CONTEXT: Language barriers affect health care interactions. Large, randomized studies of the relative efficacy of interpreter modalities have not been conducted.
OBJECTIVE: To compare the efficacy of telephonic and in-person medical interpretation to visits with verified bilingual physicians.
METHODS: This was a prospective, randomized trial. The setting was an urban pediatric emergency department at which ∼20% of visits are by families with limited English proficiency. The participants were families who responded affirmatively when asked at triage if they would prefer to communicate in Spanish. Randomization of each visit was to (1) remote telephonic interpretation via a double handset in the examination room, (2) an in-person emergency department–dedicated medical interpreter, or (3) a verified bilingual physician. Interviews were conducted after each visit. The primary outcome was a blinded determination of concordance between the caregivers' description of their child's diagnosis with the physician's stated discharge diagnosis. Secondary outcomes were qualitative measures of effectiveness of communication and satisfaction. Verified bilingual providers were the gold standard for noninferiority comparisons.
RESULTS: A total of 1201 families were enrolled: 407 were randomly assigned to telephonic interpretation and 377 to in-person interpretation, and 417 were interviewed by a bilingual physician. Concordance between the diagnosis in the medical record and diagnosis reported by the family was not different between the 3 groups (telephonic: 95.1%; in-person: 95.5%; bilingual: 95.4%). The in-person–interpreter cohort scored the quality and satisfaction with their visit worse than both the bilingual and telephonic cohorts (P < .001). Those in the bilingual-provider cohort were less satisfied with their language service than those in the in-person and telephonic cohorts (P < .001). Using the bilingual provider as a gold standard, noninferiority was demonstrated for both interpreter modalities (telephonic and in-person) for quality and satisfaction of the visit.
CONCLUSIONS: Both telephonic and in-person interpretation resulted in similar concordance in understanding of discharge diagnosis compared with bilingual providers. In general, noninferiority was also seen on qualitative measures, although there was a trend favoring telephonic over in-person interpretation.
WHAT'S KNOWN ON THIS SUBJECT:
Language barriers between patients and providers adversely affect quality of care. The gold standard is considered provider fluency in the patient's own language. Efficacy of approaches such as in-person medical interpretation and telephonic interpretation has not been rigorously compared.
WHAT THIS STUDY ADDS:
This study revealed that neither in-person medical interpretation nor remote telephonic interpretation is less efficacious than provider fluency in the families' native language. No “quality penalty” exists in substituting these modalities.
Sixty million people in America speak a language other than English at home. Less than half of these speak English “very well.”1 Language barriers often lessen satisfaction with medical care.2,–,4 Understanding and adherence to follow-up instructions may be diminished.5,6 Such barriers have been described as the single greatest barrier to health care access.7,–,9 In the United States, Spanish interpreters have been shown to mitigate this, but physicians with inadequate Spanish skills often underuse such services for families with limited English proficiency (LEP).10
Quality effects of untrained, ad hoc interpreters were demonstrated in a study that showed the extent of clinically significant errors.11 Barriers can result in unnecessary testing and increased interventions and rates of hospitalization.12,13 Differences in adherence of patients with LEP with medication and follow-up for asthma have been shown14; patients with LEP were less likely to receive analgesia when being seen for orthopedic complaints.15
US guidelines require that families with LEP have access to linguistically appropriate care.16 The optimal methods of interpretive service to satisfy this ethical and legal requirement have never, to our knowledge, been investigated rigorously. In the absence of bilingual health care providers, in-person, professional medical interpreters have become an intuitive “standard of care.” Previous studies to assess the effectiveness of modalities such as telephonic interpretation have involved small, nonrandomized samples that relied primarily on qualitative measures and lacked power to make direct comparisons possible.7,17,–,19 These studies have raised some interesting questions. We sought to supplement the literature with a more rigorous approach. To our knowledge, no other comparisons have used such randomization and partial blinding in their measure of language-service efficacy.
In a randomized, controlled, and semiblinded fashion, we explored the efficacy of 3 language-service modalities (verified bilingual physicians, in-person interpreters and telephonic service) by using an objective measure to reflect quality of communication. We hypothesized that families with LEP would differ on objective measures of understanding of the provider's diagnosis as well as qualitative indicators of satisfaction with that provider's care based on the type of language-service modality used.
Family Selection and Enrollment
Participants were recruited among families who presented to the Children's Hospital (TCH) (Denver, CO) emergency department (ED). Caregivers were asked at triage what the language spoken in their home was and the language they preferred to communicate with a physician during the visit. Preferences appeared on an electronic tracking board before any physician interaction. A bilingual research assistant or the bilingual primary investigator (Dr Crossman) approached all families who indicated they would prefer to communicate with their provider in Spanish and obtained informed consent to participate in the study. All families who presented to TCH ED during the hours of enrollment were eligible for inclusion, including ambulance arrivals. Families were subsequently excluded if the patient presented for a psychiatric evaluation or a suspicion of child abuse existed.
One cohort consisted of families with LEP who presented on enrollment days managed by a verified bilingual physician in the course of standard ED workflow. Bilingual providers were required to see the next eligible patient in accordance with rigid operational rules (based exclusively on acuity and waiting times). Bilingual providers were not permitted to alter this sequence on the basis of LEP status.
Families not seen by a bilingual provider on an enrollment day were randomly assigned to either the in-person– or telephonic-interpreter arms on the basis of a calendar-day randomization. There were equal numbers of each weekday and weekend days in both cohorts. Before consent, subjects were not informed of which modality was being used on that day. The hours of enrollment were held constant on the telephonic- and in-person–interpreter days. The treating physicians had no discretion in the cohort assignments. The interview was conducted in Spanish with the randomly assigned interpreter modality.
If a caregiver who preferred a Spanish encounter chose not to enroll, he or she received the same interpretive services (or bilingual provider) available to study families. The decision of the caregivers to enroll had no impact on the services available for their child. The study protocol was approved by the Colorado Multiple Institutional Review Board.
On in-person–interpreter days, interpretation was provided by an ED-dedicated, Spanish-language-speaking, Peruvian medical interpreter who was available promptly by activating a portable communication device. Telephonic services were provided by a proprietary service, CyraCom International, Inc (Tucson, AZ). This service provides double-headset telephones and interpretation in multiple languages with a single-touch button for Spanish. The dual-handset interpreter telephones were placed in triage as well as in all patient care rooms in the ED. Both interpretative services were sequential. Families seen by a bilingual physician communicated directly with their provider. Each bilingual provider's language skills were verified by the ED's professional interpreter during formal audits of their Spanish-language skills during actual patient interviews. No providers classified as “bilingual” failed these voluntary audits. Providers who declined to be audited by the interpreter were required to use either telephonic or on-site interpreters with all visits for families with LEP.
Participants were interviewed immediately after their ED visit or before transfer to their inpatient bed by a Spanish-speaking research assistant. Questions were administered verbally to avoid bias caused by literacy issues. Families were asked to state their child's discharge diagnosis or admission diagnosis in their own words. The interviewer asked families, “What did the doctor tell you was wrong with your child?” The reply was recorded in Spanish and interpreted by the primary investigator (Dr Crossman). Responses were compared with the official discharge diagnosis or admission diagnosis from the medical record. The primary outcome measure was concordance of diagnoses between the family and provider. Concordance was assessed by the principal investigator (Dr Crossman) and reviewed independently by the co-investigator (Dr Hampers) to establish interrater reliability. Investigators were blinded to the cohort assignment when making these determinations.
As a qualitative measure, a previously validated family-satisfaction questionnaire4,20 was adapted and translated into Spanish. Families were asked to rate their primary ED physician. (A complete questionnaire is available from the corresponding author on request.) Study participants were asked in Spanish to rank the following items on a 4-point scale (1, excellent; 2, good; 3, fair; 4, poor):
Was the physician attentive to what you said? (attentive)
Did the physician speak in words that you understood? (words understood)
Did the physician treat you with respect? (treat respect)
Did the physician make you feel comfortable? (feel comfortable)
How satisfied were you with your provider? (satisfaction with physician)
What was your overall satisfaction with the visit? (overall satisfaction)
What was your overall satisfaction with the quality of language services? (satisfaction with language services)
The χ2 test was used for categorical variables. Continuous variables, age and Likert scales, were analyzed with the Kruskal-Wallis test, because the distributions were not normal. Posthoc analysis was performed by using paired Mann-Whitney U tests. The κ statistic was used to assess interrater reliability.
For noninferiority analysis, the verified bilingual physician was considered the gold standard to which the interpreter modalities (in-person and telephonic) were compared. Likert-scale scores were grouped by combining excellent and good results and combining acceptable and bad results to create a dichotomous variable. Differences between proportions with 95% confidence intervals were calculated. Prestudy power calculations and sample sizes were chosen to ensure that noninferiority differences of ≥5% from the bilingual gold standard would be detected. Data were analyzed by using SPSS 16.0 (SPSS Inc, Chicago, IL).
A total of 1201 families were enrolled; there were no refusals and no exclusions. The randomized telephonic cohort consisted of 407 visits, and the in-person cohort consisted of 377. There were 417 families treated by verified bilingual physicians during enrollment times and comprised the random bilingual cohort. Demographic characteristics are displayed in Table 1. The demographic characteristics displayed in Table 1 demonstrate that there was no statistical significance between the cohorts with respect to age, gender, or the parent's country of origin. There was also no difference in day of week (weekend versus weekday), time of day of visit, or discharge diagnoses (categorized by International Classification of Diseases, Ninth Revision [ICD-9]groupings) between the 3 cohorts.
The families' reports of their children's diagnoses were compared with the discharge diagnoses recorded in the medical record. There were no statistically significant differences across the cohorts: 95.1% of families in the telephonic-interpreter arm, 95.5% of families in the in-person–interpreter arm, and 95.4% of families seen by a bilingual physician accurately reported their discharge diagnoses.
Concordance results determined by the principle investigator (Dr Crossman) were compared with the concordance determinations of a second investigator (Dr Hampers). Drs Crossman and Hampers both found 95.2% overall concordance between families' report of discharge diagnoses when compared with the medical record. The κ value was 0.4; Drs Crossman and Hampers agreed on 94.2% of the patients (P < .001).
For 6 of the questions (“Was the physician attentive to what you said?” “Did the physician speak in words that you understood?” “Did the physician treat you with respect?” “Did the physician make you feel comfortable?” “How satisfied were you with your provider?” and “What was your overall satisfaction with the visit?”), participants in the in-person–interpreter cohort scored their visit worse than those in the bilingual and telephonic cohorts (P < .001) (Table 2).
Participants in the bilingual-provider cohort were less satisfied with their language services than those in the in-person and telephonic cohorts for the question, “What was your overall satisfaction with the quality of language services?” (P < .001) (Table 2).
An overwhelming majority of the families in the study rated the quality of their provider and overall visit as “excellent” or “good” (Table 2). The results of the questionnaire data were compared by calculating the differences in proportions of the grouped results (excellent/good responses) between the interpreter-modality (in-person and telephonic) cohorts and the bilingual-physician cohort (the gold standard). Figure 1 shows noninferiority; the 95% confidence intervals of the difference between groups were always within 5% regardless of the direction of change for the questions, “Was the physician attentive to what you said?” “Did the physician speak in words that you understood?” “Did the physician treat you with respect?” “Did the physician make you feel comfortable?” “How satisfied were you with your provider?” and “What was your overall satisfaction with the visit?” For all of these questions, the confidence intervals of the difference crossed 0. Point estimates to the left of 0 suggest that the bilingual-provider cohort performed better than the interpreter-modality cohort (telephonic or in-person). Point estimates to the right of 0 suggest that the interpreter-modality cohort (telephonic or in-person) performed better than the bilingual-provider cohort.
Participants in the bilingual-provider cohort rated their satisfaction with the quality of language services inferior to both the telephonic and in-person cohorts' ratings for the question, “What was your overall satisfaction with the quality of language services?” The confidence intervals of the difference did not include 0.
For a comparison of interpreter modalities, concordance with the discharge diagnosis is the most concrete and least subjective metric reported to date. This is further strengthened by our large, randomized sample and blinded nature of our determinations. For comprehension of this aspect of the visit, we found no difference between the 2 interpreter modalities when compared with a gold standard of bilingual providers. Although more subjective, satisfaction was also high and similar across the groups.
Although ours is the first ED study, to our knowledge, to use “concordance” with the discharge diagnosis as a primary outcome, there are, of course, many other elements that need to be communicated during an ED visit, such as use of prescriptions and follow-up arrangements. However, our chosen outcome of concordance has several advantages in design: (1) it is a dichotomous variable, which enables rigorous and sensitive statistical comparisons; (2) it can be applied to a heterogeneous group of complaints; (3) the raters of concordance were blinded to cohort assignment; and (4) as a quantitative, knowledge-based measure, respondents could not be biased toward telling the interviewers what they thought they wanted to hear.21
Our study cannot establish a causal link between these measures and health outcomes. The outcomes of most pediatric ED visits are very good. However, anecdotes of adverse outcomes caused by faulty communication are too abundant to catalog here. Effective communication with a health care provider is, of itself, a legitimate quality metric. In addition, family satisfaction is a measure on which facilities compete and compare themselves. It is clear that communication regarding follow-up and prescriptions is relevant. However, the heterogeneity of these discharge instructions across groups, as well as a lack of an intuitive, broadly accepted measure of successful communication on these points, compromises the validity of such additional comparisons.
Given the high levels of satisfaction on our qualitative measures across all groups, we have chosen to emphasize the “noninferiority” of the interpreter modalities. Indeed, the majority of families responded “excellent” or “good” to our questions, and they did so at uniformly high rates. Such findings provide support to policy makers and administrators who need to know if some sort of “satisfaction or quality penalty” exists for a particular interpreter modality. We believe that it does not.
Our more detailed analysis of median scores deserves comment. Dichotomizing the responses of Latinos may be unreliable.21 Nevertheless, there was a persistent pattern of worse scores for the in-person cohort. It is possible to dismiss these findings by suggesting that our study was “overpowered” and detected differences of statistical but not clinical relevance, but other hypotheses could be entertained. Perhaps a third person in the examination room makes the family feel more removed from the physician. Perhaps during an interview addressing the in-person interpreter seems less respectful. The notion that telephonic interpretation may have out-performed in-person interpretation was unanticipated and deserves further investigation.
The families seen by a bilingual physician rated the quality of “language services” slightly lower than the families in the other 2 groups. It should be noted that despite the small but statistically significant difference, at least 90% of the families in all 3 arms rated the quality of the language service as “good” or “excellent.” The differences noted could be a result of a lack of stringent-enough criteria to eliminate semifluent Spanish-speaking providers from the “bilingual” group. Or, it could be a result of an inability of native English-speaking physicians with Spanish-language training (as most bilingual providers were) to ever perform as well as a native Spanish-speaking interpreter (both interpreter modalities used native speakers). There may also be subtle differences in other aspects of cultural competency between non–native-speaker bilingual providers and native speakers.
Submission to an audit for inclusion as a verified “bilingual” provider was voluntary. An alternative explanation to the one discussed above is that, rather than the auditing process being too lax, perhaps it was viewed as too stringent or intimidating (the fact that no providers failed suggests that only a confident subset of providers subjected themselves to auditing). It is possible that some semifluent providers with fairly strong Spanish skills preferred not to risk failing an audit and, thus, forced themselves into a randomized interpreter cohort. Their own language skills may have supplemented the apparent efficacy of the interpretation.
When other advantages of truly bilingual providers over interpretative services are considered (chiefly related to operational efficiency), it would seem that properly screened bilingual providers remain an appropriate standard for comparisons. Requiring professional interpreter use for verified bilingual providers is a poor allocation of resources, and the small differences reported in our study do not justify such an approach.
Our cohort of families seen by a bilingual provider was not randomized in precisely the same fashion as the other 2 cohorts. This is unlikely to be an explanation for the differences we observed. Rules required providers to see patients in order of acuity and waiting times. If any bilingual providers violated this protocol to see families with LEP out of sequence (or return visitors with LEP already known to them), it would bias our results toward higher satisfaction in that cohort. Although an identical randomization process for all 3 cohorts would have been ideal, it was impractical. In practice, our randomization protocol precluded both families with LEP and providers from influencing their cohort assignments, which is the core advantage to randomization in any clinical trial.
There may remain a popular conception that in-person interpretation is superior to telephonic interpretation in ways that were not studied here. If a patient was required to visit multiple sites in a hospital during a single visit (clinic, ED, radiology, inpatient unit), help with way-finding could only be provided in-person. In-person interpreters also may provide other benefits such as continuity during visits, feedback to providers after interpretation, and explanations of the particular cultural relevance of a patient's symptoms. Providers may, indeed, prefer the “personal touch” of an in-person interpreter. Our study was conducted in a busy tertiary ED; extrapolation to other settings such as ambulatory clinics or inpatient units should be done with caution. In addition, we must stress that our findings establish noninferiority of the interpreter modalities from the families' perspective in our pediatric ED; a study to determine providers' preferences is currently a topic of further investigation by our team.
We used a single interpreter for each day of in-person interpretation throughout this study. Although certified through our institution, this interpreter may have been a particularly good one or a bad one. Our impression was that he was good. Still, there is undoubtedly a range of skill and effectiveness, even among qualified professionals. Also, our study interpreter is from Peru, which may limit his ability to communicate with non-Peruvian families. The overwhelming majority of our families with LEP were from Mexico. However, one presumes that this limitation would also apply to telephonic services, depending on the cultural background of the remote interpreter.
It is conceivable that “contamination” via contact with multiple providers during the ED visit biased our results toward the null. Queries regarding satisfaction were directed only at the primary treating ED physician. We cannot know how much contact families may have had with other, bilingual providers in informal ways and how that may have affected our outcomes. However, it is important to recognize that ours was a “real-world” trial of interpreter modalities, and decision-makers may actually find our design more relevant to their own ED settings. Few ED visits involve contact with only 1 ED staff member, so to the extent that contamination might have obscured differences between interpreter modalities, generalizability should be undiminished.
Using an in-person interview by a research assistant may not have been the best method for obtaining negative feedback from the subjects. However, we designed the study specifically to avoid literacy issues and to ensure complete follow-up. Compared with the work of Hayes and Baker,21 whose interviews were conducted by telephone, our results may have been biased toward even more positive satisfaction responses. Because this approach was applied consistently across the random cohorts, we would still expect to detect important differences if they had existed. Again, our primary outcome, concordance, was knowledge based and should not have been affected by a personal interview.
For ethical reasons, our study did not include a “true control” cohort composed of families with LEP receiving neither professional interpretation nor a verified bilingual physician. One assumes that our measures of quality would have been lower for such a hypothetical cohort. Perhaps one might argue that the uniformly high rates of concordance we saw on our objective measure suggests that we “set the bar too low” by simply asking the family to recall a discharge diagnosis. It should be stressed that all cohorts did get some form of language service; thus, these high rates are not surprising. Again, because of the heterogeneity of complaints studied, more detailed comparative explorations of discharge-instruction understanding might not have been valid. However, the qualitative satisfaction measures reported support and strengthen the conclusions on the basis of our objective measures.
Professional interpreters can ensure that adequate communication takes place between a provider and a family. The type of interpreter modality used may not be as important as the fact that one is being used. Although our study detected no differences in the objective outcomes studied, more work is needed to examine what unmeasured differences may exist. In addition, the relative differences in duration of interpretation, manpower-allocation issues, and overall cost effects of these methods of interpretation demand further investigation.
This study was supported by an unrestricted research grant from CyraCom International, Inc (Tucson, AZ). Additional support from the Children's Hospital Research Institute (TCHRI) was provided in the form of a quality service improvement (QSI) grant. A TCHRI QSI grant of $30 000 was used to pay for interpretive services (both in-person and telephonic) and research-assistant stipends during a pilot period of the project. The TCHRI approved the study protocol but had no role in the design or conduct of the study. CyraCom provided $60 000 for in-person interpretation and research-assistant stipends. In addition, telephonic interpretation was provided free of charge to TCH for enrolled families. The retail value of this “in-kind” service is estimated at $40 000. TCH received no other direct or indirect financial support from CyraCom. CyraCom had no role in the design or conduct of the study, in the collection, analysis, or interpretation of the data, or in the preparation, review, or approval of the manuscript.
Manuel Portocarrero provided in-person interpretation to the TCH ED during enrollment times. He was compensated at an hourly rate for this work. Sylvia Winder is a bilingual research assistant who conducted postvisit interviews with families. Michael Greenbaum is the former CEO of CyraCom and facilitated the acquisition of funding as well as the in-kind contribution of telephonic service; he is no longer employed by CyraCom. Sara Deakyne is a research assistant who facilitated data entry.
- Accepted October 16, 2009.
- Address correspondence to Louis C. Hampers, MD, MBA, University of Colorado School of Medicine, Section of Pediatric Emergency Medicine, 13123 E 16th Ave, B251, Aurora CO 80045. E-mail:
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
- LEP =
- limited English proficiency •
- TCH =
- the Children's Hospital •
- ED =
- emergency department
- 1.↵US Census Bureau. 2005–2007 American Community Survey: selected population profile in the United States: Hispanic or Latino (of any race). Available at: http://factfinder.census.gov. Accessed June 1, 2009
- 2.↵Office of Civil Rights. Mission of LEP.gov. Available at: www.lep.gov. Accessed November 30, 2009
- Burbano O'Leary S,
- Federico S,
- Hampers L
- Flores G,
- Laws MB,
- Mayo SJ,
- et al
- Hampers LC,
- Cha S,
- Gutglass DJ,
- Binns HJ,
- Krug SE
- Hayashi D
- Garcia EA,
- Roy LC,
- Okada PJ,
- Perkins SD,
- Wiebe RA
- Copyright © 2010 by the American Academy of Pediatrics