Background. Although an inability to speak English is recognized as an obstacle to health care in the United States, it is unclear how clinicians alter their diagnostic approach when confronted with a language barrier (LB).
Objective. To determine if a LB between families and their emergency department (ED) physician was associated with a difference in diagnostic testing and length of stay in the ED.
Design. Prospective cohort study.
Methods. This study prospectively assessed clinical status and care provided to patients who presented to a pediatric ED from September 1997 through December 1997. Patients included were 2 months to 10 years of age, not chronically ill, and had a presenting temperature ≥38.5°C or complained of vomiting, diarrhea, or decreased oral intake. Examining physicians determined study eligibility and recorded the Yale Observation Score if the patient was <3 years old, and whether there was a LB between the physician and the family. Standard hospital charges were applied for each visit to any of the 22 commonly ordered tests. Comparisons of total charges were made among groups using Mann-Whitney U tests. Analysis of covariance was used to evaluate predictors of total charges and length of ED stay.
Results. Data were obtained about 2467 patients. A total of 286 families (12%) did not speak English, resulting in a LB for the physician in 209 cases (8.5%). LB patients were much more likely to be Hispanic (88% vs 49%), and less likely to be commercially insured (19% vs 30%). These patients were slightly younger (mean 31 months vs 36 months), but had similar acuity, triage vital signs, and Yale Observation Score (when applicable). In cases in which a LB existed, mean test charges were significantly higher: $145 versus $104, and ED stays were significantly longer: 165 minutes versus 137 minutes. In an analysis of covariance model including race/ethnicity, insurance status, physician training level, attending physician, urgent care setting, triage category, age, and vital signs, the presence of a LB accounted for a $38 increase in charges for testing and a 20 minute longer ED stay.
Conclusion. Despite controlling for multiple factors, the presence of a physician–family LB was associated with a higher rate of resource utilization for diagnostic studies and increased ED visit times. Additional study is recommended to explore the reasons for these differences and ways to provide care more efficiently to non-English-speaking patients. language barriers, resource utilization, test ordering.
- LB =
- language barrier •
- ED =
- emergency department •
- NLB =
- no language barrier
An inability to speak English is regarded widely as an obstacle to both primary and emergency health services in the United States. Recent census figures estimate that ∼14 million people living in the United States have limited proficiency in spoken English.1 In many settings, practitioners are required to solve clinical problems when neither the patient nor their family members speak English well enough to give an accurate or complete history.
Because many non-English-speaking families are poorly insured,2 ,3 variations in practice patterns with respect to this population may have a disproportionate effect on overall societal costs. Despite agreement that language barriers (LBs) are undesirable, it is unclear what effect language discordance between physician and patient has on the actual clinical practice of the physician.
In their review of the subject, Woloshin et al4 propose two competing hypotheses. First, with an incomplete medical history, physicians remain unaware of the need for particular diagnostic investigations; therefore, they order fewer tests. Conversely, deprived of the diagnostic power of the medical history, clinicians compensate for this deficiency by becoming more reliant on laboratory and radiographic data. Under the former assumption, one would expect resource utilization for the care of these patients to be decreased. Under the latter assumption, resource utilization should be increased.
By prospectively comparing two clinically similar cohorts of patients presenting to our pediatric emergency department (ED), we sought to determine the net effect of a physician–family LB on resource utilization for diagnostic testing and length of ED stay.
Our investigation took place at an urban, university-affiliated pediatric ED with a total annual volume of ∼39 000 patients. In accordance with hospital policy, all patients presenting to the ED were triaged by a qualified registered nurse who assigned them to one of four triage categories (emergent, urgent high, urgent low, or nonurgent). During weekday evenings between 5 and 11 pm and weekends between 11 am and 11 pm, ED patients triaged as nonurgent were seen in an onsite urgent care unit.
From September 1997 to December 1997, a data form was attached to every patient chart at triage. The form required physician providers to identify patients who met the following criteria: 2 months to 10 years of age; absence of chronic illness (specifically, a history of immunosuppression or immunodeficiency, inborn error of metabolism, or ventriculoperitoneal shunt); and either a triage temperature ≥38.5°C or a complaint of vomiting, diarrhea, or decreased oral intake. For patients <3 years old, the form provided elements of the Yale Observation Scale5 with which the physicians were asked to evaluate the child's initial appearance.
We asked the treating physicians to determine whether (in their estimation) the patient's family could speak English. Specifically, they were asked, “Does this patient's family speak English?” To identify encounters in which the physician believed that they were adequately proficient in the non-English speaking family's language, providers were then asked, “If not, did this create a language barrier for you?” Finally, if a barrier was thought to exist, the physicians were asked to indicate whether an interpreter was used.
After the visit, the medical records of included patients were reviewed by a primary investigator (L.H. or S.C.). Information regarding patient demographics, provider experience level (postgraduate year of training), attending physicians, setting (primary ED or urgent care unit), initial vital signs, triage category, length of stay (triage to discharge from the ED), standardized charges for 22 commonly ordered tests, and patient disposition was extracted.
Data were entered and analyzed in SPSS for Windows, Version 6.1.4 (SPSS, Inc, Chicago, IL, 1996). For categorical data, χ2tests were used to compare proportions among groups. Mean lengths of stay and other continuous variables were compared using a two-tailed Student's t test. Because of the nonnormal distribution of charges for diagnostic testing in each group, comparisons were made using a Mann-Whitney U test. To isolate the effect of a LB on charges and length of stay, an analysis of covariance (ANCOVA) model incorporating race/ethnicity, insurance status, provider training levels, patient care setting, and triage category as additional main effects with patient clinical characteristics (age and vital signs) as covariates was constructed. Significance was set at P< .05.
The study protocol was approved by the hospital's Institutional Review Board.
Physician providers properly completed study forms for a total of 2467 patient visits. Although inclusion of patients with complaints of vomiting, diarrhea, or decreased oral intake was at the discretion of the treating physician, our review of daily ED records revealed that ∼90% of eligible patient visits were included appropriately. The most common reasons for inappropriate exclusion were failure of the clerical staff to attach a study form to the ED record and failure of the physicians to complete the study form. Inappropriately excluded patients did not differ from included patients on any demographic or clinical characteristics.
Physicians reported that 286 (12%) of the patient families could not speak English. For 79 (3.2%) visits, the physician reported that a family's inability to speak English did not create a barrier for them. A LB between the physician and the family was believed to be present in the remaining 209 cases (8.5%). On 159 (6.4%) of those occasions, an interpreter was involved. In 50 (2.0%) cases, a barrier was believed to exist, but no interpreter of any kind was used. The demographic and clinical characteristics of the patients for whom a LB existed are compared with the no language barrier (NLB) group in Table 1.
LB patients were far more likely to be Hispanic than NLB patients (88% vs 49%; P < .01), much less likely to have commercial insurance (19% vs 30%; P < .01), and more likely to have Medicaid (77% vs 60%; P < .01). On average, LB patients were slightly younger than NLB patients (31 months vs 36 months; P = .02). The two groups did not differ significantly on age-adjusted triage vital signs or proportion of patients <3 years old with the minimum Yale Observation Score (ie, 6). There was no difference in the proportion of patients triaged to the two highest categories (emergent and high urgent); however, in the lower categories, LB patients were more likely to be triaged as urgent low (OR: 1.4; 95% CI: 1.0, 1.8) and less likely to be triaged as nonurgent (OR: .67; 95% CI: .49, .91).
Management of patients in the two groups differed (Table 2). LB patients were more likely to be given a bolus of intravenous fluid (OR: 1.8; 95% CI: 1.2, 2.6) and slightly more likely to be admitted (OR: 1.6; 95% CI: 1.1, 2.4). LB patients stayed in the ED an average of 28 minutes longer (95% CI: 15, 42). Although a similar proportion of patients in both the LB an NLB groups had no diagnostic tests performed (44% vs 47%;P = .30), the overall mean test charge was $41 higher in the LB group ($145 vs $104; P < .01).
Because many factors may impact physician test ordering, we constructed an ANCOVA model to isolate the effect of a LB. In this multivariate analysis, race/ethnicity, insurance status, physician training level, attending physician, urgent care setting, triage category, and the presence of a LB were entered as independent variables; patient age and vital signs were added as covariates. Factors not proved to be related significantly to charges were removed subsequently (ie, race/ethnicity, age, pulse, respiratory rate, and blood pressure). Adjusting for the remaining factors, this model predicted that a $38 (F = 14.1; P < .001) difference in mean test charges and 20 (F = 9.1; P = .003) additional minutes in the ED were associated with the independent effect of a LB.
Many studies have shown that LBs decrease the quality and accessibility of health care.6–10 Non-English speaking patients are less likely to use primary and preventive services and more likely to use emergency facilities.6 ,11 ,12 In addition, LBs decrease patients' understanding of their disease processes and subsequently, impact their compliance with treatment and follow-up.11 ,13 ,14 All these effects indirectly increase the cost of health care for this population.15
To date, there has been no prospective work examining how language discordance may affect the diagnostic approach to ED patients. We found significant differences in test ordering behavior and lengths of stay when physicians believed that they were confronted by a LB. Despite controlling for multiple other demographic and clinical factors, mean test charges were $38 higher and patients remained in the ED an average of 20 minutes longer when a LB was present.
Many factors likely contributed to the overall difference in lengths of stay. Among those measured in this study were admission rates, use of intravenous fluids and diagnostic test ordering. In each instance, our outcomes suggested that LB patients utilized more hospital resources. Total length of stay in the ED served as an approximation of the combined effects of these and other differences.
The test charge results suggest that differences in physician test ordering behavior resulted in a 32% premium for the work-up of patients in the LB group. Although charges for diagnostic services may not reflect true costs, a comparison of those charges across the two groups should reflect differences in overall resource utilization. In absolute terms, more money was spent in the diagnostic evaluation of the LB group.
A possible explanation for this premium is that physicians were compensating for the diminished diagnostic power of the medical interview by increasing the intensity of laboratory and radiographic investigations.4 ,16 Providers may also have believed that LB patients, independent of socioeconomic factors, were less likely to understand and comply with follow-up instructions. Therefore, an increased degree of assurance regarding the child's condition before discharge was required.
On our study form, we left the definition of a LB necessarily vague. This allowed the provider to incorporate many factors into their determination, including their own proficiency in the patient's language. Because physician test ordering behavior was our primary interest, the provider's perception of a barrier had great relevance to our findings. However, our inability to independently verify the presence of such a barrier or test individual providers' proficiency were important limitations of our study.
We did not ask our patient families if they perceived a language barrier. It is quite possible that physicians erroneously classified some families in the NLB group, because they did not appreciate a barrier, although the families may have believed that a barrier existed. To the extent that such misclassifications occurred, we would expect this to obscure real differences in the approach to the two groups, causing our findings to underestimate the true magnitude of those differences.
The few physicians who were comfortably fluent in Spanish were slightly underrepresented in the LB group. However, this effect was likely small, as only 79 non-English speaking families (3.2% of our total study population) were classified as NLB patients rather than as LB patients because the provider could speak their language. In addition, variations in some of these physicians' practice styles were controlled for in our ANCOVA model, because attending input was included as an independent variable.
Although our cohorts of LB and NLB patients were similar, they were not identical. Objective clinical findings such as age and vital signs were comparable. Similar proportions of patients in each group were triaged to the highest categories (emergent and high urgent), suggesting that objective, physical data predominate in the identification of these most acute patients. However, in the lower categories (urgent low and nonurgent), the medical history may have weighed more heavily. It is unclear to what extent LBs between families and the nursing staff may have affected the triage designations. If, because of uncertainty regarding the history of a patient, LBs caused the nurses to assign higher categories, then we may have overestimated the complexity of our LB group. Although only partially accounted for in our ANCOVA model, this effect would have biased our results toward the null hypothesis.
If more complicated cases were overrepresented in the LB group, this might explain the variability in test ordering that we observed. Our ANCOVA model controlled for some reflections of acuity and complexity (ie, age, vital signs, and triage category), but other important differences may have gone unmeasured. By examining visits with a rather narrow range of complaints (ie, generally healthy children with documented fever or complaints of vomiting, diarrhea, or decreased oral intake), we attempted to minimize this variability. Inclusion of patients with more complex presentations and histories in both groups would be expected to result in an even larger LB premium.
It is accepted widely that well-trained, professional medical interpreters can reduce the obstacles created by LBs.4 ,13 ,17 Unfortunately, professional interpreters were available inconsistently in our ED. Even when available, it was often impractical for them to be present for the entire ED visit. Therefore, it was quite common for our providers to use other, ad hoc forms of interpretation, including nursing or clerical staff, family friends, or even other individuals present in the waiting room. The limitations of the use of ad hoc interpreters have been well-described.4 ,13 17–19 Because of the variability in the use and definition of an interpreter in our study, we chose to analyze the LB group as a whole. To the extent that these interpreters facilitated the care of LB patients, we expect our findings to be biased toward the null hypothesis and to have underestimated the true size of the LB premium.
Determination of the cost-effectiveness of professional medical interpreters will depend chiefly on three things: 1) the volume of LB patients for whose language the interpreter has been trained (this is of course institution-specific), 2) the precise size of the LB premium, 3) and the extent to which the interpreter can reduce or eliminate that premium. Our work has identified the existence of such a premium and merely estimated its magnitude. In an era of cost-consciousness, more controlled studies are needed to assess the impact that professional interpreters can have on unnecessary testing.
This study was supported in part by a Special Projects Grant from the Ambulatory Pediatric Association.
We thank Nancy Ryan and our entire ED staff for their role in implementing this study and acknowledge the assistance of Elizabeth Powell, MD, MPH; Genie Roosevelt, MD, MPH; and Karen Sheehan, MD, MPH for their assistance in the preparation of the manuscript.
- Received November 12, 1998.
- Accepted January 13, 1999.
Reprint requests to (L.C.H.) Children's Memorial Hospital, Division of Pediatric Emergency Medicine, 2356 N Lincoln Ave, No 62, Chicago, IL 60614. E-mail:
- ↵US Bureau of the Census. Statistical Abstract of the US 1990 Census. 113th ed. Washington, DC: US Bureau of the Census; 1993
- McCarthy PL,
- Lembo RM,
- Baron MA
- Rosen K,
- Sanford S,
- Scott J
- Launer J
- Ebden P,
- Bhatt A,
- Carey OJ,
- Harrison B
- Copyright © 1999 American Academy of Pediatrics