OBJECTIVE: The study goal was to determine whether preferred language for care and insurance type are associated with cost among hospitalized children.
METHODS: A retrospective cohort study was conducted of inpatients at a freestanding children’s hospital from January 2011 to December 2012. Patient information and hospital costs were obtained from administrative data. Cost differences according to language and insurance were calculated using multivariate generalized linear model estimates, allowing for language/insurance interaction effects. Models were also stratified according to medical complexity and length of stay (LOS) ≥3 days.
RESULTS: Of 19 249 admissions, 8% of caregivers preferred Spanish and 6% preferred another language; 47% of admissions were covered by public insurance. Models controlled for LOS, medical complexity, home-to-hospital distance, age, asthma diagnosis, and race/ethnicity. Total hospital costs were significantly higher for publicly insured Spanish speakers ($20 211 [95% confidence interval (CI), 7781 to 32 641]) and lower for privately insured Spanish speakers (–$16 730 [95% CI, –28 265 to –5195]) and publicly insured English speakers (–$4841 [95% CI, –6781 to –2902]) compared with privately insured English speakers. Differences were most pronounced among children with medical complexity and LOS ≥3 days.
CONCLUSIONS: Hospital costs varied significantly according to preferred language and insurance type, even adjusting for LOS and medical complexity. These differences in the amount of billable care provided to medically similar patients may represent either underprovision or overprovision of care on the basis of sociodemographic factors and communication, suggesting problems with care efficiency and equity. Further investigation may inform development of effective interventions.
In 2011, 15% of US children lived with at least 1 parent who had limited English proficiency (LEP), defined as speaking English less than “very well.”1 Language barriers in health care have been associated with decreased adherence, comprehension, and satisfaction with care,2,3 in addition to increased resource utilization in some settings,4–7 worse condition-specific outcomes,8–11 and increased risk for serious adverse events.12–15 Language barriers often co-occur with other potential barriers to high-quality care, including poverty, low health literacy, and public insurance.10 However, the degree to which these factors interact with LEP to influence utilization and outcomes is poorly understood. Understanding these interactions may help target interventions to the patient population most likely to benefit.
Health care costs may be used to identify disparities in the provision of health care by revealing differences in care patterns at the clinic, hospital, or population level. Cost serves as a proxy for the amount of billable care provided to a given patient, which should be similar for medically similar patients. Examining differences in costs on the basis of demographic factors can facilitate identification of disparities in care, which in turn facilitates investigation and intervention.
The primary objective of the present study was to examine the association between costs of hospitalization and the interaction between patient language and insurance type at a freestanding children’s hospital. We used insurance type (private versus public) as a proxy for income level because it is highly correlated with family income and was available for all patients.16 Secondary objectives included determining if cost variation was consistent across subpopulations defined according to level of medical complexity and length of stay (LOS).
Study Population and Setting
Patients aged ≤21 years discharged from the inpatient medical or surgical unit of a freestanding children’s hospital from January 1, 2011, through December 31, 2012, were eligible for this study. Patients admitted to observation status, bone marrow transplant, rehabilitation, or inpatient psychiatry were excluded because these admission types have unique patterns of resource utilization. Both medical and surgical inpatient (but not observation) admissions were included, as we expected family language, culture, and social or financial constraints to be interfacing with care delivery in similar ways for both groups. All eligible admissions were included for a given patient, clustering according to individual in the analysis. We also restricted assessments to first hospitalization in the study period in a sensitivity analysis.
The study hospital has comprehensive professional interpreter services (in-person, telephone, and video). However, the present study was unable to track the type and amount of interpreter services provided.
Predictors and Covariates
Primary Independent Variables
Caregiver-reported preferred language for medical communication was recorded in the electronic medical record at hospital registration. Patient caregivers were classified as preferring English (hereafter referred to as “English speakers”), Spanish (hereafter referred to as “Spanish speakers”), or another language.
Insurance type was obtained from hospital administrative data and categorized as private or public (ie, Medicaid). Because <1% of children hospitalized at our institution are uninsured, they were included with the publicly insured patients.
Patient race and ethnicity were collected at registration according to caregiver report. The following mutually exclusive categories were used for analysis: non-Hispanic white (“white”), non-Hispanic black (“black”), Hispanic of any race (“Hispanic”), Asian or Pacific Islander, other or mixed, and refused or unknown.
Patient medical complexity was categorized according to the Pediatric Medical Complexity Algorithm, which classifies children as having no chronic illness, noncomplex chronic illness, or complex chronic illness by using retrospective International Classification of Diseases, Ninth Revision, codes.17 The Pediatric Medical Complexity Algorithm accounts for both diagnoses and intensity of utilization, and it does not require a minimum amount of retrospective data; however, it only uses data from the previous 3 years, beginning with the date of admission.
Patient LOS was obtained from hospital administrative data. Given its skewed distribution, the variable was winsorized at the 99th percentile. Accordingly, 192 admissions with LOS >60.7 days were assigned an LOS of 60.7 days.
Patient address and geographic information systems software were used to determine distance between home address and the hospital.18 Distances from the hospital were classified as <30 miles, 30 to 60 miles, 61 to 120 miles, >120 miles, or missing. Distance provided information both about potential barriers to discharge and severity of illness, as children who reside long distances from this hospital are typically those who require tertiary and quaternary care.
We controlled for having an asthma diagnosis according to International Classification of Diseases, Ninth Revision, codes for several reasons. First, asthma is a common reason for admission that disproportionately affects lower income and minority children.19–22 Second, asthma admissions have shorter LOS and lower costs compared with other admissions, which might confound our analysis. Hospital administrative data were also used to identify the primary treating service (eg, hematology-oncology) for each admission.
Hospital charges were obtained from administrative data, including total charges and those designated as laboratory, pharmacy, and radiology. Charges for interpreter services were not included because these are not billed to families. Charges were converted to costs by using the hospital-specific cost-to-charge-ratio, then inflation-adjusted to 2012 US dollars according to the medical care component of the Consumer Price Index.23,24 Because cost data have a skewed distribution, costs within each category were winsorized at the 99th percentile, affecting 193 encounters with total cost values from $362 661 to $3 455 981.
Descriptive statistics were generated for all predictors and outcomes. Multivariate analyses used generalized linear models with a log link and γ family.25
A separate multivariate model was constructed for each cost outcome: total, pharmacy, laboratory, and radiology. The relationships between the predictors of interest (language and insurance type) and each outcome were assessed, adjusting for race/ethnicity, LOS, age, medical complexity, distance from the hospital, and asthma diagnosis; they were clustered on individual. Given the potential for collinearity between language, race/ethnicity, and insurance status, multicollinearity diagnostic analyses were performed. All variance inflation factors were <5, indicating no problematic multicollinearity within the data set.26
Interaction terms between language, insurance type, and race/ethnicity were introduced into the multivariate models. Interactions associated with the outcomes at P < .05 were retained. Marginal differences in costs, according to language and insurance type, were then predicted from the generalized linear models.
To investigate whether observed variation in costs differed according to patient medical complexity, the primary analysis of total costs was repeated after stratifying by medical complexity level: nonchronic, noncomplex chronic, and complex chronic. We also repeated the primary analysis after stratifying by LOS ≥3 days. In both cases, all covariates listed in the main analysis were controlled for, excluding the stratification variable.
To assess whether cost patterns according to demographic characteristics were due to different types of illnesses requiring hospitalization, each of 26 admitting services was assessed for association with language, insurance type, and language/insurance combinations. The 12 services that were significantly associated with any of the predictors were included as additional covariates in a sensitivity analysis, which otherwise included the variables described previously. We also reran the models looking at total costs without controlling for LOS and after restricting inclusion to each patient’s first hospitalization during the study period. Finally, we reran the analyses after winsorizing total costs and LOS at the 95th percentile (rather than the 99th percentile) to assess the influence of the most extreme values on our outcomes.
There were 19 249 hospital admissions that met study inclusion criteria, of which 47% were covered by public insurance and 14% involved caregivers preferring a non-English language for medical care (Table 1). Among admissions from Spanish-speaking families, 96% had public insurance. Overall median LOS was 2.6 days (interquartile range [IQR], 1.2 to 5.1 days; 95th percentile, 19.5 days; mean ± SD, 5.8 ± 13.5 days), and overall median hospital costs were $12 842 (IQR, $6550 to $27 011; 95th percentile, $106 195; mean, $32 542 ± $89 830).
In multivariate analysis, the interaction term between language and insurance type was statistically significant and was thus retained in the model. Because of the interaction term, the reference group for all language and insurance combinations was privately insured English speakers. In adjusted analyses, publicly insured English speakers had hospital stays that were $4841 less expensive (95% confidence interval [CI], –6781 to –2902; P < .001) (Fig 1) than the referent. Similarly, privately insured Spanish speakers had hospital stays that were $16 730 less expensive (95% CI, –28 265 to –5195; P = .004). Publicly insured Spanish speakers, in contrast, had hospital stays that were $20 211 more expensive (95% CI, 7781 to 32 641; P = .001) than the referent. There were no significant differences in total cost among either privately or publicly insured children from families preferring other languages compared with the referent.
Pharmacy, Laboratory, and Radiology Costs
In the multivariate analysis of pharmacy costs (n = 18 973), a similar pattern to overall costs was found (Fig 2). Pharmacy costs were lower for publicly insured English speakers (–$3463 [95% CI, –5228 to –1697]; P < .001) and privately insured Spanish speakers (–$11 485 [95% CI, –23 285 to 314]; P = .056) and higher for publicly insured Spanish speakers ($15 560 [95% CI, –2529 to 28 592]; P = .01) compared with privately insured English speakers. Among families preferring other languages for care, pharmacy costs were lower for those with public insurance (–$6317 [95% CI, –12 093 to –541]; P = .03).
Adjusted analysis of laboratory costs (n = 16 240) revealed lower average costs for publicly insured English speakers (–$429 [95% CI, –778 to –79]; P = .02) and higher costs for publicly insured children preferring other languages ($1351 [95% CI, 174 to 2528]; P = .02).
Analysis of radiology costs (n = 11 911) revealed lower costs for publicly insured English speakers only (–$269 [95% CI, –399 to –139]).
Stratification according to medical complexity revealed no variation in cost by insurance or language among children with no chronic illness (n = 4826). Among children with noncomplex chronic illness (n = 4754), publicly insured English speakers had significantly less expensive hospital stays (–$6240 [95% CI, –12 275 to –205]; P = .01), but there were no differences detected among non-English speakers. Among children with complex chronic illness (n = 9669), results mirrored those of our primary analysis, with less expensive hospitalizations for publicly insured English speakers and privately insured Spanish speakers, and more expensive hospital stays for publicly insured Spanish speakers (Fig 3). Stratification of the main analysis according to LOS ≥3 days revealed no differences by group among those with a short stay (Table 2). Among those with LOS ≥3 days, there were no differences for publicly insured English speakers, whereas patterns for Spanish speakers were similar to the primary analysis.
Results for the total cost model were unchanged after controlling for 12 clinical service lines or restricting to the first admission per patient during the study time period (data not shown). Results were similar, with similar to slightly attenuated effect sizes, when costs and LOS were winsorized at the 95th percentile rather than the 99th percentile (Supplemental Information). When not controlling for LOS, the total cost findings for both publicly and privately insured Spanish speakers were more pronounced, whereas the findings for publicly insured English speakers became nonsignificant (Table 2).
In this study of 19 249 hospital admissions, patterns of hospital costs varied significantly according to insurance type and preferred language for medical communication, even after controlling for potential confounders (including medical complexity, LOS, hospital service, and distance between the child’s home and hospital). Compared with privately insured English speakers, publicly insured English speakers and privately insured Spanish speakers had less costly hospital stays, whereas publicly insured Spanish speakers had more expensive stays. These patterns were most pronounced among children with hospital stays ≥3 days in length and those with complex chronic illness. Pharmacy and laboratory costs generally mirrored total cost patterns, with less variation in radiology costs. These results suggest there were differences in the amount of billable care provided to hospitalized children on the basis of demographic, rather than clinical, characteristics.
For publicly insured English speakers, we found lower hospital costs compared with their privately insured counterparts in all cost categories. However, the cost differences were not significant when we stratified costs according to LOS or did not adjust for LOS. Because LOS is the largest contributor to hospital costs, this finding suggests that publicly insured English speakers may have been staying longer but using fewer resources per day. This lower intensity of resource utilization may be due to several factors. For example, perceived barriers to discharge (eg, lack of transportation and/or access to follow-up care for publicly insured patients) may lead to keeping a child in the hospital longer than might be strictly medically necessary. Previous studies have found longer LOS among children with Medicaid in populations hospitalized for spinal fusion,27 infections,7 and asthma.28 For example, Glick et al28 reported longer LOS and similar costs (adjusted for LOS) among Medicaid-insured children with asthma; however, they found low-income patients to have longer LOS but lower LOS-adjusted costs, similar to our findings. Another mediating factor may be lower parent engagement among publicly insured families, leading to less parental advocacy for additional clinical diagnostics and intervention. Previous studies have found that minority and/or low-income children were less likely to receive potentially inappropriate antibiotic prescriptions, suggesting less parental demand for unnecessary treatment.29,30 However, the present analysis cannot determine whether there was underutilization for publicly insured patients, overutilization for privately insured patients, or some combination of both.
Hospital stays among privately insured Spanish speakers were less expensive. Although their caregivers registered as preferring Spanish for medical communication, the fact that these children had private insurance suggests that at least 1 parent may have been English-proficient, as the majority of private insurance at the time was employer based,31 and most jobs offering insurance likely required some English proficiency. The language barrier for these families may have been lower than for publicly insured Spanish speakers, but their hospital costs still significantly differed from their privately insured English-speaking counterparts. This finding highlights a central difficulty in identifying LEP families in pediatrics. Whose English proficiency matters? Even if we know that caregivers’ English proficiencies differ, how can we know which caregiver was at the bedside, and when? Controlling for LOS attenuated but did not adjust away the difference, suggesting a shorter stay contributed to but did not entirely explain the cost findings. These results could be accounted for by less parental advocacy, perhaps informed by the cultural value among Latino subjects of respect for authority figures,32 but without the delays to discharge that are associated with public insurance. These lower cost findings were primarily driven by children with medical complexity and longer LOS. Because many of the children without medical complexity at our institution receive medical care that is standardized according to diagnosis,33 there may have been fewer opportunities to provide unequal care based on family characteristics or provider biases for those children.
Publicly insured Spanish speakers had more expensive hospital stays, likely reflecting the impact of language barriers. As seen in previous studies, parental LEP may result in providers performing more tests, trying more treatments, or observing patients for longer periods to compensate for incomplete information and poor communication.4,5,7,34 These more expensive stays may also reflect poorer access to outpatient care or delayed presentation,10 although controlling for distance from home to hospital likely attenuated those associations. As we found among privately insured Spanish-speakers, these cost differences were driven by children with medical complexity and prolonged LOS, suggesting that language barriers, cultural factors, and provider biases are most likely to affect care in visible ways when that care is more complicated or less evidence-based and standardized.
We found no associations between total cost, insurance type, and preferring a non-English, non-Spanish language, likely because the “other” group was a mix of many smaller languages and cultures, each too small to evaluate individually. Consolidation of these groups for analysis, while presently unavoidable, may be obscuring associations. Evaluation with larger samples is needed.
Previous studies examining language and hospital utilization have reported mixed results, with some finding increased LOS5,7 or resource utilization,6,35 generally within more narrowly defined diagnosis groups, and some finding none.36 Public insurance and/or other markers for socioeconomic status have also been linked to increased LOS and, in some cases, increased resource utilization (including costs).37–39 In the present study, because we included children with many conditions, we chose cost as our outcome (rather than utilization of specific resources) because it provides a measure of all billable medical care that was provided during a hospitalization. We are unaware of other studies successfully able to examine the joint impact of language and insurance type on cost or utilization. Levas et al,7 in their study of children hospitalized with infections, found increased LOS associated with both parental LEP and public insurance but failed to find statistical interaction between the 2 factors; however, they were likely underpowered for such an analysis, with only 39 LEP families, of whom only 3 were privately insured. Failure to consider interactions between the multiple barriers to full engagement with the health care system that a family faces may mask the effects of factors that are exerting simultaneous, opposing pressures. For example, because the privately and publicly insured Spanish speakers had different cost patterns, assessment of cost according to language alone might miss important differences that deserve evaluation. Aside from language and insurance, other factors may create barriers to receiving equitable care, such as low health literacy, limited self-efficacy, and lack of trust in the system; these factors should also be considered when targeting interventions to improve equity.
This study had several limitations. It was conducted at a single institution, and thus the results may not be generalizable. However, preferred language is generally not available in multi-institution data sets, as few institutions routinely collect or report this information. We were unable to account for the amount and type of professional interpreter services provided in our analysis; costs may have differed by the degree to which the language barrier was effectively bridged. Another limitation was our use of insurance type as a proxy for family income. We used insurance, rather than census tract income data, because we had this information for all participants but were missing home address data for nearly 20% of subjects (mostly post office box addresses). Although insurance type provides some idea of family income for many families, it may misclassify children with public insurance due to medical complexity, and it fails to account for additional elements of socioeconomic status such as parental education. It is also possible that the identified cost differences were driven by medical needs, or the continuation of expensive home medications, rather than demographic characteristics; however, our findings were robust to adjustment for a variety of markers of complexity and illness type. It should also be noted that LOS is central to overall costs, pharmacy costs (as most medications are given daily), and, to a lesser extent, laboratory costs; although all of our analyses controlled for LOS, differences in LOS likely remained an important driver of those outcomes. Finally, it is unclear whether the observed cost differences in this study reflect overutilization or underutilization.
We found that costs of pediatric hospitalizations varied significantly based on the child’s insurance type and the family’s preferred language for care, even after controlling for LOS and medical complexity. This finding suggests that disparities may exist in the provision of medical care on the basis of demographic characteristics. Poor access to outpatient care may lead to requiring more services when children are inpatients; however, lower LOS-adjusted costs were found for publicly insured English speakers and privately insured Spanish speakers, and only publicly insured Spanish speakers had more expensive stays. These differences between language and insurance groups in the amount of billable care being provided require additional investigation to determine whether they reflect overutilization, underutilization, or both. In addition, the relative contributions of language barriers, health literacy, parental activation and advocacy, and provider bias should be explored. Identifying patterns of disparate care and their causes is essential for development of interventions to improve the equity and efficiency of inpatient pediatric care.
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: No external funding was obtained for this work. Dr Lion’s time was supported by the National Institute of Child Health and Human Development (grant K23 HD078507; Principal Investigator, Dr Lion). Dr Wright’s time was supported by the University of Washington Institute for Translational Health Sciences (UL1 TR000423). Dr Desai’s time was supported by the Agency for Healthcare Research and Quality (grant 1 K08 HS024299-01; Principal Investigator, Dr Desai). Funded by the National Institutes of Health (NIH).
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
- ↵Migration Policy Institute. The limited English proficient population in the United States. Available at: www.migrationpolicy.org/article/limited-english-proficient-population-united-states. Accessed July 26, 2015
- Flores G,
- Abreu M,
- Tomany-Korman SC
- Flores G,
- Tomany-Korman SC
- Bartlett G,
- Blais R,
- Tamblyn R,
- Clermont RJ,
- MacGibbon B
- Cohen AL,
- Rivara F,
- Marcuse EK,
- McPhillips H,
- Davis R
- Divi C,
- Koss RG,
- Schmaltz SP,
- Loeb JM
- Lion KC,
- Rafton SA,
- Shafii J,
- et al
- Crimmel B
- Simon TD,
- Cawthon ML,
- Stanford S,
- et al
- ↵Environmental Systems Research Institute. ArcGIS Desktop: Release 10. Available at: www.arcgis.com
- Halfon N,
- Newacheck PW
- ↵Bureau of Labor Statistics. Consumer Price Index. United States Department of Labor. Available at: http://data.bls.gov/timeseries/CUUR0000SAM. Accessed February 9, 2016
- Luce B,
- Manning W,
- Siegel J,
- Lipscomb J
- Cho SK,
- Egorova NN
- Glick AF,
- Tomopoulos S,
- Fierman AH,
- Trasande L
- Fleming-Dutra KE,
- Shapiro DJ,
- Hicks LA,
- Gerber JS,
- Hersh AL
- Fronstin P
- Lion KC,
- Wright DR,
- Spencer S,
- Zhou C,
- Del Beccaro M,
- Mangione-Smith R
- Hampers LC,
- Cha S,
- Gutglass DJ,
- Binns HJ,
- Krug SE
- Porter M,
- Diaz VA,
- Gavin JK,
- et al
- Duquette S,
- Soleimani T,
- Hartman B,
- Tahiri Y,
- Sood R,
- Tholpady S
- Copyright © 2017 by the American Academy of Pediatrics