Investigating Healthcare Provider Bias Toward Patients Who Use Drugs Using a Survey-Based Implicit Association Test: Pilot Study

Abstract

Objectives:

Negative bias against people who use illicit drugs adversely affects the care that they receive throughout the hospital. We hypothesized that emergency providers would display stronger negative bias toward these patients due to life-threatening contexts in which they treat this population. We also hypothesized that negative implicit bias would be associated with negative explicit bias.

Methods:

Faculty, nurses, and trainees at a midwestern tertiary care academic hospital were invited (6/26/2019–9/5/2019) to complete an online implicit association test (IAT) and explicit bias survey.

Results:

Mean IAT results did not vary across demographics (n=79). There were significant differences in explicit bias scores between departments regarding whether patients who use drugs deserve quality healthcare access (p=0.017). We saw no significant associations between implicit and explicit bias scores.

Conclusion:

Though limited by sample size, the results indicate that emergency and obstetrics/gynecology providers display more negative explicit bias toward this patient population than other providers.

Introduction

Illicit substance use remains a controversial and complicated healthcare issue in the United States (U.S.). In 2018, the Centers for Disease Control and Prevention (CDC) reported that approximately 18% of the U.S. population (48.5 million people) aged 12 years and older used illicit drugs or misused prescription drugs within the previous year.¹ In the same report, it was estimated that 547,543 Emergency Department (ED) visits occurred in 2015 for all drug-related poisonings in the U.S., not including other drug-related ED visits such as those from accidental trauma.¹

People who use illicit drugs (PWUD) and people who inject illicit drugs (PWID) are broadly perceived in society and in some health environments as dangerous, unreliable, or untrustworthy.^2–4 Several studies have shown that this stigma may negatively impact healthcare providers’ care of patients with known history of drug use. Qualitative studies by Biancarelli and Paquette^3,5 described recurrent themes by PWID such as being denied medical services, having visits cut short, having their symptoms downplayed, or being blamed for their own health problems after disclosing drug use. Providers with negative bias toward PWUD/PWID may be more likely to misinterpret signs and symptoms as being from drug use when due to a different cause, resulting in misdiagnosis and inappropriate treatment.^6,7 Other studies have shown that providers who perceive that PWUD can control their drug use are more likely to believe that these patients are at fault for poor health.^6,8

PWUD have reported delaying healthcare visits to avoid facing stigma, which results in presenting with advanced disease and use of a higher proportion of emergency medical services.^3,6,9,10 PWUD are vulnerable to nutritional, mental health, immune, and infectious disorders,¹¹ and so there is a critical need for their accessibility to quality healthcare.

Implicit bias refers to unconscious bias, of which we may be unaware yet may still influence our behaviors, compared to explicit bias, of which we are aware and consciously act on. Implicit bias is believed to have more influence on subtle and nonverbal behaviors, whereas explicit bias more broadly influences our deliberate actions toward others. The implicit association test (IAT) was initially developed to measure individual implicit associations toward groups with specific traits,¹² especially populations who are more frequently confronted with prejudice and discrimination, such as people of color; women; or gays and lesbians.^13–16 The structure of the IAT is based on analysis of both the speed at which the test-taker chooses between words or images in association with a characteristic of a population group and with the negative or positive connotation of that word with the group, compared to another group.^12,17 An individual test result is considered invalid if the responses are consistently delayed or have a high error rate.

Within the wide breadth of literature regarding stigma toward PWUD, the IAT have been used as a methodology to investigate implicit associations toward PWUD as one among other latency measures, such as: to evaluate how PWUD identify self-identity and self-schema in order to better understand the seemingly paradoxical use of substances known to be self-harming^18,19; to investigate implicit associations with respect to the intersectionality of race, gender, and PWID²⁰; and to explore how healthcare providers perceive PWUD.^21,22 Variations of the IAT were developed to evaluate how word choices or labels can trigger stigma, for example, finding that both PWUD in recovery and health professionals displayed higher levels of negative bias toward words like “addict” and “substance abuser” compared to PWUD not in recovery and non-health professionals.^23,24

The goal of our pilot study was to investigate implicit and explicit healthcare provider bias toward PWUD at our institution using an online IAT with an accompanying explicit bias survey. We hypothesized that emergency medicine (EM) providers would display higher levels of negative implicit and explicit bias toward PWUD compared to providers in other departments due to the circumstances that may cause PWUD to present to the ED (e.g. overdose, acute intoxication, trauma). We also hypothesized that IAT results would predict explicit bias survey scores.

Methods

Study Design, Setting, and Recruitment

We administered a one-time anonymous online survey to obtain information about implicit and explicit beliefs by current healthcare providers at a single, midwestern, academic tertiary care referral center. Faculty, nurses, and residents in 13 departments were invited by e-mail with a link to a secure, web-based survey between 6/26/2019 and 9/5/2019. Participants who completed an online consent form and affirmed that they were a healthcare provider currently working at our institution were allowed to continue to the study, which was offered via a Qualtrics® (Provo, UT) platform. Participants completed the IAT first, followed by a 14-question survey on explicit beliefs about PWUD and five demographic questions. This study was deemed exempt by our Institutional Review Board.

Implicit Association Test

The website iatgen.org was created and validated to offer user-friendly public access software to enable researchers to develop and distribute IATs through an online survey platform, such as Qualtrics®, without having to create one’s own code.²⁵ We were able to use this tool to easily develop a novel IAT to investigate healthcare providers’ implicit beliefs toward PWUD using previously validated methods.^12,25

Participants completed five blocks each containing 60 trials, in which they were presented with on-screen verbal stimuli. Test-takers were instructed to match each stimulus to targets (“drug user” or “non-user”) or categories (“good words” or “bad words”) as fast as possible without making errors. The list of stimuli included 20 “good words” (e.g. “honest”), 20 “bad words” (e.g. “dishonest”), 5 short phrases for drug user (e.g. “injects heroin”), and 5 short phrases for non-user (e.g. “never tried heroin”). A summary measure (D-score) for each participant was generated using Iatgen software (http://iatgen.org).²⁵ A D-score may range from −2 to +2, and we defined a positive D-score to indicate negative implicit association toward Target A (“drug users”), whereas a negative D-score would indicate negative implicit association toward Target B (“non-users”). A D-score of zero indicated no implicit association.

Explicit Bias Survey

The explicit bias survey consisted of 14 statements, of which nine were adapted from a validated survey in Brener.⁸ We also created and validated new survey questions to investigate providers’ explicit bias regarding whether PWUD should have access to quality healthcare. Overall, these statements could be categorized as “social attitudes toward drug users” (Attitudes) and whether drug users were perceived to have control of their own drug use (PCDU). We created another category containing five statements regarding beliefs on whether PWUD deserve to receive high-quality healthcare (PWUDDHC). Participants selected their level of agreement or disagreement with each statement using a 5-point Likert Scale, ranging from ‘strongly disagree’ to ‘strongly agree’. A score of 1 represents the least negative and 5 represents the most negative explicit bias. The explicit bias survey was not timed.

Other Survey Components

Participants answered demographic questions about their age, gender, department, position, and estimated degree of weekly contact with patients with known substance use disorder (SUD). The departments were categorized as Emergency Medicine, Family Medicine, Internal Medicine, Obstetrics/Gynecology, Psychiatry, or Other (including Anesthesia, Cardiothoracic Surgery, General Surgery, Hospital Dentistry, Oral Surgery, Orthopedic Surgery, Otolaryngology, and Urology). The positions were attending/staff physician, nurse, resident/medical student trainee, or other. Degree of contact referred to whether provider typically saw 0, 1–10, or more than 10 patients with SUD weekly.

Statistical Analysis

Based on a power calculation using results from the study in Brener 2007²⁶, we estimated an overall sample size of 70 participants to detect a mean D-score difference of 0.26 with 80% power. Results from the IAT, explicit bias survey, and demographic questions were linked to each participant and anonymized by individual identification numbers generated by iatgen.com code.²⁵ Scores were compared by demographic characteristics using bivariate analyses and Mann-Whitney U-tests. Associations between explicit and implicit bias scores were measured through linear regression. As we compared scores across several categories such as between department differences, we utilized Tukey multiple comparison tests. As a marker for reliability of the new explicit bias questions, we additionally analyzed whether the components of the PWUD and healthcare access questions we generated for this study were consistent with the two categories of components from previously validated questions from Brener 2008 by measuring Cronbach’s alpha.⁸

Results

Survey Respondent Characteristics

Invitations to participate were sent to 1,836 different email addresses at our institution. Of these, 108 people attempted the survey and 79 (4.3%) respondents were eligible after applying exclusion criteria and removing invalid tests (Figure 1). Of the 79 respondents who completed the study with valid IAT results, 57 (72%) were female (Table 1). Approximately 19% and 17% of respondents were from the Departments of Psychiatry and Emergency Medicine, respectively, and 38% of all respondents were staff/attending physicians.

Figure 1.

Flow Chart of Recruitment and Analysis Sample

Table 1.

Characteristics of Respondents who Completed the Implicit Association Test and Explicit Bias Survey

Sample Characteristics	Total		D-Score1		Explicit Bias Score - All Questions2		Explicit Bias Score - Attitudes2		Explicit Bias Score - PCDU2		Explicit Bias Score - PWUDDHC2
	N	(%)	Mean	p-value	Mean	p-value	Mean	p-value	Mean	p-value	Mean	p-value
Gender3
Female	57	(72.2)	0.46	0.960	2.08	0.488	2.16	0.959	2.24	0.294	1.95	0.773
Male	21	(26.6)	0.47		1.99		2.15		2.02		1.90
Department
Emergency Medicine	13	(16.5)	0.31	0.159	2.34	0.049	2.43	0.159	2.33	0.389	2.38	0.017
Psychiatry	15	(19.0)	0.49		1.80		1.85		2.02		1.65
Family Medicine	12	(15.2)	0.56		2.01		2.13		2.00		1.92
Internal Medicine	7	(8.9)	0.54		1.79		2.11		1.86		1.46
Obstetrics/Gynecology	5	(6.3)	0.76		2.44		2.56		2.60		2.36
Other	26	(32.9)	0.40		2.05		2.14		2.20		1.90
Position
Physician - Staff/Attending	30	(38.0)	0.45	0.241	2.00	0.529	2.13	0.66	2.13	0.331	1.84	0.291
Medical Training - Student/Resident	20	(25.3)	0.37		1.98		2.1		1.92		1.95
Nurse	11	(13.9)	0.64		2.31		2.38		2.45		2.25
Other	17	(21.5)	0.45		2.03		2.13		2.28		1.84
Number of Substance Patients Seen Weekly with SUD
0	9	(11.4)	0.49	0.221	2.19	0.514	2.31	0.691	2.47	0.278	1.91	0.926
1–10	48	(60.8)	0.49		2.03		2.16		2.07		1.93
>10	22	(27.8)	0.38		2.05		2.09		2.26		1.95

¹The D-score ranged from −2 to +2, and we defined a positive score to indicate negative implicit bias toward Target A (“drug users”), whereas a negative score indicated negative implicit bias toward Target B (“non-users”). A D-score of zero indicated no implicit bias.

²Explicit bias scores are based on a 5-point Likert scale, where 5 indicates the highest negative bias. Attitudes=attitudes toward PWUD; PCDU=perceived controllability over drug use; PWUDDHC=do persons who use drugs deserve healthcare.

³One individual identified as non-binary was not included in the univariate comparison due to small sample size.

Implicit Association Test and Explicit Bias Survey Results

Mean D-scores from the IAT did not significantly vary across demographic characteristics (Table 1). The overall mean D-score was +0.46 (SD=0.37, p<0.001), indicating that most participants displayed negative implicit associations toward drug users compared to non-users (Table 2). Internal validity measures and error rates generated from iatgen.org indicated that results from our online format of the IAT were valid.²⁵

Table 2.

Summary Measures of Implicit Association Test and Explicit Bias Scores

Bias Measure	Mean	SD
D-Scorea	0.46	(0.37)
Explicit Bias Score - All Questionsb	2.05	(0.59)
Explicit Bias Score - Attitudes	2.16	(0.64)
Explicit Bias Score - PCDU	2.17	(0.72)
Explicit Bias Score - PWUDDHC	1.93	(0.78)

SD = standard deviation; PCDU = Perceived Control Over Drug Use; PWUDDHC = PWUD Deserve Health Care

^aD-Score definition: implicit association score, measured as the mean difference divided by the overall standard deviation between trial blocks, ranging from −2 to +2.

^bExplicit bias survey scores were based on 5-Point Likert scale in which a score of 1 represents the least negative and 5 represents the most negative explicit bias.

Inclusion of the new category of questions we developed for the explicit bias survey and the two previously validated question categories²¹ yielded good internal consistency (14 items, Cronbach’s alpha =0.87).⁸ Differences in explicit bias survey scores varied across demographics and categories. For example, when considering results on Attitudes and Perceived Control, there were no differences across most demographics or respondent characteristics (Table 1). However, there were differences in overall scores between departments for all explicit bias questions combined (p=0.049) and those pertaining to healthcare access (p=0.017). The Departments of Emergency Medicine and Obstetrics/Gynecology displayed the highest explicit bias scores, which were 2.34 and 2.44 respectively. When adjusting for multiple comparisons between specific departments, there were no significant differences. Linear regression comparing IAT, overall explicit bias, and category-specific explicit bias survey results indicated no significant associations.

Discussion

The results from our study suggest that providers at our institution generally share negative implicit associations toward drug users compared to non-users. We saw no differences in implicit association scores across demographics of gender, job position, department, or degree of contact with PWUD, nor did we see significant associations between IAT and explicit bias survey results.

The results did indicate a higher degree of overall negative explicit bias toward PWUD by emergency and obstetrics/gynecology providers compared to other departments and in the category regarding whether providers believed that PWUD deserved access to quality health care. This is consistent with our hypothesis that providers who work in more emergent environments might display more negative bias toward PWUD. Other studies in literature indicate that PWUD may be at increased risk of not receiving appropriate healthcare after their drug use is known to their providers.^3,5 Based on our study results, these findings may be attributable to forms of implicit or explicit bias.

A wide breadth of literature describes negative explicit bias as normative among obstetrics/gynecological providers toward pregnant mothers who use drugs, with emotions even described as “hatred” by some providers toward this patient population.^27–29 The majority of labor and delivery nurses in one study believed that pregnant mothers who use drugs should be imprisoned or sent to rehabilitation centers and also indicated that labor nurses with more experience with these patients were more likely to have increased negative bias toward them.^27,30 Yet it is recognized that mothers who use drugs are more likely to participate in prenatal care and the health risks of both mother and baby can be reduced with training interventions for providers to help them improve relationships with this patient population.²⁸ Additionally, some providers recognize that they may be the first medical contact that some of these patients may have had in some time, with mothers who are motivated to obtain care to increase their chances of having a healthy baby who might not have sought care only for their own health.^29,31, which is an opportunity to develop long-term relationships with these patients.

We identified less literature regarding whether emergency medical providers are more likely to have negative bias toward PWUD than providers in other specialties, though several studies we found might suggest such bias originates from beliefs that PWUD monopolize resources due to their higher numbers of visits to the ED.^9,10 Interestingly though, a study by Franz et al. found that EM was among a few medical specialties where physicians generally were “more willing” to work with patients with opioid use disorders compared to other specialties.³² Our study’s results suggested the opposite, though our study may not be comparable as it had different objectives and included not only physician participants but also nurses and trainees.

Based on a social psychology theory that having more exposure to a stigmatized “outgroup” would reduce stigma, a study by Brener investigated whether healthcare providers who had greater contact with PWID had less negative bias toward PWID compared to providers with less contact.²⁶ Their results indicated that providers with higher contact with PWID had correspondingly higher positive explicit bias toward PWID (i.e. explicit good associations), but also had higher negative implicit bias toward PWID (i.e. implicit bad associations), compared to providers with less contact. However, the results of our study were inconsistent with this, as we saw no differences in implicit bias across all demographics and only saw negative explicit bias toward PWUD among emergency and obstetrics/gynecology providers. We believe that the differences we saw were less due to degree of PWUD contact and more due to the higher acuity of the context in which these providers saw these patients. This could be further analyzed in future studies using a larger study sample to compare potential differences in explicit bias toward PWUD between ED providers at tertiary care institutions with different patient volumes or by severity of presentation.

Finally, we did not explore how job stress or burnout might affect the results. A study by von Hippel, Brener and von Hippel investigating implicit associations among nurses toward PWID, described that having greater exposure to “challenging behavior” by these patients was significantly associated with job stress.²¹ Therefore, it is possible that the overall implicit associations that we saw were due to compassion fatigue or burnout rather than prejudice per se. On the other hand, implicit bias is also believed to predict behavior more when our defenses are down, such as when we are rushed or fatigued, since these states reduce our capacity to filter and increase our reliance on heuristics.³³ Investigating compassion fatigue versus implicit associations toward PWUD would be an interesting aspect to pursue in future studies.

Limitations

There are several limitations to this pilot study. First, there was an overall low response rate and optional participation. The convenience sampling approach and decision to participate in the study might have yielded biased estimates if those who chose to participate in our study may have been systematically different from those who did not participate. Secondly, we noted that approximately 72% of respondents were women. This may be consistent with previous studies that identified that women are more likely than men to participate in research or survey studies.³⁴ While this may cause concern for generalizability of the results, the gender distribution may also be driven by the healthcare setting; the U.S. Census Bureau estimates that 76% of healthcare employees are women which is consistent with the gender distribution observed in this study.³⁵ We chose to recruit participants via an online method in order to keep the results of the study anonymous, believing that it would promote honest answers to the explicit bias survey questions as well as reduce any potential individual harm if the results were somehow publicly exposed.

We received feedback from three participants who started but did not complete the IAT portion of the online survey. One reported difficulty due to inability to complete the survey on a cell phone, as coding from the iatgen.org application requires the IAT be performed using a laptop or desktop computer. Two participants reported difficulty in understanding how to perform the IAT.

IAT methodology has not been without criticism. There have been many challenges toward its validity, reliability, and/or practical use, and questions whether the IAT correlates to explicit behavior.^36,37 One shortcoming of the IAT is that it requires comparison of preferences between two different categories (i.e. white skin vs. dark skin, female vs. male, drug user vs. non-drug user). It does not by itself reflect empirical like or dislike toward a single category; for example, our study indicated that our participants overall had comparably more negative implicit association toward drug users compared to non-users, but it did not give an objective score indicating the degree or severity of negative association toward drug users when not comparing them with another group. Variants of the IAT such as the SC-IAT and GNAT were both developed as attempts to address this^16,22; however, these IAT variations were not available in public open-access online survey formats at the time of our study.

Using words instead of images also could have affected the IAT results. We chose not to use images to avoid stereotyping what a “drug user” looks like. However, the task of reading a short phrase may have altered participants’ responses from being less automatic compared to responding to visual cues, and an ‘automatic’ response to the IAT is truer to the type of response for which it was designed.^17,38 Yet it was challenging to find language that would not be considered stigmatizing. We recognize that the use of phrases in the IAT such as “injects heroin” or “drug user” may also be problematic, given that each term may also be stigmatizing.^39,40 For example, literature has indicated that PWID face even higher stigma than PWUD who do not inject drugs.^5,40 Likewise, studies have shown that non-person-first labels such as “substance abuser” may trigger more stigma in medical professionals than do person-first labels such as “person with substance use disorder”.^24,39 Given the low overall error rates and high internal validation scores (Supplemental Content – Appendix C), we feel confident that the results of this IAT were valid with respect to comparison across demographics, but possibly the results indicating an overall negative association toward PWUD compared to non-PWUD was due to the inclusion of these potentially stigmatizing phrases. The phrase “drug user” was also present in the portion of questions that we adapted from “injecting drug users” from a previously developed explicit bias survey.⁸ We felt confident using these survey questions given that they were already validated. However, if we perform a study in the future including this survey, we might substitute the wording to “people who use drugs” to reduce the use of stigmatizing language.

Conclusions

Our pilot study used a novel online survey to investigate healthcare provider implicit bias toward PWUD. We developed and validated a new survey category of explicit bias about whether providers believe that PWUD deserve quality healthcare. The results from our study indicate that providers at our institution shared overall negative implicit associations toward drug users, however we saw no differences in implicit association scores across demographics. The results also indicate that there may be higher negative explicit bias toward PWUD among emergency medicine and obstetrics/gynecology providers compared to other providers at our institution; however, these results were limited by sample size. In the future, we plan to develop a qualitative study to investigate sources of bias toward PWUD in the ED through structured interviews.

Table 3.

Summary Measures of Explicit Bias Questions

Survey Question	Meana	Minimum	25th Pctl	Median	75th Pctl	Maximum
Category 1: Social attitudes toward PWUDb
Drug users should be accepted completely into our society.	1.66	1	1	2	2	3
Drug use is immoral.	2.32	1	1	2	3	5
I avoid the company of known drug users whenever possible.	3.42	1	2	4	4	5
Drug users are mistreated in our society.	1.67	1	1	2	2	3
People should feel sympathetic and understanding of drug users.	1.72	1	1	2	2	3
Category 2: Perceived control over drug useb
Drug users are responsible for their addiction.	2.58	1	2	2	4	5
People use drugs to avoid dealing with their own inadequacies.	2.75	1	2	3	4	5
Drug users have weak characters.	1.94	1	1	1	3	5
Drug users can stop using drugs whenever they decide to.	1.41	1	1	1	2	4
Category 3: Do PWUD deserve to receive quality health carec
There is no point in treating patients who use illegal drugs because they don’t care about their health anyway.	1.43	1	1	1	1	5
Patients who are drug users should be offered the same high-quality medical resources as for all other patients.	1.25	1	1	1	2	2
It makes me angry to think of Medicaid paying to treat diseases that were brought on by illicit drug use.	1.91	1	1	1	3	5
It isn’t fair to other patients when drug users monopolize the resources at the emergency department.	2.41	1	1	2	4	5
I am tired of dealing with patients who come into clinic/ED repeatedly for drug related problems.	2.67	1	2	2	4	5

^aExplicit bias survey results are based on a 5-Point Likert scale (“strongly agree”, “agree”, “neutral”, “disagree”, “strongly disagree”) converted to a numerical scale in which 1 represents the least negative and 5 represents the most negative bias.

^bQuestions obtained from Bremer and von Hippel, 2008.

^cNew explicit bias questions developed in this study.