasoning [33], 34] both have potential to improve diagnostic decision-making.

Affective bias: Clinical reasoning is inherently influenced by the heuristics and biases that are part of human cognition, and more recently emotional influences have been acknowledged as part of these subconscious tendencies, described as affective bias [35]. Examples of how affective bias can underlie diagnostic errors are appearing regularly in the literature.

Context: The context in which diagnosis takes place is another relevant factor to whether clinical reasoning will succeed or fail [36], 37]. Diagnosis is ‘situated’ in terms of where it takes place, who is involved, what resources are at hand, how much time is available, etc. Variables regarding the patient come into play as well, including how they describe their symptoms, how they emphasize particular elements (or not), how their disease manifests in them individually, and whether that disease is just becoming manifest or is well developed.

Contextual variables in the workplace also influence clinical reasoning. Some of these factors support optimal outcomes, including having enough time to think, having access to decision support tools, and to colleagues for second opinions. Other contextual elements are clearly detrimental, including time pressure and cognitive load, stress, fatigue, sleep deprivation, poor ergonomic conditions, and others.

The most conspicuous actor in the workplace in regard to its influence on diagnosis is the electronic health record (the EHR). Since their widespread adoption in the 2000s, EHR’s have transformed the diagnostic process. Many features have been beneficial: notes are legible; consults, imaging reports, and lab tests are easily found and are in chronological order; patient input is enhanced through their use of patient portals, etc.

However, multiple downsides have become evident in using EHR’s [38], [39], [40]. A study using cognitive task analysis to understand the impact of the EHR found that it did not support clinical reasoning, it interfered with the clinician’s ability to work collaboratively, and in impeded getting ‘the big picture’ of the patient and his\her problems [38]. The EHR acts to transform an individual skill set into a ‘joint cognitive system’, a mixture of human and machine. By suppressing intrinsic motivation, an inherent feature of System 1 which embraces the ‘’higher pleasures of work’ (meaning-making, exploration, inspiration and creativity), physician burn-out may increase, to the detriment of quality patient care [41].

Calibration refers to the extent to which an individual has a sense of whether their diagnoses are right or wrong, or how certain they are about this conclusion. Improving calibration, especially through the role of feedback [42] has emerged as a novel approach to improving diagnosis accuracy in practice [43].

Diagnostic excellence has been newly defined as the goal of work to improve diagnosis, and the elements of excellence have been defined [44]: Diagnosis should be accurate, timely, efficient, patient-centered, and SAFE.

Accident investigation: Root cause analysis has become a standard approach to investigating and learning from safety breakdowns throughout healthcare. The existing guidelines on conducting RCA have now been supplemented by clear guidelines on how to find and understand the cognitive contributions to diagnostic error [45]. This allows a more distal understanding of factors that may have initially triggered the error.

Next steps

Progress in appreciating the importance of cognition in diagnosis has been rapid and remarkable. There are two critical next steps in this journey:

(1) Improve education. Educational programs in the health professions need to ensure their graduates appreciate the role of cognition in diagnosis and in practice more generally [46], 47]. It requires the attainment of rationality so that the best possible decisions are made about patients given the available resources. This needs to include training on how to optimize clinical reasoning, and how to prevent, or if need be, recognize and catch diagnostic errors before they cause harm. This training can and should be done in an interprofessional manner, so that doctors, nurses [48], nurse practitioners, physician assistants, pharmacists, and everyone who is involved in diagnosis understands that diagnosis is best done as a team, and that everyone has a role to play [49].

Although pioneering programs have been developed that emphasize cognition over content [50], recent reports indicate that that only a small fraction of education programs have incorporated training on cognition in clinical reasoning [51]. This is something that every program should be doing. A key stumbling block, however, is the need for faculty development programs to equip educators with the skills and tools to provide this curriculum i.e. the curriculum needs to provide comprehensive training in cognition and how it relates to clinical practice. Given that cognitive and affective processes are considered by cognitive psychologists to be the major threat to rationality, appropriate training needs to address these issues in undergraduate medical education and continuing into postgraduate training. The requirements for attaining rationality and sound judgment, along with an awareness of how thinking can be confounded and misled, will need special attention. The need to improve education on cognition is an acute problem and remains largely unaddressed. If it is not fixed it will affect successive generations of trainees.

(2) Trial interventions. With the cognitive issues that can undermine diagnosis now well defined, it is time to begin testing interventions that might reduce the likelihood of harm. Many interventions have been suggested: Improving teamwork, engaging patients in the diagnostic process, using decision support tools, improving reflection [52], calibration, metacognition, slowing down, and so on. Each of these is reasonable and addresses one or more of the cognitive shortcomings known to be problematic. Early evidence has identified beneficial effects in several areas, including getting second opinions [53] and group-based input [54], 55], using decision support [56], and improving teamwork [57].

The explosion of interest in artificial intelligence includes many applications addressed at improving diagnosis, and refocuses attention on the central importance of knowledge as a pilar of diagnosis [58]. Systems to improve visual diagnosis in radiology and pathology are the most advanced and the most successful so far. The new large language models are being trialed for general diagnosis as well, with variable but generally impressive results. However, it is worth noting that advanced decision support for differential diagnosis has been available for several decades now, with disappointing uptake by clinicians in practice. As these new artificial intelligence modalities are refined and validated, exposure will need to begin during training [59], and they will need to be incorporated more widely in actual practice. Moreover, medical expertise will need to take on new and critical roles in the oversight of AI usage and the detection of AI bias and errors.

Conclusions

Cognitive processes have a wide application in clinical practice, and thanks to the 2015 report on Improving Diagnosis in Health Care, the imperative to improve the reliability of diagnosis is now widely recognized. Understanding the role of cognition in clinical reasoning, and how this can go astray, is central to efforts to reduce diagnostic errors. The Schola Medica Salernitana, the Salerno School of Medicine, likely got it right: Knowing HOW is just as critical as knowing THAT; both types of knowledge are essential to the diagnostic process.