Clinical Decision Support Systems and Trust in Automation: Case of a Clinical Reminder for Titration of Beta Blockers

Abstract

Trust in automation depends on more than just the automation itself, but the larger context in which the automation and the human operator are collaborating. This study takes a naturalistic approach to explore providers' trust in a Clinical Decision Support System. Primary Care Providers were shown simulated medical records and a prototype Clinical Reminder indicating that the patient should be titrated with recommended Beta Blockers to address the patient's Heart Failure with reduced ejection fraction. Analysis of responses showed three main themes: Concerns about the medical documentation used to generate the recommendation; Complexity of the patient condition and care delivery context (and how such factors limit possible courses of action); and Concerns about the Clinical Reminder and clinical guideline it is instantiating. These results align with the macrocognitive model of trust and reliance based on sensemaking and flexecution.

Keywords

decision support health information technology trust

Introduction

Trust involves a relationship based on roles and responsibilities, and an implicit agreement (Middleton 2018) for the truster to accept being vulnerable in a specific situation with the expectation that the trustee will behave in a way that benefits the truster (Hall 2001). Thus, trust in automation is about the operator assessing opportunities for collaboration. The operator must be able to evaluate the current capabilities of the artificial agent with respect to the demands of the primary task itself, and to the demands of collaboration. Designing for collaboration with automation means facilitating the ability of the human operator to establish and update justified trust and justified mistrust in the automated agent (Hoffman 2017).

Thus, it is useful to explore trust in automation via actual practitioners performing representative tasks that involve realistic (i.e., imperfect) automated agents. However, most of the literature on trust in automation is theoretical or from experimental studies; with few exceptions (e.g., Dorton & Harper 2022), naturalistic studies of trust in automation have been neglected (Hoffman 2017).

In the domain of medicine, trust in an automated system is manifest in the context of other trust relationships (Montague & Lee 2012; Alexander 2006) such as trust between patient and provider, trust in the organization, trust in practice guidelines, and issues of responsibilities and liabilities. Clinical Decision Support Systems (CDSS) touch all of these different aspects of trust in medicine.

Work focused specifically on CDSS suggests it interacts with trust-associated factors like other cognitive support automation (such as those presented in Hoff 2015; Schaefer 2016). The types of explanations provided by the CDSS affect trust (Bussone 2015). The number of alerts and the rate of false alarms are recognized as issues (Ancker 2017). Operators following incorrect recommendations of CDSS has been identified as a risk (Goddard 2011, Sutton 2020). The operator must evaluate the CDSS for appropriateness of its recommendation in any given patient.

Beta Blocker CDSS as Case for Trust in Automation

This CDSS uses Natural Language Processing and other methods to identify patients who have a particular type of Heart Failure (HF) but are not receiving the type and dosage of beta blocker medication (BB) that is recommended (Stout et al 2019). For such a patient, a component of the CDSS system screens the patient’s chart in the Electronic Health Record system (EHR) for detectable contraindications (e.g., documented allergies, asthma, systolic blood pressure < 90 mmHg) (Smith et al 2018). If no contraindications are detected in the record, it will present a Clinical Reminder (CR) to the practitioner when the patient’s chart in the EHR is accessed. The CR provides information about the guideline and its clinical merit. It also shows the patient’s recent vital signs, weight, current beta blocker prescription (if any), and ejection fraction (which indicates if the patient has the particular type of HF for which this guideline applies).

For the following reasons, this presents an informative case with which to explore issues of trust and automation.

The performance of the automation is not independent of the rest of the system. It depends on the functioning of the EHR (such as how frequently the patient’s chart is automatically updated with information from other parts of the system), and on the documentation entered by providers (e.g., Is the allergy to medication X coded in the correct field or simply mentioned in a free text note?).

The decision to titrate is very complex, involving various risks and trade-offs. Titration of BB is a process that takes several weeks and requires careful (and time-consuming) monitoring of the patient. Even patients without contraindications can experience low energy and other side-effects. Depending on the organizational context, the practitioner may be at risk of over stepping by initiating BB titration. Yet failure to titrate can result in more frequent hospital admissions, and shorter lifespan (Murphy, Ibrahim & Januzzi 2020). This decision may have to be made in a short time frame, with little time for preparation, and with other competing priorities for the 10 minutes available for the encounter with the patient.

This CDSS is designed to facilitate appropriate application of the evidence-based guideline on BB for HF patients with reduced ejection fraction (Yancy 2013). The practitioner’s interpretation of the automation is interwoven with their interpretation of the guideline.

The automation in this CDSS encompasses multiple functions. Following the four automation information processing stages of Parasuraman et al (2000), the CDSS acquires information (extracting it from the patient’s chart), analyzes information (assesses if the patient qualifies for the Clinical Reminder), and does decision selection (making the recommendation for the provider to complete the assessment for BB titration). It even facilitates - but does not initiate - action implementation (titration and referral orders are available from the Clinical Reminder itself).

These functions are ones that the providers have also done themselves and can therefore have some first-hand basis for performance expectations. In contrast, one key function of the automation is to scan through several thousand medical charts to detect patients who meet the criteria for the Clinical Reminder. This function operates at a scale possible only with automation.

Methods

Participants

The participants were all members of primary care teams in two different VA medical centers from the Great Lakes regional network (including some participants from affiliated community-based outpatient centers). There were nine participants: four PCPs (one Nurse Practitioner, three physicians); four clinical pharmacists, one Registered Nurse care manager. Institutional review board approval was obtained for this study.

Protocol

The scenario was that the provider was opening the chart of a patient in preparation for seeing that patient soon. The Clinical Reminder appeared, reflecting the data for that particular patient. The participant was asked to decide how to resolve the Clinical Reminder (i.e., make a decision about what to do regarding the alleged under-treatment of the patient’s HF).

We used a modified Wizard-of-Oz technique (Kelley 2018) to provide a mixed fidelity simulation of the Clinical Reminder and the patient’s chart in CPRS (the EHR of the VA). The participants were asked to think aloud, and were prompted to describe their thinking processes.

Cases

There were two fictional cases, each concerning a patient who was readmitted to the hospital after exacerbation of HF. The cases were generated using de-identified medical records of real patients to reflect the types of challenges that can occur with HF patients in the VA. The cases were not intended to be representative of a typical case. One case was a true positive – the patient correctly qualified for titration with recommended BB but was not on them. The other case was a false positive – the patient had HF with reduced ejection fraction, and was not on recommended BB, but had a contraindication not detectable by the CDSS. In this way our sample includes a situation in which the automation fails, potentially increasing the challenge facing the human provider (Bainbridge 1983; Brauner 2019). The sequence of presentation of the two cases was alternated across participants.

Data Collection and Analysis

The sessions were audio-recorded and transcribed. The transcriptions were reviewed individually by members of the qualitative analysis team (which included specialists in clinical and public health informatics, human factors, and industrial/organizational psychology). Emergent codes were compared, and a consensus code set was established. Two pairs of coders each coded all of the transcripts. The transcripts were reviewed as a group to establish consensus on the coding. Two coders classified the interviews and discussion phrases using a coding rubric. We calculated a kappa statistic based on a systematic sample of interview and discussion group documents. Following four rounds of classification and discussion, the percent agreement was 87.2% and the kappa was 0.8469, representing substantial interrater reliability.

Results

Eleven codes emerged that related directly to trust, which yielded 81 statements. The codes and example quotations are presented in Table 1 (next page), organized by three overarching themes: concerns about documentation, complexity of patient condition and care delivery context, and concerns about clinical reminder and/or guideline.

Table 1.

Results.

Code	sub-code	Example quotations
Concerns about Documentation
Required info not easily available	Lack of specific info about trends	“creatinine was 1.3, but I don't know what his trend has been. “ “OK how’s your stuff been since you left the hospital? …Do I have any other information available since you’ve come home?”
	Lack of info to understand available options	“I mean it doesn’t really say whether or not cardiology was consulted, but I’m curious to know if they were evaluated at cardiology. So are there any consultations. “ “I would like to titrate him up, but then I would also like to know what are some of the other factors that could be preventing maybe the previous provider….”
	Difficult to make sense of info	“ … I don’t think I got a good sense for her and then they stopped the valsartan but I don’t know why they didn’t restart it cause they said that she had hypertension later on in the hospitalization…” “There’s no summary of what happened during the hospitalization so far”
Out of date info		“Well yes but this echo [echocardiogram used to calculate ejection fraction] is way out of date”
Lack of Trust in Documentation	Potential inaccuracies in the documentation	“I also try to compare that [active problem list on the chart coversheet] with the primary care note because they may have something extra in there and may not be 100% reflective of the active problems in the chart…” …She’s definitely at home on a lower dose compared to … what they’ve got in her H&P
	Inconsistencies in the documentation	“they made it sound like everything else was back to the original dosing for the carvedilol then says it’s been decreased so I guess that’s unclear to me.” “This is interesting because they put her back on an at-home dose but the other note says that she is on a lower dose than she needs.”
	Caution in interpretation of the information	“You have more readings from inpatient, so you may not get that many weights or blood pressure or heart rate from outpatient because they're not here – inpatient … before I order or start the patient on a beta blocker, I would want to check and see what the heart rate is now, today,”
Complexity of Patient Condition and Care Delivery Context
Complexity in patient condition and care	Challenges presented by complexity of healthcare system	“Once stabilized sometimes Cardiology will say ‘OK we don’t need to return him for a year and Primary Care please manage the patient’. So it’s a little bit of a dance between Primary Care and specialty care for these folks.”
Complexity in patient condition and care	Challenges presented by complexity of patient situation	“at this point, my question is should I titrate her carvedilol today or should I just re-start her on the valsartan today, so she needs some valsartan but if we can’t restart her valsartan, we might as well titrate her carvedilol because her blood pressure is uncontrolled and she’s not on a target dose but if we can start valsartan, she needs to be on some valsartan for protection if we can do it.” [Cannot start valsartan and BB (carvedilol) at same time]
Specific Clinical Interactions to Monitor		“I would consult the pharmacist. I would say, “Is this really appropriate?” Because uh I don’t want to make their asthma worse” “they were giving her IV hydralazine [blood pressure medication] so of course sometimes it’s 119, sometimes it’s 157. It’s very, it seems like it’s difficult to see what exactly is the safest thing to do”
Concerns about Clinical Reminder and/or Guideline
Lack of Familiarity with Guideline		“…are we supposed to sort of know that you shouldn’t start a beta blocker if the rate already is 50 and you shouldn’t start any antihypertensives, or at least you should get more advice from somebody.” “I think people have a lot of questions about like ‘I’m going all the way to 200 on metoprolol succinate [type of recommended BB]. How high can I start and how, how high does that start, how does that interface with how high their pressure is ‘”
Lack of Trust in Clinical Reminder	Clinical Reminder is insufficient basis for making this big decision	“So I’m somewhat uneasy with the idea of starting a whole new medication simply on the basis of reminders.” “Although [Clinical Reminder for] aspirin is sort of the same [as BB drug administration prompted by Clinical Reminder]… But I feel more comfortable with aspirin.”
Lack of Trust in Clinical Reminder	Unclear about when, how, and to whom the CR will appear	“…but from just seeing this, I’m never going to remember that that’s—probably in 2 hours, I’m not going to remember that this is not supposed to apply to people who have an allergy or reaction to a beta blocker.” “Would it trigger that we shouldn’t start it?…Like their blood pressure was 90 over 40.”
Positive Trust in Reminder		“…and I have faith in that reminder that’s pointing to the appropriate information”

Discussion

The results reaffirm how trust in a Clinical Reminder (or other CDSS) is not just affected by the tool itself (its appearance and screening behavior), but also by: the relationship between the provider and the guideline (familiarity with the guideline, with BB treatment for HF); the accuracy and completeness of the data in the patient’s chart; and the degree of complexity in the patient’s medical condition and in the care delivery system itself.

Our results correspond to the macrocognitive model of trust and reliance proposed by Hoffman et al (2015). Trust in automation is described as emerging from the sensemaking process (gathering and interpreting data to elaborate or revise one’s understanding (Klein et al 2006; Fallon 2010)) focused on two things: the situation or process in the world, and the automation itself.

Reliance on the automation is described as emerging from the flexecution process (the elaboration or revision of plans and goals as information or situations evolve (Klein 2007) executed upon two things: the situation or process in the world, and the automation itself.

Sensemaking

By providing details on the guideline and key data from the patient chart used to generate the recommendation, the CR enables providers to assess the recommendation, to compare their own thinking with that of the automation.

However, the documentation reliability is also a factor. To make sense of the patient’s situation, providers need not just data about the current state (e.g., vitals, diagnoses, medications), but also trends in various aspects of the patient’s health (e.g., is the patient’s kidney function stable?), and the decision making of past providers (e.g., did a cardiologist try the patient on BB and decide to stop?).

Medical documentation may sometimes have deficiencies in accuracy, thoroughness, or structure (Stetson 2012). Our results indicate that providers can be sensitive to this. They expressed concerns about out-of-hospital data being infrequently updated or involving uncertain contextual factors which make it hard to interpret. Their sensemaking activities included looking for additional information to fill-in missing elements or to resolve information that does not seem to fit.

Problems that affect trust in the documentation also affect trust in the CDSS, due to the dependency of the CDSS on its input source (Alexander 2006; Kilsdonk 2017), and via a halo effect (Hoff 2015).

An additional target of sensemaking is the CDSS itself. This involves interpretation of the clinical data extracted and presented in the Clinical Reminder, including assessments of the thoroughness of the search for information, and the recency of the information. Even with complete and accurate data in the record, a CDSS could fail to access or utilize that data correctly.

The sensemaking process would be supported by access to metadata about the clinical data. This would help providers establish and revise their trust or mistrust in the clinical data as mediated by the CDSS.

Sensemaking of the CDSS itself also involves establishing and revising an understanding of how the algorithm uses records to identify which patients should receive the CR and which should not. Some providers expressed perplexity and uncertainty about the algorithm’s exclusion of patients with detectable contraindications. They sought verification that the rule ‘if contraindication then do not trigger CR’ was being followed. This inadvertently put the providers in a situation similar to the classic ‘confirmation bias’ study paradigm (Wason, 1960). Filtering out patients with detectable contraindications is a background process, not visible to the providers. Furthermore, this process is not 100% accurate, resulting in some False Positives.

To better support sensemaking and adjustment of trust and mistrust, providers should be able to access a list of which of their patients were screened and the results. Additionally, the providers should be able to run tests on known patients to observe how the algorithm interprets those patients’ records.

Yet another target of sensemaking is the guideline in the context of this patient-specific clinical reminder. Clinical guidelines can be difficult to parse (Bracha 2011) and their interpretation and application involves cognitive challenges (Klein 2016). Our CDSS specifically flags HF patients who are likely at risk of under-treatment with BB, so it helps with the challenge of characterizing the medical problem and identifying the relevant guideline. However, our results show that lack of familiarity with the guideline, or with BB and HF treatment more generally, can be an issue with some providers, and therefore affects the processes of sensemaking and collaboration between the CDSS and the provider.

Flexecution

The CDSS is intended to help the provider make changes in the treatment of applicable patients. Providers can initiate titration, issue a referral to another provider, or simply document (in a structured way) the reason why BB titration is not indicated. There is more than one way to satisfy the CR, and the provider might start with one in mind but switch to another. The provider might plan to manage the titration themselves and investigate the patient’s other medications. After seeing the number and type of medications, they decide it would be better to have the clinical pharmacist manage it, so they make a referral to the pharmacist. This is one way that flexecution (Klein 2007) is supported by the CR.

The ability to make changes in treatment is affected by dependencies with the healthcare system and with the patient. Some of the BBs may not be on the institution's formulary. There may be ambiguity about responsibility between the PCP, cardiologist, and HF clinic. The patient may have some medical condition that is insufficiently stable to initiate BB titration.

The presence of comorbidities or other interacting factors can impose constraints on what treatments are possible (Klein 2016). There may be other applicable guidelines competing for attention (Hamm & Nagykaldi 2018). Each guideline (and any associated CDSS) is focused on one particular issue by itself, but the provider must view all the different issues as an interacting, integrated set (Horsky 2017).

Thus, even if the CDSS is correct that the patient could benefit from initiation of BB titration, there may be other competing priorities for that patient at that moment. In such situations, it may be necessary to replan or reprioritize what to address now and what to address later. Individual Clinical Reminders may help inform providers of particular medical needs they might otherwise be ignorant of, but there is no good support for the cognitive work of identifying which needs are most important to focus on right now (Xiao & Gorman 2018). The ability of the human-automation team to negotiate a sacrifice of local, short-term goals (such as the focus of a particular CR) for global, longer-term goals (e.g., integrated management of the patient) is fundamental to coordination (Woods & Hollnagel 2006).

If our CR informs the provider of this hitherto unknown problem of undertreated HF, which is then taken into account during the planning of when to address various issues, then the CR has positively contributed to the care of the patient, even if it does not directly lead to BB titration of the qualifying patient at this time.

Limitations

Our cases were designed to explore the role of the Clinical Reminder in the cognitive work of the provider in interpreting the presumptive need for BB titration. As such, they included more complexity than is typical of HF patients in the VA.

Unlike the typical scenario of a provider treating a patient, in our study the providers had never seen the patient before, nor could they contact colleagues who may have known the patient. The only source of information was the simulated EHR chart.

We only looked at the decision to start titration, not issues occurring during titration. And this decision was framed only with the provider and the Clinical Reminder (with the patient’s chart); decision-making with the involvement of the patient or caregiver was not included.

Footnotes

Acknowledgements

This work was supported by VA Health Services Research and Development (VA HSRD CRE 12-037). The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs or the United States Government.

ORCID iD

Smith MW

References

Middleton

Platt

Richardson

Blumenfeld

. Recommendations for Building and Maintaining Trust in Clinical Decision Support Knowledge Artifacts Research Triangle Park, NC: Patient-Centered Clinical Decision Support Learning Network; 2018 Sep p. 21.

Hall

Dugan

Zheng

Mishra

. Trust in physicians and medical institutions: what is it, can it be measured, and does it matter? The Milbank Quarterly 2001; 79(4):613-39

Hoffman

R. R

. (2017). A taxonomy of emergent trusting in the human–machine relationship. Cognitive systems engineering: The future for a changing world, 137-163.

Dorton

S. L.

Harper

S. B

. (2022). A naturalistic investigation of trust, AI, and intelligence work. Journal of Cognitive Engineering and Decision Making, 16(4), 222-236.

Montague

Lee

J. D

. (2012). Trust in health technologies. Handbook of human factors and ergonomics in health care and patient safety, 2, 281-291.

Alexander

G. L

. (2006). Issues of Trust and Ethics in Computerized Clinical Decision Support Systems: Nursing Administration Quarterly, 30(1), 21–29. https://doi.org/10.1097/00006216-200601000-00005

Hoff

K. A.

Bashir

. (2015). Trust in Automation: Integrating Empirical Evidence on Factors That Influence Trust. Human Factors, 57(3), 407–434. https://doi.org/10.1177/0018720814547570

Schaefer

K. E.

Chen

J. Y. C.

Szalma

J. L.

Hancock

P. A

. (2016). A Meta-Analysis of Factors Influencing the Development of Trust in Automation: Implications for Understanding Autonomy in Future Systems. Human Factors, 58(3), 377–400. https://doi.org/10.1177/0018720816634228

Bussone

Stumpf

O’Sullivan

. (2015). The Role of Explanations on Trust and Reliance in Clinical Decision Support Systems. 2015 International Conference on Healthcare Informatics, 160–169. https://doi.org/10.1109/ICHI.2015.26

10.

Ancker

J. S.

Edwards

Nosal

Hauser

Mauer

Kaushal

, & with the HITEC Investigators. (2017). Effects of workload, work complexity, and repeated alerts on alert fatigue in a clinical decision support system. BMC Medical Informatics and Decision Making, 17(1), 36. https://doi.org/10.1186/s12911-017-0430-8

11.

Goddard

Roudsari

Wyatt

J. C

. (2012). Automation bias: A systematic review of frequency, effect mediators, and mitigators. Journal of the American Medical Informatics Association : JAMIA, 19(1), 121–127. https://doi.org/10.1136/amiajnl-2011-000089

12.

Sutton

R. T.

Pincock

Baumgart

D. C.

Sadowski

D. C.

Fedorak

R. N.

Kroeker

K. I

. (2020). An overview of clinical decision support systems: Benefits, risks, and strategies for success. Npj Digital Medicine, 3(1), 1–10. https://doi.org/10.1038/s41746-020-0221-y

13.

Stout

K. K.

Daniels

C. J.

Aboulhosn

J. A.

Bozkurt

Broberg

C. S.

Colman

J. M.

… Van Hare

G. F

. (2019). 2018 AHA/ACC guideline for the management of adults with congenital heart disease: executive summary: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Journal of the American College of Cardiology, 73(12), 1494-1563.

14.

Smith

M. W.

Brown

Virani

S. S.

Weir

C. R.

Petersen

L. A.

Kelly

… Garvin

J. H

. (2018). Incorporating guideline adherence and practice implementation issues into the design of decision support for beta-blocker titration for heart failure. Applied clinical informatics, 9(2), 478.

15.

Murphy

S. P.

Ibrahim

N. E.

Januzzi

J. L

. (2020). Heart failure with reduced ejection fraction: a review. Jama, 324(5), 488-504.

16.

Yancy

C. W.

Jessup

Bozkurt

Butler

Casey

D. E.

Drazner

M. H.

Fonarow

G. C.

Geraci

S. A.

Horwich

Januzzi

J. L.

Johnson

M. R.

Kasper

E. K.

Levy

W. C.

Masoudi

F. A.

McBride

P. E.

McMurray

J. J. V.

Mitchell

J. E.

Peterson

P. N.

Riegel

… Wilkoff

B. L

. (2013). 2013 ACCF/AHA Guideline for the Management of Heart Failure: A Report of the American College of Cardiology Foundation/American Heart Association Task Force on Practice Guidelines. Circulation, 128(16), e240–e327. https://doi.org/10.1161/CIR.0b013e31829e8776

17.

Parasuraman

Sheridan

T. B.

Wickens

C. D

. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, 30(3), 286–297. https://doi.org/10.1109/3468.844354

18.

Kelley

J. F

. (2018). Wizard of Oz (WoZ)—A Yellow Brick Journey. Journal of Usability Studies, 13(3), 119–124.

19.

Bainbridge

. (1983). Ironies of automation. Automatica, 19(6), 775–779.

20.

Brauner

Philipsen

Valdez

A. C.

Ziefle

. (2019). What happens when decision support systems fail? — The importance of usability on performance in erroneous systems. Behaviour & Information Technology, 0(0), 1–18. https://doi.org/10.1080/0144929X.2019.1581258

21.

Hoffman

Best

Klein

(2015). Trust and Reliance as Emergent Phenomena in Macrocognitive Work: An Integrated Model. Proceedings of the 12th International Naturalistic Decision Making Conference, 5.

22.

Klein

Moon

Hoffman

(2006). Making Sense of Sensemaking 1: Alternative Perspectives. IEEE Intelligent Systems, 21(4), 70–73. https://doi.org/10.1109/mis.2006.75

23.

Fallon

C. K.

Murphy

A. K. G.

Zimmerman

Mueller

S. T.

(2010). The calibration of trust in an automated system: A sensemaking process. 2010 International Symposium on Collaborative Technologies and Systems, 390–395. https://doi.org/10.1109/CTS.2010.5478488

24.

Klein

(2007). Flexecution as a paradigm for replanning, part 1. Intelligent Systems, IEEE, 22(5), 79–83.

25.

Stetson

P. D.

Bakken

Wrenn

J. O.

Siegler

E. L.

(2012). Assessing Electronic Note Quality Using the Physician Documentation Quality Instrument (PDQI-9). Applied Clinical Informatics, 3(2), 164–174. https://doi.org/10.4338/ACI-2011-11-RA-0070

26.

Kilsdonk

Peute

L. W.

Jaspers

M. W. M.

(2017). Factors influencing implementation success of guideline-based clinical decision support systems: A systematic review and gaps analysis. International Journal of Medical Informatics, 98, 56–64. https://doi.org/10.1016/j.ijmedinf.2016.12.001

27.

Wason

Peter C.

(1960), "On the failure to eliminate hypotheses in a conceptual task", Quarterly Journal of Experimental Psychology, 12 (3): 129–140, doi:10.1080/17470216008416717, ISSN 1747-0226

28.

Bracha

Brottman

Carlson

(2011). Physicians, Guidelines, and Cognitive Tasks. Evaluation & the Health Professions, 34(3), 309–335. https://doi.org/10.1177/0163278710392981

29.

Klein

D. E.

Woods

D. D.

Klein

Perry

S. J.

(2016). Can We Trust Best Practices? Six Cognitive Challenges of Evidence-Based Approaches. Journal of Cognitive Engineering and Decision Making, 10(3), 244–254. https://doi.org/10.1177/1555343416637520

30.

Hamm

R. M.

Nagykaldi

Z. J.

(2018). Physician Judgment and Clinical Practice Guidelines. Journal of Cognitive Engineering and Decision Making, 12(3), 209–214. https://doi.org/10.1177/1555343418782850

31.

Horsky

Aarts

Verheul

Seger

D. L.

van der Sijs

Bates

D. W.

(2017). Clinical reasoning in the context of active decision support during medication prescribing. International Journal of Medical Informatics, 97, 1–11. https://doi.org/10.1016/j.ijmedinf.2016.09.004

32.

Xiao

Gorman

P. N.

(2018). Controlling Versus Supporting in a Sociotechnical System: A Commentary on Falzer (2018). Journal of Cognitive Engineering and Decision Making, 12(3), 215–218. https://doi.org/10.1177/1555343418777342

33.

Woods

Hollnagel

(2006). Joint Cognitive Systems: Patterns in Cognitive Systems Engineering. CRC Press.