When emergency physicians meet patients displaying irritable behaviours: a randomised vignette-based experiment investigating physicians’ emotions and clinical reasoning

Abstract

Background.

Clinical encounters with patients who display challenging behaviors (e.g., rude, demanding) are increasingly common in emergency departments and can heighten negative emotions, compromise healthcare quality, and impact physician well-being. This study investigated the influence of patients’ irritable behaviors on physicians’ emotions, clinical assessments, decisions, and behaviors, and the extent to which physicians’ predispositions to experience anxiety and concern about bad clinical outcomes (i.e., stress from uncertainty) may magnify these effects. We also explored whether these effects differed if patients had a stigmatized condition (mental illness history).

Methods.

Experienced emergency physicians (n=134) completed a randomized vignette-based experiment between June and August 2022. Physicians assessed four patients in computer-simulated, interactive clinical encounters in which patient behavior (irritable versus calm) and mental illness history (present versus absent) were varied within subjects. Physicians viewed each patient’s health record, a video-recorded encounter, and physical exam results, and provided patient assessments, clinical decisions, ordered tests, and reported a diagnosis. Physicians also reported their emotional reactions to each patient and completed the Stress from Uncertainty Scale (SUS) after assessing all patients. We adhered to the Guidelines for Reporting Vignette Experiments (GROVE).

Results.

Physicians reported greater anger, anxiety, and fatigue, and less empathy, happiness, and engagement during encounters with patients displaying irritable (versus calm) behaviors (all ps<0.001), and evaluated those with irritable behaviors as poorer historians, more likely to be exaggerating pain, less cooperative, engaged, likeable, likely to adhere to treatment, and likely to return to work (all ps<0.001). Interactions revealed that SUS amplified the impact of patients’ irritable behaviors with high SUS associated with greater physician anger, anxiety, and fatigue (all ps≤0.04) and with physicians evaluating these patients as poorer historians, less likely to adhere to treatment, and less likeable (all ps≤0.03). Effects were similar regardless of patient mental illness history (all ps>.60). No effects emerged for clinical decisions, behavior, or diagnosis.

Conclusion.

Although patients’ irritable behaviors adversely influenced physicians’ emotions, engagement, and patient assessments, they did not affect the quality of care provided (i.e., clinical decisions, behavior, diagnostic accuracy). Despite this, our findings have important implications. Physicians’ reactions to patients displaying irritable behaviors may reduce the quality of physician-patient relationships (e.g., reduce trust, effective communication) and contribute to physician burnout – both of which can compromise patient care. Additional research is needed to investigate the multiple ways in which emotional factors can impact patient care quality and safety.

INTRODUCTION

Clinical encounters with patients who display uncivil, rude, and challenging behaviors (e.g., demanding, verbally or physically abusive) are common and have been rising in recent years[1–4]. Such patients are at increased risk of receiving lower quality healthcare. In two written vignette studies, for example, patients’ disruptive behaviors reduced diagnostic accuracy among medical residents[5,6]. Although such behaviors are hypothesized to elicit anger, frustration, and irritation in clinicians[7–9], few studies have examined emotions directly[10–12]. Qualitative[13,14] and mixed methods[15] research has linked these emotions to reduced clinician engagement with patients, negative patient perceptions, and clinical decision-making. In one study, for example, 63% of emergency physicians and nurses who described a recent patient encounter in which they felt angry, frustrated, or irritated reported their emotions influenced their clinical assessments, decision-making, or patient care[15]. Nearly all (92%) reported a detrimental impact (e.g., acting less professionally, delaying or not providing a full exam). Thus, encounters with patients displaying irritable behaviors may heighten risks to patient healthcare quality and may also have implications for clinician well-being.

Uncertainty, like emotion, is ubiquitous in medicine, and can influence clinical reasoning and decision making throughout the diagnostic process[16,17] impacting how clinicians gather, integrate, and interpret information[16,18]. Uncertainty varies across clinical situations in response to multiple factors, including characteristics of patients and situations[19,20]. For example, patients who display irritable or challenging behaviors can heighten uncertainty. Given uncertainty can have adverse effects if not well managed[16,20–22], relatively stable trait-like factors that predispose clinicians to respond to uncertainty in particular ways can play an important role in how uncertainty influences their emotional experiences, clinical reasoning, and decision making[20,21,23]. Individual differences in anxiety and concerns about bad outcomes (i.e., stress from uncertainty[24,25]) capture the affective component of uncertainty intolerance[20,21]. This component is particularly relevant in examining physicians’ affective and clinical responses to patients who display irritable behaviors. That is, individuals predisposed to have negative affective reactions to uncertainty (e.g., those who score high on the Stress from Uncertainty Scale [SUS]), are particularly likely to be impacted by situational factors[20,24–27] that can elicit emotion.

Using a novel interactive experimental vignette paradigm, we tested two hypotheses. First, we hypothesized that patients displaying irritable (versus calm) behaviors would adversely influence physicians’ emotions, clinical assessments, decisions, and behaviors during simulated patient encounters. Second, we hypothesized that the effects of patient behaviors may be amplified among physicians predisposed to experience high stress from uncertainty. Additionally, as an exploratory aim, we examined whether the relationship between patient behaviors and SUS differed if patients had a history of mental illness – a condition frequently stigmatized by healthcare providers[14,28] that can trigger both negative affect and increased uncertainty. Finally, based on two vignette studies that found medical residents made more diagnostic errors when patients engaged in disruptive behaviors (assumed to trigger negative emotions)[5,6], our second exploratory aim examined whether patients’ irritable behaviors may have downstream consequences and impact diagnostic accuracy.

Providing care to patients in the emergency department (ED) presents unique challenges, including highly unpredictable conditions, high variability in patient needs, lack of pre-existing relationships with patients, limited patient information, high patient volume, challenging behaviors from patients, limited time, and limited resources[29,30]. These real-world factors can contribute to heightened emotions and uncertainty with significant consequences. For this reason, we tested our hypotheses with emergency physicians in a simulated ED context, which provides a high degree of control over variables frequently confounded in real-world settings.

METHOD

Study Design

We conducted a randomized vignette-based experiment online between June and August 2022. The experiment was developed and administered using Qualtrics survey software (https://www.qualtrics.com ). Participants assessed four simulated ED patient cases presented in multimedia format, with patient behavior (irritable vs calm) and mental illness history (present vs absent) systematically varied within subjects. For each case, participants reported patient assessments, clinical judgments and behaviors, and emotional reactions to the patient. We followed the Guideline for Reporting Vignette Experiments (GROVE)[31].

Selection of Participants

Participants were recruited from a list of randomly selected attending emergency physicians for each US state purchased from the American College of Emergency Physicians. We sent 1,000 invitations (via postal mail) describing the study’s focus on medical decision-making. Invitations included a Quick-Response (QR) code that physicians could scan to provide their email address. We then emailed each physician a secure, one-time-use link to complete the study online (see Supplement for recruitment materials). Participants received a $200 electronic gift card for completing the 60-minute study.

Development of Clinical Cases and Vignettes

Two board certified emergency physicians (PS and HB) developed four multimedia patient cases in which patients presented to the ED with symptoms of physical illness. For each case, we developed a brief vignette describing the patient’s chief complaint and interaction with a triage nurse, an electronic health record (EHR), a video-recorded interaction between the patient and physician, results of a physical exam, and results of clinical tests and imaging studies that participants ordered. The cases included clinical data to support specific diagnoses for each of the following chief complaints: fatigue (myocardial infarction), fever and sore throat (malaria), migraine headache (venous thrombosis), and abdominal pain (adrenal insufficiency).

We experimentally manipulated patient behavior (irritable versus calm) and mental illness history (present versus absent) in three ways. First, to introduce patients, we developed a brief written vignette that listed patients’ chief complaint and noted that they saw a triage nurse with whom they behaved in either an irritable manner (e.g., yelled at nurse) or a calm manner (e.g., thanked nurse). Second, we presented the patient’s EHR, which included the patient’s chief complaint, brief medical history, medications, and basic medical information. To manipulate history of mental illness, a mental health diagnosis was included in the medical history, and medications taken to treat the condition were listed. Mental illness histories for each clinical case included schizophrenia (fatigue), depression/substance use (fever and sore throat), anxiety/depression (migraine headache), and bipolar disorder (abdominal pain). This information was excluded for patients without a mental illness history.

Third, for each case, we created video-recorded patient encounters that correspond to the four experimental conditions. We recruited four standardized patient actors (two White males, two White females) from a professional medical simulation center and trained each to play the role of a patient in one clinical case. Informed by research identifying patient behaviors that elicit irritation and anger in emergency physicians[13], we created video-recorded encounters in which patients engaged in several of these behaviors (e.g., questioning physician’s experience, using irritated voice). For calm conditions, we developed comparable videos that excluded these behaviors and presented patients speaking in a calm voice. Identical medical information was presented in all versions of each case. To maximize experimental control, we manipulated whether patients had a history of mental illness by editing one brief segment near the end of each video. Specifically, when the physician asked patients what medications they were taking, patients accurately reported information in their EHR. Thus, experimental materials depicted a history of mental illness, not a clinical presentation of mental illness.

Videos were 3–4 minutes long and showed the patient on an examination table in a hospital gown interacting with an out-of-frame male physician (PS). The physician asked common questions (e.g., can you describe your symptoms in more detail?)[32]. Questions were specific to each clinical case; however, they were the same across the four experimental conditions within a case. Detailed results of a physical exam followed the encounter video and were identical across conditions. (See Supplement for sample case materials.)

We developed clinical cases, measures, and all experimental materials and procedures using a highly iterative and rigorous process that was informed by extensive feedback from six emergency physicians, two patient safety experts, and a pilot study with 82 emergency physicians. Throughout this process (detailed in Supplement), we aimed to create cases that were complex (a source of uncertainty[20,22]) and challenging for attending emergency physicians.

Experimental Procedure

Participants were randomly assigned to one of four patient case sets. Each set contained all four clinical cases and included one patient from each of the four experimental conditions (patient behavior [irritable vs. calm] x mental illness history [present vs. absent]; see Supplement for case set composition). Patients were presented in random order, and participants were instructed to evaluate them “as if they were in the ED” with the goal of diagnosing them. We collected three sets of outcome measures during each patient encounter: (A) physician’s emotions and engagement, (B) assessments of patients and clinical cases, and (C) clinical decisions, behaviors, and diagnosis. See Table 1 for measures that comprise each set of outcomes. The development of all measures and procedures was informed by a literature review, consultations with experts, and extensive pilot testing (detailed in Supplement) to ensure we captured emotions, assessments, and decisions physicians make when evaluating patients and engaged in the diagnostic process during actual clinical practice. See Figure 1 for study flow and timing of measures that comprise each set of outcomes (i.e., A, B, and C shown in Table 1).

Table 1.

Overview of Key Study Variables

Outcome Variables	Measures
A. Physicians’ Emotions and Engagement
Anger, anxiety, fatigue, empathy, happy, sad, Engaged	To what extent did you feel each of the following while watching this patient talk to the doctor? Items: angry, frustrated, irritated, anxious, uneasy, calm, fatigued, empathic, happy, sad, engaged. Scale: 0 (not at all) to 100 (very much). [15,13,32] A composite anger measure was created by averaging response to angry, frustrated, and irritated (Cronbach’s α = .91); a compo site anxiety measure was created by averaging responses to anxious, uneasy, and calm (reverse scored) (Cronbach’s α = .78). †
B. Assessments of Patients and Clinical Cases
Patient Assessments
Pain exaggeration	On a scale from 0 to 10 with 0 being no pain at all and 10 being the worst pain imaginable, how do you think this patient might rate their pain? Using the same 0 to 10 pain scale, how much pain do you believe this patient is experiencing? A difference score was computed to assess perceptions of pain exaggeration (response to first question minus response to second)[1].
Good historian	To what extent did you find this patient to be a good historian? To what extent do you believe this patient reported their history (a) honestly? (b) accurately? To what extent do you believe this patient reported their symptoms (a) honestly? (b) accurately? Scale: 0 (not at all) to 100 (very much). a composite measure was created by averaging responses to these items (Cronbach’s α = .92)[32].
Seriousness of patient condition	To what extent do you think this patient’s condition is serious? Scale: 0 (not serious at all) to 100 (very serious)
Patient cooperative Patient engaged	To what extent do you believe the patient was (a) cooperative? (b) engaged? Scale: 0 (not at all) to 100 (very much)
Patient likeable	How likable did you find this patient? Scale: 0 (very unlikable) to 50 (neither likeable nor unlikeable) to 100 (very likeable)
Challenging case	To what extent did you find this case challenging? Scale: 0 (not at all challenging) to 100 (very challenging)
Expected Patient Outcomes
Treatment adherence, Recovery, Return to work	How optimistic are you that this patient will... (a) adhere to treatment? (b) recover from the chief complaint that led to this visit? (c) return to work soon? Scale: 0 (not at all optimistic) to 100 (very optimistic)
C. Clinical Decisions, Behavior, and Diagnosis
Clinical Decisions and Behavior
Patient priority level	What level of priority would you assign this patient? Scale: 1 (high priority, potentially dangerous health problem) to 5 (low priority). measure is based on the Emergency Severity Index used in EDs in the USA[33].
Necessity of emergency care	To what extent do you believe that this patient requires emergency care? Scale: 0 (does not require emergency care at all) to 100 (definitely requires emergency care)
Total clinical tests ordered	Number of clinical tests ordered during two rounds of test ordering (see Supplement for test ordering procedure.)
Perceived knowledge of case	To what extent do you think you know what is going on in this case? Scale: 0 (I do not know what is going on in this case at all) to 100 (I completely understand what is going on in this case)
Likelihood of seeking consultation	How likely would you be to seek consultation on this case? Scale: 0 (extremely unlikely) to 50 (neither likely nor unlikely) to 100 (extremely likely)
Likelihood of admitting patient	How likely would you be to admit this patient? Scale: 0 (extremely unlikely) to 50 (neither likely nor unlikely) to 100 (extremely likely)
Diagnoses
Differential diagnosis	Please provide your differential diagnosis for this patient. Responses independently scored 0 (incorrect) or 1 (correct) by PS and VK blinded to experimental condition (Kappa = 0.96)
Final diagnosis	What is your diagnosis for this patient? Response scored as above (Kappa= 0.94)
Confidence in final diagnosis	How confident are you in your diagnosis of this patient? Scale: 0 (not confident at all) to 100 (extremely confident)

^†Participants were presented with a photo of each patient when reporting their emotional responses. Emotion items were selected following a review of emotion scales used in psychological research. We found no scales adequately captured emotions likely to be experienced by physicians during patient encounters, includ-ing the widely used Positive and Negative Affect Schedule (PANAS)[34], which includes guilty, scared, hostile, and afraid, and does not include angry or frus-trated. Our scales are based on prior research in which physicians described their emotional experiences in clinical practice[13] or rated emotions they experi-enced during patient encounters[15].

Figure 1.

Study Flow Diagram

Note. Gray boxes indicate patient information provided to participants; green boxes indicate questionnaires and tasks relevant to patient encounters. Measures were collected in the order listed. Letters following each measure correspond to the set of outcome variables assessed (see Table 1): A = Physicians’ Emotions and Engagement, B = Assessments of Patients and Clinical Cases, C = Clinical Decisions, Behaviors, and Diagnosis

Experimental Manipulation Checks.

To assess the effectiveness of our patient behavior manipulation (irritable versus calm), participants rated the extent to which they believed the patient was angry, frustrated, irritated, anxious, worried, calm, and sad using a sliding scale from 0 (not at all) to 100 (very much). We averaged responses to angry, frustrated, and irritated items to create a composite irritability score (Cronbach’s α = .90). To assess the effectiveness of the mental health history manipulation, participants rated the extent to which they believed the patient was mentally ill and physically ill using the same scale.

Stress from Uncertainty Scale (SUS)

To assess participants’ predisposition to experience stress from uncertainty, participants completed the validated 8-item SUS (see Supplement), which consists of two subscales from the Revised Physicians' Reactions to Uncertainty Scale:[25]: (1) anxiety due to uncertainty (e.g., “Uncertainty in patient care makes me uneasy”) and (2) concern about bad outcomes (“I fear being held accountable for the limits of my knowledge”). Participants responded to each statement using a 6-point Likert-type scale (1=strongly disagree, 6=strongly agree). Items were averaged to create a SUS score (Cronbach’s α = 0.91).

Additional Measures

Participants completed other measures for an unrelated study (see Supplement for details), including the 6-item Attitudes of Healthcare Providers Towards People with Mental Illness Scale (a subscale of the Opening Minds Stigma Scale for Healthcare Providers)[28]. Relevant to our first exploratory aim, this measure allowed us to assess stigmatizing attitudes. Participants responded to statements (e.g., “I struggle to feel compassion for a person with a mental illness”) using a 6-point Likert-type scale (1=strongly disagree, 6=strongly agree). Items were averaged to create an overall score (Cronbach’s α = 0.77).

Finally, participants provided demographic information (see Table 2).

Table 2.

Participant Characteristics

	M (SD)
Age
Mean (SD)	45.9 (10.6)
Range	30 – 76
Years of Experience in Emergency Medicine since Completing Residency
Mean (SD)	14.22 (9.89)
Range	1 – 42
Stress from Uncertainty (SUS)
Mean (SD)	4.04 (1.12)
Range	1 – 6
	N (%)
Gender
Male	98 (73.1)
Female	35 (26.1)
Prefer not to respond	1 (0.7)
Race/Ethnicity
White	113 (84.3)
Asian	7 (5.2)
Black	7 (5.2)
Hispanic	2 (1.5)
White, Hispanic	2 (1.5)
American Indian or Alaskan Native	1 (0.7)
Other	1 (0.7)
Prefer not to respond	1 (0.7)
Education
Doctor of Medicine (MD)	123 (91.8)
Doctor of Osteopathic Medicine (DO)	9 (6.7)
MD/DO Not Specified	2 (1.5)
Practice Type
Community Hospital	90 (67.2)
University Hospital	24 (17.9)
University Hospital & Community Hospital	10 (7.5)
Community Hospital & Private Practice	4 (3.0)
Private Practice	4 (3.0)
Other	2 (1.5)

Data Analysis

We used descriptive statistics to characterize the sample, and within-subjects analyses of variance to analyze experimental manipulation check questions.

For all continuous primary outcome variables, we used multilevel modeling given the nested structure of observations within participants. Predictors included SUS, patient behavior (irritable vs. calm), patient mental illness history (present vs. absent), and their interactions. To account for the variability in outcomes introduced by differences between participants and patient health conditions (i.e., case sets), we included these as random intercepts in the models. Prior to running the models, we mean-centered SUS by subtracting its overall mean from each SUS value. We adjusted the coding of patient behavior and patient mental illness by subtracting 0.5 from the original coding (i.e., −0.5=calm behavior, 0.5=irritable behavior; −0.5=no mental illness; 0.5=mental illness) to enhance the interpretability of model coefficients[35]. We estimated the models using the ‘lme4’ package in RStudio (version: 2023.12.1+402, base R version: 4.3.2). For binary dependent variables (e.g., diagnostic accuracy), we employed multilevel logistic regression using the ‘glmer’ function from the ‘lme4’ package and reported odds ratios (ORs).

We controlled for multiple testing using the Benjamini–Hochberg procedure, which controls the false discovery rate (FDR) within each outcome family shown in Table 1.[36] We reported FDR-adjusted p-values in the results; raw p-values are in the Supplement.

To quantify the detectable effect size of our design, we conducted a post-data sensitivity power analysis (GPower 3.1) including all model terms (SUS, patient behavior, mental illness history, all two-way and three-way interactions; n=134). The minimum detectable interaction effect was f² =0.059, suggesting our sample was adequately powered to detect interaction effects in the small-to-medium range[37].

RESULTS

Sample Characteristics

Twenty study invitations were returned as undeliverable. Of the remaining 980, 208 physicians (21.2%) requested a link to participate and 134 (64.4%) completed the study. Most physicians who requested a link but did not complete the study never accessed it. Data were collected from physicians in 46 US states and the District of Columbia, with between 1 and 7 physicians from each state. Participant characteristics are in Table 2.

Experimental Manipulation Checks

We confirmed that participants perceived patients in the irritable (versus calm) behavior conditions to be more irritable (M=80.85 v. 30.83, p<0.001) and less calm (M=24.23 v. 55.23, p<0.001). Patients displaying irritable behaviors were also perceived to be more anxious (M=60.79 v. 54.45, p=0.02), but not significantly more worried (M=65.03 v. 67.86, p=0.10) or sad (M=46.48 v. 48.99, p=0.31). Thus, our manipulation led participants to perceive the patients as intended, with little or no effect on other emotions.

Next, we confirmed that participants were more likely to perceive patients with (versus without) a history of mental illness as mentally ill (M=45.59 v. 34.91, p<0.001). Patients in irritable (versus calm) behavior conditions were also perceived as more likely to be mentally ill (M=45.08 v. 35.42, p<0.001). Lastly, consistent with the design of our clinical cases, participants’ perceptions of patients’ likelihood of having a physical illness did not differ as a function of patient behavior (irritable versus calm; M=71.71 v. 70.79, p=0.59), or history of mental illness (present versus absent; M=70.60 v. 71.90, p=0.48).

Main Results

Means and standard deviations for all encounter-related variables are in the Supplement (Table S1). Results for main analyses are in Table 3.

Table 3.

Patient Behavior (Calm versus Irritable), Stress from Uncertainty (SUS), and their Interaction as Predictors of Physicians’(A) Emotions and Engagement, (B) Assessments of Patients and Clinical Cases, and (C) Clinical Decisions, Behavior, and Diagnosis

	Main Effects of SUS			Main Effects of Patient Behavior			SUS x Patient Behavior Interaction Effects
	Coef.(SE) ††	t	p	Coef.(SE) ††	t	p	Coef. (SE) ††	t	p
A. Physicians’ Emotions and Engagement
Anger†	2.67 (1.23)	2.17	0.11	31.61 (1.56)	20.21	<0.001	4.36 (1.41)	3.10	0.01
Anxiety†	2.66 (1.35)	1.98	0.12	15.25 (1.29)	11.81	<0.001	2.73 (1.16)	2.35	0.04
Fatigue	4.61 (1.61)	2.86	0.03	14.47 (1.67)	8.67	<0.001	4.31 (1.50)	2.87	0.02
Empathy	0.30 (1.24)	0.24	0.83	−20.84 (1.45)	−14.38	<0.001	−1.35 (1.30)	−1.04	0.35
Happy	−0.64 (1.23)	−0.52	0.83	−11.88 (1.36)	−8.75	<0.001	0.10 (1.22)	0.08	0.93
Sad	−0.30 (1.42)	−0.21	0.83	1.47 (1.81)	0.82	0.42	3.07 (1.63)	1.89	0.10
Engaged	0.76 (1.32)	0.58	0.83	−14.98 (1.41)	−10.59	<0.001	−2.28 (1.27)	−1.79	0.10
B. Assessments of Patients and Clinical Cases
Patient Assessments
Pain exaggeration†	−0.02 (0.09)	−0.18	0.86	1.22 (0.14)	8.70	<0.001	0.10 (0.13)	0.78	0.61
Good historian†	−0.23 (0.96)	−0.24	0.86	−14.11 (1.33)	−10.61	<0.001	−3.18 (1.20)	−2.66	0.03
Seriousness of patient condition	1.64 (1.01)	1.62	0.25	3.75 (1.75)	2.15	0.04	−2.90 (1.57)	−1.85	0.15
Patient cooperative	0.61 (0.92)	0.66	0.86	−34.31 (1.47)	−23.38	<0.001	−1.53 (1.32)	−1.16	0.43
Patient engaged	−0.31 (1.18)	−0.26	0.86	−19.29 (1.65)	−11.67	<0.001	−0.28 (1.49)	−0.19	0.85
Patient likeable	−1.51 (0.85)	−1.77	0.24	−36.79 (1.45)	−25.39	<0.001	−4.50 (1.30)	−3.46	<0.01
Challenging case	4.03 (1.12)	3.59	<0.01	0.41 (1.71)	0.24	0.81	−0.69 (1.54)	−0.45	0.77
Expected Patient Outcomes
Treatment adherence	−0.79 (1.03)	−0.77	0.44	−17.48 (1.77)	−9.88	<0.001	−4.93 (1.59)	−3.10	<0.01
Recovery from chief complaint	1.07 (0.88)	1.21	0.44	−1.41 (1.43)	−0.99	0.32	−2.43 (1.28)	−1.90	0.09
Return to work	1.04 (1.32)	0.79	0.44	−7.80 (1.82)	−4.28	<0.001	−0.95 (1.64)	−0.58	0.56
C. Clinical Decisions, Behavior, and Diagnosis
Clinical Decisions and Behavior
Patient priority level	−0.03 (0.03)	−1.07	0.54	−0.05 (0.06)	−0.87	0.52	0.07 (0.05)	1.42	0.49
Necessity of emergency care	1.55 (0.86)	1.82	0.54	−1.28 (1.38)	−0.92	0.52	−1.65 (1.24)	−1.33	0.49
Total clinical tests ordered	−0.08 (0.12)	−0.70	0.65	0.24 (0.16)	1.45	0.52	−0.07 (0.15)	−0.46	0.86
Perceived knowledge of case	−1.38 (1.03)	−1.34	0.54	0.16 (1.97)	0.08	0.94	−0.29 (1.77)	−0.16	0.87
Likelihood of seeking consultation	1.33 (1.39)	0.96	0.54	4.95 (2.41)	2.05	0.32	−4.48 (2.17)	−2.07	0.32
Likelihood of admitting patient	1.78 (1.41)	1.26	0.54	3.15 (2.47)	1.28	0.52	−1.57 (2.22)	−0.71	0.83
Diagnoses
	OR	z	p	OR	z	p	OR	z	p
Differential diagnosis	1.04	0.44	0.66	1.27	1.09	0.48	1.05	0.25	0.80
Final diagnosis	1.12	1.19	0.31	1.44	1.93	0.21	0.88	−0.77	0.80
	Coef.(SE) ††	t	p	Coef.(SE) ††	t	p	Coef. (SE) ††	t	p
Confidence in final diagnosis	−1.55 (1.06)	−1.46	0.29	−0.08 (1.99)	−0.04	0.97	1.02 (1.79)	0.57	0.80

^†Composite Measure (see Table 1 for details)

^††Unstandardized regression coefficients; Standardized coefficients (i.e., standardized effect sizes) are presented in the Supplement (Table S2)

Note. Coef = Coefficient; SE = Standard Error; OR = Odds Ratio. The coefficients for stress from uncertainty (SUS) and patient behavior repre-sent conditional effects. The coefficients for SUS refer to its effects averaged across both patient behavior conditions, and the coefficients for pa-tient behavior represent its effects at the mean level of SUS. All p-values are FDR-adjusted. Raw p-values are presented in the Supplement (Table S3)

Analyses of all outcome variables as a function of our primary predictors (SUS, patient behavior, mental illness history, and their interactions) yielded no interactions involving all three variables (all ps>.60) and no consistent effects of mental illness history. For this reason, our results focus primarily on SUS and patient behavior. Results for full models are in the Supplement (Table S3).

Influence of Patient Behavior.

We hypothesized that participants would respond to patients displaying irritable (versus calm) behaviors with more negative emotions, less engagement, more negative patient evaluations, and be associated with lower quality clinical decisions and behaviors. Consistent with this, participants reported experiencing less engagement (p<0.001), empathy (p<0.001) and happiness (p<0.001), and higher levels of anger (p<0.001), anxiety (p<0.001), and fatigue (p<0.001) in response to patients displaying irritable (versus calm) behaviors. Participants also assessed these patients to be exaggerating their pain more (p<0.001), as poorer historians (p<0.001), and less cooperative (p<0.001), engaged (p<0.001), likeable (p<0.001), likely to adhere to treatment (p<0.001), and likely to return to work (p<0.001). Additionally, participants perceived patients’ medical conditions as more serious when patients displayed irritable behaviors (p=0.04). We found no significant effects of patient behavior on clinical decisions, behaviors, or diagnosis (all ps>0.30).

Role of Stress for Uncertainty (SUS).

Our second hypothesis was that SUS would magnify the effects of patients’ behaviors. As expected, interactions between SUS and patient behavior revealed that as SUS increased, participants reported feeling more anger (p=0.01), anxiety (p=0.04), and fatigue (p=0.02) in response to patients displaying irritable (versus calm) behaviors (see Figure 2A). As shown in Figure 2B, these variables also interacted to influence participants’ patient assessments. As SUS increased, participants judged patients displaying irritable (versus calm) behaviors to be less likeable (p<0.01), as poorer historians (p<0.01), and less likely to adhere to treatment (p<0.01). (See Table S4 in Supplement for pairwise comparisons for interactions.) We found no significant effects of SUS on clinical decisions, behaviors, or diagnosis (all ps>0.30).

Figure 2.

Physicians’ Emotional Reactions to Patients (Panel A) and Patient Assessments (Panel B) as a function of Stress from Uncertainty (SUS) and Patient Behavior Condition

Role of Mental Illness History.

Finally, we explored whether effects of patient behavior and SUS would vary depending on whether patients had a history of mental illness. Although we found no significant three-way interactions on any outcome variables (all ps>0.60), we found two-way interactions between mental illness history and SUS. Participants low in SUS perceived they had less case knowledge if a patient had a history (versus no history) of mental illness (M=67.7 v. 75.7, p=0.04) and expressed lower confidence in their final diagnosis (M=66.7 v. 76.5, p=0.01). Participants high in SUS reported low knowledge and confidence regardless of patient mental illness histories. (See Figures S6 and S7 in Supplement for graphs). A main effect of mental illness history also revealed physicians expected patients with a history of mental illness would be less likely to adhere to treatment (p=0.04).

Additional Results

Clinical Decision, Behavior, and Diagnosis.

Participants perceived the cases to be challenging across experimental conditions (M=60.72; SD=15.15; ps>0.30). A main effect of SUS revealed that participants high (versus low) in SUS perceived the cases to be more challenging (p<0.01).

Overall, participants evaluated the patients as requiring emergency care (M=74.62; SD=10.93), ordered many tests (M=7.97; SD=2.13), were likely to seek consultations (M=75.02; SD=17.94) and admit patients (M=70.57; SD=18.33). Participants included the correct diagnosis in their differential in 37.1% of cases and provided the correct final diagnosis in 53.2% of cases. We found no significant differences across experimental conditions on any of these variables (ps>.10). (See Tables 3 and S2).

Stigmatizing Attitudes.

Participants’ scores on the Attitudes of Healthcare Providers Towards People with Mental Illness scale were low (M=2.71; SD=0.870; median=2.83), revealing that they generally disagreed with statements endorsing stigma. Only 5.2% had a mean score of 4 (slightly agree with statements) or higher, suggesting our sample did not harbor much stigma.

DISCUSSION

Our results contribute to a small body of research investigating the influence of patients’ irritable or challenging behaviors on physicians’ emotions and patient assessments. We found physicians reported experiencing higher levels of negative emotions (anger, anxiety, fatigue), lower levels of positive emotions (empathy, happiness) and decreased engagement during encounters with patients displaying irritable (versus calm) behaviors. Physicians also assessed these patients to be poorer historians, exaggerating their pain, and less cooperative, engaged, likeable, likely to adhere to treatment and likely to return to work, despite identical clinical information. In addition, relatively stable individual differences in affective responses to uncertainty in clinical situations (i.e., SUS) magnified the influence of patients’ irritable behaviors on physicians’ negative emotions and perceptions of these patients, assessing them to be poorer historians, less likely to adhere to treatment, and less likable. These findings provide partial support for our hypotheses.

Evidence for our hypotheses is limited to two of the three sets of outcomes we tested: physicians’ emotions and patient assessments. We found no effects of SUS or patients’ behaviors on physicians’ clinical decisions, behaviors, or diagnoses. Prior research similarly found no relationship between SUS and clinical behaviors (e.g., imaging orders, admissions)[38,39]; however, a recent study reported a positive association between [24] SUS and hospital admission rates[40]. Further, our findings deviate from Mamede, Schmidt and colleagues, who found that descriptions of patients’ disruptive behaviors were associated with diagnostic error among residents[5,6]. There are several possible explanations for our null findings.

First, our clinical cases were challenging. Physicians evaluated all patients as having serious conditions and in need of emergency care, ordered a considerable number of tests, and reported a high likelihood of seeking consultations and admitting patients. Such findings suggest that case difficulty may have led physicians to engage in careful, deliberative, and effortful information processing (i.e., system 2 processing)[41–43], which may have reduced or eliminated effects on clinical decisions, behaviors, and diagnosis. Additionally, compared to other studies of clinical reasoning, which frequently included residents or less experienced physicians[5,6,44], our sample consisted of highly experienced physicians with an average of 14.22 years of practice experience. This too may have contributed to more careful processing given case complexity and a greater ability for physicians to mitigate possible influences of patients’ irritable behaviors on diagnosis and other clinical outcomes. Of note, our finding that physicians provided correct diagnoses in only about half the cases is compatible with system 2 processing. If physicians fail to access or apply appropriate knowledge, errors occur regardless of how information is processed[45–47]. Given our patients presented symptoms in vague and subtle ways, and some had conditions infrequently seen in clinical practice (e.g., malaria), low diagnostic accuracy likely resulted from physicians not retrieving appropriate knowledge. Finally, given we found no effects of patient behaviors or SUS on physicians’ clinical behaviors, the absence of downstream effects on diagnostic accuracy is not surprising.

Importantly, our results demonstrate that physicians’ emotional reactions to patients and their patient assessments were not immune to the effects of patients’ irritable behaviors. Unfortunately, these effects can have significant consequences for patients and physicians, which underscores how low-quality care can manifest in multiple ways beyond reduced diagnostic accuracy. Negative beliefs and expectations about patients can act as self-fulfilling prophecies in which these beliefs and expectations can be subtly communicated to patients and result in patient behaviors that lead to negative outcomes (e.g., nonadherence to treatment, failure to seek follow-up care)[48–50]. Even subtle shifts in physicians’ tone, body gestures, or responsiveness can communicate negative attitudes, adversely affect the quality of interpersonal communication, and reduce patient comfort and trust[51]. Effective communication and trust are critical in physician-patient relationships and breakdowns can increase risks to patient healthcare quality and safety[18].

Our finding that negative emotional responses were magnified for physicians predisposed to experience anxiety and concern about bad outcomes in response to uncertainty (high SUS) has important implications for physician well-being and patient care. SUS is positively associated with burnout[26,27], and other measures of uncertainty intolerance are linked to use of maladaptive emotion regulation strategies including suppression[52]. ED providers in a qualitative study reported frequent suppression or disengagement from their emotions[13]. These strategies are among the least effective ways to regulate emotions[53,54], can negatively influence cognitive functioning[55,56], result in emotional rebound [57,58], and are associated with burnout[55], increased cardiovascular activation[59], chronic disease[60], and mortality[61]. Training to improve emotional intelligence may promote more effective regulation strategies [11,62,63] and buffer against burnout. Importantly, many sources of negative emotion result from systemic challenges that permeate the US healthcare system[13,14]. Thus, system-wide interventions are likely to have notable impacts on physicians’ emotions, stress, and burnout. Given that patient care quality and safety are compromised by physician burnout[64,65], efforts to reduce burnout should benefit patients.

In testing our hypotheses, we explored whether the relationships between SUS and patient behaviors varied by patient mental illness history but found no evidence. Although providers frequently hold stigmatizing attitudes toward individuals with mental illness[14,28], our sample reported low stigma. This likely accounts for why patients elicited similar emotions regardless of mental illness history. Our manipulation was also subtle and the information provided (e.g., medications) may have led physicians to assume these conditions were being appropriately managed, which may have reduced uncertainty that may accompany patient mental illnesses. Finally, given patients had symptoms of serious physical illnesses, their mental illness histories may have had little clinical relevance. Research is needed to further explore influences of patients’ mental illness histories.

Future research should also investigate relationships between uncertainty-related measures other than the SUS, and physician’s emotions, clinical reasoning, and behavior. The SUS assesses the affective component of a more complex multidimensional construct: uncertainty (in)tolerance[19–22]. This broader construct encompasses affective, cognitive, and behavioral responses to uncertainty. A recent scoping review examining the relationship between uncertainty intolerance and clinical reasoning identified a few studies that linked uncertainty intolerance to increased investigative behaviors (e.g., clinical testing, referrals) and increased prescribing behaviors (i.e., antibiotics)[21]; however research is scarce and findings are mixed[20,23,66,67]. Thus, our failure to find effects of SUS on clinical decisions, behavior, and diagnosis does not preclude the possibility that other measures related to uncertainty (in)tolerance may predict these outcomes.

Our results highlight the long overdue need for cultural shifts in medicine that normalize experiences of emotions and uncertainty. In 1978, Groves warned of the challenges of caring for “hateful patients”[7], Croskerry drew attention to affective issues in patient safety in 2008[68,69], and only a decade ago, Simpkin and Schwartzstein proposed a need for “revolutionary change in medicine's cultural attitude and approach to uncertainty” (p. 1714)[70]. Despite research progress, much work is needed to achieve change. Fostering frank and open discussions about emotions, uncertainty, and how to best manage these experiences may contribute to an important cultural shift[71]. Further, diagnosis should be actively reframed as a dynamic, evolving, and iterative process that takes place in inherently uncertain medical environments[70]. This should shift trainees’ and physicians’ expectations regarding care and, in turn, increase tolerance of uncertainty and reduce negative emotions. Institutions and professional societies can also better support physicians by offering targeted training to improve clinicians' ability to manage these experiences[23].

Strengths and Limitations

Our study employed a novel randomized experiment in which physicians evaluated patients in interactive simulated encounters. This allowed us to control many variables that are confounded and difficult to study in real clinical encounters. We also collected a considerable amount of data throughout encounters, which deviates from many vignette studies that only provide brief written information[5,6,44,72]. This enabled us to investigate a broad range of variables throughout the diagnostic process, and move beyond investigations of specific outcomes (e.g., utilization of imaging)[73] common in many studies.

Despite these strengths, our use of simulated patients precluded us from capturing the full complexity and variability of real-world clinical encounters, which may limit generalizability of our findings. In the ED, emotions may be experienced more intensely than in our study, and predispositions to experience stress from uncertainty may be more consequential. Our methods also allowed participants unlimited time to assess patients without interruptions, which may have optimized conditions compared to those in real-world encounters. Thus, our results may underestimate the influence of patients’ irritable behaviors and physicians’ stress from uncertainty in actual clinical practice.

Our development and use of novel measures was rigorous; however, these measures have not been validated, which is a limitation. Additionally, our self-report measure may be subject to social desirability biases. Our use of White standardized patients may also limit generalizability. Finally, although our sample included physicians from throughout the United States, we are unable to determine if findings would generalize to other geographic locations.

Conclusion

Physicians’ emotional reactions to and assessments of patients displaying irritable behaviors can pose threats to patient healthcare quality and physician well-being, with greater risks among physicians predisposed to experience high stress from uncertainty. Interventions that target medical education, training, and culture are needed, along with broader changes to the healthcare system that will ease system-related factors (e.g., overcrowding) that can magnify effects of emotions, stress, and uncertainty. Ultimately, a multi-faceted approach is needed; however, interventions enacted on a smaller scale should yield measurable effects on healthcare quality and the well-being of clinicians – many of whom are experiencing burnout[65,74].

What is already known on this topic

• Emotions and uncertainty are ubiquitous in medicine; however, little is known about how patients’ irritable behaviors influence physicians’ emotions and patient assessments, or how physicians’ predispositions to experience anxiety and concern about bad clinical outcomes (i.e., stress from uncertainty) may play in a role in shaping these influences. Yet, both have implications for patient healthcare quality and physician well-being. This is particularly important in emergency departments where encounters with patients displaying challenging behaviors have increased in recent years.

What this study adds

• Using a novel research method (i.e., highly controlled interactive experiment with computer simulated patients), this study is the first to demonstrate that patients displaying irritable behaviors adversely influence physicians’ emotions and patient assessments, and that physicians predisposed to experience high stress from uncertainty are particularly vulnerable to these effects.

How this study might affect research, practice or policy

• • High quality healthcare and physician well-being are priorities for patients and physicians but are threatened by common emotional factors that are amendable to change. The findings support changes to medical education and training, along with broader cultural shifts in medicine, to help clinicians better manage emotional responses and reactions to uncertainty, improve patient healthcare quality, and support clinician well-being.