Clinical reasoning (CR) has been defined as a way of thinking and decision making in professional practice (Higgs and Jones, 2000), requiring physicians to sort through a cluster of features and accurately assign a diagnosis and develop a treatment strategy (Eva K, 2005). The context in medical education has previously been defined as a complex system evolving over time, with the outcome being driven by interactions and feedback among patient, physician, setting, and their interactions (Durning et al., 2010). ‘Context specificity’, the phenomenon of a physician’s performance varying on a case by case or situation to situation basis, has been well described (Eva, 2003). 

CR is important: diagnostic error contributes significantly to medical errors (Kuhn, 2002), and complaints to governing bodies often concern clinical competence (Gmcuk.org, 2019). A medical remediation programme showed that one third of referrals were due to CR deficits (Guerrasio and Aagaard, 2014). Many errors are not solely due to lack of knowledge but also due to difficulties in physicians’ thinking in the real-world setting (Kohn, 2009).

Research supports the view that CR skills can be developed and improved (Scott, 2009; Thompson, 2004). Enhancing the way CR is performed by physicians in training has the potential to both reduce clinical errors and aid the transition from novice to expert.

This review aims to aid physicians and educators in recognising the effect of context on CR. It also suggests that this is an exciting area for research to build upon the existing literature and further examine how we might improve CR, thereby minimising clinical error, and how educators can help physicians in training enhance their CR.

We conducted a narrative literature review searching key databases (PubMed, Psycinfo, EMBASE via Ovid) for peer reviewed empirical research articles in English published from January 2010 - September 2017. The search terms ‘clinical OR critical AND reasoning OR judgement OR thinking’ were used. Titles and abstracts were reviewed by the first author and a practising psychiatrist. Reference and citation searches yielded additional results. Any differences in opinion were discussed and resolved.

A total of 22,296 articles were retrieved, reduced to once duplicates and articles older than 2010 were removed. A further 17,385 were removed as they were not relevant, 625 studies were removed as they were not empirical research, and 385 papers were removed as the studies were with allied health professions. Twenty-four studies were removed as they focused on the assessment of CR. An additional seven studies were added from reviewing references and citations. Twenty-five studies remained. Our search strategy is summarised in Figure 1.

Figure 1. Search strategy for this review

Of the 25 studies analysed, the contexts varied fell into three broad categories: patient factors, physician factors and environmental factors. Contexts varied in the studies which were shown to affect CR are outlined in Figure 2.

Figure 2. Contexts which affect CR

Patient Factors

Eight studies varied patient factors. One study showed 25 physicians videotaped encounters. They altered one or more contextual factor: low English proficiency, emotional volatility, incorrect diagnosis suggestion or atypical presentation. Physicians thought aloud and wrote their diagnosis, management plan and the effect contextual factors had on their CR processes. Participants discussed the need for additional resources due to the contextual factor. Some participants dismissed the contextual factor whilst other viewed it as potentially beneficial – incorrect suggestion and emotional volatility were most likely to be dismissed, the presence of more than one factor was always viewed negatively. Participants were more likely to miss key data if two contextual factors were present. The research team postulated that multiple contextual factors led to increased cognitive load, leading to a negative perception of the situation and missed key data (Durning et al., 2011).

McBee et al. showed ten physicians videotaped encounters and asked them to think aloud about their CR. Videos had a contextual factor altered – low English proficiency, emotional volatility or both. In the presence of both contextual factors, participants experienced higher levels of diagnostic uncertainty and the desire for further information; including additional tests (McBee et al., 2015). A follow on study asked 14 physicians to watch videotaped encounters. Each had one or more contextual factor altered (low English proficiency, emotional volatility or both). A higher cognitive load was not consistently associated with a poorer performance (Ratcliffe et al., 2017).        

Another study looked at patient’s disruptive behaviour and diagnostic difficulty of presentation. In case vignettes, patients were depicted as difficult (displaying distressing behaviours) or neutral. Case vignettes were deemed diagnostically difficult or simple. Sixty-three physicians evaluated the cases. Diagnostic accuracy was significantly lower for both the difficult patients (p=0.017) and the diagnostically difficult cases (p<0.001). Participants then reflected on the cases. Reflection improved diagnostic rates (p=0.002) (Schmidt et al., 2016). The researchers repeated the study, this time investigating why difficult patients reduced diagnostic accuracy. They gave 74 physicians case vignettes, half with ‘difficult patients’ and half with ‘neutral patients’. Diagnostic scores were significantly lower for difficult patients (p < 0.01) and participants recalled fewer clinical findings and more behaviour observations from the difficult patients (p < 0.001) (Mamede et al., 2016).

One study looked at patient appearance and the effect on CR. Forty-six physicians where given case-based scenarios with a patient picture. Investigators classed patient pictures as ‘poor and dirty’ in appearance or as ‘rich and clean’. There was no significant difference in diagnostic accuracy, however participants reported processing the case more extensively if the patient appeared ‘rich and clean’ (p = 0.04) (Mohamed et al., 2016).

A study looked at the influence of salient distracting features, a feature strongly associated with a certain disease which may catch a physician’s attention triggering pattern recognition. They gave 72 physicians case vignettes; some had a salient distracting feature at the beginning, some at the end and some had none. Half the cases were deemed simple and half complex. For the complex cases a distracting salient feature at the beginning of the case reduced diagnostic accuracy (p < 0.001). There was no significant difference when the distracting salient feature was at the end (Mamede et al., 2014).

A study looked at the bias effect of clinical history on physical examination. They gave 180 physicians a clinical history and asked for a diagnosis. Participants were then randomised into three groups. The first group were given physical findings which fit with the likely diagnosis from the history, the second group were given the physical findings of an alternative diagnosis with indistinct physical findings, the third were given physical findings of an alternative diagnosis with distinct physical findings. Physicians showed a significant decrease in diagnostic accuracy when faced with discordant physical findings (p < 0.0001) (Sibbald and Cavalcanti, 2011).

Physician

Ten studies varied physician factors. One study assessed availability bias. Researchers asked 18 first-year and 18 second-year physicians to evaluate six written clinical cases. Then participants diagnosed eight further clinical cases as quickly as possible aiming to induce non-analytical reasoning. Four of these clinical cases had similar clinical manifestations to the initial cases, but different diagnosis. They found that second-year physicians, who had been exposed to cases with similar clinical manifestations but different diagnosis in round one, were more likely to give the incorrect diagnosis (p=0.03). There was no significant difference for the first-year physicians. Finally, participants evaluated a further four cases, with two of the cases having similar clinical manifestations to previous cases, but different diagnosis. The participants were given a structure to follow, aiming to induce reflection. This improved correct diagnosis rates compared with nonanalytic reasoning for all participants (p=0.006) (Mamede et al., 2010).

Recent media exposure to a disease and the effect on a physician’s diagnostic skills was analysed in one study. Investigators asked 38 physicians to read a Wikipedia entry about a rare condition. Six hours later participants evaluated eight written clinical cases. Two cases had similar clinical manifestations to the Wikipedia entry they had read. Participants were more likely to give the incorrect diagnosis if the case described was like the Wikipedia diagnosis they had read earlier (p=0.16). The participants then evaluated four of the cases again in a structured format. This subsequent reflection increased diagnostic accuracy (p=0.27) (Schmidt et al., 2014).

A study looking at the influence of diagnostic suggestion asked 24 physicians to evaluate case vignettes, half with a correct diagnostic suggestion and half with an incorrect diagnostic suggestion. Participants were able to accept the correct diagnosis more easily than rejecting the incorrect diagnosis (p < 0.05) (van den Berge et al., 2012a). In a similar study 38 physicians evaluated case vignettes, half had a correct diagnosis suggestion and half had an incorrect diagnosis suggestion. In this study, diagnostic performance did not differ between cases(van den Berge et al., 2012b).

A study investigated the effect of physician age on diagnostic accuracy. Eighty-nine physicians were split into age bands: under 30, 30-39, 40-49 and over 50. They worked through case vignettes and gave a diagnosis. Participants younger than 30 performed better than participants aged over 40 (p<0.01) St-Onge et al., 2015). One study looked at management plans of physicians and medical students. The study asked 20 fourth-year and sixth-year medical students and ten physicians to review written clinical cases and provide management plans. The physicians provided more accurate management plans as compared to the medical students, the format of the management plans of the sixth-year medical students were more like the physicians as compared to the fourth-year medical students (Monajemi, Schmidt and Rikers, 2012).

Another study asked 191 physicians to use a checklist and assessed the effect on diagnostic accuracy. The participants performed a cardiology examination on a simulator patient. Half of participants then completed a checklist of the different components of the cardiology exam and repeated the examination. Completion of the checklist improved diagnostic accuracy (p=0.04) (Sibbald et al., 2013). Another study looked at the use of a mnemonic TWED (T = threat, W = what else, E = evidence and D = dispositional factors) created to facilitate metacognition. Forty medical students were split into two groups, with the intervention group using the mnemonic. Both groups’ clinical decision making was assessed based on their performance in case-based scenarios. The intervention groups’ mean score was significantly higher as compared to the control group (p < 0.001) (Chew, Durning and van Merriënboer, 2016).

Environment

Eight studies looked at the environment and CR. A study retrospectively interviewed 21 general practitioners working out of hours. They described two recent cases and the influences on their CR were explored. Participants perceived out of hours practice differently, they referred to ‘firefighting’: working in unfamiliar surroundings under time pressure. They dealt with the immediate situation, then stopped collecting information and ensured there was reliable safety netting. They also reported a lack of feedback on which to reflect and learn (Balla et al., 2012).

Another study sought to assess the effect of time pressure on diagnostic accuracy. They gave 42 physicians case vignettes. The intervention group were manipulated to make them feel that they were under time pressure and falling behind. The control group received no such manipulation. The mean response time for the intervention group was significantly quicker, but with significantly lower diagnostic accuracy (ALQahtani et al., 2016).  However, two other studies did not mirror this result. One study recruited 46 emergency medicine physicians and 152 less experienced physicians. They were given case vignettes; the first ‘slow conditions’ group were encouraged to give due consideration to each case, the second ‘fast conditions’ group were asked to imagine they were in an emergency department working quickly and were given less time. Diagnostic accuracy was significantly higher for the emergency physicians compared to the less experienced physicians (p<0.0001). However, although the ‘faster conditions’ group did respond quicker than the ‘slow conditions’ group there was no significant difference in diagnostic accuracy (Monteiro, 2015). In a similar study 204 physicians were split into the ‘speed cohort’ and the ‘reflect cohort’. They were given case vignettes, the ‘speed cohort’ were encouraged to work quickly and accurately and had a visible count down timer. The ‘reflect cohort’ were encouraged to be thorough, consider all data and had no count down timer. There was no significant difference in the diagnostic accuracy, although the reflect group were significantly slower (p<0.0001) (Norman et al., 2014). A separate study also limited participants’ time. Twenty-five physicians were shown videotaped encounters. Participants thought aloud and completed a post encounter form. The time given to complete the post encounter form was randomly varied. During the ‘think aloud’ process researchers measured the cognitive load on the participant. Increased cognitive load and reduced time to complete the post encounter form had a negative impact on performance (Durning et al., 2011).

Three studies looked across the board at patient, physician and environmental factors. Forty consultant dermatologists were interviewed regarding what influenced their decision to discharge patients from clinic. The list of influences was extensive; of the patient-based influences most participants highlighted the disease type and the patient’s ability to cope with their disease. Of the physician-based influences, most physicians highlighted their own amount of experience and their confidence in the decision. Of environment-based influences, physicians felt a supportive, well run department and confidence in local primary care services facilitated earlier discharge. Pressure from hospital managers also influenced their decision (Harun et al., 2015). Another study interviewed ten infectious disease experts asking them to recall a complex antibiotic-prescribing incident. They focused on what made the incident complex.  Several themes emerged; the overall clinical picture not matching the clinical pattern – an unexpected outcome, risky patient characteristics and unusual cases. A lack of comprehension of the situation – lack of or conflicting data, lack of evidence for treatment effectiveness, no diagnosis and gaps in the physician’s knowledge. Participants also highlighted the effect of social and emotional pressures – frustration and regret, liability or fear and the pressure and potential conflict of multiple care providers (Islam et al., 2015). In another study investigators interviewed 36 newly qualified physicians, asking what influenced their behaviour when identifying and managing acutely deteriorating patients. They described difficulties translating theoretical knowledge into practice and making decisions in uncertain circumstances. They referred to the fear of being wrong or causing harm. They described the strong presence of hierarchy deterring new physicians from asking for help, fearing they would fall short of their senior’s expectations. They also described the pressure of the acutely unwell patient; time pressure, high-stakes outcome, heavy information load and a changing situation; causing them to freeze and having an emotionally negative impact (Tallentire et al., 2011).

Limitations of the study include difficulty in measuring CR, with most studies using diagnosis as a surrogate marker. Most studies were done with a small number of participants, often with inexperienced clinicians and, so it may not be possible to draw broad conclusions based on this group. Few studies were done in the real-world clinical setting, but rather with paper, online or simulated cases, this limits the validity of conclusions regarding the effect of context. There are a very small number of investigators doing this work which could be a potential source of bias. Many of the studies depended on participant self-reporting of the effects of context on CR, physicians’ own understanding of context on their CR may be limited. The search was restricted to medical physicians only, much work has been done within other fields, particularly nursing. Work other fields could provide useful insights which could help inform further research into physicians’ CR.  None of the trials looked at long-term follow up data, and most of the studies were performed by a small number of research groups. This paper is a narrative review and therefore open to selection and interpretation bias.

Contextual factors in CR which have been researched in the literature fall into the categories of patient, physician and environment. Studies have used various methodologies including case vignettes, videotaped encounters, simulation and interviews. Interventional studies mainly used correct diagnosis as an outcome measure, some studies looked at management and physician reflections.

Several patient factors, both personal to the patient and physical disease manifestations, appear to have an impact on a physician’s CR. Some studies demonstrated that altering patient factors, including low English proficiency, an incorrect diagnostic suggestion, emotional volatility and disruptive behaviour can lead to poorer physician CR performance (Durning, 2011; Schmidt, 2017; Mamede 2016). One study showed that patient appearance did not affect diagnostic accuracy, although physicians processed the case more extensively if the patient appeared ‘rich and clean’ (Mohamed, 2016). The condition with which the patient presents can have an impact, with complex or atypical presentation of disease leading to a reduction in CR skills (Durning, 2011; Schmidt, 2017; Harun, 2015; Islam, 2015). One study suggested that the addition of patient factors led to physicians desiring more information, including investigations (Durning, 2011; McBee, 2015). If patient factors, such as distressing behaviours and poor English proficiency, have a negative impact on CR then it suggests that these patients could be at higher risk of misdiagnosis, incorrect treatment and additional investigation.

There are physician factors which influence CR beyond theoretical knowledge. Studies have suggested that many factors can bias physicians and reduce their CR performance; including recent exposure to a disease presenting in a similar way (Mamede, 2010; Schmidt, 2014). Salient distracting features at the beginning of a complex case reduced diagnostic accuracy (Mamede, 2014). One study suggested that physicians’ diagnostic accuracy may decrease with age (St-Onge, 2016). Physicians’ emotions were reported to negatively influence CR, with fear of causing harm to the patient and fear of disappointing seniors being identified (Harun, 2015; Islam, 2015, Tallentire 2011). One study suggested that physicians found it difficult to reject an incorrect diagnostic suggestion (Van den Berge, 2012a); however another study did not support this finding (Van den Berge, 2012b). Some factors can also positively influence a physician’s CR, including higher experience level (Monajemi, 2012; Monteiro, 2015), the use of checklists (Sibbald, 2013) and mnemonics (Chew, 2016), and reflection (Schmidt, 2017; Sibbald, 2011; Schmidt 2014). If physicians are influenced by recent exposure to similar cases, salient distracting features and incorrect diagnostic suggestion, then work to aid physicians’ recognition of these biasing factors could help mitigate against their effect. Experience seems to have a positive influence on CR, however, one study suggested that older physicians have reduced diagnostic accuracy. If this is the case older physicians may have to implement additional methods to maintain their CR abilities. The use of reflection, checklist and mnemonics seem to have a positive influence on CR and implementing these strategies in clinical practice could improve CR.

Some studies suggested that time pressure has a negative effect on CR (Alqahtani, 2014; Durning, 2012), while other studies showed a neutral impact (Alqahtani, 2014; Durning, 2012), this area would benefit from further research. Working in unfamiliar surroundings and a rapidly changing situation were identified as having a negative effect on CR (Balla, 2012; Tallentire, 2011). This suggests that physicians in new departments or in settings where situations change rapidly may require additional support.

Lastly some studies have tried to assess the impact of more than one contextual factor upon CR and have suggested that this leads to a reduction in CR performance (Durning, 2011; Durning, 2012), although one study did not show this consistently (Ratcliffe, 2017), and that this may also lead to an increased desire for information (McBee, 2015; Durning, 2012). 

In this original review article there is a strong theme emerging from the literature that contexts – patient, physician and environment - affect physician’s CR. In certain patient groups CR may be more challenging and so those groups are potentially vulnerable to poorer care. Physicians may be vulnerable to bias from recent experiences and exposures and their own emotional state. Reflection techniques may help to mitigate against this. The effect of time on CR has shown conflicting results and would benefit from further research. Rapidly changing and unfamiliar surroundings have been identified by physicians as having a negative effect on CR suggesting that CR may be more challenging in certain placements. Finally, combinations of the factors may interact, adding cognitive load on physicians, reducing CR performance and leading to the use of additional health resources. 

This area needs further research to assess the impact of these contexts in a real-world setting. There was a lack of research in the literature looking at whether there is variability amongst specialities or locations; it may be that certain specialities or working environments are more supportive for CR.

Medical institutions wish to produce physicians with good CR, improving care for patients and minimising medical error. Our review shows that contexts can influence physicians’ CR, suggesting that altering the context can enhance physicians’ CR. This must be considered in teaching and assessing CR, rather than simply focussing on the individual physician.

It is clear that many factors influence physicians’ CR, and that it is likely that these factors both interact and change from moment to moment. What the literature does suggest is that focusing solely the individual physician is too simplistic and that we need to recognise the interplay between patient, physician and environment.