Access to Care and Chronic Disease Outcomes Among Medicaid-Insured Persons Versus the Uninsured

Abstract

Objectives. We sought to determine the association between Medicaid coverage and the receipt of appropriate clinical care.

Methods. Using the 1999 to 2012 National Health and Nutritional Examination Surveys, we identified adults aged 18 to 64 years with incomes below the federal poverty level, and compared outpatient visit frequency, awareness, and control of chronic diseases between the uninsured (n = 2975) and those who had Medicaid (n = 1485).

Results. Respondents with Medicaid were more likely than the uninsured to have at least 1 outpatient physician visit annually, after we controlled for patient characteristics (odds ratio [OR] = 5.0; 95% confidence interval [CI] = 3.8, 6.6). Among poor persons with evidence of hypertension, Medicaid coverage was associated with greater awareness (OR = 1.83; 95% CI = 1.26, 2.66) and control (OR = 1.69; 95% CI = 1.32, 2.27) of their condition. Medicaid coverage was also associated with awareness of being overweight (OR = 1.30; 95% CI = 1.02, 1.67), but not with awareness or control of diabetes or hypercholesterolemia.

Conclusions. Among poor adults nationally, Medicaid coverage appears to facilitate outpatient physician care and to improve blood pressure control.

Lack of health insurance is associated with lower rates of preventive care, delays in necessary care, forgone care, medical bankruptcy, and increased mortality.^1–5 The Affordable Care Act (ACA; Pub L No. 111–148) expanded Medicaid insurance for people with low incomes (< 138% of the federal poverty level [FPL]) in 31 states. However, whether Medicaid coverage improves health outcomes remains controversial. Several studies described differences in chronic disease prevalence and control between uninsured persons and those with Medicaid, but have not been designed or powered to explore whether Medicaid coverage might cause these differences.^6–8

Some have suggested that Medicaid’s low reimbursement rates discourage physician acceptance of Medicaid patients, limiting access to care and resulting in poor health outcomes.^9,10 Recently, the Oregon Health Insurance Experiment (OHIE), a randomized, controlled trial, found that Medicaid coverage increased health care use, improved patients’ financial security and self-reported health, lowered depression rates, and raised diabetes diagnosis rates.^11–13 However, the OHIE did not find improvements in other important health outcomes such as control of other chronic diseases, fueling Medicaid’s critics.^14,15

The rigorous design of the OHIE provides strong evidence on the impact of Medicaid in the Portland, Oregon, metropolitan area where it was conducted. However, Portland’s relatively robust medical safety net for the uninsured^16,17 may have attenuated the potential for health improvements from Medicaid expansion compared with other locales, or the United States as a whole.

We used the nationally representative National Health and Nutrition Examination Survey (NHANES) to compare outpatient physician visit frequency among the uninsured and comparable persons with Medicaid coverage. We also assessed whether individuals with major chronic conditions had been previously diagnosed with the condition, and whether it was under control.

METHODS

The NHANES obtains health information from a nationally representative sample of the civilian, noninstitutionalized US population. It is administered by the National Center for Health Statistics, a branch of the Centers for Disease Control and Prevention. Participants are interviewed and undergo physical examinations and laboratory tests. We compiled 14 years of data (1999–2012) from the NHANES, including interview data from 71 916 individuals (79.4% response rate) and physical examination data on 68 705 individuals (75.8% response rate).¹⁸

We included survey respondents aged 18 to 64 years, living at or below the FPL; we identified them by using NHANES’s poverty index ratio (the ratio of family income to poverty level).¹⁹ Of the 68 705 individuals examined in the NHANES between 1999 and 2012, 31 360 were aged 18 to 64 years, of whom 4460 had incomes at or below the FPL. Of this group, 1485 individuals had Medicaid and 2975 were uninsured. We excluded those with private health insurance and those covered by the Indian Health Service, military, or other government insurance (Appendix 1, available as a supplement to the online version of this article at http://www.ajph.org).

Variable Description

Demographic characteristics.

We coded race and ethnicity as non-Hispanic White, non-Hispanic Black, Hispanic, or non-Hispanic other race. We analyzed age in 5 categories: 18 to 24 years, 25 to 34 years, 35 to 44 years, 45 to 54 years, and 55 to 64 years. Because having a disability can be a basis to qualify for Medicaid, we elected to control for self-reported disability in our analysis. We defined participants as disabled if they responded yes to the question “Does a physical, mental, or emotional problem now keep you from working at a job or business?”¹⁸ We performed sensitivity analyses that excluded disability from our models, which generated numerically similar results. We created a chronic disease indicator, which included whether the respondent reported having 1 or more of the following: a history of cardiovascular disease or previous stroke, emphysema or active asthma, any previous cancer (excluding nonmelanoma skin cancer), diabetes, hypertension, or hypercholesterolemia.

Outpatient health care use

We focused on outpatient medical provider visits as a key measure of access to care because it is included in the National Health Disparities Report List of Access to Care Measures issued by the Agency for Healthcare Research and Quality.²⁰ We assessed responses to the question:

During the past 12 months, how many times have you seen a doctor or other health care professional about your health at a doctor’s office, a clinic, hospital emergency room, at home or at some other place. Do not include times you were hospitalized overnight.¹⁸

Response categories were none, 1, 2 to 3, 4 to 9, 10 to 12, 13 or more, or refused. For all analyses, we dichotomized this indicator into zero versus 1 or more visits. We excluded from the analysis individuals who refused to answer the question. We conducted a sensitivity analysis to test for differences by using alternative thresholds including 2 or more visits and 4 or more visits, which yielded similar results.

Previously diagnosed chronic conditions.

We focused on diabetes, hypertension, hypercholesterolemia, and overweight because of their population health salience. We considered respondents to have previously diagnosed diabetes, hypertension, or hypercholesterolemia if they either reported being told by a health professional that they had the condition or were taking a medication indicated for treatment of the condition in question. We elected to include medication usage to indicate “previous diagnosis” to minimize bias attributable to underreporting of diagnoses among low-literacy patients.²¹ Using the NHANES Drug Files, we identified patients taking any drug commonly used to treat hypertension and counted them as having hypertension. We followed a similar protocol for diabetes and hypercholesterolemia. The list of medications is included in Appendix 2 (available as a supplement to the online version of this article at http://www.ajph.org). In this article, we use the terms “unaware” and “undiagnosed” interchangeably, although our definition of “aware” differs from that used in other studies that required that respondents specifically name the condition for which they took medications. We performed a sensitivity analysis excluding disease defined by use of prescription drugs, which yielded similar results to our main analysis. For overweight, we defined previous diagnosis on the basis of whether respondents reported ever being told by a health professional that they were overweight.

Undiagnosed chronic conditions.

We used clinical guidelines, physical examination, and laboratory data to identify the presence of undiagnosed chronic conditions in respondents who had not been previously diagnosed as defined. For blood pressure, NHANES procedures have the patient rest quietly in a sitting position for 5 minutes, then 3 consecutive blood pressure readings are obtained. When a blood pressure measurement was interrupted or incomplete, a fourth attempt may be made. We identified respondents as having high blood pressure if the average of the measured values for systolic blood pressure was 140 millimeters of mercury (mm Hg) or higher or if the average of the measured diastolic blood pressure was 90 mm Hg or higher, consistent with concurrent guidelines.²² We defined respondents’ hypertension as “not previously diagnosed” if the respondent had no history of diagnosed hypertension and was not taking any medication for blood pressure control.

We used a similar process to define respondents with undiagnosed diabetes mellitus, hypercholesterolemia, and overweight. Using the fasting sample from the mobile examination center, we defined a participant as having undiagnosed diabetes if the individual did not report a previous diagnosis of diabetes, was not taking any medications to treat diabetes, and had a fasting blood sugar greater than 125 milligrams per deciliter or a hemoglobin A1c greater than or equal to 6.5%.

Because the guidelines defining hypercholesterolemia changed during the study period, we used Adult Treatment Panel (ATP) II definitions for the 1999–2000 and 2001–2002 surveys and ATP III guidelines for all subsequent surveys (Appendix 3, available as a supplement to the online version of this article at http://www.ajph.org).^23,24 We defined a patient as having undiagnosed hypercholesterolemia if the individual did not report a previous diagnosis of hypercholesterolemia, was not taking any medications for hypercholesterolemia, and had a lipid panel that mandated treatment on the basis of concurrent (either ATP II or ATP III) guidelines. We conducted sensitivity analyses changing the lab criteria for hypercholesterolemia to total cholesterol greater than 240 milligrams per deciliter without any significant changes in results.

Lastly, we defined overweight as undiagnosed if a respondent reported never having been informed of being overweight and had a body mass index (BMI; defined as weight in kilograms divided by the square of height in meters) of 25 or higher.

Control of chronic conditions.

We considered respondents’ chronic conditions controlled on the basis of concurrent clinical guidelines specific to each condition. We defined respondents as having controlled hypertension if they met criteria for previously diagnosed hypertension and had a systolic blood pressure of less than 140 mm Hg and a diastolic blood pressure less than 90 mm Hg. We defined a respondent as having controlled diabetes if they met criteria for previously diagnosed diabetes and had hemoglobin A1C less than or equal to 9%. We performed sensitivity analysis by using hemoglobin A1C cut-points of 7% and 8% with similar results (data not shown). We defined respondents as having controlled hypercholesterolemia if they met criteria for previously diagnosed hypercholesterolemia and met ATP II or III guidelines in their NHANES measured lipid panel. Persons with previously undiagnosed hypertension, diabetes, and hypercholesterolemia were all considered uncontrolled. Because Medicaid provides coverage for large numbers of pregnant women, we performed a sensitivity analysis including and excluding pregnant women from all of the chronic disease outcome analyses, with similar results.

Statistical Analysis

We used multivariable logistic regression models to examine the association between Medicaid coverage (vs being uninsured) on measures of disease diagnosis, disease control, and medical visit frequency. We controlled for gender, age, race and ethnicity, presence of a chronic medical condition, self-identified disability, and later year of survey. We selected these covariates on the basis of previously published literature.²⁵

We explored the relationship between having Medicaid and having 1 or more outpatient visits in the 12 months before the NHANES survey. Although we controlled for chronic conditions and disability in this model, we also conducted a subgroup analysis of those with no chronic conditions or disability to further diminish residual confounding by disease burden, which may differ between Medicaid and uninsured respondents. We then tested for the association between having Medicaid and the likelihood of being previously diagnosed with diabetes, hypertension, hypercholesterolemia, or overweight, as well as the likelihood of control of disease for individuals with diabetes, hypertension, and hypercholesterolemia. We could not evaluate the likelihood of control for obesity because too few respondents with any history of overweight were controlled (i.e., had a BMI < 25).

We performed all statistical analysis with SAS version 9.3 (SAS Institute, Cary, NC). We accounted for the complex survey design by using appropriate survey procedures, and applied weights provided by the National Center for Health Statistics to derive national estimates.

RESULTS

Table 1 displays the demographic characteristics of the weighted population. Compared with the uninsured, respondents with Medicaid were more often female (68.7% vs 45.0%) and Black (28.9% vs 14.0%), and more had a chronic medical condition (55.7% vs 33.2%) or disability (39.0% vs 8.4%). These observed differences in demographics between Medicaid-covered and uninsured individuals are consistent with previous studies.²⁶

TABLE 1—

Demographic Characteristics of US Adults Aged 18–64 Years Living Below the Poverty Line: 1999–2012 US National Health and Nutritional Examination Surveys

	Medicaid Population		Uninsured Population
Characteristic	Medicaid Population in Millions (95% CI)	Percentage of Medicaid Population (95% CI)	Uninsured Population in Millions (95% CI)	Percentage of Uninsured Population (95% CI)
Gender
Men	3.1 (2.7, 3.5)	31.3 (29.0, 33.5)	22.2 (21.0, 23.5)	55.0 (53.7, 56.2)
Women	6.9 (6.3, 7.6)	68.7 (66.5, 71.0)	18.2 (17.0, 19.4)	45.0 (43.8, 46.3)
Age, y
18–24	2.4 (2.1, 2.8)	24.3 (22.1, 26.5)	9.0 (8.3, 9.6)	22.1 (20.7, 23.5)
25–34	2.3 (2.0, 2.7)	23.0 (20.5, 25.5)	11.0 (10.2, 11.9)	27.3 (25.9, 28.7)
35–44	2.1 (1.8, 2.4)	20.9 (18.6, 23.3)	9.3 (8.6, 10.0)	23.0 (21.6, 24.3)
45–54	1.9 (16.4, 22.1)	19.1 (16.9, 21.3)	7.2 (6.6, 7.8)	17.7 (16.6, 18.8)
55–64	1.3 (1.1, 1.5)	12.7 (10.8, 14.5)	4.0 (3.5, 4.5)	9.9 (8.9, 10.9)
Race/ethnicity
Non-Hispanic White	4.7 (3.9, 5.5)	46.4 (40.9, 52.0)	19.9 (17.7, 22.0)	49.3 (45.2, 52.9)
Non-Hispanic Black	2.9 (2.5, 3.3)	28.9 (24.7, 33.0)	5.6 (5.0, 6.3)	14.0 (12.1, 15.8)
Hispanic	1.9 (1.5, 2.3)	18.6 (15.0, 22.3)	12.2 (10.7, 13.9)	30.3 (26.5, 34.1)
Other	0.6 (0.4, 0.8)	6.0 (4.1, 7.9)	2.7 (2.2, 3.2)	6.7 (5.6, 7.9)
Disability status
No disability present	6.1 (5.5, 6.8)	61.0 (57.8, 64.2)	37.1 (35.1, 39.1)	91.6 (90.8, 92.5)
Disability present	3.9 (3.4, 4.4)	39.0 (35.8, 42.2)	33.8 (29.9, 37.8)	8.4 (7.5, 9.2)
Health status
No chronic illness	4.5 (4.0, 4.9)	44.3 (41.6, 47.1)	27.0 (25.5, 28.5)	66.7 (64.9, 68.6)
Chronic illness	5.6 (5.0, 6.2)	55.7 (52.9, 58.4)	13.5 (12.3, 14.6)	33.2 (31.4, 35.1)
Total	10.0 (9.1, 11.0)	100	40.5 (38.3, 42.7)	100

Note. CI = confidence interval.

Outpatient Health Care Use

The unadjusted frequencies of outpatient visits in the sample population are shown in Table 2 and Figure 1, and those for the subgroup with chronic disease or disability are shown in Table 2. Among uninsured individuals, 38.4% (95% confidence interval [CI] = 36.8%, 39.9%) had not been seen even once in the outpatient setting in the past year, compared with only 8.2% (95% CI = 6.7%, 9.6%) of those with Medicaid. Among uninsured persons with a chronic disease or disability, 19.6% (95% CI = 15.5%, 23.7%) had not seen a provider in the past year, compared with 4.5% (95% CI = 2.7%, 6.4%) of those with Medicaid.

TABLE 2—

Outpatient Visits Among Nonelderly Adults Living Below the Federal Poverty Line, for Full Sample Population and Subgroup With a Chronic Disease or Disability (Unadjusted n = 4461): 1999–2012 US National Health and Nutritional Examination Surveys

No. of Outpatient Visits per Year	Uninsured, % (95% CI)	Medicaid, % (95% CI)
Full sample
0	38.4 (36.8, 39.9)	8.2 (6.7, 9.6)
1	24.2 (22.9, 25.4)	12.8 (11.2, 14.5)
2–3	19.9 (18.7, 21.0)	22.9 (20.8, 25.1)
4–9	11.4 (10.3,12.4)	27.4 (25.0, 29.8)
10–12	2.9 (2.4, 3.4)	11.6 (9.9, 13.4)
≥ 13	3.3 (2.8, 3.8)	17 (14.8, 19.1)
Subgroup with a chronic disease or disability
0	19.6 (15.5, 23.7)	4.5 (2.7. 6.4)
1	16.0 (12.6, 19.3)	6.6 (4.1, 9.0)
2–3	20.2 (16.8, 23.6)	15.1 (12.4, 17.8)
4–9	21.8 (18.0, 25.5)	33.2 (29.1, 37.3)
10–12	8.6 (5.8, 11.4)	16.6 (13.3, 19.8)
≥ 13	13.6 (10.5, 16.6)	24 (20.4, 27.6)

Note. CI = confidence interval.

FIGURE 1—

Frequency of Outpatient Visits Among Nonelderly Adults Living Below the Federal Poverty Line (Unadjusted n = 4461): 1999–2012 US National Health and Nutritional Examination Surveys

Next, we determined the odds ratio of having any outpatient care for Medicaid-insured individuals, compared with the referent group of uninsured individuals. In unadjusted analyses, Medicaid recipients had 8.4 (95% CI = 6.4, 10.8) times greater odds of having 1 or more visits compared with uninsured individuals. After we controlled for age, gender, race/ethnicity, presence of chronic medical condition, self-reported disability, and later year of survey, Medicaid-insured individuals still had a 5.0-fold greater odds (95% CI = 3.8, 6.6). In the subgroup analysis of individuals without disability or a chronic illness, Medicaid-insured individuals continued to have a 6.8 increased odds (95% CI = 5.2, 9.0), adjusted for age, gender, and race/ethnicity.

Clinical Outcomes

Figure 2 displays the adjusted odds ratios for various chronic condition measures for Medicaid recipients compared with the uninsured. We found no difference between insurance groups on awareness (i.e., previous diagnosis of their diabetes) or control of diabetes, and small but insignificant differences for hypercholesterolemia. However, among individuals with objective evidence of hypertension, those with Medicaid insurance had 1.83 (95% CI = 1.26, 2.67) greater odds of being previously diagnosed and 1.69 (95% CI = 1.32, 2.66) greater odds of having their hypertension controlled. Similarly, for overweight individuals, those with Medicaid had 1.30 (95% CI = 1.02, 1.67) greater odds of being previously diagnosed.

FIGURE 2—

Chronic Disease Measures for Medicaid vs Uninsured: 1999–2012 US National Health and Nutritional Examination Surveys

Note. CI = confidence interval. A separate model was generated for each predictor. Each model adjusted for gender, age, race/ethnicity, disability, and year of survey.

DISCUSSION

Our findings suggest that, nationally, Medicaid was associated with improved access to outpatient medical care, as well as awareness and control of important chronic conditions. Medicaid recipients visited health care providers much more frequently than comparable uninsured individuals, and were more likely to be aware of their hypertension and overweight. In addition, Medicaid recipients were more likely to have their blood pressure controlled, a clinical goal known to reduce all-cause mortality by as much as 17%.²⁶ However, we found no differences in the diagnosis or control of diabetes, and only nonsignificant differences among those with hypercholesterolemia. We theorize that the lack of findings for diabetes may be because control requires more significant diet and lifestyle changes compared with other chronic conditions, and these changes may not be easily remedied through access to medical care. One possible explanation for the lack of association between Medicaid and hypercholesterolemia diagnosis or control may be the changing definitions of goal lipid values over time. Although we used a broad definition for lipid control, the changing guidelines may cause confounding.

Observational studies about the impact of Medicaid coverage compared with having no insurance have suggested improvements in access to care, use of preventive care, and mortality.^27–32 However, previous studies regarding the effect of Medicaid on chronic disease outcomes have had mixed findings.^33–37 Whereas national studies have demonstrated that Medicaid is associated with increased diabetes diagnosis, awareness and treatment of hypertension, and control of chronic conditions (such as diabetes, hypertension, coronary heart disease, congestive heart failure), Massachusetts did not observe changes after reform in cholesterol, blood pressure, or glycated hemoglobin among the previously uninsured. Decker et al. found that hypertension and hypercholesterolemia control rates were better for Medicaid recipients (virtually all of whom had incomes below poverty) than for uninsured persons who might become eligible for Medicaid under the ACA (including persons with incomes up to 138% of poverty, but excluding noncitizens).⁶ The contrast between their findings and ours may arise from differences in the uninsured populations studied, and the fact that our larger sample size allowed us to control for important demographic differences such as gender and race/ethnicity.

Our findings differ from those in the OHIE. First, in the Oregon study, outpatient visits among uninsured individuals averaged 5.5 per year, whereas we found that more than 60% of uninsured individuals nationally had zero or 1 outpatient visit per year. Second, whereas the OHIE observed a 50% increase in outpatient visits among the Medicaid population, we found substantially larger national effects (a 5.0-fold increased odds). Despite methodological differences that preclude exact comparisons with the OHIE data (the NHANES coded outpatient visits into categories), our results suggest that outpatient health care use by the uninsured nationally is probably lower than that of the OHIE’s control population and the boost from Medicaid somewhat bigger. Third, the OHIE found significantly increased detection of diabetes but not hypertension or obesity, and did not show improvement in any physical health measures.¹¹

Our observational study has important limitations. First, residual confounding, especially related to illness or disabilities that sometimes qualify individuals for Medicaid coverage, might explain the association between Medicaid coverage and increased physician visits. However, our use of physical examination and laboratory data from the NHANES survey allowed us to control for the presence of several important chronic illnesses. Moreover, our subgroup analysis of those without chronic disease or disability suggests that residual confounding is an unlikely explanation for our results. Second, there may be reverse causation because individuals who are chronically ill are more likely to seek provider visits or gain Medicaid coverage. Our subgroup analysis of individuals with a chronic disease or disability demonstrates similar health care use disparities between uninsured versus Medicaid-covered individuals as for the full sample, thus suggesting that reverse causation does not explain our findings. Third, using self-reported insurance status introduces recall bias, although previous research suggests that surveys provide reasonable estimates of Medicaid coverage.³⁸ Lastly, there may also be recall bias for diagnosis of disease, as uninsured people may be less likely to recall a previous diagnosis because of longer intervals between visits to health care providers.

For states that expanded Medicaid under the ACA, enrollment began only 1 year ago; several states are considering expanding Medicaid in the future. Thus, it will likely take several years before nationally representative data become available on the ACA’s impacts. Our data on the association of Medicaid coverage with both outpatient visit frequency and chronic disease care in the pre-ACA era help inform projections of the impact of the largest expansion of Medicaid since its founding.