Effect of Socioeconomic Status Bias on Medical Student–Patient Interactions Using an Emergency Medicine Simulation

Abstract

Objectives

Implicit bias in clinical decision making has been shown to contribute to healthcare disparities and results in negative patient outcomes. Our objective was to develop a high‐fidelity simulation model for assessing the effect of socioeconomic status (SES) on medical student (MS) patient care.

Methods

Teams of MSs were randomly assigned to participate in a high‐fidelity simulation of acute coronary syndrome. Cases were identical with the exception of patient SES, which alternated between a low‐SES homeless man and a high‐SES executive. Students were blinded to study objectives. Cases were recorded and scored by blinded independent raters using 24 dichotomous items in the following domains: 13 communication, six information gathering, and five clinical care. In addition, quantitative data were obtained on the number of times students performed the following patient actions: acknowledged patient by name, asked about pain, generally conversed, and touching the patient. Fisher's exact test was used to test for differences between dichotomous items. For continuous measures, group differences were tested using a mixed‐effects model with a random effect for case to account for multiple observations per case.

Results

Fifty‐eight teams participated in an equal number of high‐ and low‐SES cases. MSs asked about pain control more often (p = 0.04) in patients of high SES. MSs touched the low‐SES patient more frequently (p = 0.01). There were no statistically significant differences in clinical care or information gathering measures.

Conclusions

This study demonstrates more attention to pain control in patients with higher SES as well as a trend toward better communication. Despite the differences in interpersonal behavior, quantifiable differences in clinical care were not seen. These results may be limited by sample size, and larger cohorts will be required to identify the factors that contribute to SES bias.

Healthcare provider bias negatively affects patient outcomes1 and contributes to poor physician–patient interactions.2 The Institute of Medicine 2003 report “Unequal Treatment” suggests the need for research on how patients’ race, ethnicity, sex, and social class may influence the decision making of healthcare providers.3 While prior research on bias has concentrated on race, the impact of socioeconomic status (SES) bias on healthcare delivery has not been studied.4, 5

SES is an important variable with potential influence on physicians’ perceptions of and attitude toward patients, impacting both patient communication and quality of care.6, 7, 8 Previous studies have demonstrated implicit9 and explicit10 bias by medical students (MSs) against low socioeconomic classes. However, these studies have not examined behavioral manifestations of such biases in the patient care setting.

There are multiple subtle behaviors associated with bias that may be observed during patient interactions (e.g., maintaining a distance or avoiding eye contact.)11, 12, 13, 14 Although these biases are largely outside of personal awareness, patients, especially minority patients, are adept at perceiving such biases.15

Medical schools have recently started to address healthcare disparities in their curriculum.16, 17, 18 These curricula focus on exposure to different patient populations in the community and assess the success of the curriculum based on student attitudes toward different patient populations. There is an assumption that exposure and improved attitude will lead to improved patient outcomes and decreased health disparities. However, while attitudes are being measured, the impact on clinical outcomes has yet to be studied.

Using a simulation laboratory, we tested the hypothesis that SES would change the behaviors of MSs during their patient interaction. Emphasis was placed on subtle behaviors that indicate the presence of implicit bias. In addition, the impact of patient SES on quality of care was assessed by comparing patient management across both SES cases.

Methods

All aspects of this study were reviewed and approved by the institutional review board. The study was exempt, and written informed consent for participants was waived. We used a prospective observational study design to assess care outcomes associated with socioeconomic bias. This study was conducted in the simulation center at the authors’ institution using MSs during the months of June to November 2013 and January to April 2014.

Participants

Study participants were fourth‐year MSs in their emergency medicine (EM) block, which is a required senior clerkship at the authors’ institution that includes a mandatory 2‐hour simulation session. MSs participating in the simulation sessions during the months of the study were automatically enrolled. Two MSs participated in each simulated patient encounter. In addition, a nursing student also participated in each encounter, which is standard practice during MS simulation.

Procedure

Student teams participated in a high‐fidelity, mannequin‐based simulation of a patient with an acute myocardial infarction. Cases were identical with the exception of patient SES. Case A featured a high‐SES patient and Case B featured a low‐SES patient. SES was simulated using visual and location‐based cues. The high‐SES simulated patient was a mannequin dressed in a clean button down shirt, dress pants, and a tie and reported a history of having chest pain while walking to his office building. The low‐SES simulated patient was the same mannequin dressed in a dirt‐covered t‐shirt, dirt‐covered pants, and flip‐flops. This patient reported a history of having chest pain while walking to a homeless shelter. Other aspects of the cases were identical including a history of hypertension and noncompliance with medication. For all cases the mannequin was voiced by the same simulation technician who was trained to maintain case standardization.

Each month, six teams of learners completed the simulation; three teams completed the high‐SES case, and three teams, the low‐SES case. The order of the cases was alternated monthly. During each simulation session, individual teams participated in only one of three possible cases, the study case and two nonstudy cases. Student teams were randomly assigned to one of the three cases, resulting in a random sample of one‐third of all eligible students participating in the study. Students were aware that they were being observed as part of the standard simulation experience and that a study was taking place, but were kept blind to the nature of the study.

Data Collection

All simulations were recorded using Simcapture (B‐line Medical). Six blinded reviewers (two EM residents and four nursing students who had not participated in the simulation) independently viewed the videos and recorded responses from the data collection sheet using REDCap (Research Electronic Data Capture), an electronic data capture tool hosted at the authors’ institution.19 REDCap is a secure, Web‐based application designed to support data capture for research studies, providing: 1) an intuitive interface for validated data entry, 2) audit trails for tracking data manipulation and export procedures, 3) automated export procedures for seamless data downloads to common statistical packages, and 4) procedures for importing data from external sources. Each simulation was reviewed by two of the six reviewers. Each reviewer was assigned to either the high‐SES cases or the low‐SES cases to keep them blind to the nature of the study. Prior to beginning the study, investigators determined that for the study cases, an action would be considered completed if either reviewer scored an action as having been done. As a measure of inter‐rater reliability, all six reviewers also recorded responses from six additional nonstudy simulations. These cases represented a standard chest pain case.

Reviewers used a two‐part data collection instrument designed by the study investigators. First, 24 dichotomous items were scored in the following domains: 13 communication, six information gathering, and five clinical care. Second, data were collected on the following five observable MS behaviors: how often the patient was acknowledged by name, how many times the patient was asked about pain, how many times the patient was physically touched, the total amount of time the student spent conversing with the patient, and the percentage of time that the student faced the patient. The data collection instrument was designed prior to the start of the study and items were chosen based on elements considered best practices by a team of EM physicians.

Data Analysis

To evaluate inter‐rater reliability, all six reviewers reviewed six additional standard chest pain cases. Scores were compared and, in general, the agreement was high for all dichotomous items. Cases of disagreement were generally one reviewer scoring an item as “no” that the majority had scored as “yes.” This validated our decision to mark an item as completed for disagreements that occurred. Fisher's exact test was used to test for differences for dichotomous items. For continuous measures, group differences were tested using a mixed‐effects model with a random effect for case to account for multiple observations per case. To adjust for the multiple tests, we adjusted p‐values for multiple comparisons. Specifically, all p‐values within each domain were adjusted using Bonferonni's step‐up method of Hochberg's.20 Inter‐rater reliability was calculated using percent agreement and kappa. The software program used for analysis was SAS v9.4.

Results

A total of 58 teams (116 MSs) participated in simulation cases. Twenty‐nine teams had a low‐SES patient and 29 teams had a high‐SES patient. Inter‐rater agreement for the individual simulation scores were determined by calculating percent agreement (Table 1). For communication and interpersonal skills, there was substantial agreement of at least 87% on the following three items: acknowledging the patient by name, introducing themselves, and explaining their role. The remaining communication items exhibited moderate agreement, with the percent agreement ranging from 55% to 72%. All data gathering items except past medical history exhibited percent agreement ranging from 82% to 98%. All items from patient management exhibited percent agreement ranging from 88% to 100%.

Table 1.

Inter‐rater Reliability Scores for Assessment of MS Interactions During Simulation Cases

	Kappa (95% CI)	% Agreement
Communication/interpersonal skills
Introduce themselves	0.663 (0.436 to 0.891)	87.9
Explain their role	0.394 (0.030 to 0.758)	87.9
Acknowledge patient by name	0.545 (0.088 to 1.000)	94.8
Tell about additional workup	0.145 (–0.106 to 0.396)	55.2
EKG would be obtained	0.176 (–0.074 to 0.427)	58.6
Would be given aspirin	0.128 (–0.127 to 0.382)	56.9
Would be given additional pain meds	0.280 (0.030 to 0.531)	65.5
Cardiology would be consulted	0.387 (0.149 to 0.625)	72.4
Told patient is having a heart attack	0.121 (–0.068 to 0.310)	62.1
Catheterization would be done	0.263 (0.028 to 0.498)	69.0
Told what catheterization involves	0.382 (0.146 to 0.617)	69.0
Did pain meds help/reassess pain	0.394 (0.152 to 0.636)	75.9
Did student thank patient	0.000 (0.000 to 0.000)	98.3
Data gathering
Past medical history	0.196 (–0.064 to 0.456)	75.9
Medication use	0.483 (0.217 to 0.749)	82.6
Allergies	0.590 (0.355 to 0.826)	84.5
Cocaine use	0.617 (0.312 to 0.923)	89.8
Smoking history	0.946 (0.841 to 1.000)	98.3
History of alcohol use	0.815 (0.567 to 1.000)	96.6
Patient management
Order aspirin	0.000 (0.000 to 0.000)	96.6
Order additional pain meds	1.000a	100.0
Order repeat EKG	0.659 (0.036 to 1.000)	98.3
Consult cardiology	0.413 (0.015 to 0.811)	91.4
Send patient for cardiac catheterization	0.531 (0.244 to 0.819)	87.9

EKG = electrocardiogram; MS = medical student.

^aUnable to calculate a CI when the kappa is 1.000

Table 2 presents our main findings. We observed no difference for the majority (26/29) of items. There were no statistically significant differences for any of the items in the communication, information gathering, and clinical care domains. The three variables that showed statistical significance prior to adjustment were items from the MS behaviors domains. The three items were the number of times MSs acknowledged the patient by name, asked about pain medication, and physically touched the patient. After adjustment for multiple comparisons, the data demonstrate that MSs were more likely to ask the simulated patient with high SES about pain control (p = 0.04) and more likely to touch the low‐SES patient (p = 0.01).

Table 2.

Assessment of MS Interactions During Simulation Cases

	Low SES (n = 29)	High SES (n = 29)	p‐value	Adjusted p‐valuea
Communication/interpersonal skills, % (n)
Introduce themselves	31.0 (9)	27.6 (8)	1.000	1.000
Explain their role	17.2 (5)	17.2 (5)	1.000	1.000
Acknowledge patient by name	93.1 (27)	100.0 (29)	0.491	1.000
Tell about additional workup	72.4 (21)	89.7 (26)	0.180	1.000
EKG would be obtained	62.1 (18)	79.3 (23)	0.248	1.000
Would be given aspirin	58.6 (17)	72.4 (21)	0.408	1.000
Would be given additional pain meds	69.0 (20)	86.2 (25)	0.207	1.000
Cardiology would be consulted	72.4 (21)	89.7 (26)	0.179	1.000
Told patient is having a heart attack	89.7 (26)	96.6 (28)	0.612	1.000
Catheterization would be done	86.2 (25)	89.7 (26)	1.000	1.000
Told what catheterization involves	62.1 (18)	72.4 (21)	0.577	1.000
Did pain meds help/reassess pain	79.3 (23)	93.1 (27)	0.253	1.000
Did student thank patient	3.4 (1)	0.0 (0)	1.000	1.000
Data gathering, % (n)
Past medical history	93.1 (27)	96.6 (28)	1.000	1.000
Medication use	79.3 (23)	96.6 (28)	0.102	0.612
Allergies	27.6 (8)	37.9 (11)	0.577	1.000
Cocaine use	17.2 (5)	17.2 (5)	1.000	1.000
Smoking history	17.2 (5)	24.1 (7)	0.747	1.000
History of alcohol use	6.9 (2)	17.2 (5)	0.423	1.000
Patient management, % (n)
Order aspirin	100.0 (29)	100.0 (29)	1.000	1.000
Order additional pain meds	100.0 (29)	100.0 (29)	1.000	1.000
Order repeat EKG	96.6 (28)	100.0 (29)	1.000	1.000
Consult cardiology	96.6 (28)	96.6 (28)	1.000	1.000
Sent patient for cardiac catheterization	93.1 (27)	89.7 (26)	1.000	1.000
MS behaviors, mean (±SD)
No. of times acknowledged patient by name	3.9 (2.6)	5.2 (2.7)	0.036	0.108
No. of times asked about pain	3.6 (2.0)	4.7 (2.3)	0.010	0.040
No. of minutes directly communicating	4.8 (1.3)	5.1 (1.9)	0.471	0.734
No. of times physically touch patient	6.9 (4.3)	3.6 (2.8)	0.002	0.010
Percentage of time facing patient	78.7 (11.2)	77.5 (22.4)	0.734	0.734

EKG = electrocardiogram; MS = medical student; SES = socioeconomic status.

^aAdjusted p‐values calculated using Bonferonni step‐up method of Hochberg.20

Discussion

Implicit bias can have a wide‐reaching effect across many areas of medicine, with potential detrimental effect on both patients and medical care providers. In a simulation environment, the authors were able to investigate what effect a single variable, namely, SES, had on both patient care, communication, and healthcare delivery. We found that MSs were more likely to touch the low‐SES patient model and less likely to ask about pain control. One explanation for increased physical touching is that the student was attempting to display increased compassion toward the low‐SES patient. However, research has suggested that this form of contact can also be perceived as a display of power.21, 22 Despite these differences in interpersonal interactions, the data gathering and patient management was similar for both the low‐ and the high‐SES patient. We used rigorous methods to assess these differences including interobserver testing and statistical adjustment for multiple comparisons. These data are the first to quantify SES bias and manifestation in an experimental setting.

Limitations

Before application of these results, limitations must be considered. The Hawthorne Effect slightly limits this study. Learners were aware that the encounters were being observed and recorded, but were unaware of what was being studied. Additionally, the learners were familiar with the simulation center and the process of simulation. According to recent literature, the effect of participant reactivity was possibly minimal.23 Next, while high‐fidelity simulation provides realism, there are some limitations as the mannequin lacks the facial expressions and gestures of an actual patient. In addition, the providers worked as a team and not individually, so that the bias of one team member may affect the bias of another. Finally, the study team did not control for learners’ prior training and exposure to implicit bias principles. However, the MSs are almost all from the same medical school with similar medical education curriculum. Because of this, we expected their baseline training to also be similar. In addition, case assignments were random, which potentially negates some of this effect.

While many were not statistically significant, for each of eight data points regarding explaining the diagnosis and care plan to the patients (telling about additional workup, that an electrocardiogram would be obtained, that aspirin would be given, that additional pain medications would be given, that cardiology would be consulted, that the diagnosis was a heart attack, that a catheterization would be done, and explaining what a catheterization is), the MSs communicated with the high‐SES patient a larger percentage of the time than with the low‐SES patient. This potentially could represent implicit bias that leads providers to involve higher‐SES patients in medical decision making more than low‐SES patients. Unfortunately, because this pilot study was run over the course of only one academic year, we did not have enough student‐simulated patient encounters to determine if this was by chance. We are currently conducting additional studies to investigate this possibility.

Conclusion

This study indicates there may be some implicit bias within healthcare providers interacting with low‐socioeconomic‐status patients. There is promise that despite biased interpersonal treatment, patients received the appropriate medical care for the presenting complaint (i.e., patient received aspirin and was sent for a cardiac catheterization) in both groups. More investigation is needed to elicit how this occurred or whether the providers accommodated for their biases or performed well despite these biases. Future studies focused on debriefing these cases with specific questions related to both learner bias and patient care decision making may provide insight. Future work could assess bias across multiple healthcare professions to better understand this topic. These research opportunities will help develop new strategies and education methods. Improved education of subtle behavioral and implicit bias has the potential to benefit the healthcare community, providers, and patients.

The authors thank Rachel M. Davis, Katherine E. Holle, Emily A. McKenzie, and Natalie S. Reed for their time spent reviewing and scoring the taped simulation encounters. The authors also thank Dr. Jeffrey Kline for his mentoring and assistance in editing the manuscript.

AEM Education and Training 2017;1:126–131