Do Words Matter? Bias in the EHR

Interviewer: Laurie Scudder, DNP, NP; Interviewees: Anna P. Goddu, MSc; Mary Catherine Beach, MD, MPH


April 02, 2018

Implicit bias is a well-recognized contributor to the ongoing dilemma of healthcare disparities. Although most people probably bring their own inherent biases with them when they enter the healthcare workforce, it is also known that they may acquire biases along the way as a result of interactions with other clinicians, including during education and training.

Could the electronic health record (EHR)—specifically, the notes written by clinicians about their patients—be another way in which clinicians may unwittingly transmit bias to each other? To try to answer this question, Anna P. Goddu, MSc, a medical student at Johns Hopkins University, and her colleagues examined how medical students and residents at a large urban academic medical center responded to vignettes written as clinician chart notes: one deliberately designed to be neutral and the other to be stigmatizing, using language drawn from the medical records at that institution.

To discuss the study and its implications,[1] Medscape spoke with Goddu and Mary Catherine Beach, MD, MPH, a professor of medicine in the School of Medicine at Johns Hopkins University and the principal investigator.

Medscape: Could you begin by providing a brief summary of your methodology?

Anna P. Goddu, MSc

Goddu: In an earlier study,[2] a medical student here at Hopkins, Katie O'Conor, who has a background in linguistics, examined language that was found within EHRs and categorized words and phrases that were deemed to be potentially stigmatizing. We built on her work and wrote two chart notes about a hypothetical patient with sickle cell disease presenting to the emergency department in a pain crisis. One was deliberately written to be potentially stigmatizing in language and tone, using all phrases abstracted from medical records at our institution, and the second was written to be medically equivalent and neutral.

We recruited medical students and residents, over 400 altogether, to participate in an online survey where they were randomly assigned to read either the stigmatizing-language or neutral-language chart note. They then answered questions about how they would manage the patient's pain, including such options as ketorolac versus hydromorphone. Participants also completed a previously validated survey measuring their attitudes toward the patient that was described in the chart note.

Our analysis compared the group of residents and students who had read the stigmatizing-language chart note with the group that had read the neutral-language chart note. We assessed whether they had different attitudes toward the patient and managed the patient's pain differently solely on the basis of which chart note they had read.

Medscape: You noted that you used actual language drawn from your institution's EHR. Can you describe some of that language that was deemed to be stigmatizing? Is there an objective way to determine stigma within language? Or is it a case of "you know it when you see it"?

Goddu: The researchers of the study I referred to earlier developed a framework to characterize factors that make language in medical records stigmatizing. Included in that framework were ideas about casting doubt on the patient's pain—for example, documenting that "the patient insists his pain is 'still a 10'" versus a note indicating "the patient reported that he still has 10 out of 10 pain," presented as a fact.

Another potentially stigmatizing use of language is portraying the patient negatively with irrelevant or unnecessary indicators of socioeconomic status. Such statements as "the patient was hanging out with his friends outside of McDonald's" can be potentially stigmatizing.

A third indicator of bias is implying patients' responsibility for their own condition with references to them being uncooperative: for example, documentation that states "the patient is refusing his oxygen mask" versus a statement that "the patient is not tolerating the oxygen mask."

Then, finally, there are small linguistic variations. For example, reporting that the patient is taking a "narcotic" versus "opioid" pain medication is a small word change that can be potentially stigmatizing, because "narcotic" is a term that typically has legal connotations.

I think this framework helps us characterize language that we see commonly in medical records. But I don't think we would ever argue that there's an objective way to measure whether language is stigmatizing or not.

To your point about whether you know it when you see it, many of our participants, when asked to reflect on the chart note after completing the hypothetical case, had really different ideas about such statements as "the patient was hanging outside McDonald's" or that "his girlfriend needs a bus token to go home." Are those kinds of statements empathetic or potentially stigmatizing? There really is a gray area.

Our study is definitely not trying to identify exactly which phrases may be having an impact in transmitting bias from one clinician to another. We're not commenting on whether you should or should not include McDonald's or the word "narcotics." That's not what we set out to do.

Instead, we were attempting a proof-of-concept study to show that the language we use in medical records can affect our attitudes and our decision-making as clinicians. We think that future work should try to identify which dimensions of stigmatizing language are most likely to perpetuate bias and adversely affect patient care.

Medscape: What were your key findings?

Goddu: We found that our physicians-in-training who read the stigmatizing-language chart note had more negative attitudes toward the patient described in the chart note than those who had read the neutral-language note. That was the case with both medical student respondents and resident respondents.

But furthermore, we found that the residents who had read the stigmatizing-language chart note were prescribing pain medication less aggressively than those who read the neutral-language chart note.

I think these findings are demonstrating that language can play a powerful role in influencing clinicians. This suggests that language is directly affecting the attitudes and medication-prescribing behavior of subsequent physicians who read the chart. We think that is a potentially important pathway by which bias can be propagated from one clinician to another.

Medscape: It was particularly interesting that you noted more bias in residents further along in their training, potentially attributing that to a hidden curriculum in medical school and residency that fosters bias. Can you describe what you mean?

Mary Catherine Beach, MD, MPH

Beach: When trainees enter the clinical environment, they are surrounded by other trainees who are 1 to as much as 5 years ahead of them. These physicians are working hard, have had more interactions with patients, and may have become more cynical.

There can be a transient cynicism that occurs during medical residency training, which some have suggested will resolve when people finish their medical training. Medical students and residents subsequently have greater exposure to the clinical setting where the cynicism has developed that can become contagious, because it is part of the culture.

When physicians get out of that environment, their overriding values take over, and they are found to be more compassionate and less cynical on objective measures.

Medscape: That is interesting. Is it a factor of the intensity of training and just being tired? We're all a little more cynical and cranky when stressed out and tired.

Beach: I definitely think fatigue, exhaustion, work, and not having great balance in your own life can make you a more cynical person. As that becomes part of the culture and socially acceptable, it then creates the hidden curriculum.

Medscape: An interesting finding was about the use of quotations in documenting things that a patient or family actually said. What are the pros and cons of this technique?

Goddu: It is a tension that we found really interesting and worth further study. There's no doubt that quotation marks can be a great way to bring a patient's voice into the medical record in a way that can inspire empathy and bring really important clinical information. For example, noting that a patient says he is experiencing pain "like an elephant is sitting on my chest" provides a really vivid description that is clinically useful.

On the other hand, quotes can be used to signify a patient's socioeconomic status by showing what words they use or how they describe something. The example we use in our chart note is that the patient describes pain "all up in my arms and legs." I think a phrase like that, quoting the patient in that context, can activate a reader's bias about a particular patient population.

Quotes can also potentially be used to cast doubt on a patient's report. Can you imagine the air quotes we use in conversation? Someone reports his pain as "still a 10," versus simply stating that the patient still has 10 out of 10 pain. Quotation marks are a really great example of a linguistic feature that can be used to portray a patient in an empathetic light or a suspicious light.

This tension highlights that we cannot use this study to dictate what words or phrases or punctuation clinicians should or should not use in their notes. Rather, what we're trying to highlight is that these tensions exist in medical records, and we need to be aware of them both as we write in the patient's chart and as we read other notes in a patient's chart.

Medscape: What are the main messages for practice that you would like clinicians to take away from your research?

Goddu: My hope for this study is that it raises awareness about a mechanism of perpetuating bias and disparities that's probably widespread, but can be overlooked. In short, I hope it makes clinicians think twice.

As an example, I hope that somebody who reads the study might more carefully consider whether they record certain nonessential points about, for example, a patient's demeanor during their encounter. Are they going to note that in the chart? Do they need to? If they read charged language in a patient's chart, they might actively attempt to give the patient a chance to make a new first impression.

We hope that further studies will refine which aspects of stigmatizing language are the most impactful and what kinds of interventions could reduce this method of transmitting bias. I think one promising result of the study from our perspective is that, when prompted, the participants were able to reflect on how the words they were using in the notes were potentially communicating respect or like or empathy—or the opposite—for a patient. To us, this capacity for reflection on language seems like a promising point of intervention.

There is an assumption that the medical record represents an objective space. I think our study calls that into question. For some patients, the medical record is the only source of information that a new doctor has about them. In this age where notes written about a patient are often copied forward, either in part or entirety, we need to be very careful about what explicit or implicit messages become amplified into a permanent record of the patients. My hope is that this makes us a little more vigilant.

Beach: I would like to reiterate that this is not to say that any particular language or phrase is right or wrong. But the attitude that we have toward patients can be consciously or unconsciously conveyed by the language that we use.

Improving the attitudes that we have toward people will help reduce the stigmatizing language that we might use. Some of the primary ways we can improve our attitudes that we have toward patients is to first take better care of ourselves. We can develop practices of mindfulness and recognize that we're experiencing emotions as clinicians. The patient is not to be blamed for those emotions, but rather they may have prompted some uncomfortable feelings in us because we are frustrated or angry or judgmental.

If we can recognize our own feelings, and recognize that these feelings have no place in defining who that other person is, then we can take a pause and prevent ourselves from recording in the medical record in a way that is biased.

Medscape: Do you think that the movement toward open records will be potentially positive in helping to address the issues of stigmatizing language?

Beach: That's a great question. I think it will help, actually. I think to the extent that our writing is deliberate and we are aware of what we are doing, we will adjust how we document to account for the fact that our patients might read it.

However, given that we are often unaware of our own attitudes and how they're affecting our documentation, it could be difficult for the open notes to have an impact. But if we start using open notes and patients start seeing what's in their records and complaining, then maybe clinicians will become more aware.

Goddu: Many physicians feel some discomfort about this idea of open notes, of patients reading their own records. I think the fact that people feel uncomfortable about it points to this underlying feeling that the medical record is not an objective space. We are putting things into a patient's medical record that are charged. That is one of the findings that I hope our study highlights. And I think this helps explain in part the emotional reaction physicians have to the idea of open notes. I hope this move to open notes calls attention to how patient care may be affected by the language we use in medical records, and helps us question the implicit biases that contribute to healthcare disparities.

Follow Laurie Scudder on Twitter
Follow Medscape on Facebook, Twitter, and Instagram


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.