Sociodemographic disparities in epilepsy care: Results from the Houston/New York City health care use and outcomes study
Address correspondence to Charles E. Begley, The University of Texas Health Science Center at Houston, School of Public Health, PO Box 20186, Houston, TX 77225, U.S.A. E-mail: email@example.com
Purpose: The purpose of this study was to identify sociodemographic disparities in health care use among epilepsy patients receiving care at different sites and the extent to which the disparities persisted after adjusting for patient characteristics and site of care.
Methods: Three months of health care use data were obtained from baseline interviews of approximately 560 patients at four sites. One-half of the patients were from a Houston site and two NYC sites that serve predominantly low-income, minority, publicly insured, or uninsured patients. The other half were at the remaining site in Houston that serves a more balanced racial/ethnic and higher sociodemographic population. Differences in general and specialist visits, hospital emergency room (ER) care, and hospitalizations were associated with race/ethnicity, income, and coverage. Logistic regression was used to assess the extent to which the differences persisted when adjusting for individual patient characteristics and site of care.
Results: Compared to whites, blacks and Hispanics had higher rates of generalist visits [odds ratio (OR) = 5.3 and 4.9, p < 0.05), ER care (OR = 3.1 and 2.9, p < 0.05) and hospitalizations (OR = 5.4 and 6.2, p < 0.05), and lower rates of specialist visits (OR = 0.3 and 0.4, p < 0.05). A similar pattern was found related to patient income and coverage. The magnitude and significance of the disparities persisted when adjusting for individual characteristics but decreased substantially or were eliminated when site of care was added to the model.
Discussion: There are sociodemographic disparities in health care for people with epilepsy that are largely explained by differences in where patients receive care.
A number of studies provide evidence that minorities with epilepsy in the United States receive different amounts of health care compared to nonminority whites (see literature reviews by Szaflarski et al., 2006 and Theodore et al., 2006). This is a concern for advocates and policy makers who seek to eliminate inequalities in epilepsy care and improve the health of high-risk populations. However, there are no studies examining the extent to which the disparities can be explained by such factors as differences in individual patient characteristics or variations in provider practices. We are conducting a 1-year longitudinal study of epilepsy care at four sites, two in Houston and two in New York City, to identify sociodemographic disparities in epilepsy care and what factors might explain the disparities. The questions we address include:
- • Do disparities in care exist with respect to sociodemographic factors such as race/ethnicity, income, and insurance coverage?
- • To what extent are disparities explained by clinical and other personal characteristics of patients?
- • To what extent are disparities associated with practice settings where patients seek care?
Answers to such questions are needed to help inform program administrators, providers, and policy makers of the underlying reasons for disparities so that effective strategies might be designed to reduce or eliminate them.
This report describes the methods used to conduct the study and provides preliminary results based on initial baseline data from the patient sample.
The magnitude of the differences in patterns of care that are reported in the epilepsy literature varies by type of service. With respect to hospital care, a U.S. Centers for Disease Control and Prevention (CDC) study of National Hospital Discharge Survey data during 1988–1992 indicated that the rate of hospitalization for minority race/ethnicity groups was higher than whites (51 vs. 35 per 100,000) (CDC, 1995). This study was based on a nationally representative survey of hospital admissions and used the International Classification of Diseases, Ninth Revision, Clinical Modification, codes 345.0–345.9 to identify epilepsy. The sample data were extrapolated to all hospital facilities in the country and age-adjusted to the 1980 U.S. resident population to derive national age- and race-specific hospitalization rates. Age-specific race/ethnicity rates were similar for the youngest age group, but rates were higher for nonwhites in other age groups.
A study at an outpatient pediatric epilepsy clinic in Mississippi compared 91 patients with undetectable antiepileptic drug (AED) blood levels (indicating several consecutive doses missed) to a comparison group of 100 patients verified as compliant by acceptable blood levels (Snodgrass et al., 2001). The noncompliant group had significantly fewer whites (only 9 of 91) and insured patients (only 9 of 91) compared to the compliant group (44 whites of 100, 32 with insurance of 100). There were no differences found between the two groups in terms of patient age, seizure type, and seizure control.
Two studies examining surgery use have shown that Hispanics and blacks are significantly less likely than non-Hispanic whites to undergo epilepsy surgery. In a study of 565 surgery candidates who received standardized presurgical evaluations at six surgical centers in the Northeast and Midwest United States and were determined to be eligible for respective surgery, Berg et al. (2003) found that race/ethnicity and marital status were associated with surgery. Bivariate analyses determined that Hispanic and black patients were 62% and 63%, respectively, less likely to have surgery compared to whites and Asians (73% and 74%, respectively) (overall p-value < 0.05).
In a retrospective study of surgical candidates discharged from an epilepsy surgery center in Alabama, the characteristics of 130 surgical candidates with temporal lobe epilepsy and medial temporal sclerosis determined with magnetic resonance imaging (MRI) who had surgery were determined (Burneo et al., 2005). Using logistic regression modeling to control for differences in income, insurance coverage, and clinical characteristics, blacks were found to be 70% less likely than non-Hispanic whites to undergo surgery [odds ratio (OR) = 0.3; 95% confidence interval (CI), 0.2–0.8].
A 2004–2005 prevalence study in a multiracial, multiethnic community in New York City (NYC) identified racial/ethnic disparities both in drug treatment and the use of emergency room (ER) care (Kelvin et al., 2007). The study used random-digit dialing to identify people with epilepsy in the Washington Heights/Inwood/Central Harlem area of NYC. The majority of cases reported receiving care for epilepsy in a neurology clinic (29% among active cases and 31% among lifetime cases) or from a primary care physician in private practice (24% among active cases and 23% among lifetime cases). Overall 20% of active cases and 18% of lifetime cases indicated their source of care was the ER. No whites reported the ER as their source of care, whereas 50% of blacks and 20% of Hispanics indicated they sought care in the ER. About 52% of the people with active epilepsy were taking AEDs at the time of the interview. Eighty percent of Hispanics but only 50% of both blacks and whites with active epilepsy reported they were taking AEDs.
A number of studies in other disease areas have examined the extent to which factors such as regional variations, local health markets, health care settings, and provider practices can explain the disparities in health care use (see review by Zaslavsky & Ayanian, 2005). A recent study by Hasnain-Wynia et al. (2007) examined quality of care related to treatment of acute myocardial infarction, congestive heart failure, and community acquired pneumonia in hospitals across the country. In a multivariate regression model, they found differences in 13 of 17 condition-specific quality-of-care measures comparing minority and nonminority patients. When they introduced a variable distinguishing hospitals that served mostly minority patients from hospitals that served mostly nonminority patients, the differences were largely eliminated. Similar findings were reported by Barnato et al. (2005) comparing minority/nonminority mortality from heart attacks in U.S. hospitals, and Howell et al. (2008) comparing very-low-birth-weight neonatal mortality among minority/nonminority infants in NYC. In the Howell et al. study, black babies had higher mortality than whites and were also more likely to be born in hospitals that had higher neonatal death rates.
Contrary evidence was recently provided by Sequist et al. (2008) comparing outcomes of patients with diabetes mellitus. Outcome measures of almost 7,000 patients treated by 90 primary care physicians in a multispecialty group practice in eastern Massachusetts were examined using electronic medical record data and linear regression models. The baseline model determined that white patients were significantly more likely than black patients to be in control of hemoglobin A1e, low-density lipoprotein cholesterol, and blood pressure. In models adding sociodemographic, clinical, physician, and health care center variables, physician effects (differences within the same physician’s panel) accounted for 66–75% of the race/ethnicity disparities.
We have employed similar methods to examine sociodemographic disparities in care among epilepsy patients and determine the relative importance of patient and provider characteristics in explaining the differences.
In measuring disparities in health care it is useful to consider the variety of different environmental, behavioral, and health system factors that can act independently and together to influence patterns of health care use and outcomes (Andersen & Aday, 1978; Andersen, 1995; Shi & Stevens, 2005). Predisposing factors include environmental and patient characteristics such as age, gender, race/ethnicity, marital status, education, occupation, environmental risk factors, and health attitudes and beliefs. These factors affect a patient’s health care–seeking behavior and are largely immutable.
Enabling factors refer to patient and health system characteristics that affect the means for obtaining health care services. They include resources for accessing services such as income, health insurance coverage, transportation, and language proficiency. They also include health system characteristics such as the availability of resources and the training and practice styles of providers, which may vary on a regional basis or over time. These factors are often targeted by health policies and programs aimed at improving access.
Professional and patient perceptions of need are also important predictors of health care use. Need characteristics include the severity of illness, presence of comorbidities, and quality-of-life, among others, that influence patient and provider choices independently and in connection with predisposing and enabling factors.
Using this sociobehavioral framework of health care use, disparities are defined when environmental, sociodemographic, or health system factors that predispose or enable patients to get treatment are more important than need factors in explaining who gets care. In measuring health care disparities, population surveys or administrative databases are used to describe health service use and examine associations between these factors and use.
We have recruited and are surveying patients who are attending clinics at four sites: two in Houston and two in NYC. At baseline we collected data on patients’ sociodemographic characteristics, clinical conditions, epilepsy-related health care use, and quality of life. At quarterly follow-up interviews, patients were asked to report the number of seizures they experienced during the prior 3 months and their health care use related to epilepsy. The exit interview was similar to the baseline: asking patients to provide information on sociodemographic characteristics, clinical conditions, health care use, and quality of life. In addition to the interviews we reviewed medical charts and collected billing data on a subset of the cohort to verify subject’s recall of care and to develop a cost model.
All patients who met the inclusion criteria were recruited for the study during their regular visits at each site for a period of 1 year. Participants were limited to patients who were 12 years of age and older, and had no progressive cerebral disease or any other progressively degenerative neurologic disorder. Clinicians identified patients who met the inclusion criteria and offered them or their surrogates the chance to participate in the study. Interested patients or surrogates were referred to a private setting where a trained interviewer explained the study, enrolled patients, and conducted the interviews. A financial incentive of $10 per interview was offered for participation. Interviews were conducted on a quarterly basis in a uniform manner by trained interviewers. Questionnaires were administered either during clinic visits or by telephone. They were conducted in English and Spanish.
The sites were selected because they are in two different regions of the country and serve sociodemographically diverse populations. One of the Houston sites and both NYC sites serve predominantly low-income, minority, publicly insured, or uninsured communities. The remaining Houston site serves a more balanced racial/ethnic and higher level socioeconomic population.
Kelsey-Seybold Clinic is the largest multispecialty medical organization in Houston, with 22 clinics and more than 300 physicians. Care is provided to an ethnically diverse population of more than 325,000 patients. Patients are largely from middle-class, employed, populations with private insurance coverage primarily through HMO- or PPO-type plans. Epilepsy patients seeking neurologic care are referred to a specialty clinic location where there are three general neurologists, one epileptologist, and a nurse epilepsy specialist. Access to the epileptologist is possible by appointment 5 days a week. Neurosurgeons and pediatric neurologists affiliated with Kelsey-Seybold are also available through referral. The neurology department cares for approximately 400 patients with epilepsy per year, aged 3 and older.
The Ben Taub General Hospital (BT) is one of two public hospitals in the Harris County Hospital District (HCHD) that serve approximately 275,000 primarily low-income, uninsured and Medicaid-covered patients a year in Houston. The hospital-based epilepsy clinic, which is in the department of neurology, primarily serves Hispanics and black adults. The clinic is open 1 day each week for regular patients and 1 day for new patients. Approximately 30–40 patients are seen on a typical clinic day by medical residents and students who rotate through the clinic under the supervision of physician faculty. Many patients are also managed by a primary physician in one of 11 community health centers operated by the HCHD system.
The Harlem Hospital Center (HH) and its outreach clinics are affiliated with Columbia University in NYC. They are the primary sources of medical care for central Harlem, located in Upper Manhattan. The area contains about 200,000 people with the following racial/ethnic makeup: 59% black, 33% Hispanic, and a much smaller proportion of Asian and white individuals. Most residents of this region are receiving public assistance and utilize Medicaid as their primary source of health care insurance. The seizure clinic at Harlem Hospital serves about 230 adult patients, convenes one afternoon per week, and is staffed by four attending physicians. The general neurology clinic, which also sees epilepsy patients, convenes one afternoon per week and is staffed by six attending physicians, neurology residents, and medical students on a rotating basis. The referral basis for the clinics includes the outreach clinics of the Harlem Hospital system, other inpatient and outpatient departments at Harlem Hospital, and the Harlem Hospital emergency room.
New York-Presbyterian Hospital (CP) is affiliated with Columbia University in NYC. Although it is a tertiary-care center, the epilepsy clinic within the hospital primarily serves the Washington Heights/Inwood community, which comprises about 270,000 residents, with the following racial/ethnic makeup: 71% Hispanic (predominantly Caribbean Hispanic), 19% black. Most patients have Medicaid coverage (>90%) and are receiving income assistance. This clinic convenes one afternoon per week and is staffed by two attending physicians, and two to four epilepsy fellows and neurology residents on a rotating basis. The clinic services about 230 adults and 25–30 pediatric patients. The referrals include the other Medicaid clinics at CP, the hospital ER, and occasionally private physicians in the community.
The questionnaire was developed for this study to obtain sociodemographic characteristics of patients including age, gender, income, health insurance coverage, education, marital status, housing, and employment. There were also a series of questions on seizure frequency and type that ask whether a patient had “small” (partial, absence, or myoclonic) and/or “big” (generalized) seizures during the prior 3 months and, if so, how many seizures they had during each month. Health care use questions were asked including whether the patient received over the prior 3 months hospital ER services, visited a general practice or family doctor, visited a neurologist or epilepsy specialist, had been admitted to a hospital overnight as an inpatient, or received specified tests or investigations because of their epilepsy. Patients were also asked to list the AEDs they had taken and the number of days of use over the prior 3 months. These service items were selected because they are readily recalled by most patients; provide a fairly complete picture of the types of health care commonly used by patients with epilepsy; and are influenced by access issues, adherence, and changes in medical condition. Patients were also asked a series of questions regarding drug side effects and other chronic illnesses and conditions. Finally, there was a series of questions related to quality of life that were drawn from existing validated scales including the SF-36 Depression Scale (Ware & Sherbourne, 1992), the Jacoby Stigma Scale (Jacoby, 1994), the World Health Organization Health Care Responsiveness Scale (Valentine et al., 2003), and the Liverpool Quality of Life Battery (Jacoby et al., 1993).
English and Spanish versions of the questionnaire were piloted with a total of 40 patients, 12 from KS, 12 patients from BT, and 8 each from CP and HH. Sample patients were selected from consecutive patients visiting the clinic on randomly selected days who met inclusion criteria. Patients were approached by the clinicians and asked to participate. Consent of the patient was obtained before conducting the interview. During the pilot interviews the interviewer would circle and note problematic questions. Following the interview, patients were asked to evaluate the clarity, completeness, and difficulty of the questions. An incentive of $10 was given to the patient at the end of the interview. The investigators analyzed the responses and modified the questionnaire and procedures to address issues that arose in the pilot.
For this study, we examined self-reported health care use for the 3 months prior to the baseline interview to determine if there are disparities in patterns of care associated with sociodemographic characteristics. Bivariate statistics were derived comparing differences in the likelihood of receiving a particular type of service (generalist visit, specialist visit, ER care, hospitalization, and new-generation AED use) by race/ethnicity, income, and insurance coverage.
We then estimated logistic regression models to determine the relative importance of patient characteristics, need factors, and site of care in explaining the sociodemographic differences in health care use. Three models were estimated of the likelihood of patients receiving a particular service. Model 1 examined the magnitude and significance of health care disparities with regard to specific sociodemographic factors (race/ethnicity, income, or type of coverage) represented as dummy variables. Model 2 examined the importance of sociodemographic disparities after adjusting for the effects of other individual characteristics (age and gender) and clinical characteristics (seizure frequency and comorbid illness). Model 3 determined the magnitude and significance of sociodemographic disparities after adding a dummy variable to Model 2 for whether the patient was treated at KS (the Houston site serving a largely middle-class population) versus a non-KS site (the Houston and NYC sites serving lower sociodemographic communities).
About half of the 563 patients enrolled in the study were under treatment at KS (287) in Houston and half were seen at the three university-affiliated sites serving lower sociodemographic populations in Houston and NYC (276) (Table 1). There was a total of 13 new cases. The rest were prevalent cases. The average age was 42 with age at onset of 23. The patients were of relatively evenly mixed race/ethnicity and about half had some college or college degrees. Almost 40% were married and almost 60% were unemployed.
Table 1. Predisposing characteristics by site
| White non-Hispanic||35.2||61.3||8.0||12.1||3.9||3.4|
| Black non-Hispanic||27.7||20.6||35.1||39.3||15.6||50.9|
| <High school degree||21.8||5.6||39.0||36.7||41.7||41.4|
| High school degree||26.2||20.4||32.3||33.1||31.9||31.0|
| College degree||13.4||21.4||4.8||2.9||8.3||5.2|
| Graduate school ||8.7||12.6||4.5||2.9||4.2||8.6|
| Not married||19.0||13.3||25.0||28.2||19.7||24.6|
| Never married||41.8||37.9||45.9||41.5||52.6||47.4|
| Employed fulltime||32.3||50.5||13.1||11.9||5.2||27.3|
| Employed part time||8.2||8.2||8.2||11.9||3.9||5.5|
Compared to the KS patients, the patients at the non-KS sites were significantly younger, mostly black and Hispanic, had less education, were not married, and were more frequently unemployed. There were also differences in the characteristics of patients among the three non-KS sites with respect to some of these characteristics, but the differences were less frequent and much smaller in magnitude than those comparing KS and non-KS patients.
There were significant differences between the patients at the KS and non-KS sites with respect to family income and insurance coverage (Table 2). The patients in the non-KS sites were disproportionately low-income, with 76% living in poverty compared to 13% of KS patients. Only 3% of the patients at the non-KS sites had commercial insurance compared to 74% at KS. There was a difference among the non-KS sites in patient coverage, in that the majority of the patients in NYC had Medicaid coverage (75% at Columbia Presbyterian and 55% at Harlem), whereas at BT only 22% were covered by Medicaid and almost two-thirds (65%) were uninsured. This difference reflects the variability in Medicaid coverage of low-income populations between the two states.
Table 2. Enabling characteristics by site
| <100% FPL||41.9||13.4||76.2||76.3||75.5||76.7|
| 100–<200% FPL||17.9||19.1||16.5||17.8||15.1||14.0|
| 200–<400% FPL||18.1||28.2||6.1||5.2||5.7||9.3|
| ≥400% FPL||22.1||39.4||1.3||0.7||3.8||0.0|
The patients at the non-KS sites had greater need for services as reflected by higher seizure frequency (21 per month at non-KS vs. 12 at KS), lower seizure remission (19% no seizures in last year at non-KS vs. 41% at KS), and greater number with other chronic illness (36% at non-KS vs. 26% at KS) (Table 3). There were no differences in seizure type or age at diagnosis, and there were no significant differences in the quality-of-life scores across the sites as measured by the Liverpool Quality of Life Battery (Jacoby et al., 1993).
Table 3. Clinical conditions by site
|Number of seizures in last 3 months (SE)a,b||16.2 (2.8)||11.6 (2.1)||21.2 (5.5)||22.2 (6.1)||31.2 (14.7)||2.1 (0.5)|
| Partial, absence, myoclonic ||29.3||28.2||30.5||24.3||34.2||40.4|
| Generalized ||24.3||23.0||25.7||32.4||14.5||24.6|
|Age at diagnosis||22.5||22.2||22.8||21.9||22.2||25.9|
|Years of epilepsy||18.7||18.4||19.0||19.1||19.2||18.2|
|Seizures in last yeara,b|
| None ||29.9||40.6||18.7||13.7||16.0||33.9|
| One or >/month ||36.3||26.2||46.9||51.1||53.3||28.8|
| Don’t know||4.5||0.7||8.4||14.4||4.0||0.0|
|Liverpool Quality of Life Batteryb||3.2||3.2||3.2||3.1||3.5||3.3|
|Presence of comorbid conditiona,b||31.0||25.8||36.4||28.6||52.6||33.9|
Unadjusted sociodemographic disparities in health care use were found by race/ethnicity, income, and coverage for most of the services (Table 4). Over the 3-month period, minority status was significantly associated with higher rates of generalist visits, ER care, and hospitalization, and lower rates of specialist visits. Similar patterns were found comparing lower and upper income groups (Table 4, Income columns). Compared to patients with higher incomes, lower-income patients tended to have higher rates of generalist visits, ER care, and hospitalization. Lower-income groups had lower rates of specialist visits and higher rates of new generation AED use. Patterns of care were also associated with insurance coverage (Table 4, Insurance columns). Compared to privately insured patients, those with Medicare, Medicaid, combination of both, and the uninsured, had higher rates of generalist visits, ER care, and hospitalizations, and, except for those with combination coverage, lower rates of specialist visits.
Table 4. Health care use by sociodemographic characteristics
| Generalist visitsa||6.6||27.3||23.9||20.0||–|
| Specialist visitsa||93.4||81.8||83.5||90.0||–|
| ER visitsa||9.6||24.5||23.3||20.0||–|
| Hospital admissionsa||2.5||12.3||13.8||5.0||–|
| New AEDs||62.5||64.5||65.9||70.0||–|
| ||>400% FPL||200–<400% FPL||100–<200% FPL||<100% FPL||–|
| Generalist visitsa||5.4||5.4||8.9||29.9||–|
| Specialist visitsa||93.8||90.2||85.6||81.0||–|
| ER visitsa||11.6||9.8||14.3||28.3||–|
| Hospital admissionsa||2.7||5.4||8.9||11.4||–|
| New AEDsa||51.4||55.6||68.6||69.6||–|
| Generalist visitsa||2.7||39.7||20.0||29.8||18.3|
| Specialist visitsa||94.5||86.5||85.0||94.7||67.0|
| ER visitsa||9.6||32.5||17.5||17.5||21.9|
| Hospital admissionsa||4.1||16.0||10.0||8.8||9.6|
| New AEDs||58.1||67.7||59.0||71.9||69.3|
Bivariate logistic regressions (Model 1) showed the magnitude of the disparities in terms of unadjusted ORs and 95% CIs for each service type (Table 5). Compared to whites, minorities had higher odds of generalist visits (range from 3.6–5.3, p < 0.05 for all groups), ER care (range from 2.4–3.1, p < 0.05 for two of three groups), and hospitalization (range from 2.0–6.2, p < 0.05 for two of three groups), and lower odds of specialist visits (range from 0.3–0.6, p < 0.05 for two of three groups). There were no differences with regard to being on new generation AEDs (ORs ranged from 1.1–1.4, all group differences not significant).
Table 5. Bivariate odds ratios of health care use by sociodemographic characteristics
| Black non-Hispanic||5.3a (2.7–10.4)||0.3a (0.2–0.6) ||3.1a (1.7–5.6)||5.4a (2.0–14.9)||1.1 (0.7–1.7)|
| Hispanic ||4.5a (2.3–8.6) ||0.4a (0.2–0.7)||2.9a (1.6–5.1)||6.2a (2.3–16.5)||1.2 (0.8–1.8)|
| Other (ref = white non-Hispanic)||3.6a (1.0–12.2)||0.6 (0.1–3.0)||2.4 (0.7–7.8)||2.0 (0.2–18.3)||1.4 (0.5–3.8)|
| 200–<400% FPL||1.0 (0.3–3.4)||0.6 (0.2–1.7)||0.8 (0.3–2.0)||2.1 (0.5–9.0)||1.2 (0.7–2.1)|
| 100–<200% FPL||1.7 (0.6–5.2)||0.4 (0.2–1.0)||1.3 (0.6–2.9)||3.5 (0.9–13.8)||2.1a (1.1–3.7)|
| <100% FPL (ref ≥ 400% FPL)||7.5a (3.1–18.0)||0.3a (0.1–0.7)||3.0a (1.6–5.7)||4.7a (1.4–15.9)||2.2a (1.3–3.5)|
| Medicare||23.4a (9.6–56.7)||0.4a (0.2–0.8)||4.5a (2.5–8.2)||4.4a (2.0–10.1)||1.5 (1.0–2.4)|
| Medicaid||8.9a (2.9–27.3)||0.3a (0.1–0.9)||2.0 (0.8–5.1)||2.6 (0.8–8.9)||1.0 (0.5–2.1)|
| Combination||15.1a (5.6–40.6)||1.0 (0.3–3.8)||2.0 (0.9–4.5)||2.2 (0.7–7.0)||1.9 (1.0–3.5)|
| None (ref = private)||7.9a (3.1–20.6)||0.1a (0.1–0.2)||2.6a (1.4–5.0)||2.5 (1.0–6.3)||1.6 (1.0–2.7)|
Compared to patients with incomes above 400% of the federal poverty level (FPL), patients living in poverty (income <100% of the FPL) were 7.5 times more likely to have received a generalist physician visit (p < 0.05), 0.3 as likely to receive a specialist visit (p < 0.05), 3.1 times more likely to visit a hospital ER (p < 0.05), and 4.7 times more likely to have been admitted to a hospital (p < 0.05). Their odds of being on new AEDs was also higher than upper income groups (unadjusted OR 2.2, p < 0.05).
The pattern of disparities was comparable for Medicaid and uninsured groups compared to those with private insurance. Patients who were uninsured or had Medicaid coverage had a greater likelihood of generalist visits (unadjusted OR 7.9–8.9, p < 0.05), lower likelihood of specialist visits (unadjusted OR 0.1–0.3, p < 0.05), greater likelihood of ER visits (unadjusted OR 2.7–4.6, p < 0.05), similar likelihood of hospitalizations (unadjusted OR 2.5–2.6, not significant for either group), and similar likelihood of being on new AEDs (unadjusted OR 1.6–1.0, nonsignificant for all groups).
When adjusting for individual clinical characteristics (seizure frequency and presence of another chronic condition besides epilepsy) and other patient characteristics (age and gender) (Model 2), the magnitudes and significance of the ORs of health care use by sociodemographic group remained about the same (Table 6). Compared to whites, blacks and Hispanics continued to have significantly higher odds of generalist visits, ER visits, and hospital admissions, and lower likelihood of specialist visits. Lower-income patients continued to have higher odds of having generalist visits and ER visits, lower odds of specialist visits, and higher odds of new-generation AEDs. Compared to those with private coverage, the uninsured had significantly higher likelihood of generalist visits and ER visits, and lower odds of specialist visits. Those with Medicaid had significantly lower odds of specialist visits compared to the privately insured.
Table 6. Multivariate odds ratios of health care use by sociodemographic characteristics adjusted for clinical characteristics, age, and gender
| Black non-Hispanic||4.9a (2.3–10.3)||0.3a (0.1–0.7)||3.3a (1.7–6.4)||5.7a (2.0–16.2)||1.0 (0.6–1.6)|
| Hispanic||5.9a (2.8–12.3)||0.3a (0.2–0.8)||3.0a (1.6–5.7)||6.4a (2.3–17.8)||1.1 (0.7–1.8)|
| Other (ref = WNH)||4.7a (1.3–17.5)||0.4 (0.1–2.1)||2.5 (0.7–9.1)||1.8 (0.2–17.5)||0.9 (0.3–2.5)|
| 200–<400% FPL||0.5 (0.1–2.3)||0.3a (0.1–1.0)||0.7 (0.3–1.8)||1.7 (0.4–7.3)||1.2 (0.6–2.3)|
| 100–<200% FPL||1.7 (0.5–5.5)||0.2a (0.1–0.8)||1.0 (0.4–2.4)||2.6 (0.7–9.9)||1.8 (0.9–3.4)|
| <100% FPL (ref ≥ 400% FPL)||5.8a (2.2–15.5)||0.1a (0.0–0.5)||2.2a (1.1–4.3)||2.9 (0.9–9.9)||2.1a (1.2–3.7)|
| Medicare ||16.5a (6.6–41.4)||0.4 (0.1–1.2)||4.1a (2.2–7.8)||3.8a (1.6–9.4)||1.4 (0.8–2.4)|
| Medicaid||3.2 (0.9–11.1)||0.2a (0.1–0.6)||1.1 (0.4–3.3)||2.3 (0.6–8.8)||1.1 (0.5–2.6)|
| Combination ||7.6a (2.4–24.4)||0.6 (0.2–2.6)||1.6 (0.6–4.0)||2.3 (0.6–8.3)||2.8a (1.3–6.0)|
| None (ref = private)||7.3a (2.8–19.3)||0.1a (0.0–0.2)||2.5a (1.3–4.8)||2.1 (0.8–5.5)||1.8 (1.0–3.4)|
The magnitudes and significance of the sociodemographic disparities in health care use decreased substantially when we adjusted for site of care (Model 3) (Table 7). Only the odds of being hospitalized with epilepsy remained significantly different for blacks and Hispanics compared to whites, and its magnitude was reduced by about 39%. All of the differences related to income were eliminated after adjusting for site of care, and the differences comparing Medicaid and the uninsured to privately insured were also largely eliminated. Those on Medicaid and the uninsured continued to have a lower likelihood of specialist visits compared to patients with private coverage. Those with a combination of two or more types of coverage continued to have a significantly higher likelihood of receiving new AEDs.
Table 7. Multivariate odds ratios of health care use by sociodemographic characteristics adjusted for clinical characteristics, age, gender, and site
| Black non-Hispanic||1.4 (0.6–3.4)||0.7 (0.3–1.8)||1.7 (0.8–3.6)||3.5a (1.1–10.9)||0.7 (0.4–1.2)|
| Hispanic||1.3 (0.5–3.1)||0.9 (0.4–2.3)||1.3 (0.6–2.7)||3.3a (1.1–10.0)||0.7 (0.4–1.2)|
| Other (ref = WNH)||1.5 (0.3–8.0) ||0.8 (0.2–4.0)||1.4 (0.4–5.2)||1.2 (0.1–10.1)||0.7 (0.2–1.8)|
| 200–<400% FPL||0.4 (0.1–1.7)||0.3 (0.1–1.3)||0.6 (0.2–1.5)||1.4 (0.3–6.0)||1.2 (0.6–2.3)|
| 100–<200% FPL||0.5 (0.1–2.4)||0.5 (0.1–2.2)||0.4 (0.2–1.1)||1.1 (0.3–4.9)||1.6 (0.8–3.2)|
| <100% FPL (ref ≥ 400% FPL)||1.0 (0.2–4.0)||0.6 (0.2–2.4)||0.5 (0.2–1.4)||0.8 (0.2–3.3)||1.7 (0.8–3.6)|
| Medicare||2.6 (0.8–8.6)||0.7 (0.2–2.7)||1.1 (0.4–2.8)||0.7 (0.2–2.9)||0.9 (0.4–2.0)|
| Medicaid||0.8 (0.1–4.7)||0.2a (0.1–0.8)||0.5 (0.1–1.5)||0.7 (0.2–3.4)||0.9 (0.4–2.2)|
| Combination ||3.0 (1.0–9.1)||0.8 (0.2–3.5)||0.8 (0.3–2.1)||0.9 (0.2–3.1)||2.4a (1.1–5.3)|
| None (ref = private)||1.1 (0.3–3.7)||0.1a (0.0–0.4)||0.6 (0.2–1.7)||0.4 (0.1–1.8)||1.2 (0.5–2.7)|
Low-income, minority patients, and those with public insurance or no insurance had higher rates of generalist visits, ER care, and hospitalization, and lower rates of specialty care. In multivariate regression analysis, the effect of income, minority status, and insurance coverage persisted after adjusting for individual patient characteristics (age, gender, seizure frequency, and comorbidities). The magnitude and significance of the effects of income, minority status, and insurance coverage were significantly reduced or eliminated when site of care was added to the model. The associations with site of care persisted despite the different geographic location of the clinics serving patients with lower sociodemographic status.
These findings are consistent with studies indicating that sociodemographic disparities in patterns of care for people with epilepsy may result more from differences in where they seek care than who they are (Wennberg et al., 2004; Barnato et al., 2005;Hasnain-Wynia et al., 2007;Howell et al., 2008). They also suggest that health care patterns could be modified with strategies designed to change the practices of providers at the sites that are serving lower sociodemographic groups with a goal to equalize the service patterns among patient cohorts.
The magnitude and consistency of the effects of sites on disparities in health care raise questions about the university-affiliated clinics serving lower sociodemographic patients:
- • Are patients receiving adequate care from primary care physicians who appear to complement the specialists in the university-affiliated clinics in providing epilepsy care?
- • Is the lower rate of specialty care at the university-affiliated sites appropriate, or does it result from access barriers or resource shortages in these settings?
- • Are practice styles different among university-affiliated and private physicians and do these differences affect quality and efficiency of care?
- • Are the additional ER visits and hospitalizations at the university-affiliated sites preventable?
In trying to explain disparities in epilepsy care, our study has several limitations that should be considered in interpreting the results. Like many health care use studies, we rely on surveys because of the difficulty of tracking all health care utilization across multiple providers of care. Although the electronic medical record (EMR) holds promise for more efficient, accurate, and complete data collection in the future, it may be years before researchers can rely on the EMR for health services research. There are a few studies on the accuracy of survey questionnaires for collecting service-use data and none to our knowledge has focused on patients with epilepsy. The concordance between medical records and survey data has been shown to be relatively high when key service items are self-reported over relatively short recall periods (Mirandola et al., 1999; Byford et al., 2006).
The differences in patterns of care do not necessarily imply differences in the quality of care at these sites. Although there are well-known gaps between clinical recommendations and practice regarding AED treatment, there are currently no objective measures of health care use for evaluating the quality of care provided to adults with epilepsy in the United States. (Pugh et al., 2007). The fact that insured patients visited a specialist more often than noninsured patients does not necessarily mean that the former patients are receiving better care. The use of newer antiepileptics is not necessarily indicative of better care. Higher rates of ER visits and hospitalizations are suggestive of lower quality of care. At the completion of our study, we will relate the differences in patterns of care to patient outcomes to provide a clearer picture of the significance of the disparities.
All subjects are clinic attendees, which raises concerns about selection bias. In a population-based study referenced earlier (Kelvin et al., 2007), a higher proportion of minority patients received care primarily through hospital ERs. In light of these results, by limiting our study to clinic patients, we may be underestimating racial/ethnic disparities in care at the population level.
We were able to compare only four sites, thereby limiting the generalizability of the findings. The sites serving predominantly low-income, minority, Medicaid/uninsured populations are in different regions of the country, which provides a degree of control for regional variation in patterns of care. However, we cannot conclude unambiguously that disparities between the KS and non-KS sites are primarily caused by characteristics of the sites because of the correlation of site with patient sociodemographic characteristics. We reestimated the regression models for subgroups of the sample that were more evenly distributed across the sites. For race/ethnicity, we examined disparities between minority subgroups that were served by all the sites—blacks, Hispanics, and Asians—and eliminated whites. For income, we examined the relationships only for patients below 100% FPL compared to those with incomes between 100% and 200% FPL, thereby eliminating upper-income groups. For coverage, we compared those without any coverage to those with public coverage, eliminating those covered by private insurance. With these configurations, the regression results did not change significantly. We still found that site significantly mediated disparities with respect to income and coverage. With respect to race/ethnicity, there were no disparities found between the three minority groups.
The disparities may result from differences in unmeasured clinical characteristics (e.g., seizure severity) or other personal characteristics of patients at the sites (e.g., patient attitudes and beliefs). It is possible that low sociodemographic patients at the KS site are more like higher sociodemographic patients in their use of care than those at the university clinics, because the KS patients may be more knowledgeable, better able to manage their condition, or are better at communicating with their caregiver.
The results also depend upon the assumption that patients used the same site for all of their care throughout the study period. This assumption is justified given the relatively short period of the study and evidence that patients with serious chronic illness usually rely on a single source of care.
Future research is needed to examine the specific site-related factors that are associated with disparities in care for people with epilepsy. Are the factors related to resource scarcities, patient–provider relations, provider practices, or patient characteristics? This information is needed to help inform program administrators, providers, and policy makers of the type of strategies that would be most effective in reducing and/or eliminating disparities.
This project was sponsored by the U.S. Centers for Disease Control and Prevention (CDC), Association of Schools of Public Health Cooperative Grant #S3031-23/34. The authors are particularly grateful to David Thurman at the CDC for facilitating the project and providing technical assistance. We also thank Beverly McGowan, Raquel Wright, and Carlos Ramos for their assistance in conducting patient interviews. We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.
Disclosure: None of the authors has any conflict of interest to disclose.