Transmission of negative biases through social commentary included in neonatal intensive care unit progress notes

Abstract

Objective

To determine how the perception of families elicited after reading progress note social commentary differs by patient race.

Study design

We retrospectively performed content analysis of social commentary in physician progress notes for neonatal intensive care unit patients hospitalized from 2018–2019. Neonatologists blinded to patient race rated how commentary impacted their perception of the patient’s family on a 5-point Likert scale. Frequency of negative ratings was compared across reported race using chi-squared tests.

Results

We reviewed charts of 460 neonates. In total, 225 (49%) contained social commentary beyond parents’ names. Twelve neonatologists rated how commentaries impacted their perception of the patient’s family; 79%, 18%, and 3% were rated neutrally, negatively, and positively, respectively. Frequency of negative ratings was significantly greater among American Indian/Alaska Native than other patients (35% vs. 22%, p < 0.001).

Conclusions

Physician documentation of social commentary in patient notes may reflect and perpetuate implicit biases that contribute to race-based healthcare disparities.

Introduction

Race is a social construct that creates structural obstacles to health and well-being and disadvantages infants who are labeled as Black, Indigenous, and Hispanic even before they are born [1]. Black and American Indian/Alaska Native (AIAN) infants are more likely to be born preterm [2] and to subsequently have higher infant mortality rates [3] and poor outcomes [4]. Racial disparities in preterm birth are reduced among African-born women who give birth in the United States [5, 6] or Black women living in Canada [7], highlighting that American-born Black women experience structural inequity and racism with intergenerational impacts on health [8]. In the neonatal intensive care unit (NICU), studies suggest that Black and Hispanic families receive less education and support [9–12] during their stay, which likely contributes to worse outcomes on several measures [13], including rates of breastfeeding [14, 15], referral to high-risk infant follow-up [16, 17], readmission [18], and mortality [19, 20]. Notably, Black infants have better outcomes when cared for in NICUs with Black physicians [21], underscoring the role of racism in these disparities, whether at a structural, interpersonal, or implicit level.

Though deeply-rooted structural inequities drive disparities in neonatal health outcomes [8, 22, 23], individual providers can play a role in interrupting the effect of these disparities through acknowledging and challenging their own implicit biases [24]. Implicit bias arises when ideas about an individual are unconsciously and automatically activated based on characteristics, or stereotypes, of the group that the individual has been categorized into [25]. While automatic thinking based on categorization can help humans to make sense of a complex world, it is imperfect, and stereotypes are often incorrectly applied to individuals [26]. Implicit biases are likely to be triggered in busy, high-stress, low-certainty environments typical of healthcare settings [26]. These biases are more commonly applied to outgroup members, while ingroup members are more often considered with their individual characteristics [26, 27]. In the United States, where most healthcare providers are White, this disadvantages Black, Hispanic, and Indigenous patients even though providers’ explicit beliefs may be at odds with implicit biases [26]. The Implicit Association Test has repeatedly demonstrated that healthcare providers are not immune to experiencing and acting upon implicit biases, with effects on provider perception of patient cooperativeness, communication and rapport with patients, and treatment plans [25, 28].

Written medical records are essential for facilitating collaborative patient care, but documented content has the potential to transmit biases [27]. Previous studies demonstrate that medical records frequently carry stigmatizing language about patients, more so when these patients are Black [29–32]. Language choice has the potential to impact management, as shown in studies comparing the use of stigmatizing and neutral terms to convey identical medical information about individuals with sickle cell disease [33] and substance use disorders [34, 35]. This work has not yet been explored in the neonatal setting, where the patient’s family is particularly integral to care. We therefore performed an exploratory analysis of free-text social commentary in attending progress notes to describe included content, evaluate how commentary influences reader perception of the patient’s family, and assess racial disparities in documented content and elicited perceptions. We hypothesized that social commentary about Black and Indigenous neonates would be more likely to evoke negative perceptions of their families compared to commentary about other neonates.

Methods

Study population

This study included all infants hospitalized in the Seattle Children’s Hospital Level IV NICU for at least 15 consecutive days between January, 2018 and December, 2019.

Data extraction

We retrospectively reviewed charts of included patients and extracted race and social commentary for each patient. During the study period, two electronic medical records were in use, one a hospital-wide system (Cerner PowerChart) and the other a tailored system for tracking neonatal data (NeoData). Hospital protocol required that parent-reported race be recorded in Cerner PowerChart, while the source of race reported in NeoData was less clear; therefore, we extracted race data from Cerner PowerChart, using NeoData only to fill in gaps. Of note, neither electronic medical record in use during the study period tracked ethnicity, so our analysis was restricted to recorded race.

Social commentary was extracted from the neonatal attending progress note written on hospital day 15. In our hospital, attending physicians spend one week on service at a time, so we selected the hospital day 15 progress note in order to capture multiple physicians’ contributions to the note. Progress notes are copy-forwarded and edited daily; by hospital day 15, two to four neonatologists had contributed to each note. In most notes, social commentary was included under a “Social” heading in the patient’s free-text problem list, however, notes were reviewed in their entirety to capture commentary included elsewhere. Commentary ranged from non-existent to a few words to several lines of text, each written at the author’s discretion. We exclusively extracted social commentary because we believed it had the most potential for transmitting social biases as the principal narrative component of our notes, which otherwise focus on clinical data.

We selected 2018–2019 as our study period because including social documentation became common practice at our institution in early 2018, and we hoped to reduce confounding from the COVID-19 pandemic. As a retrospective review of de-identified medical records, this study was exempted from review by the Seattle Children’s Hospital Institutional Review Board (STUDY 00002996).

Content documented in social history

We performed content analysis [36] of social commentaries using 20 categories that the research team agreed upon through a combination of deductive and inductive processes with expert input from NICU social workers. One researcher (HS) read each history and tallied mentions of items in these categories.

Reader perception of families based on social commentary

For this analysis, we excluded histories that contained only parent first names. We emailed all neonatology faculty at the University of Washington to recruit a volunteer sample of attending neonatologists, to whom we relayed that we were trying to characterize transmission of bias through medical documentation. We asked each neonatologist to read each one-to-three-line social commentary (each requiring an estimated 10–20 seconds) and respond in an anonymous survey to the prompt, “How is your perception of this family impacted by this social history?” Responses were on a 5-point Likert scale of “very negatively,” “negatively,” “neutrally,” “positively,” and “very positively.” Respondents were provided with only the social commentary from the note without other patient or clinical information. Any names included in social commentaries were shortened to the first letter. Because implicit biases are automatic, subconscious thought processes, we asked respondents to rate each commentary based on their immediate reaction without taking more time to reflect upon the author’s intention or justify their own reaction [25, 37].

Each of the five rating options was assigned a score from −2 (very negatively) to 2 (very positively), and we calculated a mean score for each history. Histories were divided into those with mean scores that were very negative (score <−1), negative (score −1 to <−0.5), neutral (score −0.5 to 0.5), positive (score >0.5 to 1), or very positive (score >1). One researcher (HS) reviewed commentaries to identify common themes by mean score category; these themes were then reviewed and agreed upon on with a second researcher (AH).

Racial discrepancies in documented content and elicited perceptions

During the study period, race was classified in our electronic medical records as: American Indian/Alaska Native, Asian, Black, Native Hawaiian/Other Pacific Islander, White, did not wish to indicate, declined to specify, or other. We grouped patients with more than one race selected into a category called “more than 1 race.” We grouped patients with “did not wish to indicate,” “declined to specify,” “other,” and those with the field left blank into a category named “unknown race.”

We used chi-squared goodness of fit tests (each cell n ≥ 5) or Fisher’s exact tests (any cell n < 5) to compare frequency of included content and perception of the family between each individual race category and all other categories, excluding commentaries from patients in the “unknown race” category. To assess discrepancies in evoked perception of the family, we compared proportions of ratings in composite categories of “negative valence” (including very negative or negative rating) and “neutral/positive valance” (including neutral, positive, or very positive ratings).

Results

Social histories retrieved

In total, 460 infants were hospitalized for at least 15 days during our study period and were included in our study. Races for these patients were reported as: American Indian/Alaska Native (AIAN) (22, 5%), Asian (37, 8%), Black (48, 10%), more than 1 race (25, 5%), Native Hawaiian/Other Pacific Islander (6, 1%), White (280, 61%), and unknown race (42, 9%).

Social history content

Social commentary was included in 265 (58%) of hospital day 15 progress notes. The distribution of reported patient races did not significantly differ between patients with and without social commentary. Commentary included parent’s: first names (185, 70%), place of residence (110, 42%), other children (79, 30%), language preference or immigration background (43, 16%), family medical history (40, 15%), marital status (37, 14%), substance use (21, 8%), Child Protective Services (CPS) involvement or custody matters (20, 8%), communication between care team and family (20, 8%), employment (19, 7%), presence or absence of family supports (16, 6%), and parent ages (15, 6%). Parental coping or emotional or mental health disposition, where parents were staying during the hospitalization, housing instability or institutionalized living, social work involvement, prenatal care, paternity, and other stressors (financial, deaths in the family) were each mentioned in <5% of histories.

Perceptions elicited from social commentary

After removing 40 social commentaries containing only parent names, we assessed the remaining 225 for perception of the family elicited among neonatologists in our division. Twelve neonatologists volunteered to complete our survey, responding to the single question, “How is your perception of this family impacted by this social history?” for each commentary. Most histories (178/225, 79%) had a mean neutral rating (mean score −0.5 to 0.5), while 18% (40/225) had a mean negative rating (mean score < −0.5), and 3% (7/225) had a mean positive rating (mean score >0.5).

Table 1 shows themes identified from histories rated negatively, neutrally, and positively. “Very negatively” rated histories often mentioned CPS, substance use, and potential challenges with parents’ emotional coping. Notably absent from these histories were personal details, like parent names. Histories describing treatment for substance use tended to be rated “somewhat” rather than “very” negatively. “Somewhat negative” histories also included references to disagreement with the medical team. “Neutrally” rated histories included other children, interpreter needs, and recent care conferences. “Positively” rated histories included expressly positive descriptors and more humanizing language.

Table 1.

Content themes from negatively, neutrally, and positively rated social commentaries in neonatal progress notes (N = 225).

Themes	Examples
Very negative rating (N = 18, 8%)
•Loss of custody or CPS involvement •Substance use •Potential challenges with emotional coping •Omission of parent names	“Abandoned by mother, CPS involved.” “Maternal polysubstance use including heroin. Old utox…positive for cocaine, amphetamine, opiates and methadone. Maternal GM planning to be primary caretaker at discharge.” “Maternal drug use, per report mother aggressive to staff at delivery hospital, CPS involved.”
Somewhat negative rating (N = 22, 10%)
•Treatment for substance use disorder •Disagreement with medical team •Young parents	“Mom A, Maternal history of drug usage (methamphetamines and marijuana); is a smoker; last tox screen negative by report—maternal drug screen negative at delivery and she is in treatment (inpatient and outpatient).” “Mother B, not married but in relationship with FOB, lives in X; often difficult to reach; father works in Y and plans to call med provider daily at 1 p.m.; both want transfer back to Z as soon as it can be done safely even though there is a moderate risk that he will need transfer back to [this hospital] for specialty services.”
Neutral rating (N = 178, 79%)
•Place of residence •Other children •Interpreter needs/immigration background •Care conferences & communication •Family medical history	“Mother speaks a rare Mayan dialect, not married. Dad speaks both Spanish and the Mayan dialect, and understands basic English. Updated in Family Conference on 10/31 with Neurology.”
Somewhat (N = 6, 2.6%) or very (N = 1, 0.4%) positive rating
•Occupations •Humanizing language •Positive descriptors	“C and D live in X with 4-year-old who was born LGA at 33 weeks. C was on vacation in Y when she delivered prematurely; she has a college degree in psychology and has worked recently as an office manager; D is a computer programmer at Z.” “Father E, mother F, 4.5 yo daughter; parents are insightful and prefer short/medium term updates over long-term hypotheticals for now.”

Racial inequities in included content and emotional valence

Figure 1 shows the frequency of social history items mentioned across races. Compared to non-AIAN patients, AIAN patients had significantly less frequent mention of parent names (8/17 vs. 161/224, p = 0.022) and significantly more frequent mention of parent ages (4/17 vs. 9/224, p = 0.008). There were also non-significant trends toward greater frequency of mention of social work involvement (2/17 vs. 4/224, p = 0.060) custody or CPS (3/17 vs. 16/224, p = 0.139) and marital status (5/17 vs. 32/224 p = 0.095). Black patients had more frequent mention of immigration background (11/28 vs. 26/213, p < 0.001) and marital status (9/28 vs. 27/213, p = 0.007). Asian patients also had significantly more frequent mention of immigration status (7/21 vs. 30/220, p = 0.017). White patients had significantly less frequent mention of immigration background (19/162 vs. 18/79, p = 0.025) and marital status (19/162 vs. 17/79, p = 0.045). White patients trended toward more frequent mention of care communication (16/162 vs. 3/79, p = 0.128) and less frequent mention of social work involvement (2/162 vs. 4/79, p = 0.092).

Fig. 1. Frequency of categories mentioned in neonatal progress note social commentaries by patient race.

Colors correspond to Z-score for comparison of frequency of mention within each race group to frequency among all patients not in that race group, excluding patients with unknown race. * denotes p < 0.05. Bold denotes additional values thought to be significant by the research team, although not statistically significant. Commentary from one patient with race reported as Native Hawaiian/Pacific Islander is not depicted independently but is included in reporting of “All” patients. AIAN American Indian/Alaskan Native, CPS Child Protective Services. 1. Categories refer to parent characteristics unless otherwise stated. 2. Residence includes city or state where family lives. 3. Immigration includes language preference, country of origin, visa status. 4. Marital status with an infant’s biological mother determines ability to consent for an infant’s treatment in Washington state. 5. Care communication refers to notes about care conferences, the family’s communication preferences, or details of recent communications. 6. Employment includes employment status or occupation. 7. Family support refers to presence or absence. Categories with <5% frequency across all patient groups included: parental coping, housing instability, where parents are staying during infant’s hospitalization, maternal prenatal care, other stressors, paternity, and religion.

Figure 2 demonstrates the perceptions elicited from reading social commentary stratified by patient race. For this analysis, 12 neonatologists each rated commentaries from 15 AIAN, 18 Asian, 23 Black, 9 multiple race, 1 Native Hawaiian/Pacific Islander, 140 White patients, and 19 patients with unknown race. Excluding patients with unknown race, there were a total of 2472 ratings. There was a significantly higher proportion of negative ratings among AIAN patients compared to all other patients (64/180 vs. 484/2292 p < 0.001). Black and Asian patients had significantly lower proportions of negative ratings compared to all other patients (41/276 vs. 517/2196, p = 0.001; 34/216 vs. 524/2256, p = 0.012).

Fig. 2. Distribution of reader perceptions of families after reading social commentary in progress notes.

Social commentary from patients reported to have AIAN race had significantly more negative ratings compared to non-AIAN patients. Social commentary from patients reported to have Asian or Black race had significantly fewer negative ratings compared to non-Asian or non-Black patients, respectively. *Denotes p < 0.05 in chi-squared test comparing proportion of negative ratings in each racial group to all other patients with known race. N refers to number of individual ratings, i.e., number of patients multiplied by 12 raters. Rating of perception is not reported independently for commentary from one patient with race reported as Native Hawaiian/Pacific Islander but is included in reporting of “All” patients. AIAN American Indian or Alaskan Native.

Discussion

Our review of free text social commentary in neonatal attending physician progress notes revealed wide variation in documentation with more commentaries evoking a negative than a positive perception of the family. Our findings suggest that commentaries mentioning custody issues, Child Protective Services (CPS), or substance use and those pertaining to patients with reported American Indian/Alaska Native (AIAN) race are more likely to evoke a negative perception of the patient’s family. Consistent with descriptions of linguistic bias in psychology literature [27], language used in progress notes may reflect implicit racial biases and has the potential to perpetuate the detrimental effects of these biases among healthcare providers.

Compared to all patients, social commentaries for AIAN patients were more likely to evoke a negative than a neutral or positive perception of the family. Commentaries for AIAN patients seemed less likely to contain parent names, a simple humanizing detail, and more likely to contain potentially stigmatizing details, such as younger or older parent ages, parental marital status, social work involvement, and custody issues or Child Protective Services (CPS) involvement. AIAN families face important imposed structural vulnerabilities that may increase rates of CPS involvement, substance use, and other factors that may be viewed negatively [38]. While our study cannot discern the relative contributions of race and social factors to negative perceptions of families, the concept of intersectionality encourages us to consider these racial and social factors together [39].

Selective inclusion of “negative” details in social commentaries serves to strengthen negative cultural stereotypes [27] which Implicit Association Tests suggest already have a strong hold among healthcare workers [40]. It is notable that commentaries for AIAN patients contrasted those for White patients, which seemed to have more frequent explanation of care communication measures and preferences, details that, when shared between providers, are likely to improve families’ experiences of care [41]. AIAN patients are known to experience disproportionately worse healthcare outcomes, including greater rates of prematurity and its related complications [4]. Based on our findings, we call on physicians to carefully consider how they document social details for families in order to reduce rather than perpetuate marginalization.

While our findings relating to AIAN patients are of particular concern, it is also striking that negative perceptions of families were evoked far more frequently than positive perceptions among neonatologists reading social commentaries. This is in agreement with analyses of medical documentation showing frequent use of stigmatizing language [29–31] and, more specifically, narratives of “mother blame” in the perinatal period, which are particularly pronounced for Black mothers [42]. Though disappointing, this is unsurprising, as it likely reflects the “hidden curriculum” in how physicians are taught to take social histories: unlike social workers, who are trained to gather a holistic picture of patients and families, including cultural formulations, strengths, and challenges, physicians are encouraged to focus on a narrow range of risk behaviors [43, 44]. While physician awareness of their patientsʼ socioeconomic background and particular risk factors, such as CPS involvement, is essential, these details ought to be communicated carefully to minimize potential for transmitting bias. For example, highlighting family strengths [45, 46] alongside structural vulnerabilities [43] may better promote patients’ long-term thriving. It may also be worth considering leaving documentation of risk factors to social workers, who are trained to convey sensitive information [44]. In our analysis, all commentaries referring to substance use or custody/CPS evoked a mean negative perception from readers. Word choice matters [47] and can be carefully selected and potentially standardized to reduce bias, as evidenced by studies looking at differential use of terms like “substance abuser” or individual with a “substance use disorder” [34, 35]. Physicians must be aware of the potential for transmitting bias in the details they choose to document and how they choose to document them.

Several factors limit our ability to draw conclusions about racial biases and intersectionality. We had a small sample of social commentaries from a single site to draw from, limiting the generalizability of our findings. Our analysis of racial disparities is limited by how race was entered into the medical record, which we acknowledge does not always agree with individuals’ self-description. Our sample contained small proportions of patients identified as AIAN, Asian, or Black, limiting our ability to make inferences based on statistical tests. A larger sample size may have also supported our ability to discern confounding between race and other factors that may be viewed negatively, such as CPS involvement or substance use. Additionally, we were unable to report on patients with Hispanic ethnicity owing to the limitations of our medical record. Our analysis of perceptions of families evoked from reading social commentaries is inherently subjective and may be influenced by rater characteristics that we did not control for, such as age, gender, race or ethnicity, or prior life experiences. Furthermore, this analysis is based on ratings from only twelve neonatologists, all of whom volunteered to participate in this equity-centered work and are likely to have pre-existing notions about health equity that influenced their responses.

Of note, our hypothesis that Black patients and their families would also be disproportionately affected by commentaries evoking negative perceptions did not bear out. This may have been related to the low number of Black patients in our sample and the high proportion of Black patients in our geographic region with an immigrant background, a factor known to mitigate healthcare-related racial disparities [5, 6]. Other emerging studies suggest that Black patients are disproportionately described negatively in medical documentation [29], highlighting the importance of continued investigation in this area with larger, more representative samples, potentially harnessing natural language processing tools.

Our review of social commentary in neonatology progress notes suggests racial disparities in documentation that influence reader perceptions of families. This effect was most pronounced for AIAN patients, who already experience harmful inequities in care. Using standardized language, harnessing collaborations with social work colleagues, and highlighting families’ strengths may help to reduce disparities in documentation and care. Targeted workshops can also help healthcare providers be more mindful about their word choice in medical documentation [48] and their own biases that are called up as they read patient notes. In our institution, we used our findings to guide division-wide discussions on implicit bias, and we have since anecdotally noticed more careful consideration of what social details are included as attending physicians write their progress notes. Medical documentation gives physicians “testimonial power” [30] over patients and families, and physicians are responsible for using this power to reduce rather than deepen disparities.

Author contributions

HS, ZB, DV, EMW, MB and ABH conceptualized the study. HS and ABH reviewed social commentary and contributed to analysis. HS drafted the manuscript. All authors provided significant edits to the manuscript and approved the final manuscript.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.