Skip to content
Publicly Available Published online by De Gruyter August 22, 2022

Improving diagnosis: adding context to cognition

Mark Linzer EMAIL logo , Erin E. Sullivan , Andrew P. J. Olson ORCID logo , Maram Khazen , Maria Mirica and Gordon D. Schiff
From the journal Diagnosis



The environment in which clinicians provide care and think about their patients is a crucial and undervalued component of the diagnostic process.


In this paper, we propose a new conceptual model that links work conditions to clinician responses such as stress and burnout, which in turn impacts the quality of the diagnostic process and finally patient diagnostic outcomes. The mechanism for these interactions critically depends on the relationship between working memory (WM) and long-term memory (LTM), and ways WM and LTM interactions are affected by working conditions.


We propose a conceptual model to guide interventions to improve work conditions, clinician reactions and ultimately diagnostic process, accuracy and outcomes.


Improving diagnosis can be accomplished if we are able to understand, measure and increase our knowledge of the context of care.

Where we work is how we think

While who we are and what we know are crucial for being good diagnosticians, the environment in which we provide care and think about our patients is a crucial and undervalued component of the diagnostic process. Our current work examining the quality of diagnostic assessments in primary care, in which we defined key domains of diagnostic construction that ultimately can lead to a high-quality diagnosis, suggests important shortcomings and opportunities-both conceptual and practical. Early findings suggest this process can be impacted by matters such as time allotted for visits, distractions, as well as clinician stress and burnout. While this may seem obvious, there is a paucity of evidence supporting such associations in the literature and it has not been easy to design real-world studies to confirm these associations. Based upon our view that the context in which diagnosis occurs is one of the most important factors that can influence the diagnostic process and its outcomes, we propose a structure for understanding diagnostic process quality within the context of the environment in which that process occurs, setting the stage for future studies and refinement of a conceptual model to guide us.

Decades of work in educational theory inform this approach. Sweller, Paas and van Merrienboer described components of cognitive load [1]. In brief, two kinds of memory, Working Memory (WM) and Long Term Memory (LTM), do the work of integrating new knowledge (in WM) with previously collected and organized knowledge stored in LTM. As Schiano clearly points out [2], workplace distractions (interruptions, chaos, or poorly designed computer screens/navigation) can interfere with WM ability to process new information, then retrieve what it needs from LTM to synthesize new information with prior knowledge. This dovetails well with what we have shown in the demand-control model of job stress in clinicians, where interruptions, chaos, lack of control, and time pressure all lead to stress and burnout and are associated with poor performance in terms of quality and safety [3, 4]. Schiano also emphasizes the importance of taking time needed for thought, “I just need a minute to process what you have told me … ” Time must be available in clinical settings for such integration of WM and LTM. Paas goes on to say that WM can only store 5–9 items at a time, and often for no longer than 20 s. This has profound implications for how the diagnostic evaluation of patients is structured (Figure 1), and highlights the importance of allowing sufficient time for accessing LTM. Avoiding the crowding of cognitive load into WM via interruptions, noise or other distractions (e.g. computer over-alerting) has rarely been formally conceptualized in efforts to improve diagnosis. Because stress has been postulated to diminish resources available for working memory, efforts to minimize stress will likely pay significant dividends in diagnostic accuracy (a hypothesis to be tested). Croskerry recently framed the importance of the work environment within other aspects of diagnostic cognitive excellence [5]. Building on the multiple domains he proposes, we further incorporate environmental workload factors and stresses along with new measures of the process of diagnosis and patient outcomes.

Figure 1: 
Proposed impact of work environment on interaction between working memory (WM) and long term memory (LTM).
Figure 1:

Proposed impact of work environment on interaction between working memory (WM) and long term memory (LTM).

The conceptual model for this process (Figure 2) is based upon two decades of studying the role of work conditions as they relate to clinician reactions and patient outcomes [6], as well as our more recent work linking work conditions to diagnostic process [7]. Most categories mentioned in the model have validated, parsimonious measures. “Work conditions” represent the environment in which clinical reasoning occurs and as such require deep exploration as covariates which can upend a typical non-linear, iterative diagnostic process. Furthermore, there is emerging evidence that inherent biases may be more likely to surface if reasoning is rushed [8, 9], and as such, equity may suffer in time pressured, chaotic environments [10]. As we move from left to right in the model, with the work environment leading to clinician outcomes of stress and burnout, then to the diagnostic process and outcomes (diagnostic assessment domains), the impacts are not likely to be linear. That is, the relationships between work conditions, provider factors, and diagnostic process and diagnostic outcomes are likely complex, and seemingly small factors may have large effects while others may have smaller effects, and multiple factors are likely advantageously or disadvantageously synergistic in their effect on outcomes. Even if clinicians do not suffer visibly, their diagnostic processes may still be impaired. An adverse work environment will almost certainly contribute directly to impaired diagnostic processes, e.g. chaos may not allow for testing of possible diagnoses (via accessing LTM) to proceed smoothly, or an emphasis on productivity instead of on the patient may push people to move more quickly through diagnostic reasoning in rote or predetermined and potentially biased manners. The increasing availability of time stamped data and keystroke recordings from electronic medical records (EMRs) may help untangle and allow deeper understanding of these multiple environmental contributors to diagnostic inaccuracy.

Figure 2: 
Conceptual model: Mediators of diagnostic quality. Relationships between work condition/infrastructure, diagnostic assessment, clinician reactions, patient factors, and final diagnosis. Most variables in the model can be readily measured for improvement. Dx = diagnosis; EMR = electronic medical record; IT, information technology.
Figure 2:

Conceptual model: Mediators of diagnostic quality. Relationships between work condition/infrastructure, diagnostic assessment, clinician reactions, patient factors, and final diagnosis. Most variables in the model can be readily measured for improvement. Dx = diagnosis; EMR = electronic medical record; IT, information technology.

New research provides support for this model [11] and suggests that making a high vs low quality diagnosis may have consequences for clinicians, patients who receive the diagnosis, as well as for other patients seen the same day when the clinician has fallen behind in their schedule. There is likely greater pressure and constraints in making diagnoses in shorter visits later in the day by potentially exhausted clinicians. This resonates with recent work by Neprash and others about suboptimal quality care occurring later in the day/session, with more potentially inappropriate antibiotic and opiate prescribing [12]. To avoid a clinic workday where better diagnoses are made earlier in the day and worse ones later, we will need to be more intentional about how the day is structured and how time is allotted, perhaps in more flexible manners, as seen in the advanced access model in Quebec [13] and the flexible time allotments for primary care visits at Southcentral Foundation in Alaska [14]. To do this well requires knowing which visits are diagnostic in nature and scheduling those differently than those for management alone, allotting adequate time and scheduling preference for harder cognitive load visits. Hospital nursing units do this by using acuity scoring to create assignments. Also, in the inpatient setting, distractions and stress, heightened during the pandemic, have created a need for excess time for PPE donning and doffing with worries of viral exposure and transmission, all of which can contribute to cognitive load interference and inhibiting the typical processing of WM and LTM to arrive at diagnoses.

Situated cognition: what is known, what remains to be known?

Much of the literature on situated cognition and cognitive load, interestingly, is set within the educational context. The concept, proposed by Sweller in 1988, is that information whizzing by in a classroom with numerous distractions is unlikely to be adequately stored, integrated and utilized by the distracted student, even if they are trying hard to learn [15]. This implies that it is not the student (or in our case, the clinician) who is at fault for not being able to process what is needed to learn: it is the environment in which that processing occurs. When confronted with a more complex patient, with an unusual and not readily diagnosed medical problem, and a distracted clinician with insufficient time, cognitive overload/breakdown occurs and misdiagnosis is more likely. Add to this a patient with a language barrier, or cultural differences, or a patient who may have experienced a long wait while in pain, and we have a familiar and challenging scenario [16]. The good news is that the environment is malleable with more readily available interpreters, better time management, teamwork that can expand clinician time for quality cognition, bias training, pauses allowing clinicians to reflect on impending diagnostic pitfalls, and access to just-in-time consultation with consultants, peers or online reference materials [17, 18]. Time is present for pausing and processing [19], and decision support is accessible when desired. Studies demonstrating the relative successes of straightforward solutions (such as allotting 50% more time when there are language differences) would be valuable.

How to test the model?

Focus groups with patients and clinicians, followed by a longitudinal observational study, would be important first steps, cataloguing work environment issues that correspond with high and lower quality diagnoses, and examining the relative weights of which variables appear to matter most in predicting better outcomes for clinicians and patients. Refining the measures is an important step to help validate and facilitate their use for both scholars and health systems seeking efficient ways to evaluate and better understand how clinical care is or is not working and what changes could be implemented. A measure of cognitive load, such as the single-item measure described in Konopasky [20], would be a useful asset in such studies. We have distilled key elements of good diagnostic assessments [7] that can be found and measured in the EMR which we have included in our model in Figure 2. Logistic regressions with structural equation modeling could be used to confirm the structure and weights within the model. Once the model is validated, randomized trials of interventions to reduce work stress and improve diagnostic accuracy could be conducted.

Next steps: using “where we work is how we think” to allow us to think better

By focusing on principles enumerated by Merkebu and colleagues [21], such as the embedded mind (thoughts relate to the environment) and ecological psychology (how participants and the environment relate to diagnosis), we have an opportunity to improve diagnoses while we simultaneously improve work conditions for a workforce that is struggling under the weight of two years of a pandemic and work overload. Teamwork, an important aspect of burnout reduction, can be a valuable asset in diagnosis [17], with varied team types (e.g. template (ward) teams and knotwork (consultation) teams) contributing to improved diagnostics. Because much has been written about means to reduce burnout [22], we can now put that to use in assessing impact of work condition improvement on diagnosis [7]. For example, minimizing interruptions is a part of work control; as interruptions lead to distraction, recent work shows that experienced doctors working with distractions perform more like residents than experienced clinicians [23]. Likewise, distractions lead to more expressions of uncertainty by clinicians [24]. Thus, interruptions which relate to stress and burnout, also appear to relate to poorer diagnostics. These data could provide a strong incentive for organizations to assess their workflows and reduce interruptions (e.g. pop up alerts) so that doctors are not frequently interrupted as they try to think through complex cases under time pressure in hectic environments. Efforts to improve workflow could be deeply beneficial to the thought processes of individual clinicians and the teams engaged in solving these problems, as well as signal to clinicians the organization’s priority for improved diagnosis [25].

Because there is pent-up demand for care of other (non-Covid related) medical problems, efficient, accurate diagnosis has never been more important in primary care, hospital medicine, emergency medicine, and urgent care, among other sites. For health systems straining under the weight of Covid-related as well as non-Covid diagnoses, the time is right to truly support better diagnosis [26, 27] and to ensure that requisite work conditions to support good diagnosis are hard-wired into practice so healthcare workers aiming to do the right thing for their patients can do so.

Corresponding author: Dr. Mark Linzer, MD, Department of Medicine and the Institute for Professional Worklife, Hennepin Healthcare and University of Minnesota Medical School, Minneapolis, 701 Park Ave, 55415 MN, USA. Phone: 612-873-6963, E-mail:

  1. Research funding: CRICO Malpractice Insurance Co, Boston, MA, for support of the MD-SOS (Medical Diagnosis Stress or Safety) Project.

  2. Author contributions: All authors have accepted responsibility for the entire content of this manuscript and approved its submission.

  3. Competing interests: Dr. Linzer was paid to consult on this project; funds were donated to Hennepin Healthcare Foundation. Dr. Linzer is also supported through his employer for burnout reduction projects for ACP, AMA, ABIM, IHI, Optum Office for Provider Advancement, Gillette Children’s Hospital and Essentia Health System. He is supported for scholarly studies in shared decision making and burden of treatment through the NIH and as a PI on a Learning Health System (LHS) K12 award by AHRQ. Drs. Schiff and Mirica for funding by the Gordon and Betty Moore Foundation for the PRIDE (Primary Care Improvement in Diagnostic Error) Project.

  4. Informed consent: Not applicable.

  5. Ethical approval: Not applicable.


1. Sweller, J, Van Merrienboer, JJ, Paas, FG. Cognitive architecture and instructional design. Educ Psychol Rev 1998;10:251–96. in Google Scholar

2. Schiano, B. Reducing cognitive overload while teaching. Harvard Business Publishing. Available from: in Google Scholar

3. Linzer, M, Manwell, LB, Williams, ES, Bobula, JA, Brown, RL, Varkey, AB, et al.. Working conditions in primary care: physician reactions and care quality. Ann Intern Med 2009;151:28–36. in Google Scholar PubMed

4. Linzer, M, Gerrity, M, Douglas, J, McMurray, JE, Williams, ES, Konrad, TR. Physician stress: results from the physician worklife study. Stress Health 2002;18:37–42. in Google Scholar

5. Croskerry, P. The rational diagnostician and achieving diagnostic excellence. JAMA 2022;327:317–8. in Google Scholar PubMed

6. Linzer, M, Bitton, A, Tu, SP, Plews-Ogan, M, Horowitz, KR, Schwartz, MD, et al.. The end of the 15–20 minute primary care visit. J Gen Intern Med 2015;30:1584–6. in Google Scholar PubMed PubMed Central

7. Khazen, M, Sullivan, EE, Ramos, J, Mirica, M, Linzer, M, Schiff, GD, et al.. Anatomy of diagnosis in a clinical encounter: how clinicians discuss uncertainty with patients. BMC Prim Care 2022;23:1–9.10.1186/s12875-022-01767-ySearch in Google Scholar PubMed PubMed Central

8. Burgess, D, van Ryn, M, Dovidio, J, Saha, S. Reducing racial bias among health care providers: lessons from social-cognitive psychology. J Gen Intern Med 2007;22:882–7. in Google Scholar PubMed PubMed Central

9. Burgess, DJ. Are providers more likely to contribute to healthcare disparities under high levels of cognitive load? Med Decis Making 2010;30:246–57. in Google Scholar PubMed PubMed Central

10. Audi, C, Poplau, S, Freese, R, Heegaard, W, Linzer, M, Goelz, E. Negative experiences due to gender and/or race: a component of burnout in women providers within a safety-net tospital. J Gen Intern Med 2021;36:840–2. in Google Scholar PubMed PubMed Central

11. Olson, APJ, Linzer, M, Schiff, GD. Measuring and improving diagnostic safety in primary care: addressing the “twin” pandemics of diagnostic error and clinician burnout. J Gen Intern Med 2021;36:1404–6. in Google Scholar PubMed PubMed Central

12. Neprash, HT, Barnett, ML. Association of primary care clinic appointment time with opioid prescribing. JAMA Netw Open 2019;2:e1910373. in Google Scholar

13. Breton, M, Maillet, L, Paré, I, Abou Malham, S, Touati, N. Perceptions of the first family physicians to adopt advanced access in the province of Quebec, Canada. Int J Health Plann Manag 2017;32:e316–2. in Google Scholar

14. Driscoll, DL, Hiratsuka, V, Johnston, JM, Norman, S, Reilly, KM, Shaw, J. Process and outcomes of patient-centered medical care with Alaska Native people at Southcentral Foundation. Ann Fam Med 2013;11:S41–9. in Google Scholar

15. Sweller, J. Cognitive load during problem solving: effects on learning. Cognit Sci 1988;12:257–85. in Google Scholar

16. Schiff, G. Finding and fixing diagnosis errors: can triggers help? BMJ Qual Saf 2012;21:89–92.10.1136/bmjqs-2011-000590Search in Google Scholar PubMed

17. Olson, APJ, Durning, SJ, Branson, CF, Sick, B, Lane, KP, Rencic, JJ. Teamwork in clinical reasoning – cooperative or parallel play? Diagnosis 2020;7:307–12. in Google Scholar PubMed

18. Zwaan, L, Olson, APJ. When measuring Is more important than measurement: the importance of measuring diagnostic errors in health care. J Pediatr 2021;232:14–6. in Google Scholar PubMed

19. Prasad, K, Poplau, S, Brown, R, Yale, S, Grossman, E, Varkey, AB, et al.. Time pressure during primary care office visits: a prospective evaluation of data from the Healthy Work Place study. J Gen Intern Med 2020;35:465–72. in Google Scholar PubMed PubMed Central

20. Konopasky, A, Antino, A, Battista, A, Ohmer, M, Hemmer, PA, Torre, D, et al.. Understanding context specificity: the effect of contextual factors on clinical reasoning. Diagnosis 2020;7:257–64. in Google Scholar PubMed

21. Merkebu, J, Battistone, M, McMains, K, McOwen, K, Witcop, C, Konopasky, A, et al.. Situativity: a family of social cognitive theories for clinical reasoning. Diagnosis 2020;7:169–76. in Google Scholar PubMed

22. Linzer, M, Poplau, S. Eliminating burnout and moral injury: bolder steps required. EClin Med 2021;39:101090. in Google Scholar PubMed PubMed Central

23. Soh, M, Konopasky, A, Durning, SJ, Ramani, D, McBee, E, Ratcliffe, T, et al.. Sequence matters: patterns in task-based clinical reasoning. Diagnosis 2020;7:281–9. in Google Scholar PubMed

24. Ramani, D, Soh, M, Merkebu, J, Durning, SJ, Battista, A, McBee, E, et al.. Examining the patterns of uncertainty across clinical reasoning tasks. Diagnosis 2020;7:299–305. in Google Scholar PubMed

25. Schiff, G, Ruan, EL. The elusive and illusive quest for diagnostic safety metrics. J Gen Intern Med 2018;33:983–5.10.1007/s11606-018-4454-2Search in Google Scholar PubMed PubMed Central

26. Berk, SI. Time to care: primary care visit duration and value-based healthcare. Am J Med 2020;133:655–6.10.1016/j.amjmed.2019.12.046Search in Google Scholar PubMed

27. Freedman, S, Golberstein, E, Huang, T-Y, Satin, J, Smith, B. Docs with their eyes on the clock? The effect of time pressures on primary care productivity. J Health Econ 2021;77:102442. in Google Scholar PubMed PubMed Central

Received: 2022-05-24
Accepted: 2022-07-26
Published Online: 2022-08-22

© 2022 Walter de Gruyter GmbH, Berlin/Boston

Downloaded on 5.12.2022 from
Scroll Up Arrow