Racial and Ethnic Differences in Communication Quality During Family-Centered Rounds

We conducted a cross-sectional observational study of racial, ethnic, and language differences in measures of communication quality between clinicians and caregivers of children admitted to the general pediatric service at a tertiary care center. Our institution averages approximately 2500 admissions to general pediatrics per year, of whom 31% self-identify as Black, 49% as white, 10% as Latinx, and 10% as another race or ethnicity. We oversampled Black and Latinx families for greater power to detect racial and ethnic differences in communication with a goal of 80 non-Latinx Black (Black), 80 non-Latinx white (white), and 40 Latinx (any race) caregivers. We approached caregivers whose child’s race and ethnicity was Black, Latinx, or white in the electronic medical record. Caregivers were eligible if they were adults (≥18 years of age), fluent in English or Spanish, and identified as a primary caregiver (parent or guardian) of a hospitalized child on the general pediatrics team. We limited languages to English and Spanish, given these are the predominant languages in our racial and ethnic groups of interest. We excluded caregivers of children undergoing evaluation for child abuse or neglect and those admitted for a primary psychiatric condition.

We identified caregivers through daily monitoring of the inpatient general pediatric team lists through the electronic health record Monday through Friday. We approached caregivers on the child’s first hospital day (within 24 hours of admission) and consented before rounds. Enrolled caregivers completed a survey immediately following rounds. Because of coronavirus disease 2019 (COVID-19) policies, most children had a single caregiver present with them during this study. In rare cases where more than 1 caregiver was present, we asked them to identify the primary caregiver to answer the survey questions. We selected the first hospital day to sample communication as we hypothesized that team communication on the first day would be more standardized, include greater medical decision-making, and involve more questions from caregivers than subsequent days.³¹  Once per week we performed audio-recordings of rounds on each of the 2 general pediatrics teams with a goal of recruiting 20 Black, 20 Latinx (half with Spanish as preferred language), and 20 white caregivers. Our institution has an established system for requesting an in-person interpreter during rounds, which is the general practice for caregivers with a preferred language other than English. Caregivers were approached and completed written consent to perform audio-recordings before rounds. Healthcare providers were given information about the study and the opportunity to opt out of being recorded 1 week before enrollment.

We used REDCap, a secure database with survey capability, to collect and store participant data.^32,33  This study was approved by our Institutional Review Board and all caregivers underwent written consent before participating. Caregivers were compensated $10.00.

We collected caregiver age, gender, highest educational level, financial security, child insurance, child age, and whether their child had a previous hospitalization (excluding birth admission). In analyses, we transformed age into a binary variable, less than 10 years or 10 years or greater, as we reasoned children 10 years or older may become more involved in conversations with doctors and influence how teams communicate with caregivers. Financial security was defined as “high” if caregivers selected “after paying the bills you still have enough money for special things you want;” “medium” if caregivers selected “you have enough money to pay the bills, but little spare money to buy extra or special things;” or “low” if caregivers selected “you have money to pay the bills, but only because you cut back on things” or “you have trouble paying the bills no matter what you do.” Caregivers were asked to self-identify their race, and if they selected more than 1 race, we asked them the race with which they most strongly identified. We also assessed ethnicity (ie, Hispanic or Latinx). Caregivers were excluded from analyses if they declined to answer the questions on race or identified as a race or ethnicity other than Black, white or Latinx. We coded Latinx caregivers as preferring Spanish if they had a Spanish interpreter during rounds.

Consented caregivers completed an 84-item survey immediately following rounds (Supplemental Information 1: Caregiver Survey). Two independent bilingual Spanish-speakers translated and back-translated our survey. The survey took approximately 10 to 15 minutes to complete and included the following measures:

• Interpersonal Processes of Care (IPC): an instrument to measure patient perceptions of care that has been validated in English-speaking and Spanish-speaking Latinx adult patients. The IPC has not been studied or validated in pediatric populations; we modified items to make them appropriate for this use (ie, “your child’s doctors” rather than “your doctors”). The measure includes several subdomains, each of which serves as an independent predictor.26 Average scores for each domain were calculated and dichotomized to top score (ie, average score 5 for positive domain and 1 for negative domain) yes or no because of large ceiling and floor effects.

• Perceived autonomy support: caregivers responded to a 15-item scale to assess how well physicians supported their autonomy using the validated health care climate questionnaire.³⁴  Perceived autonomy support was dichotomized to a response of “strongly agree” for all 15 items versus all other responses.

• Empowerment: we used a 5-point Likert scale to measure agreement with a statement of caregiver empowerment and comfort to participate in rounds (1 = strongly disagree and 5 = strongly agree) and dichotomized to top score response.

On hospital day 1, we obtained audio recordings via a research-encrypted iPad. All files were stored on a secure research folder behind an institutional firewall. We transcribed audio-recordings verbatim and developed a codebook a priori using validated instruments reported by Street et al.³⁵  We included counts of medical team behaviors, caregiver participatory behaviors, and global ratings of communication (Supplemental Information 2: Objective Measures Codebook). Bilingual audio-recordings were transcribed, coded, and analyzed in English. Two authors (V.P. and H.R., attending hospitalist and medical student) coded the encounters. Coders were blinded to caregiver race and ethnicity. V.P. and H.R. compared codes until they achieved reliability (n = 14 transcriptions) and then, the remainder were coded by a single individual. In 3 instances, 2 caregivers were present, and their participatory behaviors were tallied together. Patient participatory behaviors were not included in the count of caregiver participatory behaviors.

We summarized medical team behaviors and caregiver participatory behaviors as total counts over the encounter. We examined associations between caregiver race, ethnicity, and language and these count outcomes via generalized linear models, with appropriate link outcome (either Poisson or Negative Binomial where there was evidence of overdispersion) and offset term accounting for encounter length in minutes. We used descriptive statistics and goodness of fit test statistics to determine whether overdispersion was present. Additionally, we assessed outcomes for presence of clustering by attending physician. Because of the wide variation in number of caregivers per physician in our sample, we followed recommendations by Li and Redden (2015) using the Fay and Graubard-corrected sandwich estimator for SE estimation where evidence of clustering was found.³⁶  Incident rate ratios (IRR) with corresponding 95% confidence intervals summarized the relative mean number of behaviors per encounter time minute between the groups. We performed a sensitivity analysis, where we calculated the IRR using the rate of each behavior per time in a caregivers preferred language to compare rates of behaviors in our Spanish-speaking caregivers (time in Spanish) to non-Latinx white caregivers (time in English).

We examined associations between caregiver race and ethnicity and outcomes of global ratings of the conversation via general linear models accounting for clustering by attending physician. Unadjusted models only included the race, ethnicity, and language variable, whereas adjusted models additionally included if the child had any prior hospitalizations and child age (≤10 vs 11 years or more). We were unable to include covariates of socioeconomic status (education, income, or insurance) in the audio-recording analyses because of a lack of heterogeneity in our Latinx population (Supplemental Tables 6 and 7).

In the full sample of caregivers, we assessed associations between caregiver race and ethnicity and the dichotomized survey outcomes via logistic regression. Again, we examined outcomes for the presence of clustering by attending physician, and where possible, model estimation incorporated the Fay and Graubard-corrected sandwich estimator. Covariates in adjusted models included insurance (Medicaid), previous hospitalization (yes), and older children (11 or older). We conducted analyses utilizing SAS v9.4 (Cary, NC).

We identified 341 eligible caregivers during study enrollment from August 2020 to June 2021 (Fig 1). Of the 341 eligible caregivers, 63 (19%) were not at the bedside, 26 (8%) were unavailable (eg, off the floor for procedures or imaging), or not approached because of infectious risk (eg, COVID-19 positive), and 39 (11%) declined to participate. A total of 213 (62%) of the 341 eligible caregivers completed the survey. We excluded 11 caregivers who completed the survey for declining to answer the race and ethnicity question or identifying as a race or ethnicity other than Black, white or Latinx resulting in a sample of 202 caregivers. On audio-recording days, 61 (79%) of the 77 eligible caregivers consented to the recordings; 5 declined to participate, 8 were not present, and 3 were unavailable (2 COVID-19 positive, 1 in a procedure).

Of the 202 caregivers included in the study, 82 (41%) identified as white, 80 (40%) identified as Black, 40 (20%) identified as Latinx (Table 1). Of the 40 Latinx caregivers, 29 (73%) identified as “other” racial group. Most caregivers were the child’s parents (94%), female (87%), and 22 to 45 years old (77%). There were differences in the percent of children with Medicaid insurance by race and ethnicity and language; the children of 38% of white, 83% of Black, 88% of English-speaking Latinx, and 71% of Spanish-speaking Latinx caregivers had Medicaid. Admission diagnosis was variable (data not shown) with seizure or seizure-like activity being the most common (11%), and approximately half of the children had a previous hospitalization (Table 1).

Medical team behaviors, including eliciting questions and empathic statements, were similar in terms of absolute values while controlling for encounter length across white, Black, and English-speaking Latinx caregivers; however, teams asked Spanish-speaking caregivers fewer open-ended questions and responded with fewer reflections than in encounters with white caregivers (IRR 0.3; 95% CI 0.1–0.7 and IRR 0.3; 95% CI 0.1–0.8, respectively) (Table 2). Black, English-speaking Latinx, and Spanish-speaking Latinx caregivers demonstrated significantly fewer caregiver participatory behaviors than white caregivers; the mean number of participatory behaviors was 8.3 for white caregivers (reference group), 5.8 for English-speaking Latinx caregivers (IRR 0.5; 95% CI 0.3–0.8), 5.1 for Black caregivers (IRR 0.6; 95% CI 0.4–0.8), and 3.3 for Spanish-speaking Latinx caregivers (IRR 0.3; 95% CI 0.2–0.5). In our global measures of medical team communication, Spanish-speaking Latinx and Black caregivers experienced lower scores of respect and partnership compared with white caregivers as well as borderline lower attentiveness to Spanish-speaking Latinx compared with white caregivers (Table 3 and Fig 2). In rounds with an interpreter present, time in Spanish ranged from 10% to 48% of the total time, with a median of 39% (Supplemental Table 8). In our sensitivity analyses, we used talk time in the caregiver’s preferred language rather than total encounter time as the offset term for calculating the IRR. In these analyses, we no longer found significant differences in rates of behaviors between Spanish-speaking caregivers and white caregivers. Of note, analyzing just this truncated section “rewards” encounters where less was interpreted, giving an open-ended question more weight in a short encounter with less time in Spanish, whereas in an absolute sense, this might not be clinically significant nor is it consistent with our observed differences between Spanish-speaking and white caregivers in the total count of behaviors (Table 3).

In the caregiver survey, a smaller proportion of Black caregivers strongly agreed with the statements “I had adequate time to ask questions” and strongly disagreed with the statement “I was afraid to ask questions” compared with their white counterparts, although these differences were not significant in adjusted models. A smaller proportion of Spanish-speaking Latinx caregivers strongly agreed with the statements “I felt comfortable participating in rounds,” “I had adequate time to ask questions,” and strongly disagreed with the statement “I was afraid to ask questions” (Tables 4 and 5). Spanish-speaking Latinx caregivers were also less likely to report always feeling that doctors were respectful compared with white caregivers (OR 0.4; 95% CI 0.2–0.9) and that their concerns were always elicited (OR 0.4; 95% CI 0.1–0.9).

In our study, we found racial and ethnic differences in subjective and objective measures of communication quality by race and ethnicity during family-centered rounds. Global ratings of medical team communication demonstrated lower respect and partnership for Black and Spanish-speaking Latinx caregivers than white caregivers. Additionally, medical teams asked fewer questions and responded with fewer reflections to Spanish-speaking Latinx caregivers. We also found Black and Latinx caregivers demonstrated fewer participatory behaviors compared with white caregivers. Concordantly, Black and Spanish-speaking Latinx caregivers were more likely to report not having adequate time to ask questions and fear of asking questions compared with white caregivers.

Evidence suggests clinician communication is one of the strongest drivers of caregiver participation.³⁷  Clinician behaviors such as information giving, supportive talk, and partnership-building statements can all promote patient participation.¹⁷  Further, communication can be a reinforcing cycle where clinicians provide more information when patients are more active participants, and patients are more active when partnership-building statements are used by clinicians.^12,17,38  In our study, the reasons for lower participation among Black and Spanish-speaking caregivers are unknown, yet consistent with prior studies.^14,17,39,40  We found Spanish-speaking Latinx caregivers were asked fewer open-ended questions by the medical team, which has been associated with lower caregiver participation during rounds.⁴¹  In contrast, we did not find differences in medical team behaviors in encounters of Black versus white caregivers.

These findings may be explained in several ways. First, rounds were delivered the same to all families, yet the current approach worked better in empowering white caregiver participation because of cultural, historical, and other social group differences. Caregivers of color have been historically under-represented in research, and thus our current knowledge of caregiver preferences for communication is likely more representative of white families.^42–45  Additionally, values of autonomy and assertiveness may not be equally endorsed across cultural backgrounds and can influence caregiver participation.^46,47  A second explanation is that lower participation was driven by subtle communication differences, as captured in our global ratings of lower respect and partnership in Black compared with white caregivers. Lastly, it is possible there were differences in communication behaviors that we did not capture in our codes. For instance, we were limited to measures of verbal communication and it is possible nonverbal communication impacted caregiver participation.⁴⁸  Importantly, communication behaviors can be taught, refined, and modeled to strengthen communication across differences of race, ethnicity, culture, and power.⁴⁹  Some potential strategies to improve participation in rounds might include inviting caregivers to participate early and often, providing complete interpretation of rounds, active-listening to caregiver contributions, treating caregivers with dignity and respect, and requiring provider trainings on bias and cultural humility.^50–53 

Language-specific barriers may operate independently, or possibly synergistically, with implicit bias to impact communication quality. Although language barriers likely contribute to these differences, Spanish-speaking caregivers may also experience intersectional bias with skin color, class, gender, country of origin, immigration status, religion, cultural norms, or other identities.⁵⁴  These dimensional nuances and heterogeneity within the Latinx population are not fully captured in our methods and require attention in future studies.⁵⁵  Additionally, there was variability in the use of interpreters during rounds. For example, some teams would complete rounds in English, then summarize to the caregiver in Spanish, whereas others summarized the entire discussion with frequent pauses for interpretation. The former approach values efficiency over family-centeredness and has implications for caregiver trust, participation, and decision making.^56–58 

This study has several limitations. First, we do not know the demographics of caregivers not enrolled and thus our sample may not be representative of the total general pediatrics population at our institution. Second, we were unable to adjust for socioeconomic status on the audio-recording measures because of our sample size. This is important as there is conflicting evidence as to whether racial inequities in communication persist after adjusting for income and education.^12,14,39,59  Third, we did not control for clinician-caregiver racial concordance or Spanish fluency and these factors have been shown to impact communication quality.^10,13,60,61  However, there were no bilingual providers captured on the audio-recordings and only 2 attending physicians of color. Further, it would be hard to define concordance in this setting given there were multiple team members in the room. Fourth, it is possible that our method of coding Spanish-speaking did not capture caregivers who would have preferred Spanish but did not have an interpreter. Lastly, medical teams knew they were being recorded and thus might have altered their behavior.⁶²  However, we would expect this potential bias to result in type II errors, as teams are more likely to be on their “best behavior” when being recorded.

In conclusion, our team found racial and ethnic differences in the quality of communication during family-centered rounds. Although family-centered rounds may promote humanistic care,⁶³  our study suggests that family-centered rounds as currently practiced may not achieve inclusive and equitable communication outcomes for all families.