Five decades of research and theorization on clinical reasoning: a critical review.

Clinical reasoning is a complex cognitive process that is essential to evaluate and manage a patient's medical problem. The aim of this paper was to provide a critical review of the research literature on clinical reasoning theories and models. To conduct our study, we applied the process of conducting a literature review in four stages in accordance with the approach of Carnwell and Daly. First, we defined the scope of the review as being limited to clinical reasoning theories and models in medical education. In the second stage, we conducted a search based on related words in PubMed, Google Scholar, PsycINFO, ERIC, ScienceDirect and Web of Science databases. In the third stage, we classified the results of the review into three categories, and in the fourth stage, we concluded and informed further studies. Based on the inclusion and exclusion criteria, 31 articles were eligible to be reviewed. Three theories and two models were recognized and classified into three categories. Several theories and models have been proposed in relation to clinical reasoning, but it seems that these theories and models could only explain part of this complex process and not the whole process. Therefore, to fulfill this gap, it may be helpful to build a Meta-model or Meta-theory, which unified all the models, and theories of clinical reasoning.

Introduction

Clinical reasoning is a complex cognitive process that is essential to evaluate and manage a patient’s medical problem.1 It includes the diagnosis of the patient problem, making a therapeutic decision and estimating the prognosis for the patient.2 In describing the importance of clinical reasoning, it has been acknowledged that clinical reasoning is the central part of physician competence,3 and stands at the heart of the clinical practice,4 it has an important role in physicians’ abilities to make diagnoses and decisions.1 Clinical reasoning has been the subject of academic and scientific research for decades;5 and its theoretical underpinning has been studied from different perspectives.6 Clinical reasoning is a challenging, promising, complex, multidimensional, mostly invisible,7 and poorly understood process.8 Researchers have explored its nature since 1980,9 but due to the lack of theoretical models, it remains vague. Most used theoretical models have limited explanatory power, and are based on certain assumptions about what constitutes clinical reasoning.10 In the literature of clinical reasoning, several competing theories and models have been raised.1,11–13 Although most of the theoretical contributions on clinical reasoning belong to the 20th century, proposing new models are well continued into the 21st century, for example, Haring and her colleagues proposed a conceptual model for expert judgment of clinical reasoning of medical students.14 However, there is no general agreement as to which of these is the best.15 The purpose of this paper is to provide a critical review of the research literature on clinical reasoning theories and models and present a comprehensive view of main models and theories of clinical reasoning in medical education. A clearer understanding of clinical reasoning models and theories help medical teachers for teaching, planning, and assessment of clinical reasoning. This paper tries to clarify the current knowledge regarding the clinical reasoning models and theories and present a classification of the main theories and models.

Materials and methods

Grant noted, “A critical review aims to demonstrate that the writer has extensively researched the literature and critically evaluated its quality.”16 It included a degree of analysis and conceptual innovation.16 In this study, we applied the process of conducting a literature review according to Carnwell and Daly.17 They proposed a “four-stage” method that included: 1) “Defining the scope of the review,” 2) “Identifying and selecting the sources of relevant information,” 3) “Organizing the results of the review into categories,” and 4) “Concluding and informing further studies.”17

Step 1: defining the scope of the review

The scope of this review was limited to the main clinical reasoning theories and models proposed in medical education literature. We concentrated our review on published works in scholarly journals between the years 1970 and 2018.

Step 2: identifying and selecting the sources of relevant information

At this stage, each of the two reviewers conducted a separate search based on keywords – “clinical reasoning,” “diagnostic reasoning,” “clinical reasoning theory,” “clinical reasoning model,” “script theory,” “hypothetico–deductive model,” “cognitive continuum theory,” and “dual processing theory” – in PubMed, Google Scholar, PsycINFO, ERIC, ScienceDirect and Web of Science databases. The results were pooled and extensive literature were found (n=305) which was from 1970 to 2018, but due to lack of access to the full text of some articles, and after removing duplicated studies, the title and abstract of articles from 1974 up to 2018 have been reviewed by researchers (n=170) (Table 1).

Table 1

Scientific studies about clinical reasoning that has been title and abstract reviewed

First decade	Second decade	Third decade	Fourth decade	Fifth decade
Feinstein (1974)41Rubin (1975)42Elstein et al (1978)43Elstein & Bordage (1979)44	Kraytman et al (1981)45Feltovich et.al (1984)46Kuipers & Kassirer (1984)47Schmidt & de Volder (1984)23McGuire (1985)48Patel et al (1986)19Barrows et al (1987)31Case et al (1988)32Hamm (1988)39Groen & Patel et al (1988)49Girotto & Legrenzi (1989)50	Elstein et al (1990)18Joseph & Patel (1990)51Patel et al (1990)52Norman et al (1990)53Schmidt et al (1990)25Ericsson (1991)54Kaufman (1991)55Mattingly (1991)56Custers et al (1992)57Evans & Patel (1992)58Florance (1992)59Henny Boshuizen et al (1992)60Higgs (1992)20Arocha et al (1993)61Hassebrock et al (1993)62Patel et al (1993)63Schmidt et al (1993)64Higgs (1993)65Joseph Arocha et al (1993)66Bordage (1994)67Elstein (1994)22Patel et al (1994)68Jones (1995)69Joseph Arocha & Patel (1995)70Custers (1995)71Custers et al (1996)26,28Hammond (1996)38Mandin et al (1997)30Van de Wiel (1997)72Allen et al (1998)73Chang & Bordage (1998)74Charlin et al (1998)75Custers et al (1998)24Cuthbert (1999)76Kaufman et al (1999)77Round (1999)78	Charlin et al (2000)12Carter & Robinson (2001)79Harries & Harries (2001)80Round (2001)6Elstein & Schwarz (2002)81Nendaz (2002)82Patel et al (2002)83Bleakley et al (2003)84Coderre et al (2003)85Norman & Eva (2003)86Eshach & Bitterman (2003)87Groves et al (2003)88Hardin (2003)89Charlin &Van der Vleuten (2004)90Eva (2004)91Rikers et al (2004)92Verkoeijen et.al (2004)93Holyoak & Morrison (2005)21Norman (2005)3Tamayo-Sarver (2005)94Anderson (2006)95Bowen (2006)96Eva & Cunnington (2006)97Loftus (2006)5Montgomery (2006)98Novak et al (2006)99Thornton (2006)100Auclair (2007)101Schmidt & Rikers (2007)102Norman et al (2007)11Banning(2008)103Evans (2008)13Harasym et al (2008)29Heiberg (2008)104Higgs (2008)7Humbert (2008)105	Aleluia et al (2010)113Corcoran (2010)114Mariasin (2010)115Lee et al (2010)116Omana et al (2010)117Wilhelmsson (2010)118Thomson et al(2010)8Amini et al (2011)119Durning (2011)120Franklin et.al.(2011)121Pelaccia et al (2011)1Thomson et al (2011)8Adams (2012)122Ashoorion et al (2012)123Van Bruggen (2012)124Charlin et al (2012)125Demirören & Palaoğlu (2012)126Braude (2012)127Braude (2012)128Khatami et al (2012)129Lucchiari & Pravettoni (2012)10Loftus (2012)4Marcum (2012)34Nouh et al (2012)130Shaban (2012)131Adams (2013)9Custers (2013)40Audétat et al (2013)132Da Silva (2013)133Evans & Stanovich (2013)36Gigante (2013)134Kriewaldt (2013)135Lubarsky et al (2013)136Munshi et al (2013)137Smith (2013)138Weiss et al (2013)139Ilgen et al (2013)140
			Noreen et al (2008)106Vertue & Haig (2008)107Braude(2009)108 Bissessur et al (2009)109Carrière et al (2009)110Croskerry (2009)35,37Elstein (2009)33Stempsey (2009)15Rehder & Woo Kim (2009)111Vosniadou (2009)112	Bowen & Ilgen (2014)141Delany & Golding (2014)142Freiwald et al (2014)143Geisler et al (2014)144Gordon (2014)145Holmboe & Durning (2014)146Hrynchak et al (2014)147Hochberg et al (2014)148,149Monajemi (2014)150Roots (2014)151Salkeld (2014)152Smith et al (2014)153Capaldi (2015)154Custers (2015)27Gaba (2015)155Islam et al (2015)156Lafleur & Leppink (2015)157Lubarsky et al (2015)158Park et al (2015)159Lisk (2016)160McBee et al (2016)161Gruppen (2017)162Haring et al (2017)14Jarodzka et al (2017)163Norman et.al (2017)164Ten Cate et.al (2017)165Zamani et al (2017)166Bowen & ten Cate (2018)167Custers (2018)168Daly (2018)2King et al (2018)169Keemink et al (2018)170Lopes et al (2018)171Yazdani & Hoseini (2017)172Yazdani et al (2018)173Higgs et al (2018)174

Then, the articles that presented theories or models of clinical reasoning in medicine or provided evidence in relation to them were selected to full-text study. Studies were eligible for this critical review if they presented a model or a theory of clinical reasoning, or related critiqued models and theories or the studies that add some features to the theories and models of clinical reasoning (n=47). The inclusion criteria of selecting studies were: 1) published articles in English and Persian and 2) published articles in the field of medicine. Studies were excluded if they provided clinical reasoning models or theories in other fields (like nursing and optometry), examined the clinical reasoning in the field of artificial intelligence (like clinical decision support systems), and/or examined brain biology and brain functions (like fMRI studies).

Step 3: organizing the results of the review into categories

After excluding irrelevant studies, a total of 31 documents were initially selected for review which is shown in PRISMA flowchart below (Figure 1).

Figure 1

PRISMA flowchart.

Models and theories which were extracted from studies, classified to three categories and each category, based on Carnwel and Daly approach, reviewed in three steps: first, we present a summary of the models and theories, and then reflect other author’s views and finally, we present our view (Table 2).17 If a model or theory explains about the process of clinical reasoning our first category owns it while models and theories which clarified the formation of knowledge structures and their application during the clinical reasoning process belongs to the second category, and our third category consisted models and theories which consider more than one processing modes of clinical reasoning.

Table 2

The list of 31 related literatures that included in the review

Category no.	Theory/model	Author	Year	Aspects related to supporting/criticizing the model
First category: theories and models based on the process of clinical reasoning	Hypothetico-deductive model	Elstein	199018	The description of the hypothetico-deductive modelThe advantages of hypothetico-deductive model
		Patel	198619	Lack of consistency of hypothetico-deductive model with other domains
		Higgs	199220	Clinical reasoning process is not sequential
		Charlin	200012	Unfamiliarity of psychological mechanisms involved in this model
		Holyoak	200521	No differentiation between novice and expert clinical reasoning
		Loftus	20065	Clarification of the role of hypothesis in clinical reasoning process
		Higgs	20087	Adequate description of the process of clinical reasoning
Elstein	199422	Application of hypotheses for framing of clinical problems
Second category: theories and models based on the knowledge structure	Illness script theory	Schmidt	198423	The description of the illness script theoryFormation and development of the illness script
		Custers	199824	The description of the illness script theory
		Schmidt	199025	The structure of illness scriptFormation and development of the illness script
		Custers	199626	The structure of illness script
		Custers	201527	Script concordance testFormation and development of the illness script
		Custers	199628	The structure of illness script
		Harasym	200829	Formation and development of the illness script
	Mandin	199730	Distinction between the concept of the script and the schema
	Pattern recognition model	Barrows	198731	The description of the model
		Case S	198732	It used by experienced practitioners
		Norman	200711	The most usual form of nonanalytic processes
		Elstein	200933	Unanswered questions about pattern recognition model
		Marcum	201234	The complexity of cognitive processes involved in clinical reasoning to be ignored.
Higgs	20087	Pattern recognition model examined in limited field of expertise.
Third category: compilation theories and mode	Dual processing theory	Evans	200836	The description of the theory
		Croskerry	200935	Advantages of this theory
		Croskerry	200937	Proposing a model based on dual processing theory
		Pelaccia	20111	Clarification of the place of pattern recognition and hypothetico-deductive models in dual processing theory.
		Evans	201336	Criticized this theory in five major themes
	Lucchiari	201210	Models based on dual processing theory
	Cognitive continuum	Hammond	199638	The description of the theory
		Hamm	198839	The description of the theory
Custers	201340	Advantages of this theory

Results

First category: theories and models based on the process of clinical reasoning

This category includes the models and theories that explain the clinical reasoning process, between models and theories that we reviewed, only hypothetico-deductive model was eligible to get placed in the first category as the most reputed model that explains the clinical reasoning process. This model was proposed by Elstein (1978), and, according to this model, the physicians primarily generate a limited number of diagnostic hypotheses or problem formulations in the process of solving a diagnostic problem and then testing them. These hypotheses guide further patient information.18,19 Unlike the findings of hypothetico-deductive model that claim: “primarily generated and tested hypotheses by expert and novice are the same,” Patel believed that it is not consistent in other domains, like physics.19

Higgs argued that this model posits the idea that the process of clinical reasoning is largely a sequential process.20 Charlin pointed out that the psychological mechanisms involved in the generation and testing of relevant hypotheses are unfamiliar,12 and Holyoak argued that this model does not distinguish between novice and expert clinical reasoning strategies.21 Loftus believed that the collected information and the way they interpreted, distorted by the used hypothesis.5 This model as an adequate description of the process of clinical reasoning has challenged by the case specificity findings.7

Nevertheless, some researchers defend hypothetico-deductive model, Elstein argued that the small set of solutions that generated in this model transformed an unstructured problem to structured one and it is an effective way to solve diagnostic problems.18 This model is recommended by medical experts as a useful reasoning strategy for medical students.22 Hypothetico-deductive model is applicable when data are vague or reveal over time,22 and is a representation of clinical reasoning.20 This model represents a description of the mental processes used by physicians and has repeatedly been validated by empirical studies and is the basis for modern clinical education.12

Hypothetico-deductive model assumes the physician starts hypothesizing after collecting patient information and then tests hypotheses, while we believe the physician starts hypothesizing initially from his/her clinical encounter. The initial hypotheses can be strong or weak, depending on whether the physician is an expert or novice, the difference between the novice and the expert lies in the quality of the hypotheses they made. Therefore, since the initial hypothesis of an expert has good quality, hypothesis testing will be fast and efficient. The simplicity of this model in describing the process of clinical reasoning is both strength and the weakness of it, as a strength, because it simply portrays the start point of the process of clinical reasoning so it can be used to design the teaching plan and evaluate clinical reasoning. As a weakness, because it considers the process of diagnostic reasoning very simple, while even for a novice, this process does not occur so easily, and other factors (such as the individual’s knowledge structure, the context, the health system, etc.) affect the process of clinical reasoning, but this model does not consider these factors.

Second category: theories and models based on the knowledge structure

For this category, we considered theories and models that explain the formation of knowledge structures in the clinical reasoning process, by this description and the inclusion criteria just one theory and one model of clinical reasoning gain eligibility to include, the “illness script theory” and the “pattern recognition model.”

The illness script theory proposed by Barrows and Feltovich consists of three components: 1) enabling conditions, 2) fault, and 3) consequences.23 The first component is the factors such as age, sex, current medication, previous medical history, occupation, risk behavior, hereditary, and environment affect the probability of someone gets a disease, are the patient’s contextual and background factors that refer to “Enabling conditions.” These “Enabling conditions” can cause the latter pathophysiological malfunctioning that called “fault” which is the second component of illness script. Consequences of this fault are complaints, signs, and symptoms that consist of the third component.24 Illness scripts are the list-like structures,25 which conceptualized as a specific representation of clinical knowledge.26 Script concordance test designed according to this theory.27

While we were studying about Illness script theory, we realized that we could categorize studies into two broad groups. The first group is the studies that deal with the concept of the script, the schema, and the illness script, and their features, distinctions, and components (Table 3).24,26–28

Table 3

The studies that deal with the concept of “script,” “schema,” and “illness script,” and their features, distinctions, and components

Author	year
Custers et al.	(1996)26
Custers et al.	(1996)28
Custers et al.	(1998)24
Custers et al.	(2015)27
Loftus	(2006)5
Charlin	(2000)12

The second group deals with the formation and development of the illness script during the acquisition of expertise and changes in the physician’s knowledge structure (Table 4).23,25,26,29

Table 4

The studies that deal with the formation and development of “illness script” during the acquisition of expertise and changes in the physician’s knowledge structure

Author	year
Schmidt et al.	(1990)25
Schmidt et al.	(1984)23
Harasym et al.	(2008)29

The first group of studies also looked at the distinction between the concept of the script and the schema, but this distinction was not clear in the literature. The schemas and scripts are stored in long-term memory.30

Schema as a knowledge structure has an “if/then” formatting and occurs sequentially, in the sense that this sequence divided into two branches: “if” and “then,” so we can claim out that its format is algorithmic. This algorithm starts with a hypothesis in a person’s mind or something that a person thinks about and then continues with inquiries and searches that a physician has performed and then with the findings that a physician has reached, and finish with the decisions that he/she has finally taken (Figure 2).

Figure 2

Generic flow of events in a typical schema. D1: Decision No 1; D5: Decision No 6; F1: Finding No 1; F5: Finding No 5; H1: Hypothesis No 1; H2: Hypothesis No 2; I1: Inquiry No 1; I3: Inquiry No 3.

In terms of the structure of the script, we also agree with Schmidt’s view that the scripts are list-like structures, but unlike Charlin, who believed that “the script describes the structure of clinical knowledge,” we believe that the script is not necessarily the structure of clinical knowledge, but a knowledge structure that has clinical applications. The script is schemas for common situations, which include a packet or a list of expectations of what people see or do at a given location. The schemas and scripts are stored in long-term memory, and if physicians encounter a clinical case that matches with them, they will retrieve it from long-term memory and move it to short-term memory (Figure 3).

Figure 3

Script as a routinized pathway of previously used schema. D1: Decision No 1; D5: Decision No 6; F1: Finding No 1; F5: Finding No 5; H1: Hypothesis No 1; H2: Hypothesis No 2; I1: Inquiry No 1; I3: Inquiry No 3.

Pattern recognition model

In the pattern recognition model, a physician directly compares the pattern of the patient’s problem with disease patterns and if found them similar to each other, then select the pattern that matches it.31 Experienced practitioners often use pattern recognition to achieve a medical diagnosis.32 Norman and his colleagues argue that pattern recognition is the most usual form of nonanalytic processes.11 However, Elstein proposed some questions about this model, as followed:

When dose a person use a pattern recognition model?

When is this method preferable to the hypothetico-deductive method?

What guarantees that the choice of a pattern or an illness script is correct?

What happens if the pattern or script stored in the knowledge base is wrong?33

This model considers the complexity of cognitive processes involved in clinical reasoning to be insignificant.34

Based on the definition of the pattern recognition model, it only mentions the existence of patterns in mind, but does not explain how the construction of these patterns occurred. The studies which designed to prove that the pattern recognition model happens in reality are in a limited field of expertise, like radiology, dermatology, and pathology. 7 So the pattern recognition model is not extendable to all medical specialties.

Third category: compilation theories and model

Some of the included documents were about “dual processing” and “cognitive continuum” theories that explain two modes of reasoning – “analytical” and “non-analytical,” these modes are the characteristics of both first and second category, so we cannot involve them in one of them, therefore they form our third category.

The dual-processing theories commonly have two different processing modes in which they refer to: System 1 and System 2.13 System 1 described as a fast, automatic and intuitive mode, which shares similarities through perception, while System 2 is slow and analytic mode that applies rules without inferring emotions.7

Croskerry believed that dual-processing theory is an applicable model in multiple domains of health care like decision-making and it can be useful in teaching decision theory or in making a platform to future research.35 Pelaccia et al noted that in the framework of this theory, the pattern recognition and hypothetico-deductive models are the basis of the intuitive system and the analytic system, respectively.1

Evans and Stanovich criticized this theory in five major themes: 1) various theorists have proposed multiple and vague definitions for this theory, 2) there is no consistency in associated attribute clusters with dual systems, 3) distinctions are referred to the continuum of processing and not to discrete processing; 4) the apparent dual-process phenomenon can present by single-process accounts; and 5) the evidence base for the dual-processing theory is ambiguous or unconvincing.36

In the reviewing of the literature, we found out that some of the researchers established their models based on dual-processing theory like Marcum,34 Croskerry,35,37 and Lucchiari and Pravettoni.10

Dual-processing theory employs many of the seemingly contradictory features that have been proposed for clinical reasoning in the literature (such as fast, slow, reflective, etc.). It seems that, in reality, a physician does not use just intuitive or analytic systems and the mind of physician operates in the space between them, while the dual-processing theory ignores this.

The theories and models that have been proposed following this theory have led to the introduction of cognitive concepts such as metacognition and perception and their role in the process of clinical reasoning. This theory has relatively clarified the role of emotions and their place of influence in the process of clinical reasoning, and has also contributed to clarifying the concept of intuition in clinical reasoning.

The second theory that has placed in third categories is cognitive continuum, as Hammond claimed, this theory considered two poles, an intuitive cognition and an analytical cognition, in which various modes or forms of cognition have relational order on a continuum, and this assumption rejected the dual-processing approach.38 Hamm believed, this theory does not explain the information processing that is the basis of analysis and intuition, but based on analytical and intuitive cognitive attributes it gives us various techniques in describing cognitive modes. Also, he believed that this theory did not offer an instruction about thinking analytically or intuitively, and it just presented a general framework. Cognitive continuum theory described the features of cognition and their correlation with the characteristics of the task.39 Custers noted that this theory illustrates the cognitive processes and the cognitive tasks on a continuum, and this theory can be used to provide advice on how to structure clinical tasks in an educational setting.40 In criticizing cognitive continuum theory, we did totally agree with Hamm and Custers.

Conclusion

The present study was conducted to critically review theories and models of clinical reasoning that have often been raised in the medical education literature within five decades (1970–2018). Several theories and models presented in relation to clinical reasoning and it seems that they can explain only part of the complex process, but not the whole process. For example, the models and theories of the first category in our study just address the process of clinical reasoning and do not pay attention to knowledge structures and cognition; in the second category, they just focused on knowledge structures and their formations during clinical reasoning process and do not clarify the process of clinical reasoning. In addition, the dual processing and cognitive continuum theories that form the third category just covered the cognition part of the clinical reasoning. Therefore, to fulfill this gap, it may be helpful to build a Meta-model or Meta-theory, which unified all the models, and theories of clinical reasoning. Although our focus was on the main models and theories of clinical reasoning in the field of medical education, but we acknowledge that there are other models and theories of clinical reasoning in the literature and their absence can be the bias of this study.